jbd: Fix lock ordering bug in journal_unmap_buffer()
authorJan Kara <jack@suse.cz>
Fri, 23 Nov 2012 13:03:04 +0000 (14:03 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 3 Dec 2012 20:59:14 +0000 (12:59 -0800)
commit1c4e871038f17769d943d4473c339171fed70f45
treed91dbffdd5c5068898e955e33767f204a5354415
parentf6a8d4ccb6a92e815669e65dd1140280876f3bd1
jbd: Fix lock ordering bug in journal_unmap_buffer()

commit 25389bb207987b5774182f763b9fb65ff08761c8 upstream.

Commit 09e05d48 introduced a wait for transaction commit into
journal_unmap_buffer() in the case we are truncating a buffer undergoing commit
in the page stradding i_size on a filesystem with blocksize < pagesize. Sadly
we forgot to drop buffer lock before waiting for transaction commit and thus
deadlock is possible when kjournald wants to lock the buffer.

Fix the problem by dropping the buffer lock before waiting for transaction
commit. Since we are still holding page lock (and that is OK), buffer cannot
disappear under us.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/jbd/transaction.c