[LU-11896] Memory leak occurs when lfs mkdir fails

• Type: Bug
• Resolution: Duplicate
• Priority: Minor
• Affects Version/s: Lustre 2.10.5
• Severity: 3
• Assignee: Tatsushi Takamura
• Reporter: Tatsushi Takamura

Description

When lfs mkdir fails, llog_cat_cancel_records() is executed for error handling. But, in a condition of !llog_exist(loghandle), llog_handle_put() is not executed (that is, lgh_refcount is not decremented), so llog_free_handle() will not be executed in umount. This cause memory leak.

 

 682 int llog_cat_cancel_records(const struct lu_env *env,
 683 struct llog_handle *cathandle, int count,
 684 struct llog_cookie *cookies)
 685 {
 686 int i, index, rc = 0, failed = 0;
 687
 688 ENTRY;
 689
 690 for (i = 0; i < count; i++, cookies++) {
 691 struct llog_handle *loghandle;
 692 struct llog_logid *lgl = &cookies->lgc_lgl;
 693 int lrc;
 694
 695 rc = llog_cat_id2handle(env, cathandle, &loghandle, lgl);
 // llog_cat_id2handle(llog_handle_get) increments lgh_refcount
<snip>
 705 if ((cathandle->lgh_ctxt->loc_flags &
 706 LLOG_CTXT_FLAG_NORMAL_FID) && !llog_exist(loghandle)) {
 707 /* For update log, some of loghandles of cathandle
 708 * might not exist because remote llog creation might
 709 * be failed, so let's skip the record cancellation
 710 * for these non-exist llogs.
 711 */
 712 lrc = -ENOENT;
 713 CDEBUG(D_HA, "%s: llog "DFID":%x does not exist"
 714 ": rc = %d\n",
 715 cathandle->lgh_ctxt->loc_obd->obd_name,
 716 PFID(&lgl->lgl_oi.oi_fid), lgl->lgl_ogen, lrc);
 717 failed++;
 718 if (rc == 0)
 719 rc = lrc;
 // llog_handle_put() should be called to decrement lgh_refcount
 720 continue;
 721 }

 

Issue Links

• is related to: LU-11924 Combine cancellation of llog records at osp_sync_process_committed