[LU-7473] Enlarge ACL entry limitation

• Type: New Feature
• Resolution: Fixed
• Priority: Critical
• Fix Version/s: Lustre 2.10.0
• Affects Version/s: Lustre 2.8.0
• Assignee: nasf (Inactive)
• Reporter: Li Xi (Inactive)

Description

The current limitation of ACL entries is 32, which is not enough for some use cases. The limitation is inherited since Lustre-1.6 (or even earlier verion). Is there any special reason we keep this small limitation? The only reason I can think of, is to prevent space from exhausting before inodes exhausts on MDT. But, still 32 is too small, can we enlarge or even remove the limitation?

Issue Links

• is related to: LU-9514 sanity test 51f times out due to out of memory
• is related to: LU-10513 Prepare small buffer for ACL RPC reply

Comments

[Andreas Dilger]

Reopen to remove 2.11 fix version.

[nasf (Inactive)]

The patch https://review.whamcloud.com/28116 will be landed via LU-10513.

[Andreas Dilger]

Reopen to track landing of last patch.

[Gerrit Updater]

Fan Yong ([e-mail]) uploaded a new patch: https://review.whamcloud.com/28116
Subject: LU-7473 acl: prepare small buffer for ACL in RPC reply
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 971dbecce884acb07baccc175ea1e26336fff11c

[Peter Jones]

Landed for 2,10

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/19790/
Subject: LU-7473 acl: increase ACL entries limitation
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 64b2fad22a4eb4727315709e014d8f74c5a7f289

[Gerrit Updater]

Fan Yong ([e-mail]) uploaded a new patch: http://review.whamcloud.com/19790
Subject: LU-7473 acl: increase ACL entries limitation
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 3d33ba9e5d402c48bc8b76b9d4427b391ee85f65

[Andreas Dilger]

The reason that there is a limit on the ACLs is that the client needs to preallocate the reply buffer to receive the ACL data from the server, so it cannot be unlimited. The kernel limit in the past was also 32 entries, which the Lustre code also used.

It seems it would be possible to increase the ACL limit to 64 (each entry is only 8 bytes). Since the ACL entries are sharing the same space as the LOV layout there is an upper limit on how much space the two of them can use, so this shouldn't make the replies too large.

There will be some compatibility issues if this size is increased, if older and newer clients are mixed in the same cluster, and newer clients set a larger ACL that older clients cannot access. Besides this, I don't think there will be any protocol compatibility issues, since the ACL buffer size is described in the request and the reply, and older servers should be able to handle the larger ACL from newer clients.

Please ensure any patch for this is tested with both older and newer clients against older and newer servers. It would also be useful to enhance the conf-sanity test_32 cases to include ACLs, if that is not already the case today.