[LU-11102] lock revoke may not take effect

• Type: Bug
• Resolution: Fixed
• Priority: Critical
• Fix Version/s: Lustre 2.12.0
• Affects Version/s: Lustre 2.10.0, Lustre 2.11.0
• Severity: 3
• Assignee: Lai Siyao
• Reporter: Lai Siyao

Description

Under some situations lock revoke may not take effect, suppose a concurrent setattr on a striped directory and rename under this directory:
1. setattr takes UPDATE lock of directory, but not unlock it yet (i.e., this lock is not downgraded to COS lock).
2. a concurrent 'mv' under this directory will first getattr file by name, this getattr will revoke UPDATE lock of this directory, since this getattr is not a distributed operation, this revoke does nothing.
3. setattr unlocks this UPDATE lock and downgrade it to COS mode, which notifies getattr that its lock is granted.
4. rename tries to lock UPDATE lock of this directory, but this lock was revoked before (though nothing did), it won't revoke again, rename will wait until setattr transaction commit.

Issue Links

• is related to: LU-11170 sanity test 415 fails with 'rename took N > M sec'
• is related to: LU-11270 Racer is stuck on ldlm lock
• is related to: LU-11082 stuck threads on MDS

Comments

[Peter Jones]

Landed for 2.12

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/33458/
Subject: LU-11102 ldlm: don't skip bl_ast for local lock
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 75a417fa0065d52a31215daaaaf41c0fa9751a89

[Gerrit Updater]

Mike Pershin ([e-mail]) uploaded a new patch: https://review.whamcloud.com/33458
Subject: LU-11102 ldlm: don't skip bl_ast for local lock
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 78a841012f0a7c47ac88ef03a7e92f4832d65d57

[Gerrit Updater]

Mike Pershin ([e-mail]) uploaded a new patch: https://review.whamcloud.com/33199
Subject: LU-11102 ldlm: don't skip bl_ast for local lock completely
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 0cd33ff7d13718aa61428d7988dd3f019b84c1c2

[Mikhail Pershin]

Commit 954cc6754b19a5eb4b9f717f79037c40baa87f3f contains the following issue - a local lock on server may be returned to a client via mdt_intent_lock_replace() and if it is blocking lock for some other waiting lock it must has LDLM_FL_AST_SENT flag while returned to the client, so client will call bl_ast on that lock. Otherwise the waiting lock will wait until timeout. That means we cannot just don't skip bl_ast for local locks until we sure it will not be sent to the client.

[Mikhail Pershin]

reopen due to problems with patch 954cc6754b19a5eb4b9f717f79037c40baa87f3f

[Gerrit Updater]

Mike Pershin ([e-mail]) uploaded a new patch: https://review.whamcloud.com/33066
Subject: LU-11102 ldlm: revert commit 954cc675
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: c0fe2f58b6c68896248087604031d311a6905afd