[LU-8353] mdt unlink should lock parent before lookup child

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.8.0
• Fix Version/s: Lustre 2.9.0
• Type: Bug
• Priority: Critical
• Reporter: Lai Siyao
• Assignee: Lai Siyao
• Resolution: Fixed
• Labels: dne2
• Epic/Theme: dne2
• Severity: 3

Description

In DNE Commit-on-Sharing implementation, mdt_reint_unlink() changed code to lookup child before lock parent, as is to find whether current operation is distributed, if so it needs to trigger commit of local changes to eliminate dependencies. But this may cause stale child used in later code. We should always follow lock-before-lookup semantic.

Issue Links

• is related to: LU-7117 replay-single test_70d: timeout and m...
• is related to: LU-6844 replay-single test 70b failure: 'rund...

Comments

[Gerrit Updater]

Lai Siyao ([e-mail]) uploaded a new patch: http://review.whamcloud.com/21088
Subject: LU-8353 mdt: mdt unlink should lock before lookup
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: cd90c77d1216c988126c86ec09de1b9242ab96e7

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch http://review.whamcloud.com/21088/
Subject: LU-8353 mdt: mdt unlink should lock before lookup
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 5e5dabac9eca8bc6ea516fb3ad53b8d74fe80a37

[Andreas Dilger]

Should this ticket be closed? Is it related to LU-7117 and should be linked there, or is this unrelated?

[Alex Zhuravlev]

I think it's one of the root causes for LU-7117, but it can pop up w/o 7117, w/o any recovery. basically we were doing lookups w/o LDLM lock and that can result in a corrupted namespace if another "unlink+create" can win race.

[Joseph Gmitter (Inactive)]

Patch landed to master for 2.9.0