[LU-16478] faulty MDT connection can leak a reference to export

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.16.0
• Severity: 3
• Assignee: Alex Zhuravlev
• Reporter: Alex Zhuravlev

Description

if target_handle_connect() races with an export eviction then the following scenario can happen:

• mdt_obd_reconnect() -> .. nodemap_add_member() grabs a reference to the export
• then target_handle_connect() finds the export invalide and exits with -ENODEV:

  	if (export->exp_disconnected) {
  		spin_unlock(&export->exp_lock);
  		GOTO(out, rc = -ENODEV);
  

  after umount won't be able to complete with the following symptoms:

  00000020:02000400:1.0:1673726953.882508:0:8583:0:(genops.c:1792:obd_exports_barrier()) lustre-MDT0000 is waiting for obd_unlinked_exports more than 7 seconds. The obd refcount = 4. Is it stuck?
  00000020:02000400:1.0:1673726953.889142:0:8583:0:(genops.c:1758:print_export_data()) lustre-MDT0000: UNLINKED 000000002760c5c2 5a4bd497-6ace-43e0-8e46-c6b0e7dc84ba 0@lo 1 (0 0 0) 1 0 1 0: 00000000e9920e55  4294967301 stale:0
  

Issue Links

• is related to: LU-11388 replay-single test_131b: test timeout