Workaround the brokenness of realpath() a bit more.

At least on android it blindly writes to the end of the buffer passed to it assuming it's sufficiently. It wasn't in our case, resulting in a buffer overflow (and breakage).
This should fix strange problems relating to database initialization on application targets.

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@29147 a1c6a512-1295-4272-9138-f99709370657
This commit is contained in:
Thomas Martitz 2011-01-27 20:17:55 +00:00
parent 3a1bc3cfdd
commit b703d251be

View file

@ -4226,8 +4226,17 @@ static bool add_search_root(const char *name)
#ifndef WIN32
struct search_roots_ll *this, *prev = NULL;
char target[MAX_PATH];
char _abs_target[MAX_PATH];
char * abs_target;
/* Okay, realpath() is almost completely broken on android
*
* It doesn't accept NULL for resolved_name to dynamically allocate
* the resulting path; and it assumes resolved_name to be PATH_MAX
* (actually MAXPATHLEN, but it's the same [as of 2.3]) long
* and blindly writes to the end if it
*
* therefore use sufficiently large static storage here
* Note that PATH_MAX != MAX_PATH
**/
static char abs_target[PATH_MAX];
ssize_t len;
len = readlink(name, target, sizeof(target));
@ -4235,9 +4244,7 @@ static bool add_search_root(const char *name)
return false;
target[len] = '\0';
/* realpath(target, NULL) doesn't work on android ... */
abs_target = realpath(target, _abs_target);
if (abs_target == NULL)
if (realpath(target, abs_target) == NULL)
return false;
for(this = &roots_ll; this; prev = this, this = this->next)
@ -4258,6 +4265,7 @@ static bool add_search_root(const char *name)
logf("Error at adding a search root: %s", this ? "path too long":"OOM");
free(this);
prev->next = NULL;
return false;
}
this->path = ((char*)this) + sizeof(struct search_roots_ll);
strcpy((char*)this->path, abs_target); /* ok to cast const away here */