[LU-12309] write fails with -EOPNOTSUPP with zfs 0.8.0-rc5 using 0.7.x pool

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.10.7
• Fix Version/s: Lustre 2.13.0, Lustre 2.12.3
• Type: Bug
• Priority: Minor
• Reporter: Olaf Faaland
• Assignee: Nathaniel Clark
• Resolution: Fixed
• Labels: llnl
• Severity: 3

Description

Testing Lustre 2.10.7 against zfs 0.8.0-rc5

If the file system is stored in pools created under previous versions of zfs and the project quota feature has not yet been enabled on those pools, writes fail with -EOPNOTSUPP

Debug logs on the OST seem to indicate that the -EOPNOTSUPP is returned from osd_declare_attr_set(), which requires that if Lustre has been built with the ZFS project quota feature enabled, that the pool must have the feature enabled.

#ifdef ZFS_PROJINHERIT
       if (attr && attr->la_valid & LA_PROJID) {
               if (!osd->od_projectused_dn)
                       GOTO(out, rc = -EOPNOTSUPP);

It's unclear why this is necessary. In addition, every user who upgrades from an earlier version of lustre will initially be in this state, where their zfs software supports project quotas, but the feature is not enabled on the pool. As long as they do not enable the feature on the pool, they can revert to the earlier zfs version if they encounter issues.

Comments

[Olaf Faaland]

The stack is 2.10.7 plus a few build-related patches. For the exact source, see
https://github.com/ofaaland/lustre/commits/b-lustre210-zfs08
commit 2331955

[Peter Jones]

Nathaniel

Could you please investigate?

Peter

[Olaf Faaland]

-1 debug logs from client and one of the OSTs attached.

[Olaf Faaland]

I haven't tried it, but I believe you could reproduce this with a new file system easily, just by using option "-d" to zpool create, and then explicitly enabling features other than project quotas.

[Gerrit Updater]

Nathaniel Clark ([e-mail]) uploaded a new patch: https://review.whamcloud.com/34888
Subject: LU-12309 osd-zfs: Support disabled project quotas
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 74733dd3aa761dbe0477b599b166a3791e8582c5

[Andreas Dilger]

I think the intent is that someone trying to set the projid on a file where the storage does not support (or enable) project quota should get }{-ENOTSUPP}}. Definitely it shouldn't refuse to mount in that case.

Is there something that you are running in userspace that is trying to set projid, or is this just normal usage?

Possibly the check could have a caveat that it shouldn't try to set projid if it is zero and not enabled, because that is just the default value?

[Olaf Faaland]

Is there something that you are running in userspace that is trying to set projid, or is this just normal usage?

Mount (all the servers and the client) is successful. I see the failure with an ordinary write, e.g. using bash redirect or using dd.

Possibly the check could have a caveat that it shouldn't try to set projid if it is zero and not enabled, because that is just the default value?

Good idea.

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/34888/
Subject: LU-12309 osd-zfs: Support disabled project quotas
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 291e7196d39365739f9daa02efd25535b5415174

[Peter Jones]

Landed for 2.13

[Gerrit Updater]

Minh Diep ([e-mail]) uploaded a new patch: https://review.whamcloud.com/35009
Subject: LU-12309 osd-zfs: Support disabled project quotas
Project: fs/lustre-release
Branch: b2_12
Current Patch Set: 1
Commit: 259c14db010a0733c4e467e451250a89c5efac3e

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/35009/
Subject: LU-12309 osd-zfs: Support disabled project quotas
Project: fs/lustre-release
Branch: b2_12
Current Patch Set:
Commit: 965af09947473129c45e8201e9f9b4cd4da77d8b