rockbox/firmware/common/file.c

1231 lines
34 KiB
C
Raw Normal View History

/***************************************************************************
* __________ __ ___.
* Open \______ \ ____ ____ | | _\_ |__ _______ ___
* Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ /
* Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <
* Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \
* \/ \/ \/ \/ \/
* $Id$
*
* Copyright (C) 2002 by Björn Stenberg
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
* Copyright (C) 2014 by Michael Sevakis
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version 2
* of the License, or (at your option) any later version.
*
* This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
* KIND, either express or implied.
*
****************************************************************************/
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
#define RB_FILESYSTEM_OS
#include "config.h"
#include "system.h"
#include <string.h>
#include <errno.h>
#include "debug.h"
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
#include "file.h"
#include "fileobj_mgr.h"
#include "disk_cache.h"
#include "dircache_redirect.h"
#include "string-extra.h"
/**
* These functions provide a roughly POSIX-compatible file I/O API.
*/
/* structure used for open file descriptors */
static struct filestr_desc
{
struct filestr_base stream; /* basic stream info (first!) */
file_size_t offset; /* current offset for stream */
file_size_t *sizep; /* shortcut to file size in fileobj */
} open_streams[MAX_OPEN_FILES];
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* check and return a struct filestr_desc* from a file descriptor number */
static struct filestr_desc * get_filestr(int fildes)
{
struct filestr_desc *file = &open_streams[fildes];
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if ((unsigned int)fildes >= MAX_OPEN_FILES)
file = NULL;
else if (file->stream.flags & FDO_BUSY)
return file;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
DEBUGF("fildes %d: bad file number\n", fildes);
errno = (file && (file->stream.flags & FD_NONEXIST)) ? ENXIO : EBADF;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
return NULL;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
#define GET_FILESTR(type, fildes) \
({ \
file_internal_lock_##type(); \
struct filestr_desc * _file = get_filestr(fildes); \
if (_file) \
FILESTR_LOCK(type, &_file->stream); \
else \
file_internal_unlock_##type(); \
_file; \
})
/* release the lock on the filestr_desc* */
#define RELEASE_FILESTR(type, file) \
({ \
FILESTR_UNLOCK(type, &(file)->stream); \
file_internal_unlock_##type(); \
})
/* find a free file descriptor */
static int alloc_filestr(struct filestr_desc **filep)
{
for (int fildes = 0; fildes < MAX_OPEN_FILES; fildes++)
{
struct filestr_desc *file = &open_streams[fildes];
if (!file->stream.flags)
{
*filep = file;
return fildes;
}
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
DEBUGF("Too many files open\n");
return -1;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* return the file size in sectors */
static inline unsigned long filesize_sectors(file_size_t size)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* overflow proof whereas "(x + y - 1) / y" is not */
unsigned long numsectors = size / SECTOR_SIZE;
if (size % SECTOR_SIZE)
numsectors++;
return numsectors;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* flush a dirty cache buffer */
static int flush_cache(struct filestr_desc *file)
{
int rc;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
struct filestr_cache *cachep = file->stream.cachep;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
DEBUGF("Flushing dirty sector cache (%lu)\n", cachep->sector);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (fat_query_sectornum(&file->stream.fatstr) != cachep->sector)
{
/* get on the correct sector */
rc = fat_seek(&file->stream.fatstr, cachep->sector);
if (rc < 0)
FILE_ERROR(EIO, rc * 10 - 1);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
rc = fat_readwrite(&file->stream.fatstr, 1, cachep->buffer, true);
if (rc < 0)
{
if (rc == FAT_RC_ENOSPC)
FILE_ERROR(ENOSPC, RC);
else
FILE_ERROR(EIO, rc * 10 - 2);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
cachep->flags = 0;
return 1;
file_error:
DEBUGF("Failed flushing cache: %d\n", rc);
return rc;
}
static void discard_cache(struct filestr_desc *file)
{
struct filestr_cache *const cachep = file->stream.cachep;
cachep->flags = 0;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* set the file pointer */
static off_t lseek_internal(struct filestr_desc *file, off_t offset,
int whence)
{
off_t rc;
file_size_t pos;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
file_size_t size = MIN(*file->sizep, FILE_SIZE_MAX);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
switch (whence)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
case SEEK_SET:
if (offset < 0 || (file_size_t)offset > size)
FILE_ERROR(EINVAL, -1);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
pos = offset;
break;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
case SEEK_CUR:
if ((offset < 0 && (file_size_t)-offset > file->offset) ||
(offset > 0 && (file_size_t)offset > size - file->offset))
FILE_ERROR(EINVAL, -1);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
pos = file->offset + offset;
break;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
case SEEK_END:
if (offset > 0 || (file_size_t)-offset > size)
FILE_ERROR(EINVAL, -1);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
pos = size + offset;
break;
default:
FILE_ERROR(EINVAL, -1);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
file->offset = pos;
return pos;
file_error:
return rc;
}
/* Handle syncing all file's streams to the truncation */
static void handle_truncate(struct filestr_desc * const file, file_size_t size)
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
{
unsigned long filesectors = filesize_sectors(size);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
struct filestr_base *s = NULL;
while ((s = fileobj_get_next_stream(&file->stream, s)))
{
/* caches with data beyond new extents are invalid */
unsigned long sector = s->cachep->sector;
if (sector != INVALID_SECNUM && sector >= filesectors)
filestr_discard_cache(s);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* files outside bounds must be rewound */
if (fat_query_sectornum(&s->fatstr) > filesectors)
fat_seek_to_stream(&s->fatstr, &file->stream.fatstr);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* clip file offset too if needed */
struct filestr_desc *f = (struct filestr_desc *)s;
if (f->offset > size)
f->offset = size;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
}
/* truncate the file to the specified length */
static int ftruncate_internal(struct filestr_desc *file, file_size_t size,
bool write_now)
{
int rc = 0, rc2 = 1;
file_size_t cursize = *file->sizep;
file_size_t truncsize = MIN(size, cursize);
if (write_now)
{
unsigned long sector = filesize_sectors(truncsize);
struct filestr_cache *const cachep = file->stream.cachep;
if (cachep->flags == (FSC_NEW|FSC_DIRTY) &&
cachep->sector + 1 == sector)
{
/* sector created but may have never been added to the cluster
chain; flush it now or the subsequent may fail */
rc2 = flush_cache(file);
if (rc2 == FAT_RC_ENOSPC)
{
/* no space left on device; further truncation needed */
discard_cache(file);
truncsize = ALIGN_DOWN(truncsize - 1, SECTOR_SIZE);
sector--;
rc = rc2;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
else if (rc2 < 0)
FILE_ERROR(ERRNO, rc2 * 10 - 1);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
rc2 = fat_seek(&file->stream.fatstr, sector);
if (rc2 < 0)
FILE_ERROR(EIO, rc2 * 10 - 2);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
rc2 = fat_truncate(&file->stream.fatstr);
if (rc2 < 0)
FILE_ERROR(EIO, rc2 * 10 - 3);
/* never needs to be done this way again since any data beyond the
cached size is now gone */
fileobj_change_flags(&file->stream, 0, FO_TRUNC);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* else just change the cached file size */
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (truncsize < cursize)
{
*file->sizep = truncsize;
handle_truncate(file, truncsize);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* if truncation was partially successful, it effectively destroyed
everything after the truncation point; still, indicate failure
after adjusting size */
if (rc2 == 0)
FILE_ERROR(EIO, -4);
else if (rc2 < 0)
FILE_ERROR(ERRNO, rc2);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
file_error:
return rc;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* flush back all outstanding writes to the file */
static int fsync_internal(struct filestr_desc *file)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* call only when holding WRITER lock (updates directory entries) */
int rc = 0;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
file_size_t size = *file->sizep;
unsigned int foflags = fileobj_get_flags(&file->stream);
/* flush sector cache? */
struct filestr_cache *const cachep = file->stream.cachep;
if (cachep->flags & FSC_DIRTY)
{
int rc2 = flush_cache(file);
if (rc2 == FAT_RC_ENOSPC && (cachep->flags & FSC_NEW))
{
/* no space left on device so this must be dropped */
discard_cache(file);
size = ALIGN_DOWN(size - 1, SECTOR_SIZE);
foflags |= FO_TRUNC;
rc = rc2;
}
else if (rc2 < 0)
FILE_ERROR(ERRNO, rc2 * 10 - 1);
}
/* truncate? */
if (foflags & FO_TRUNC)
{
int rc2 = ftruncate_internal(file, size, rc == 0);
if (rc2 < 0)
FILE_ERROR(ERRNO, rc2 * 10 - 2);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
file_error:;
/* tie up all loose ends (try to close the file even if failing) */
int rc2 = fat_closewrite(&file->stream.fatstr, size,
get_dir_fatent_dircache());
if (rc2 >= 0)
fileop_onsync_internal(&file->stream); /* dir_fatent is implicit arg */
if (rc2 < 0 && rc >= 0)
{
errno = EIO;
rc = rc2 * 10 - 3;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
return rc;
}
/* finish with the file and free resources */
static int close_internal(struct filestr_desc *file)
{
/* call only when holding WRITER lock (updates directory entries) */
int rc;
if ((file->stream.flags & (FD_WRITE|FD_NONEXIST)) == FD_WRITE)
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
{
rc = fsync_internal(file);
if (rc < 0)
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
FILE_ERROR(ERRNO, rc * 10 - 1);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
rc = 0;
file_error:;
int rc2 = close_stream_internal(&file->stream);
if (rc2 < 0 && rc >= 0)
rc = rc2 * 10 - 2;
return rc;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* actually do the open gruntwork */
static int open_internal_inner2(const char *path,
struct filestr_desc *file,
unsigned int callflags)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
int rc;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
struct path_component_info compinfo;
rc = open_stream_internal(path, callflags, &file->stream, &compinfo);
if (rc < 0)
{
DEBUGF("Open failed: %d\n", rc);
FILE_ERROR_RETURN(ERRNO, rc * 10 - 1);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
bool created = false;
if (rc > 0)
{
if (callflags & FF_EXCL)
{
DEBUGF("File exists\n");
FILE_ERROR(EEXIST, -2);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (compinfo.attr & ATTR_DIRECTORY)
{
if ((callflags & FD_WRITE) || !(callflags & FF_ANYTYPE))
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
DEBUGF("File is a directory\n");
FILE_ERROR(EISDIR, -3);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
compinfo.filesize = MAX_DIRECTORY_SIZE; /* allow file ops */
}
}
else if (callflags & FF_CREAT)
{
if (compinfo.attr & ATTR_DIRECTORY)
{
DEBUGF("File is a directory\n");
FILE_ERROR(EISDIR, -5);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* not found; try to create it */
callflags &= ~FO_TRUNC;
rc = create_stream_internal(&compinfo.parentinfo, compinfo.name,
compinfo.length, ATTR_NEW_FILE, callflags,
&file->stream);
if (rc < 0)
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
FILE_ERROR(ERRNO, rc * 10 - 6);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
created = true;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
else
{
DEBUGF("File not found\n");
FILE_ERROR(ENOENT, -7);
}
fat_rewind(&file->stream.fatstr);
file->sizep = fileobj_get_sizep(&file->stream);
file->offset = 0;
if (!created)
{
/* size from storage applies to first stream only otherwise it's
already up to date */
const bool first = fileobj_get_flags(&file->stream) & FO_SINGLE;
if (first)
*file->sizep = compinfo.filesize;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (callflags & FO_TRUNC)
{
/* if the file is kind of "big" then free some space now */
rc = ftruncate_internal(file, 0, *file->sizep >= O_TRUNC_THRESH);
if (rc < 0)
{
DEBUGF("O_TRUNC failed: %d\n", rc);
FILE_ERROR(ERRNO, rc * 10 - 4);
}
}
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
rc = 0;
file_error:
if (rc < 0)
close_stream_internal(&file->stream);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
return rc;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* allocate a file descriptor, if needed, assemble stream flags and open
a new stream */
static int open_internal_inner1(const char *path, int oflag,
unsigned int callflags)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
DEBUGF("%s(path=\"%s\",oflag=%X,callflags=%X)\n", __func__,
path, oflag, callflags);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
int rc;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
struct filestr_desc *file;
int fildes = alloc_filestr(&file);
if (fildes < 0)
FILE_ERROR(EMFILE, -1);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
callflags &= ~FDO_MASK;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (oflag & O_ACCMODE)
{
callflags |= FD_WRITE;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if ((oflag & O_ACCMODE) == O_WRONLY)
callflags |= FD_WRONLY;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (oflag & O_APPEND)
callflags |= FD_APPEND;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (oflag & O_TRUNC)
callflags |= FO_TRUNC;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
else if (oflag & O_TRUNC)
{
/* O_TRUNC requires write mode */
DEBUGF("No write mode but have O_TRUNC\n");
FILE_ERROR(EINVAL, -2);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* O_CREAT and O_APPEND are fine without write mode
* for the former, an empty file is created but no data may be written
* for the latter, no append will be allowed anyway */
if (oflag & O_CREAT)
{
callflags |= FF_CREAT;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (oflag & O_EXCL)
callflags |= FF_EXCL;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
rc = open_internal_inner2(path, file, callflags);
if (rc < 0)
FILE_ERROR(ERRNO, rc * 10 - 3);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
return fildes;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
file_error:
return rc;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
static int open_internal_locked(const char *path, int oflag,
unsigned int callflags)
{
file_internal_lock_WRITER();
int rc = open_internal_inner1(path, oflag, callflags);
file_internal_unlock_WRITER();
return rc;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* fill a cache buffer with a new sector */
static int readwrite_fill_cache(struct filestr_desc *file, unsigned long sector,
unsigned long filesectors, bool write)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* sector != cachep->sector should have been checked by now */
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
int rc;
struct filestr_cache *cachep = filestr_get_cache(&file->stream);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (cachep->flags & FSC_DIRTY)
{
rc = flush_cache(file);
if (rc < 0)
FILE_ERROR(ERRNO, rc * 10 - 1);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (fat_query_sectornum(&file->stream.fatstr) != sector)
{
/* get on the correct sector */
rc = fat_seek(&file->stream.fatstr, sector);
if (rc < 0)
FILE_ERROR(EIO, rc * 10 - 2);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (!write || sector < filesectors)
{
/* only reading or this sector would have been flushed if the cache
was previously needed for a different sector */
rc = fat_readwrite(&file->stream.fatstr, 1, cachep->buffer, false);
if (rc < 0)
FILE_ERROR(rc == FAT_RC_ENOSPC ? ENOSPC : EIO, rc * 10 - 3);
}
else
{
/* create a fresh, shiny, new sector with that new sector smell */
cachep->flags = FSC_NEW;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
cachep->sector = sector;
return 1;
file_error:
DEBUGF("Failed caching sector: %d\n", rc);
return rc;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* read or write to part or all of the cache buffer */
static inline void readwrite_cache(struct filestr_cache *cachep, void *buf,
unsigned long secoffset, size_t nbyte,
bool write)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
void *dst, *cbufp = cachep->buffer + secoffset;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (write)
{
dst = cbufp;
cachep->flags |= FSC_DIRTY;
}
else
{
dst = buf;
buf = cbufp;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
memcpy(dst, buf, nbyte);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* read or write a partial sector using the file's cache */
static inline ssize_t readwrite_partial(struct filestr_desc *file,
struct filestr_cache *cachep,
unsigned long sector,
unsigned long secoffset,
void *buf,
size_t nbyte,
unsigned long filesectors,
unsigned int flags)
{
if (sector != cachep->sector)
{
/* wrong sector in buffer */
int rc = readwrite_fill_cache(file, sector, filesectors, flags);
if (rc <= 0)
return rc;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
readwrite_cache(cachep, buf, secoffset, nbyte, flags);
return nbyte;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* read from or write to the file; back end to read() and write() */
static ssize_t readwrite(struct filestr_desc *file, void *buf, size_t nbyte,
bool write)
{
DEBUGF("readwrite(%p,%lx,%lu,%s)\n",
file, (long)buf, (unsigned long)nbyte, write ? "write" : "read");
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
const file_size_t size = *file->sizep;
file_size_t filerem;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (write)
{
/* if opened in append mode, move pointer to end */
if (file->stream.flags & FD_APPEND)
file->offset = MIN(size, FILE_SIZE_MAX);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
filerem = FILE_SIZE_MAX - file->offset;
}
else
{
/* limit to maximum possible offset (EOF or FILE_SIZE_MAX) */
filerem = MIN(size, FILE_SIZE_MAX) - file->offset;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (nbyte > filerem)
{
nbyte = filerem;
if (nbyte > 0)
{}
else if (write)
FILE_ERROR_RETURN(EFBIG, -1); /* would get too large */
else if (file->offset >= FILE_SIZE_MAX)
FILE_ERROR_RETURN(EOVERFLOW, -2); /* can't read here */
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (nbyte == 0)
return 0;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
int rc = 0;
struct filestr_cache * const cachep = file->stream.cachep;
void * const bufstart = buf;
const unsigned long filesectors = filesize_sectors(size);
unsigned long sector = file->offset / SECTOR_SIZE;
unsigned long sectoroffs = file->offset % SECTOR_SIZE;
/* any head bytes? */
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (sectoroffs)
{
size_t headbytes = MIN(nbyte, SECTOR_SIZE - sectoroffs);
rc = readwrite_partial(file, cachep, sector, sectoroffs, buf, headbytes,
filesectors, write);
if (rc <= 0)
{
if (rc < 0)
FILE_ERROR(ERRNO, rc * 10 - 3);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
nbyte = 0; /* eof, skip the rest */
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
else
{
buf += rc;
nbyte -= rc;
sector++; /* if nbyte goes to 0, the rest is skipped anyway */
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* read/write whole sectors right into/from the supplied buffer */
unsigned long sectorcount = nbyte / SECTOR_SIZE;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
while (sectorcount)
{
unsigned long runlen = sectorcount;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* if a cached sector is inside the transfer range, split the transfer
into two parts and use the cache for that sector to keep it coherent
without writeback */
if (UNLIKELY(cachep->sector >= sector &&
cachep->sector < sector + sectorcount))
{
runlen = cachep->sector - sector;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (runlen)
{
if (fat_query_sectornum(&file->stream.fatstr) != sector)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* get on the correct sector */
rc = 0;
/* If the dirty bit isn't set, we're somehow beyond the file
size and you can't explain _that_ */
if (sector >= filesectors && cachep->flags == (FSC_NEW|FSC_DIRTY))
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
rc = flush_cache(file);
if (rc < 0)
FILE_ERROR(ERRNO, rc * 10 - 4);
if (cachep->sector + 1 == sector)
rc = 1; /* if now ok, don't seek */
}
if (rc == 0)
{
rc = fat_seek(&file->stream.fatstr, sector);
if (rc < 0)
FILE_ERROR(EIO, rc * 10 - 5);
}
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
rc = fat_readwrite(&file->stream.fatstr, runlen, buf, write);
if (rc < 0)
{
DEBUGF("I/O error %sing %ld sectors\n",
write ? "writ" : "read", runlen);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
FILE_ERROR(rc == FAT_RC_ENOSPC ? ENOSPC : EIO,
rc * 10 - 6);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
else
{
buf += rc * SECTOR_SIZE;
nbyte -= rc * SECTOR_SIZE;
sector += rc;
sectorcount -= rc;
/* if eof, skip tail bytes */
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if ((unsigned long)rc < runlen)
nbyte = 0;
if (!nbyte)
break;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (UNLIKELY(sectorcount && sector == cachep->sector))
{
/* do this one sector with the cache */
readwrite_cache(cachep, buf, 0, SECTOR_SIZE, write);
buf += SECTOR_SIZE;
nbyte -= SECTOR_SIZE;
sector++;
sectorcount--;
}
}
/* any tail bytes? */
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (nbyte)
{
/* tail bytes always start at sector offset 0 */
rc = readwrite_partial(file, cachep, sector, 0, buf, nbyte,
filesectors, write);
if (rc < 0)
FILE_ERROR(ERRNO, rc * 10 - 7);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
buf += rc;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
file_error:;
#ifdef DEBUG
if (errno == ENOSPC)
DEBUGF("No space left on device\n");
#endif
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
size_t done = buf - bufstart;
if (done)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* error or not, update the file offset and size if anything was
transferred */
file->offset += done;
DEBUGF("file offset: %ld\n", file->offset);
/* adjust file size to length written */
if (write && file->offset > size)
*file->sizep = file->offset;
if (rc > 0)
return done;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
return rc;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/** Internal interface **/
/* open a file without codepage conversion during the directory search;
required to avoid any reentrancy when opening codepages and when scanning
directories internally, which could infinitely recurse and would corrupt
the static data */
int open_noiso_internal(const char *path, int oflag)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
return open_internal_locked(path, oflag, FF_ANYTYPE | FF_NOISO);
}
void force_close_writer_internal(struct filestr_base *stream)
{
/* only we do writers so we know this is our guy */
close_internal((struct filestr_desc *)stream);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/** POSIX **/
/* open a file */
int open(const char *path, int oflag)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
DEBUGF("open(path=\"%s\",oflag=%X)\n", path, (unsigned)oflag);
return open_internal_locked(path, oflag, FF_ANYTYPE);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* create a new file or rewrite an existing one */
int creat(const char *path)
{
DEBUGF("creat(path=\"%s\")\n", path);
return open_internal_locked(path, O_WRONLY|O_CREAT|O_TRUNC, FF_ANYTYPE);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* close a file descriptor */
int close(int fildes)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
DEBUGF("close(fd=%d)\n", fildes);
int rc;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
file_internal_lock_WRITER();
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* needs to work even if marked "nonexistant" */
struct filestr_desc *file = &open_streams[fildes];
if ((unsigned int)fildes >= MAX_OPEN_FILES || !file->stream.flags)
{
DEBUGF("filedes %d not open\n", fildes);
FILE_ERROR(EBADF, -2);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
rc = close_internal(file);
if (rc < 0)
FILE_ERROR(ERRNO, rc * 10 - 3);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
file_error:
file_internal_unlock_WRITER();
return rc;
}
/* truncate a file to a specified length */
int ftruncate(int fildes, off_t length)
{
DEBUGF("ftruncate(fd=%d,len=%ld)\n", fildes, (long)length);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
struct filestr_desc * const file = GET_FILESTR(READER, fildes);
if (!file)
FILE_ERROR_RETURN(ERRNO, -1);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
int rc;
if (!(file->stream.flags & FD_WRITE))
{
DEBUGF("Descriptor is read-only mode\n");
FILE_ERROR(EBADF, -2);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (length < 0)
{
DEBUGF("Length %ld is invalid\n", (long)length);
FILE_ERROR(EINVAL, -3);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
rc = ftruncate_internal(file, length, true);
if (rc < 0)
FILE_ERROR(ERRNO, rc * 10 - 4);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
file_error:
RELEASE_FILESTR(READER, file);
return rc;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* synchronize changes to a file */
int fsync(int fildes)
{
DEBUGF("fsync(fd=%d)\n", fildes);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
struct filestr_desc * const file = GET_FILESTR(WRITER, fildes);
if (!file)
FILE_ERROR_RETURN(ERRNO, -1);
int rc;
if (!(file->stream.flags & FD_WRITE))
{
DEBUGF("Descriptor is read-only mode\n");
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
FILE_ERROR(EINVAL, -2);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
rc = fsync_internal(file);
if (rc < 0)
FILE_ERROR(ERRNO, rc * 10 - 3);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
file_error:
RELEASE_FILESTR(WRITER, file);
return rc;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* move the read/write file offset */
off_t lseek(int fildes, off_t offset, int whence)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
DEBUGF("lseek(fd=%d,ofs=%ld,wh=%d)\n", fildes, (long)offset, whence);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
struct filestr_desc * const file = GET_FILESTR(READER, fildes);
if (!file)
FILE_ERROR_RETURN(ERRNO, -1);
off_t rc = lseek_internal(file, offset, whence);
if (rc < 0)
FILE_ERROR(ERRNO, rc * 10 - 2);
file_error:
RELEASE_FILESTR(READER, file);
return rc;
}
/* read from a file */
ssize_t read(int fildes, void *buf, size_t nbyte)
{
struct filestr_desc * const file = GET_FILESTR(READER, fildes);
if (!file)
FILE_ERROR_RETURN(ERRNO, -1);
ssize_t rc;
if (file->stream.flags & FD_WRONLY)
{
DEBUGF("read(fd=%d,buf=%p,nb=%lu) - "
"descriptor is write-only mode\n",
fildes, buf, (unsigned long)nbyte);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
FILE_ERROR(EBADF, -2);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
rc = readwrite(file, buf, nbyte, false);
if (rc < 0)
FILE_ERROR(ERRNO, rc * 10 - 3);
file_error:
RELEASE_FILESTR(READER, file);
return rc;
}
/* write on a file */
ssize_t write(int fildes, const void *buf, size_t nbyte)
{
struct filestr_desc * const file = GET_FILESTR(READER, fildes);
if (!file)
FILE_ERROR_RETURN(ERRNO, -1);
ssize_t rc;
if (!(file->stream.flags & FD_WRITE))
{
DEBUGF("write(fd=%d,buf=%p,nb=%lu) - "
"descriptor is read-only mode\n",
fildes, buf, (unsigned long)nbyte);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
FILE_ERROR(EBADF, -2);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
rc = readwrite(file, (void *)buf, nbyte, true);
if (rc < 0)
FILE_ERROR(ERRNO, rc * 10 - 3);
file_error:
RELEASE_FILESTR(READER, file);
return rc;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* remove a file */
int remove(const char *path)
{
DEBUGF("remove(path=\"%s\")\n", path);
file_internal_lock_WRITER();
int rc = remove_stream_internal(path, NULL, FF_FILE);
file_internal_unlock_WRITER();
return rc;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* rename a file */
int rename(const char *old, const char *new)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
DEBUGF("rename(old=\"%s\",new=\"%s\")\n", old, new);
int rc, open1rc = -1, open2rc = -1;
struct filestr_base oldstr, newstr;
struct path_component_info oldinfo, newinfo;
file_internal_lock_WRITER();
/* open 'old'; it must exist */
open1rc = open_stream_internal(old, FF_ANYTYPE, &oldstr, &oldinfo);
if (open1rc <= 0)
{
DEBUGF("Failed opening old: %d\n", open1rc);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (open1rc == 0)
FILE_ERROR(ENOENT, -1);
else
FILE_ERROR(ERRNO, open1rc * 10 - 1);
}
/* if 'old' is a directory then 'new' is also required to be one if 'new'
is to be overwritten */
const bool are_dirs = oldinfo.attr & ATTR_DIRECTORY;
/* open new (may or may not exist) */
unsigned int callflags = FF_FILE;
if (are_dirs)
{
/* if 'old' is found while parsing the new directory components then
'new' contains path prefix that names 'old'; if new and old are in
the same directory, this tests positive but that is checked later */
callflags = FF_DIR | FF_CHECKPREFIX;
newinfo.prefixp = oldstr.infop;
}
open2rc = open_stream_internal(new, callflags, &newstr, &newinfo);
if (open2rc < 0)
{
DEBUGF("Failed opening new file: %d\n", open2rc);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
FILE_ERROR(ERRNO, open2rc * 10 - 2);
}
#ifdef HAVE_MULTIVOLUME
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (oldinfo.parentinfo.volume != newinfo.parentinfo.volume)
{
DEBUGF("Cross-device link\n");
FILE_ERROR(EXDEV, -3);
}
#endif /* HAVE_MULTIVOLUME */
/* if the parent is changing then this is a move, not a simple rename */
const bool is_move = !fat_file_is_same(&oldinfo.parentinfo.fatfile,
&newinfo.parentinfo.fatfile);
/* prefix found and moving? */
if (is_move && (newinfo.attr & ATTR_PREFIX))
{
DEBUGF("New contains prefix that names old\n");
FILE_ERROR(EINVAL, -4);
}
const char * const oldname = strmemdupa(oldinfo.name, oldinfo.length);
const char * const newname = strmemdupa(newinfo.name, newinfo.length);
bool is_overwrite = false;
if (open2rc > 0)
{
/* new name exists in parent; check if 'old' is overwriting 'new';
if it's the very same file, then it's just a rename */
is_overwrite = oldstr.bindp != newstr.bindp;
if (is_overwrite)
{
if (are_dirs)
{
/* the directory to be overwritten must be empty */
rc = test_dir_empty_internal(&newstr);
if (rc < 0)
FILE_ERROR(ERRNO, rc * 10 - 5);
}
}
else if (!strcmp(newname, oldname)) /* case-only is ok */
{
DEBUGF("No name change (success)\n");
rc = 0;
FILE_ERROR(ERRNO, RC);
}
}
else if (!are_dirs && (newinfo.attr & ATTR_DIRECTORY))
{
/* even if new doesn't exist, canonical path type must match
(ie. a directory path such as "/foo/bar/" when old names a file) */
DEBUGF("New path is a directory\n");
FILE_ERROR(EISDIR, -6);
}
/* first, create the new entry so that there's never a time that the
victim's data has no reference in the directory tree, that is, until
everything else first succeeds */
struct file_base_info old_fileinfo = *oldstr.infop;
rc = fat_rename(&newinfo.parentinfo.fatfile, &oldstr.infop->fatfile,
newname);
if (rc < 0)
{
DEBUGF("I/O error renaming file: %d\n", rc);
FILE_ERROR(rc == FAT_RC_ENOSPC ? ENOSPC : EIO, rc * 10 - 7);
}
if (is_overwrite)
{
/* 'new' would have been assigned its own directory entry and
succeeded so at this point it is treated like a remove() call
on the victim which preserves data until the last reference is
closed */
rc = remove_stream_internal(NULL, &newstr, callflags);
if (rc < 0)
FILE_ERROR(ERRNO, rc * 10 - 8);
}
fileop_onrename_internal(&oldstr, is_move ? &old_fileinfo : NULL,
&newinfo.parentinfo, newname);
file_error:
/* for now, there is nothing to fail upon closing the old stream */
if (open1rc >= 0)
close_stream_internal(&oldstr);
/* the 'new' stream could fail to close cleanly because it became
impossible to remove its data if this was an overwrite operation */
if (open2rc >= 0)
{
int rc2 = close_stream_internal(&newstr);
if (rc2 < 0 && rc >= 0)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
DEBUGF("Success but failed closing new: %d\n", rc2);
rc = rc2 * 10 - 9;
}
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
file_internal_unlock_WRITER();
return rc;
}
/** Extensions **/
/* get the binary size of a file (in bytes) */
off_t filesize(int fildes)
{
struct filestr_desc * const file = GET_FILESTR(READER, fildes);
if (!file)
FILE_ERROR_RETURN(ERRNO, -1);
off_t rc;
file_size_t size = *file->sizep;
if (size > FILE_SIZE_MAX)
FILE_ERROR(EOVERFLOW, -2);
rc = (off_t)size;
file_error:
RELEASE_FILESTR(READER, file);
return rc;
}
/* test if two file descriptors refer to the same file */
int fsamefile(int fildes1, int fildes2)
{
struct filestr_desc * const file1 = GET_FILESTR(WRITER, fildes1);
if (!file1)
FILE_ERROR_RETURN(ERRNO, -1);
int rc = -2;
struct filestr_desc * const file2 = get_filestr(fildes2);
if (file2)
rc = file1->stream.bindp == file2->stream.bindp ? 1 : 0;
RELEASE_FILESTR(WRITER, file1);
return rc;
}
/* tell the relationship of path1 to path2 */
int relate(const char *path1, const char *path2)
{
/* this is basically what rename() does but reduced to the relationship
determination */
DEBUGF("relate(path1=\"%s\",path2=\"%s\")\n", path1, path2);
int rc, open1rc = -1, open2rc = -1;
struct filestr_base str1, str2;
struct path_component_info info1, info2;
file_internal_lock_WRITER();
open1rc = open_stream_internal(path1, FF_ANYTYPE, &str1, &info1);
if (open1rc <= 0)
{
DEBUGF("Failed opening path1: %d\n", open1rc);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (open1rc < 0)
FILE_ERROR(ERRNO, open1rc * 10 - 1);
else
FILE_ERROR(ENOENT, -1);
}
info2.prefixp = str1.infop;
open2rc = open_stream_internal(path2, FF_ANYTYPE | FF_CHECKPREFIX,
&str2, &info2);
if (open2rc < 0)
{
DEBUGF("Failed opening path2: %d\n", open2rc);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
FILE_ERROR(ERRNO, open2rc * 10 - 2);
}
rc = RELATE_DIFFERENT;
if (open2rc > 0)
{
if (str1.bindp == str2.bindp)
rc = RELATE_SAME;
else if (info2.attr & ATTR_PREFIX)
rc = RELATE_PREFIX;
}
else /* open2rc == 0 */
{
/* path1 existing and path2's final part not can only be a prefix or
different */
if (info2.attr & ATTR_PREFIX)
rc = RELATE_PREFIX;
}
file_error:
if (open1rc >= 0)
close_stream_internal(&str1);
if (open2rc >= 0)
close_stream_internal(&str2);
file_internal_unlock_WRITER();
return rc;
}
/* test file or directory existence */
bool file_exists(const char *path)
{
file_internal_lock_WRITER();
bool rc = test_stream_exists_internal(path, FF_ANYTYPE) > 0;
file_internal_unlock_WRITER();
return rc;
}