Remove E_NOTSUPP define. Use ENOTSUP directly where required.
(vfs_clone_file): use ENOTSUP instead of EOPNOTSUPP.
(mc_open): likewize. Fix sign.
Signed-off-by: Andrew Borodin <aborodin@vmail.ru>
The mc's built-in samba library which is used to access data across
smb links is taken from an old samba version.
Signed-off-by: Andrew Borodin <aborodin@vmail.ru>
Fix out-of-boundary access to gname if it's shorter than 255 bytes.
Found by Coverity.
Coverity id #331844.
Fix macro name that is used for array size.
Signed-off-by: Andrew Borodin <aborodin@vmail.ru>
Fix out-of-boundary access to uname if it's shorter than 255 bytes.
Found by Coverity.
Coverity id #331835.
Signed-off-by: Andrew Borodin <aborodin@vmail.ru>
...to free or not to free the string representation
of vfs_path_t object.
It allows to get rid of string duplication in following cases:
vfs_path_t *vpath;
char *path;
...
vpath = vfs_path_from_str (...);
path = g_strdup (vfs_path_as_str (vpath));
vfs_path_free (vpath);
Now we can write:
vfs_path_t *vpath;
char *path;
...
vpath = vfs_path_from_str (...);
path = vfs_path_free (vpath, FALSE);
Signed-off-by: Andrew Borodin <aborodin@vmail.ru>
In the function vfs_expire(), curr_time and exp_time are declared
guint64. curr_time is initialised with a timestamp and exp_time
with this timestamp minus vfs_timeout seconds. Later there is if
(stamping->time <= exp_time). Prior to commit
a94dd7d2de curr_time was initialised with
a value larger than vfs_timeout seconds, so everything was fine. This
commit changed the initialisation to a timer starting when mc is
started. So for the first vfs_timeout seconds, the result of the
subtraction is negative, but it is a guint64, so we just get a VERY
large unsigned value and the if (stamping->time <= exp_time) is always
true. So mc thinks the vfs hasn't been used recently and goes into an
infinite loop.
If one opens a .rpm file with mc and goes into the CONTENTS.cpio and
then tries to go into the .tar.gz there (this is the usual structure of
a .rpm) after waiting vfs_timeout seconds, everything is fine. However,
before vfs_timeout seconds, mc hangs.
Solution: use g_get_real_time() instead of mc_timer_elapsed().
Thanks nvwarr at hotmail.com for finding out the reason for this bug.
Signed-off-by: Andrew Borodin <aborodin@vmail.ru>
Avoid limitation of file name length.
(vfs_dirent): redefined to use instead of standard "struct direct"
to hold file name of any length.
(vfs_class::readdir): return newly allocated vfs_dirent structure.
Related changes.
Signed-off-by: Andrew Borodin <aborodin@vmail.ru>
Each VSF entry is added to VFS using vfs_s_insert_entry() via
g_list_append(). For long lists, a lot of walking through entire list
is performed. To get rid that, change type of vfs_s_inode::subdir from
GList to GQueue.
Signed-off-by: Andrew Borodin <aborodin@vmail.ru>
* (vfs_s_opendir): test if path is invalid or not ASAP.
* (vfs_s_readlink): likewise.
* (vfs_s_open): likewise.
Signed-off-by: Andrew Borodin <aborodin@vmail.ru>
