xfs_file_last_byte() needs to acquire ilock
authorLachlan McIlroy <lmcilroy@redhat.com>
Fri, 24 Apr 2009 02:18:00 +0000 (22:18 -0400)
committerFelix Blyakher <felixb@sgi.com>
Thu, 30 Apr 2009 05:25:25 +0000 (00:25 -0500)
We had some systems crash with this stack:

[<a00000010000cb20>] ia64_leave_kernel+0x0/0x280
[<a00000021291ca00>] xfs_bmbt_get_startoff+0x0/0x20 [xfs]
[<a0000002129080b0>] xfs_bmap_last_offset+0x210/0x280 [xfs]
[<a00000021295b010>] xfs_file_last_byte+0x70/0x1a0 [xfs]
[<a00000021295b200>] xfs_itruncate_start+0xc0/0x1a0 [xfs]
[<a0000002129935f0>] xfs_inactive_free_eofblocks+0x290/0x460 [xfs]
[<a000000212998fb0>] xfs_release+0x1b0/0x240 [xfs]
[<a0000002129ad930>] xfs_file_release+0x70/0xa0 [xfs]
[<a000000100162ea0>] __fput+0x1a0/0x420
[<a000000100163160>] fput+0x40/0x60

The problem here is that xfs_file_last_byte() does not acquire the
inode lock and can therefore race with another thread that is modifying
the extext list.  While xfs_bmap_last_offset() is trying to lookup
what was the last extent some extents were merged and the extent list
shrunk so the index we lookup is now beyond the end of the extent list
and potentially in a freed buffer.

Signed-off-by: Lachlan McIlroy <lmcilroy@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Felix Blyakher <felixb@sgi.com>
Signed-off-by: Felix Blyakher <felixb@sgi.com>
fs/xfs/xfs_inode.c

index e7ae08d1df485105a10f89cd8e976a826e63af53..123b20c8cbf23ddd54e697d0f3a0d058e126f44b 100644 (file)
@@ -1258,8 +1258,10 @@ xfs_file_last_byte(
         * necessary.
         */
        if (ip->i_df.if_flags & XFS_IFEXTENTS) {
+               xfs_ilock(ip, XFS_ILOCK_SHARED);
                error = xfs_bmap_last_offset(NULL, ip, &last_block,
                        XFS_DATA_FORK);
+               xfs_iunlock(ip, XFS_ILOCK_SHARED);
                if (error) {
                        last_block = 0;
                }