[LU-17203] false assertion in cfs_hash_for_each_empty()

• Type: Bug
• Resolution: Unresolved
• Priority: Minor
• Severity: 3
• Assignee: WC Triage
• Reporter: Alex Zhuravlev

Description

this assertion in cfs_hash_for_each_empty():

LASSERT(atomic_read(&hs->hs_count) == 0);

is not correct - a lock can be in a process on enqueueing so can't be released right away.
here is an evidence:

00010000:00020000:1.0:1697224811.088493:0:31361:0:(ldlm_lockd.c:1479:ldlm_handle_enqueue()) ### lock on destroyed export 00000000fc58c50a ns: mdt-lustre-MDT0001_UUID lock: 0000000088760baf/0x8b83035ffb756f74 lrc: 3/0,0 mode: PR/PR res: [0x240000bd0:0x7f:0x0].0x0 bits 0x1b/0x0 rrc: 2 type: IBT gid 0 flags: 0x50200000000000 nid: 0@lo remote: 0x8b83035ffb756f58 expref: 6 pid: 31361 timeout: 0 lvb_type: 0
00000001:00000001:0.0:1697224811.091533:0:7065:0:(hash.c:1714:cfs_hash_for_each_nolock()) Process leaving (rc=0 : 0 : 0)
00010000:00000001:0.0:1697224811.091536:0:7065:0:(ldlm_lock.c:2473:ldlm_reprocess_recovery_done()) Process leaving
00010000:00020000:0.0:1697224811.091537:0:7065:0:(ldlm_lock.c:2748:ldlm_export_cancel_locks()) Export 00000000fc58c50a, canceled 51 locks, left on hash table 1.
...

I think we can just put the export on the stale list again via obd_stale_export_put()