[Ocfs2-commits] khackel commits r2798 - branches/ocfs2-1.2-cert/patches

[svn-commits@oss.oracle.com]

Author: khackel
Signed-off-by: jlbec
Date: 2006-03-29 03:40:24 -0600 (Wed, 29 Mar 2006)
New Revision: 2798

Added:
   branches/ocfs2-1.2-cert/patches/fix-remote-lock-during-reco.patch
Modified:
   branches/ocfs2-1.2-cert/patches/series
Log:
add fix-remote-lock-during-reco.patch to handle node death of owner of recovery lockres
Signed-off-by: jlbec

Added: branches/ocfs2-1.2-cert/patches/fix-remote-lock-during-reco.patch
===================================================================
--- branches/ocfs2-1.2-cert/patches/fix-remote-lock-during-reco.patch	2006-03-29 02:24:35 UTC (rev 2797)
+++ branches/ocfs2-1.2-cert/patches/fix-remote-lock-during-reco.patch	2006-03-29 09:40:24 UTC (rev 2798)
@@ -0,0 +1,69 @@
+Index: ocfs2-1.2-cert/fs/ocfs2/dlm/dlmlock.c
+===================================================================
+--- ocfs2-1.2-cert.orig/fs/ocfs2/dlm/dlmlock.c	2006-03-29 00:01:04.294719000 -0800
++++ ocfs2-1.2-cert/fs/ocfs2/dlm/dlmlock.c	2006-03-29 01:37:56.571286000 -0800
+@@ -226,7 +226,16 @@
+ 	res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
+ 	lock->lock_pending = 0;
+ 	if (status != DLM_NORMAL) {
+-		if (status != DLM_NOTQUEUED) {
++		if (status == DLM_RECOVERING &&
++		    dlm_is_recovery_lock(res->lockname.name,
++					 res->lockname.len)) {
++			/* recovery lock was mastered by dead node.
++			 * we need to have calc_usage shoot down this
++			 * lockres and completely remaster it. */
++			mlog(ML_NOTICE, "%s: recovery lock was owned by "
++			     "dead node %u, remaster it now.\n",
++			     dlm->name, res->owner);
++		} else if (status != DLM_NOTQUEUED) {
+ 			/*
+ 			 * DO NOT call calc_usage, as this would unhash
+ 			 * the remote lockres before we ever get to use
+@@ -691,18 +700,22 @@
+ 			msleep(100);
+ 			/* no waiting for dlm_reco_thread */
+ 			if (recovery) {
+-				if (status == DLM_RECOVERING) {
+-					mlog(0, "%s: got RECOVERING "
+-					     "for $REOCVERY lock, master "
+-					     "was %u\n", dlm->name, 
+-					     res->owner);
+-					dlm_wait_for_node_death(dlm, res->owner, 
+-							DLM_NODE_DEATH_WAIT_MAX);
+-				}
++				if (status != DLM_RECOVERING)
++					goto retry_lock;
++
++				mlog(0, "%s: got RECOVERING "
++				     "for $RECOVERY lock, master "
++				     "was %u\n", dlm->name,
++				     res->owner);
++				/* wait to see the node go down, then
++				 * drop down and allow the lockres to
++				 * get cleaned up.  need to remaster. */
++				dlm_wait_for_node_death(dlm, res->owner,
++						DLM_NODE_DEATH_WAIT_MAX);
+ 			} else {
+ 				dlm_wait_for_recovery(dlm);
++				goto retry_lock;
+ 			}
+-			goto retry_lock;
+ 		}
+ 
+ 		if (status != DLM_NORMAL) {
+Index: ocfs2-1.2-cert/fs/ocfs2/dlm/dlmrecovery.c
+===================================================================
+--- ocfs2-1.2-cert.orig/fs/ocfs2/dlm/dlmrecovery.c	2006-03-29 00:01:03.768807000 -0800
++++ ocfs2-1.2-cert/fs/ocfs2/dlm/dlmrecovery.c	2006-03-29 01:07:50.417523000 -0800
+@@ -2427,6 +2427,10 @@
+ 		mlog(ML_NOTICE, "%s: reco master %u is ready to recover %u\n",
+ 		     dlm->name, dlm->reco.new_master, dlm->reco.dead_node);
+ 		status = -EEXIST;
++	} else if (ret == DLM_RECOVERING) {
++		mlog(ML_NOTICE, "dlm=%s dlmlock says master node died (this=%u)\n",
++		     dlm->name, dlm->node_num);
++		goto again;
+ 	} else {
+ 		struct dlm_lock_resource *res;
+ 

Modified: branches/ocfs2-1.2-cert/patches/series
===================================================================
--- branches/ocfs2-1.2-cert/patches/series	2006-03-29 02:24:35 UTC (rev 2797)
+++ branches/ocfs2-1.2-cert/patches/series	2006-03-29 09:40:24 UTC (rev 2798)
@@ -19,3 +19,4 @@
 ocfs2_heartbeat-better_I_O_error_handling.patch -p0
 never-purge-master.patch 
 reassert-vs-migration.patch 
+fix-remote-lock-during-reco.patch