[LU-5006] chown/chgrp doesn't work for files created by lfs setstripe

• Type: Bug
• Resolution: Fixed
• Priority: Critical
• Fix Version/s: Lustre 2.7.0
• Affects Version/s: Lustre 2.6.0, Lustre 2.5.1, Lustre 2.4.3
• Labels: HB
• Severity: 3
• Assignee: Niu Yawei (Inactive)
• Reporter: Niu Yawei (Inactive)

Description

This is easy to reproduce:

• lfs setstripe -c 1 -i 0 /mnt/lustre/a
• dd if=/dev/zero of=/mnt/lustre/a bs=1M count=1 conv=notrunc
• chown $ID /mnt/lustre/a

The uid on OST object will not be changed after chown.

Look into the code osp_attr_set():

        /* new object, the very first ->attr_set()
         * initializing attributes needs no logging
         * all subsequent one are subject to the
         * logging and synchronization with OST */
        if (o->opo_new) {
                o->opo_new = 0;
                RETURN(0);
        }

I don't quite see why this is necessary, but seems it'll break chown in above situation: OST object is created in 'dd' operation, and the first attr_set operation after create is the 'chown', and it's skipped.

Issue Links

• is related to: LU-5951 sanity test_39k: mtime is lost on close

Comments

[Gerrit Updater]

Niu Yawei ([e-mail]) uploaded a new patch: http://review.whamcloud.com/14774
Subject: LU-5006 mdd: don't call attr_set on object create
Project: fs/lustre-release
Branch: b2_4
Current Patch Set: 1
Commit: d0a2067981a16a1d536db3801d58328fbad2ff7e

[Gerrit Updater]

Niu Yawei ([e-mail]) uploaded a new patch: http://review.whamcloud.com/14773
Subject: LU-5006 mdd: don't call attr_set on object create
Project: fs/lustre-release
Branch: b2_5
Current Patch Set: 1
Commit: c79136559617ffd23415dab3d995e8aec994fab0

[Niu Yawei (Inactive)]

Ok, I'll port it soon.

[Jay Lan (Inactive)]

Can you cook up patches for 2.5.3 and 2.4.3, Niu? Cherry-picking resulted in big difference, especially in lustre/osp/osp_object.c.

[Niu Yawei (Inactive)]

Is this patch required for 2.5.3?

Yes, 2.5.3 doesn't have this fix. If your application perform the commands like the reproducer does, it'll hit the problem.

[Mahmoud Hanafi]

Is this patch required for 2.5.3?

[Jodi Levi (Inactive)]

Patch landed to Master. Please reopen ticket if more work is needed.

[Alex Zhuravlev]

forgot to mention - sanity/103 does check ACLs very well, please make sure it passes with ldiskfs and ZFS.

[Alex Zhuravlev]

please check mdd_acl_init() -> __mdd_fix_mode_acl(). it looks like the code was reorganized since that. the fingers are crossed for the testing

[Niu Yawei (Inactive)]

Thanks, Alex. I didn't see how default ACLs can change the object mode from code. (seems default ACLs is set after the additional attr_set call in current code). I cooked a patch to remove this additional attr_set: http://review.whamcloud.com/#/c/12243/