cluster: RHEL6 - rgmanager: Fix for deadlock

Ryan McCabe rmccabe at fedoraproject.org
Wed Oct 3 14:09:35 UTC 2012


Gitweb:        http://git.fedorahosted.org/git/?p=cluster.git;a=commitdiff;h=8d3265089fcb45cb6653c2c6947f78ab9e996be5
Commit:        8d3265089fcb45cb6653c2c6947f78ab9e996be5
Parent:        de3d3adfc52c0872e9576dc3119fc61d6fc414bb
Author:        Ryan McCabe <rmccabe at redhat.com>
AuthorDate:    Thu Sep 27 15:08:52 2012 -0400
Committer:     Ryan McCabe <rmccabe at redhat.com>
CommitterDate: Wed Oct 3 02:31:56 2012 -0400

rgmanager: Fix for deadlock

This patch fixes a deadlock in rgmanager that could occur when a node
starts rgmanager while a service is recovering.

Resolves: rhbz#861157

Acked-by: Lon Hohberger <lhh at redhat.com>
Signed-off-by: Ryan McCabe <rmccabe at redhat.com>
---
 rgmanager/src/daemons/rg_state.c   |    1 +
 rgmanager/src/daemons/rg_thread.c  |   19 ++++++++++++++++++-
 rgmanager/src/daemons/service_op.c |    1 +
 3 files changed, 20 insertions(+), 1 deletions(-)

diff --git a/rgmanager/src/daemons/rg_state.c b/rgmanager/src/daemons/rg_state.c
index 8c5af5b..80e8667 100644
--- a/rgmanager/src/daemons/rg_state.c
+++ b/rgmanager/src/daemons/rg_state.c
@@ -1963,6 +1963,7 @@ retry:
 			/* Deliberate */
 		case RG_EDEPEND:
 		case RG_EFAIL:
+		case RG_EDEADLCK:
 			/* Uh oh - we failed to relocate to this node.
 			   ensure that we tell the next node to start it from
 			   the 'recovering' state. */
diff --git a/rgmanager/src/daemons/rg_thread.c b/rgmanager/src/daemons/rg_thread.c
index 72b5f96..5e551c3 100644
--- a/rgmanager/src/daemons/rg_thread.c
+++ b/rgmanager/src/daemons/rg_thread.c
@@ -9,6 +9,8 @@
 #include <rg_queue.h>
 #include <assert.h>
 #include <members.h>
+#include <liblogthread.h>
+
 
 /**
  * Resource thread list entry.
@@ -735,13 +737,28 @@ rt_enqueue_request(const char *resgroupname, int request,
 			ret = 0;
 			break;
 		}
-		fprintf(stderr, "Failed to queue request: Would block\n");
 		/* EWOULDBLOCK */
 		pthread_mutex_unlock(resgroup->rt_queue_mutex);
 		pthread_mutex_unlock(&reslist_mutex);
+		logt_print(LOG_DEBUG,
+			"Failed to queue %d request for %s: Would block\n",
+			request, resgroupname);
 		return ret;
 	}
 
+	if (resgroup->rt_request == RG_START &&
+	    (request == RG_START_REMOTE || request == RG_START_RECOVER)) {
+		send_ret(response_ctx, resgroup->rt_name, RG_EDEADLCK,
+			request, 0);
+		msg_free_ctx(response_ctx);
+		pthread_mutex_unlock(resgroup->rt_queue_mutex);
+		pthread_mutex_unlock(&reslist_mutex);
+		logt_print(LOG_DEBUG,
+			"Failed to queue %d request for %s: Would block\n",
+			request, resgroupname);
+		return -1;
+	}
+
 	ret = rq_queue_request(resgroup->rt_queue, resgroup->rt_name,
 			       request, 0, 0, response_ctx, 0, target,
 			       arg0, arg1);
diff --git a/rgmanager/src/daemons/service_op.c b/rgmanager/src/daemons/service_op.c
index f094129..4b74427 100644
--- a/rgmanager/src/daemons/service_op.c
+++ b/rgmanager/src/daemons/service_op.c
@@ -62,6 +62,7 @@ service_op_start(char *svcName,
 			++dep;
 			continue;
 		case RG_EFAIL:
+		case RG_EDEADLCK:
 			++fail;
 			continue;
 		case RG_EABORT:


More information about the cluster-commits mailing list