[LU-12509] lockdep warning in ofd_precreate_objects()

• Type: Improvement
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.13.0
• Assignee: Dongyang Li
• Reporter: Dongyang Li

Description

[ 141.023079]
============================================
[ 141.023750] WARNING: possible recursive locking detected
[ 141.024443] 4.18.0-debug #8 Tainted: G O --------- - -
[ 141.025226] --------------------------------------------
[ 141.025967] ll_ost00_002/7298 is trying to acquire lock:
[ 141.030372] 000000001989e622 (&mo->oo_sem){++++}, at: osd_write_lock+0x8a/0xb0 [osd_zfs]
[ 141.031190]
[ 141.031190] but task is already holding lock:
[ 141.032872] 0000000086a83aaf (&mo->oo_sem){++++}, at: osd_write_lock+0x8a/0xb0 [osd_zfs]
[ 141.033676]
[ 141.033676] other info that might help us debug this:
[ 141.034304] Possible unsafe locking scenario:
[ 141.034304]
[ 141.034875] CPU0
[ 141.035121] ----
[ 141.035366] lock(&mo->oo_sem);
[ 141.035678] lock(&mo->oo_sem);
[ 141.035993]
[ 141.035993] *** DEADLOCK ***
[ 141.035993]
[ 141.037086] May be due to missing lock nesting notation
[ 141.037086]
[ 141.037727] 3 locks held by ll_ost00_002/7298:
[ 141.038186] #0: 000000003540612a (&m->ofd_lastid_rwsem)

{.+.+}

, at: ofd_create_hdl+0x1a9/0x2270 [ofd]
[ 141.039081] #1: 00000000ee55375f (&oseq->os_create_lock){..}, at: ofd_create_hdl+0x364/0x2270 [ofd]
[ 141.039969] #2: 0000000086a83aaf (&mo->oo_sem){++++}, at: osd_write_lock+0x8a/0xb0 [osd_zfs]
[ 141.041147]
[ 141.041147] stack backtrace:
[ 141.041568] CPU: 0 PID: 7298 Comm: ll_ost00_002 Tainted: G O --------- - - 4.18.0-debug #8
[ 141.042465] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
[ 141.043004] Call Trace:
[ 141.043256] dump_stack+0x106/0x175
[ 141.043600] validate_chain.isra.26.cold.44+0x224/0x2da
[ 141.044103] __lock_acquire+0x3df/0xa70
[ 141.044535] lock_acquire+0x13a/0x370
[ 141.044908] ? osd_write_lock+0x8a/0xb0 [osd_zfs]
[ 141.045372] down_write_nested+0x6f/0x120
[ 141.045770] ? osd_write_lock+0x8a/0xb0 [osd_zfs]
[ 141.046236] osd_write_lock+0x8a/0xb0 [osd_zfs]
[ 141.046694] ofd_precreate_objects+0x1764/0x2480 [ofd]
[ 141.047201] ofd_create_hdl+0xbec/0x2270 [ofd]
[ 141.047722] tgt_handle_request0+0xdf/0x890 [ptlrpc]
[ 141.048274] tgt_request_handle+0x3ca/0x1aa0 [ptlrpc]
[ 141.048789] ? libcfs_nid2str_r+0x12e/0x160 [lnet]
[ 141.049332] ptlrpc_server_handle_request+0x634/0x1180 [ptlrpc]
[ 141.049898] ? __wake_up+0x17/0x20
[ 141.050299] ptlrpc_main+0xd7f/0x1470 [ptlrpc]
[ 141.050794] ? ptlrpc_register_service+0x14c0/0x14c0 [ptlrpc]
[ 141.051353] kthread+0x190/0x1c0
[ 141.051668] ? kthread_create_worker+0x90/0x90
[ 141.052102] ret_from_fork+0x3a/0x50

we should use different subclass to silent the lockdep false warning.

Issue Links

• is related to: LU-12269 Support RHEL 8.0

Comments

[Peter Jones]

Landed for 2.13

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/35420/
Subject: LU-12509 ofd: ofd_precreate_objects lockdep warning
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 697dcf4e87f2dbebe57f3ccb9c0b0962b89cf1b4

[Alex Zhuravlev]

hmm, took Oleg's .config for rhel8, but still can't get the warning.. keep trying.

[Dongyang Li]

This is actually found by Oleg, see here:
http://testing.linuxhacker.ru:3333/lustre-reports/dev/7/testresults/racer-zfs-rhel8.0_x86_64-rhel8.0_x86_64/oleg103-server-console.txt

and here:
http://testing.linuxhacker.ru:3333/lustre-reports/dev/7/

[Alex Zhuravlev]

what exact kernel options do you use to cause the warning?
I'm using linux-4.18.0-32.el8 with the following options:
CONFIG_LOCKDEP_SUPPORT=y
CONFIG_LOCKDEP=y
CONFIG_DEBUG_LOCKDEP=y
but can't reproduce the warning.

[Dongyang Li]

Looks like 31293 changed the dt_write_lock order, so if we have 31293 landed, we don't need this patch anymore.

[Andreas Dilger]

Alex, does this relate to your patch https://review.whamcloud.com/31293 "LU-10048 ofd: take local locks within transaction"?