src/patches/suse-2.6.27.31/patches.fixes/xfs_file_last_byte-needs-to-acquire-ilock.patch

   1 From: Lachlan McIlroy <lmcilroy@redhat.com>
   2 Date: Fri, 24 Apr 2009 02:18:00 +0000 (-0400)
   3 Subject: xfs_file_last_byte() needs to acquire ilock
   4 Patch-mainline: 2.6.30-rc5
   5 Git-commit: def6b3ba56b637d58126ef67fc19bab57945fcc4
   6 References: SGI:PV963454 bnc#487987
   7
   8 xfs_file_last_byte() needs to acquire ilock
   9
  10 We had some systems crash with this stack:
  11
  12 [<a00000010000cb20>] ia64_leave_kernel+0x0/0x280
  13 [<a00000021291ca00>] xfs_bmbt_get_startoff+0x0/0x20 [xfs]
  14 [<a0000002129080b0>] xfs_bmap_last_offset+0x210/0x280 [xfs]
  15 [<a00000021295b010>] xfs_file_last_byte+0x70/0x1a0 [xfs]
  16 [<a00000021295b200>] xfs_itruncate_start+0xc0/0x1a0 [xfs]
  17 [<a0000002129935f0>] xfs_inactive_free_eofblocks+0x290/0x460 [xfs]
  18 [<a000000212998fb0>] xfs_release+0x1b0/0x240 [xfs]
  19 [<a0000002129ad930>] xfs_file_release+0x70/0xa0 [xfs]
  20 [<a000000100162ea0>] __fput+0x1a0/0x420
  21 [<a000000100163160>] fput+0x40/0x60
  22
  23 The problem here is that xfs_file_last_byte() does not acquire the
  24 inode lock and can therefore race with another thread that is modifying
  25 the extext list.  While xfs_bmap_last_offset() is trying to lookup
  26 what was the last extent some extents were merged and the extent list
  27 shrunk so the index we lookup is now beyond the end of the extent list
  28 and potentially in a freed buffer.
  29
  30 Signed-off-by: Lachlan McIlroy <lmcilroy@redhat.com>
  31 Reviewed-by: Christoph Hellwig <hch@lst.de>
  32 Reviewed-by: Felix Blyakher <felixb@sgi.com>
  33 Signed-off-by: Felix Blyakher <felixb@sgi.com>
  34 Acked-by: Jeff Mahoney <jeffm@suse.com>
  35 ---
  36
  37  fs/xfs/xfs_inode.c |    2 ++
  38  1 file changed, 2 insertions(+)
  39
  40 --- a/fs/xfs/xfs_inode.c
  41 +++ b/fs/xfs/xfs_inode.c
  42 @@ -1306,8 +1306,10 @@ xfs_file_last_byte(
  43          * necessary.
  44          */
  45         if (ip->i_df.if_flags & XFS_IFEXTENTS) {
  46 +               xfs_ilock(ip, XFS_ILOCK_SHARED);
  47                 error = xfs_bmap_last_offset(NULL, ip, &last_block,
  48                         XFS_DATA_FORK);
  49 +               xfs_iunlock(ip, XFS_ILOCK_SHARED);
  50                 if (error) {
  51                         last_block = 0;
  52                 }