[LU-18101] sanityn test_25a: FAIL: checkstat /mnt/lustre2/d25a.sanityn/f1 #2

• Type: Bug
• Resolution: Fixed
• Priority: Major
• Fix Version/s: Lustre 2.16.0
• Affects Version/s: Lustre 2.16.0
• Labels: mtg reviewed
• Environment: Ubuntu 24.04 client ; SLES 15 SP6 client
• Severity: 3
• Assignee: Sebastien Buisson
• Reporter: Maloo

Description

This issue was created by maloo for jianyu <[e-mail]>

This issue relates to the following test suite run: https://testing.whamcloud.com/test_sets/dbb37b62-af78-469d-b324-8b599665b4e7

test_25a failed with the following error:

== sanityn test 25a: change ACL on one mountpoint be seen on another ============================================================= 15:55:46 (1722527746)
running as uid/gid/euid/egid 500/500/500/500, groups:
 [checkstat] [-v] [/mnt/lustre2/d25a.sanityn/f1]
running as uid/gid/euid/egid 500/500/500/500, groups:
 [checkstat] [-v] [/mnt/lustre2/d25a.sanityn/f1]
 sanityn test_25a: @@@@@@ FAIL: checkstat /mnt/lustre2/d25a.sanityn/f1 #2 

Test session details:
clients: https://build.whamcloud.com/job/lustre-master/4559 - 6.8.0-35-generic
servers: https://build.whamcloud.com/job/lustre-master/4559 - 5.14.0-427.24.1_lustre.el9.x86_64

<<Please provide additional information about the failure here>>

VVVVVVV DO NOT REMOVE LINES BELOW, Added by Maloo for auto-association VVVVVVV
sanityn test_25a - checkstat /mnt/lustre2/d25a.sanityn/f1 #2

Issue Links

• is related to: LU-18070 sanity test_103a: FAIL: run_acl_subtest 'misc' failed

Comments

[Peter Jones]

Merged for 2.16

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/56552/
Subject: LU-18101 sec: fix ACL handling on recent kernels again
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 13fd5ebef3a7a1ae3574458674e16ca782b181e7

[Gerrit Updater]

"Sebastien Buisson <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/56552
Subject: LU-18101 sec: fix again ACL handling on recent kernels
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: b36c67b32d1ebbf4b85810d080556ab5c799202e

[Andreas Dilger]

Note that sanityn test_25b is also failing with the same problem, but it is normally skipped during Ubuntu 24.04 testing because the test has only one MDT.

[Andreas Dilger]

It looks like this problem has only hit on Ubuntu 24.04 and SLES 15sp6, so with newer kernels. It would appear that Lustre is correctly invalidating the ACL xattr from the xattr cache on the client, but the ACL itself is cached in the inode/VFS somewhere else and that is not being refreshed correctly.

[Andreas Dilger]

The test changes ACLs on file $DIR2/$tdir/f1 with multiple values, but does not clear the cache between changes.

Sebastien, which cache are you referring to? The xattr cache, or is there a separate level of ACL cache?

The xattr cache for security xattrs should be cleared when any security xattr/ACL/permission is changed on the MDS, otherwise a client can cache incorrect file access permissions for a long time (granting or denying access incorrectly).

If this is not happening automatically by the MDS acquiring the MDS_INODELOCK_PERM DLM lock bit during the xattr/ACL update, and this triggering client DLM lock (and related ACL/xattr) cancellation and refresh, then I would consider this a bug in the code, and not the test. Covering this failure up by having the test cancel the DLM locks, or otherwise flushing the xattrs from the client cache would just be hiding the problem until it will eventually be hit by a customer.

[Peter Jones]

Seems to be low frequency failure so can defer to a future release

[Sebastien Buisson]

I think the problem might be in the test itself. The test changes ACLs on file $DIR2/$tdir/f1 with multiple values, but does not clear the cache between changes.

[Colin Faber]

Hi sebastien can you triage this one as well?