[LU-5057] ofd_fini()) ASSERTION( atomic_read(&d->ld_ref) == 0 ) failed:

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.6.0
• Type: Bug
• Priority: Major
• Reporter: Jinshan Xiong (Inactive)
• Assignee: Mikhail Pershin
• Resolution: Duplicate
• Severity: 3

Description

I keep seeing this issue on racer test. If someone is working on this issue and need more log, please feel free to contact me.

LustreError: 24020:0:(ofd_dev.c:859:ofd_fini()) ASSERTION( atomic_read(&d->ld_ref) == 0 ) failed:
LustreError: 24020:0:(ofd_dev.c:859:ofd_fini()) LBUG
Pid: 24020, comm: umount

Call Trace:
[<ffffffffa03a3895>] libcfs_debug_dumpstack+0x55/0x80 [libcfs]
[<ffffffffa03a3e97>] lbug_with_loc+0x47/0xb0 [libcfs]
[<ffffffffa0e16b2b>] ofd_device_fini+0x23b/0x240 [ofd]
[<ffffffffa051c5d3>] class_cleanup+0x573/0xd30 [obdclass]
[<ffffffffa04f30b6>] ? class_name2dev+0x56/0xe0 [obdclass]
[<ffffffffa051e2fa>] class_process_config+0x156a/0x1ad0 [obdclass]
[<ffffffffa0517453>] ? lustre_cfg_new+0x2d3/0x6e0 [obdclass]
[<ffffffffa051e9d9>] class_manual_cleanup+0x179/0x6f0 [obdclass]
[<ffffffffa04f30b6>] ? class_name2dev+0x56/0xe0 [obdclass]
[<ffffffffa0552fbc>] server_put_super+0x5ec/0xf60 [obdclass]
[<ffffffff8118366b>] generic_shutdown_super+0x5b/0xe0
[<ffffffff81183756>] kill_anon_super+0x16/0x60
[<ffffffffa0520886>] lustre_kill_super+0x36/0x60 [obdclass]
[<ffffffff81183ef7>] deactivate_super+0x57/0x80
[<ffffffff811a21ef>] mntput_no_expire+0xbf/0x110
[<ffffffff811a2c5b>] sys_umount+0x7b/0x3a0
[<ffffffff8100b072>] system_call_fastpath+0x16/0x1b

Issue Links

• is related to: LU-6434 Object reference is not zero when umo...

Comments

[Patrick Farrell (Inactive)]

Cray has seen this recently in our testing as well.

[Andreas Dilger]

Mike, any idea on what might be causing this? Oleg has crash dumps from this situation as well, and it started a few months ago.

[Patrick Farrell (Inactive)]

For what it's worth, we hit this in a fairly clean (IE, close to Intel's) version of 2.5.1. If requested, I might be able to make the dump available, though only default debug was enabled.

[Mikhail Pershin]

well, such errors means that device has still references, objects most likely. Usually this is missed object put in some rare code path.

[Andreas Dilger]

Jinshan, Patrick, are you still seeing this bug in your testing?

[Patrick Farrell (Inactive)]

Andreas - No. We have not seen this again since my report above, and we have been running racer in the same configurations regularly since then. It seems very likely this was fixed, or is so rare as to not be a concern.

[Jinshan Xiong (Inactive)]

I saw this issue again. Please check the attachment for the log.

[Jinshan Xiong (Inactive)]

duplication of LU-6434