fix O_SYNC|O_APPEND syncing the wrong range on write()
authorAl Viro <viro@zeniv.linux.org.uk>
Sun, 9 Feb 2014 20:18:09 +0000 (15:18 -0500)
committerAl Viro <viro@zeniv.linux.org.uk>
Sun, 9 Feb 2014 20:18:09 +0000 (15:18 -0500)
commitd311d79de305f1ada47cadd672e6ed1b28a949eb
treeed5fe40264a27e1d8cc4410352585dfb34af50d8
parent38dbfb59d1175ef458d006556061adeaa8751b72
fix O_SYNC|O_APPEND syncing the wrong range on write()

It actually goes back to 2004 ([PATCH] Concurrent O_SYNC write support)
when sync_page_range() had been introduced; generic_file_write{,v}() correctly
synced
pos_after_write - written .. pos_after_write - 1
but generic_file_aio_write() synced
pos_before_write .. pos_before_write + written - 1
instead.  Which is not the same thing with O_APPEND, obviously.
A couple of years later correct variant had been killed off when
everything switched to use of generic_file_aio_write().

All users of generic_file_aio_write() are affected, and the same bug
has been copied into other instances of ->aio_write().

The fix is trivial; the only subtle point is that generic_write_sync()
ought to be inlined to avoid calculations useless for the majority of
calls.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
fs/cifs/file.c
fs/ext4/file.c
fs/ntfs/file.c
fs/sync.c
fs/xfs/xfs_file.c
include/linux/fs.h
mm/filemap.c