[LU-6195] osd-zfs: osd_declare_object_destroy() calls dmu_tx_hold_zap() with wrong keys

• Type: Bug
• Resolution: Won't Fix
• Priority: Minor
• Affects Version/s: Lustre 2.7.0
• Severity: 3
• Assignee: Isaac Huang (Inactive)
• Reporter: Isaac Huang (Inactive)

Description

In osd_declare_object_destroy():

        /* declare that we'll remove object from fid-dnode mapping */
        zapid = osd_get_name_n_idx(env, osd, fid, buf);
        dmu_tx_hold_bonus(oh->ot_tx, zapid);
        dmu_tx_hold_zap(oh->ot_tx, zapid, 0, buf);

        /* declare that we'll remove object from inode accounting ZAPs */
        dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
        dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, 0, buf);
        dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
        dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, 0, buf);

The buf holds string representation of the FID, so it's wrong to use it in dmu_tx_hold_zap() calls for osd->od_iusr_oid and osd->od_igrp_oid where the keys should be obj->oo_attr.la_uid and obj->oo_attr.la_gid.

Same mistake in osd_declare_object_create().

Issue Links

• is related to: LU-6160 all test builds should enable SPL/ZFS debugging
• is related to: LU-5242 Test hang sanity test_132, test_133: umount ost

Comments

[Gerrit Updater]

Isaac Huang ([e-mail]) uploaded a new patch: http://review.whamcloud.com/13656
Subject: LU-6195 osd-zfs: dmu_tx_hold_zap() called with wrong keys
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: fd8fb80c6f71fedc476e82008fd309f9f4f8b418

[Isaac Huang (Inactive)]

Looks like just replace buf with NULL, and zap_count_write() will use worst-case estimate - should be good enough until we have in-zfs accounting.

[Isaac Huang (Inactive)]

Yes, but it's going be a while until we have in-zfs accounting.

[Alex Zhuravlev]

didn't we plan to replace this code with in-zfs accounting?