xfs: fix hung when transaction commit fail in xfs_inactive_ifree

Offering: HULK
hulk inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I76JSK



--------------------------------

After running unplug disk test and unmount filesystem, the umount thread
hung all the time.

 crash> dmesg
 sd 0:0:0:0: rejecting I/O to offline device
 XFS (sda): log I/O error -5
 XFS (sda): Corruption of in-memory data (0x8) detected at
 xfs_defer_finish_noroll+0x12e0/0x1cf0 (fs/xfs/libxfs/xfs_defer.c:504).
	Shutting down filesystem.
 XFS (sda): Please unmount the filesystem and rectify the problem(s)
 XFS (sda): xfs_inactive_ifree: xfs_trans_commit returned error -5
 XFS (sda): Unmounting Filesystem

 crash> bt 3368
 PID: 3368   TASK: ffff88801bcd8040  CPU: 3   COMMAND: "umount"
  #0 [ffffc900086a7ae0] __schedule at ffffffff83d3fd25
  #1 [ffffc900086a7be8] schedule at ffffffff83d414dd
  #2 [ffffc900086a7c10] xfs_ail_push_all_sync at ffffffff8256db24
  #3 [ffffc900086a7d18] xfs_unmount_flush_inodes at ffffffff824ee7e2
  #4 [ffffc900086a7d28] xfs_unmountfs at ffffffff824f2eff
  #5 [ffffc900086a7da8] xfs_fs_put_super at ffffffff82503e69
  #6 [ffffc900086a7de8] generic_shutdown_super at ffffffff81aeb8cd
  #7 [ffffc900086a7e10] kill_block_super at ffffffff81aefcfa
  #8 [ffffc900086a7e30] deactivate_locked_super at ffffffff81aeb2da
  #9 [ffffc900086a7e48] deactivate_super at ffffffff81aeb639
 #10 [ffffc900086a7e68] cleanup_mnt at ffffffff81b6ddd5
 #11 [ffffc900086a7ea0] __cleanup_mnt at ffffffff81b6dfdf
 #12 [ffffc900086a7eb0] task_work_run at ffffffff8126e5cf
 #13 [ffffc900086a7ef8] exit_to_user_mode_prepare at ffffffff813fa136
 #14 [ffffc900086a7f28] syscall_exit_to_user_mode at ffffffff83d25dbb
 #15 [ffffc900086a7f40] do_syscall_64 at ffffffff83d1f8d9
 #16 [ffffc900086a7f50] entry_SYSCALL_64_after_hwframe at ffffffff83e00085

When we free a cluster buffer from xfs_ifree_cluster, all the inodes in
cache are marked XFS_ISTALE. On journal commit dirty stale inodes as are
handled by both buffer and inode log items, inodes marked as XFS_ISTALE
in AIL will be removed from the AIL because the buffer log item will clean
it. If the transaction commit fails in the xfs_inactive_ifree(), inodes
marked as XFS_ISTALE will be left in AIL due to buf log item is not
committed, this will cause the unmount thread above to be blocked all the
time. Set inode item abort associated with the buffer that is stale after
buf item release, let ail clean up these items, that prevent inode item
left in AIL and can not being pushed.

Signed-off-by: Long Li <leo.lilong@huawei.com>