[LU-11218] osp xattr cache brought by mdd_is_parent needs to be invalidated

• Status: Closed
• Project: Lustre
• Type: Bug
• Priority: Major
• Reporter: Vladimir Saveliev
• Assignee: Vladimir Saveliev
• Resolution: Fixed
• Severity: 3

Description

mdd_is_parent() brings link xattrs of not locked objects into osp cache.
If that cache becomes stale a directory rename includes an outdated link xattr into object xattr table. That makes a directory to have more than one link. That also confuses lfsck.

Issue Links

• is related to: LU-10406 sanity-lfsck test_31c: (4) Fail to re...

Comments

[Gerrit Updater]

Vladimir Saveliev ([e-mail]) uploaded a new patch: https://review.whamcloud.com/32942
Subject: LU-11218 mdd: do not cache atributes on mdd_parent_fid
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 092ac85e6d522c5d797686f3be785c778c2d7cbc

[Alex Zhuravlev]

probably https://review.whamcloud.com/#/c/31194/ can help

 

[Vladimir Saveliev]

probably https://review.whamcloud.com/#/c/31194/ can help

Yes, it helps: the sanity.sh:807 from https://review.whamcloud.com/32942 does not fail with it.

Alex, however, that patch seems to get stuck since march. Do you have plans to move it forward?

[Andreas Dilger]

Alex, Vladimir, do you have a preference of which of these two patches to land?

Is one "more complete" than the other? It seems like Alex's LU-10406 patch will also fix the bug with LFSCK, while the patch here will only fix the one MDD case?

I've rebased Alex's patch to verify that it is still working correctly with master (there can be bad interactions if a very old patch is landed without retesting), so if this is the better one to land then it should be ready in a few hours (barring some intermittent failure).

[Alex Zhuravlev]

I think we should land the old (LU-10406) patch as it improves locking/invalidation schema in general.

 

[Vladimir Saveliev]

https://review.whamcloud.com/31194 at LU-10406 fixes this.