[Top] [All Lists]

3.6.0-rc6: inconsistent {IN-RECLAIM_FS-W} -> {RECLAIM_FS-ON-W} usage

To: xfs@xxxxxxxxxxx
Subject: 3.6.0-rc6: inconsistent {IN-RECLAIM_FS-W} -> {RECLAIM_FS-ON-W} usage
From: Christian Kujau <lists@xxxxxxxxxxxxxxx>
Date: Fri, 21 Sep 2012 17:37:43 -0700 (PDT)
User-agent: Alpine 2.01 (DEB 1266 2009-07-14)
After upgrading from 3.5 to 3.6.0-rc6, the lockdep warning below is 
printed after the machine was running for ~1 day. When it was printed, 
disk i/o load was higher than usual as the machine was running backups, so 
this might've triggered it.

I've observed a simliar lockdep warning earlier[0] for 3.5.0-rc5, but with 
a different backtrace. Also, then I was told that I may have run out of 
inode attributes. As I have not reformatted the filesystem, this might 
still be the case.

A more similar backtrace has been posted[1] for 3.5.0-rc1, but I don't 
think a consensus on its solution has been reached.

The .config, mountoptions and full dmesg are here:



[0] http://oss.sgi.com/archives/xfs/2012-07/msg00113.html
[1] https://lkml.org/lkml/2012/6/13/582

[ INFO: inconsistent lock state ]
3.6.0-rc6-00052-gc46de22 #1 Not tainted
inconsistent {IN-RECLAIM_FS-W} -> {RECLAIM_FS-ON-W} usage.
rm/12693 [HC0[0]:SC0[0]:HE1:SE1] takes:
 (&(&ip->i_lock)->mr_lock){++++?-}, at: [<c01ab424>] xfs_ilock+0x8c/0xb0
{IN-RECLAIM_FS-W} state was registered at:
  [<c0070224>] lock_acquire+0x50/0x6c
  [<c005626c>] down_write_nested+0x54/0x94
  [<c01ab424>] xfs_ilock+0x8c/0xb0
  [<c01b6334>] xfs_reclaim_inode+0x11c/0x32c
  [<c01b717c>] xfs_reclaim_inodes_ag+0x1c4/0x3c8
  [<c01b74e0>] xfs_reclaim_inodes_nr+0x38/0x4c
  [<c01b3f44>] xfs_fs_free_cached_objects+0x14/0x24
  [<c00c8368>] prune_super+0xf4/0x188
  [<c009b9bc>] shrink_slab+0x1c0/0x2b4
  [<c009ddbc>] kswapd+0x460/0x940
  [<c0050444>] kthread+0x84/0x88
  [<c000ecf8>] kernel_thread+0x4c/0x68
irq event stamp: 6402587
hardirqs last  enabled at (6402587): [<c04ca678>] 
hardirqs last disabled at (6402586): [<c04c9d80>] 
softirqs last  enabled at (6401106): [<c0037d74>] __do_softirq+0x138/0x17c
softirqs last disabled at (6401095): [<c000e7f4>] call_do_softirq+0x14/0x24

other info that might help us debug this:
 Possible unsafe locking scenario:


 *** DEADLOCK ***

4 locks held by rm/12693:
 #0:  (sb_writers#11){.+.+.+}, at: [<c00e488c>] mnt_want_write+0x24/0x58
 #1:  (&type->i_mutex_dir_key#5/1){+.+.+.}, at: [<c00d38d0>] do_rmdir+0xac/0x110
 #2:  (sb_internal#2){.+.+.+}, at: [<c01f316c>] xfs_trans_alloc+0x28/0x58
 #3:  (&(&ip->i_lock)->mr_lock){++++?-}, at: [<c01ab424>] xfs_ilock+0x8c/0xb0

stack backtrace:
Call Trace:
[d9e35b60] [c00091e4] show_stack+0x48/0x15c (unreliable)
[d9e35ba0] [c04cc4b4] print_usage_bug.part.34+0x260/0x274
[d9e35bd0] [c006e528] mark_lock+0x570/0x644
[d9e35c00] [c00708d4] mark_held_locks+0x98/0x168
[d9e35c40] [c0070fd4] lockdep_trace_alloc+0x84/0xe8
[d9e35c50] [c00c0834] kmem_cache_alloc+0x34/0x124
[d9e35c70] [c00b5088] vm_map_ram+0x228/0x5a0
[d9e35cf0] [c01a3b44] _xfs_buf_map_pages+0x44/0x104
[d9e35d10] [c01a4c1c] xfs_buf_get_map+0x74/0x11c
[d9e35d30] [c01fbd30] xfs_trans_get_buf_map+0xc0/0xdc
[d9e35d50] [c01e55bc] xfs_ifree_cluster+0x3f4/0x5b0
[d9e35de0] [c01e6c78] xfs_ifree+0xec/0xf0
[d9e35e20] [c01b9678] xfs_inactive+0x274/0x448
[d9e35e60] [c01b4930] xfs_fs_evict_inode+0x60/0x74
[d9e35e70] [c00e0858] evict+0xc0/0x1b4
[d9e35e90] [c00dc740] d_delete+0x1b0/0x1f4
[d9e35eb0] [c00d3820] vfs_rmdir+0x124/0x128
[d9e35ed0] [c00d390c] do_rmdir+0xe8/0x110
[d9e35f40] [c0010a4c] ret_from_syscall+0x0/0x38
--- Exception: c01 at 0xff4c918
    LR = 0x10001b2c
BOFH excuse #198:

Post-it Note Sludge leaked into the monitor.

<Prev in Thread] Current Thread [Next in Thread>
  • 3.6.0-rc6: inconsistent {IN-RECLAIM_FS-W} -> {RECLAIM_FS-ON-W} usage, Christian Kujau <=