[LU-12120] LustreError: 15069:0:(tgt_grant.c:561:tgt_grant_incoming()) LBUG

• Type: Bug
• Resolution: Fixed
• Priority: Major
• Fix Version/s: Lustre 2.13.0, Lustre 2.12.3
• Affects Version/s: Lustre 2.10.5
• Severity: 3
• Assignee: Patrick Farrell (Inactive)
• Reporter: Mahmoud Hanafi

Description

First time we have see this lbug. We have a crash dump.

[11168227.609396] Lustre: Skipped 4149 previous similar messages
[11168546.559600] LustreError: 15069:0:(tgt_grant.c:559:tgt_grant_incoming()) nbp2-OST0121: cli 811dc111-6b22-b447-9144-bcd1173a571d/ffff881d57bff000 dirty -610055168 pend 0 grant -643806208
[11168546.609562] LustreError: 15069:0:(tgt_grant.c:561:tgt_grant_incoming()) LBUG
[11168546.631425] Pid: 15069, comm: ll_ost_io01_084 3.10.0-693.21.1.el7.20180508.x86_64.lustre2105 #1 SMP Mon Aug 27 23:04:41 UTC 2018
[11168546.631428] Call Trace:
[11168546.631439]  [<ffffffff8103a1f2>] save_stack_trace_tsk+0x22/0x40
[11168546.650223]  [<ffffffffa08597cc>] libcfs_call_trace+0x8c/0xc0 [libcfs]
[11168546.650228]  [<ffffffffa085987c>] lbug_with_loc+0x4c/0xa0 [libcfs]
[11168546.650292]  [<ffffffffa0e8c270>] tgt_grant_prepare_read+0x0/0x3b0 [ptlrpc]
[11168546.650325]  [<ffffffffa0e8c37b>] tgt_grant_prepare_read+0x10b/0x3b0 [ptlrpc]
[11168546.650335]  [<ffffffffa0c86590>] ofd_preprw+0x450/0x1170 [ofd]
[11168546.650365]  [<ffffffffa0e702a5>] tgt_brw_read+0x975/0x1860 [ptlrpc]
[11168546.650394]  [<ffffffffa0e6deca>] tgt_request_handle+0x92a/0x1370 [ptlrpc]
[11168546.650420]  [<ffffffffa0e164bb>] ptlrpc_server_handle_request+0x23b/0xaa0 [ptlrpc]
[11168546.650445]  [<ffffffffa0e1a4a2>] ptlrpc_main+0xa92/0x1e40 [ptlrpc]
[11168546.650449]  [<ffffffff810b1131>] kthread+0xd1/0xe0
[11168546.650452]  [<ffffffff816a14f7>] ret_from_fork+0x77/0xb0
[11168546.650468]  [<ffffffffffffffff>] 0xffffffffffffffff
[11168546.650469] Kernel panic - not syncing: LBUG 
 

Issue Links

• is duplicated by: LU-11939 ASSERTION( tgd->tgd_tot_granted >= ted->ted_grant ) on OSS
• is duplicated by: LU-12322 negative grant and tgt_grant.c:561:tgt_grant_incoming() LBUG
• is related to: LU-12759 parameter grant_shrink gets reset to 1 after client reconnects

Comments

[Jay Lan (Inactive)]

The patch was applied cleanly to nas-2.10.8. No need of a backport.

[Stephane Thiell]

We used the patch from https://review.whamcloud.com/35084/ which applied against 2.10.8 with no change (or it was trivial).

[Mahmoud Hanafi]

We need a 2.10 backport. Our 2.10.7 server this lbug

 7955125.139926] LustreError: 23678:0:(tgt_grant.c:1068:tgt_grant_discard()) ASSERTION( tgd->tgd_tot_granted >= ted->ted_grant ) failed: nbp7-OST0017: tot_granted 7448219072 cli 4817e87d-d95a-1480-0cbd-835e2c3966b1/ffff880f68fdc400 ted_grant -49152
[7955125.164200] LustreError: 23678:0:(tgt_grant.c:1068:tgt_grant_discard()) LBUG

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/35084/
Subject: LU-12120 grants: prevent negative ted_grant value
Project: fs/lustre-release
Branch: b2_12
Current Patch Set:
Commit: 83d43ecc947fa8f43bd5c7f792f3788cadd0d3ef

[Gerrit Updater]

Minh Diep ([e-mail]) uploaded a new patch: https://review.whamcloud.com/35084
Subject: LU-12120 grants: prevent negative ted_grant value
Project: fs/lustre-release
Branch: b2_12
Current Patch Set: 1
Commit: 82ad1cf4fd9298f95a8f233ee18feef363b10ebb

[Peter Jones]

Landed for 2.13

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/34996/
Subject: LU-12120 grants: prevent negative ted_grant value
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 7e08317ef5cbed5cd587017cbe343eb4cc52822c

[Gerrit Updater]

Mike Pershin ([e-mail]) uploaded a new patch: https://review.whamcloud.com/34996
Subject: LU-12120 grants: prevent negative ted_grant value
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 296ac6b910880dc65c9cee8be2552bba759cea57

[Quentin Bouget (Inactive)]

Ok, thanks Mikhail!

From what I understand clients and servers maintain separate views of grants and this LU relates to pairs of (client; server) that drift apart from each other. At the next shrink request, that results in the client asking to give back more space than the server remembers granting.

—

Does anyone know if there a valid reason as to why in tgt_grant_incoming():

static void tgt_grant_incoming(...)
{
        ...
        CDEBUG(D_CACHE, "%s: cli %s/%p reports grant %llu dropped %u, local %lu\n",
               obd->obd_name, exp->exp_client_uuid.uuid, exp, oa->o_grant,
               oa->o_dropped, ted->ted_grant);
        ...
}

The value printed after "reports grant" would not be the same as the value printed after "local"?
Can this be used to track the desynchronization?

[Mikhail Pershin]

Quentin, I agree with your analysis and also think that tgt_grant_shrink should not decrease ted_grant blindly but do sanity checks there. I will prepare patch today