ext3: wait on all pending commits in ext3_sync_fs
authorArthur Jones <ajones@riverbed.com>
Thu, 6 Nov 2008 20:53:35 +0000 (12:53 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 6 Nov 2008 23:41:19 +0000 (15:41 -0800)
In ext3_sync_fs, we only wait for a commit to finish if we started it, but
there may be one already in progress which will not be synced.

In the case of a data=ordered umount with pending long symlinks which are
delayed due to a long list of other I/O on the backing block device, this
causes the buffer associated with the long symlinks to not be moved to the
inode dirty list in the second phase of fsync_super.  Then, before they
can be dirtied again, kjournald exits, seeing the UMOUNT flag and the
dirty pages are never written to the backing block device, causing long
symlink corruption and exposing new or previously freed block data to
userspace.

This can be reproduced with a script created
by Eric Sandeen <sandeen@redhat.com>:

#!/bin/bash

umount /mnt/test2
mount /dev/sdb4 /mnt/test2
rm -f /mnt/test2/*
dd if=/dev/zero of=/mnt/test2/bigfile bs=1M count=512
touch
/mnt/test2/thisisveryveryveryveryveryveryveryveryveryveryveryveryveryveryveryverylongfilename
ln -s
/mnt/test2/thisisveryveryveryveryveryveryveryveryveryveryveryveryveryveryveryverylongfilename
/mnt/test2/link
umount /mnt/test2
mount /dev/sdb4 /mnt/test2
ls /mnt/test2/
umount /mnt/test2

To ensure all commits are synced, we flush all journal commits now when
sync_fs'ing ext3.

Signed-off-by: Arthur Jones <ajones@riverbed.com>
Cc: Eric Sandeen <sandeen@redhat.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: <linux-ext4@vger.kernel.org>
Cc: <stable@kernel.org> [2.6.everything]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/ext3/super.c

index e5717a4fae67740459104b9da4ba8b1f6a79a4ef..5dec6d1356c4b07477bb40f68305c0d9de89258b 100644 (file)
@@ -2390,13 +2390,12 @@ static void ext3_write_super (struct super_block * sb)
 
 static int ext3_sync_fs(struct super_block *sb, int wait)
 {
-       tid_t target;
-
        sb->s_dirt = 0;
-       if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
-               if (wait)
-                       log_wait_commit(EXT3_SB(sb)->s_journal, target);
-       }
+       if (wait)
+               ext3_force_commit(sb);
+       else
+               journal_start_commit(EXT3_SB(sb)->s_journal, NULL);
+
        return 0;
 }