[Cluster-devel] cluster/cman-kernel/src cnxman.c membership.c

[Cluster-devel] cluster/cman-kernel/src cnxman.c membership.c

[pcaulfield]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4
Changes by:	pcaulfield at sourceware.org	2007-11-20 11:02:46

Modified files:
	cman-kernel/src: cnxman.c membership.c 

Log message:
	A fix to the last patch. The last_ackneeded_seq_recv variable needed to
	be cleared when the node went down, otherwise we end up comparing
	received sequence numbers against old ones and end up throwing all
	new messages away!
	bz#387081 shows this happening. The "inconsistent" message is slightly
	misleading in this context

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cman-kernel/src/cnxman.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.42.2.30&r2=1.42.2.31
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cman-kernel/src/membership.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.44.2.28&r2=1.44.2.29

--- cluster/cman-kernel/src/Attic/cnxman.c	2007/11/01 10:44:25	1.42.2.30
+++ cluster/cman-kernel/src/Attic/cnxman.c	2007/11/20 11:02:45	1.42.2.31
@@ -867,10 +867,10 @@
         /* Have we received this message before ? If so just ignore it, it's a
 	 * resend for someone else's benefit */
 	if (!(flags & MSG_NOACK) &&
-	    rem_node && rem_node->last_seq_recv &&
+	    rem_node && rem_node->last_ackneeded_seq_recv &&
 	    (short)((short)le16_to_cpu(header->seq) - (short)rem_node->last_ackneeded_seq_recv) <= 0) {
-		P_COMMS("Discarding message -  seq = %d, last_seen = %d\n",
-			header->seq, rem_node->last_seq_recv);
+		P_COMMS("Discarding message -  seq = %d, last_seen = %d, last_acked_seen = %d\n",
+			header->seq, rem_node->last_seq_recv, rem_node->last_ackneeded_seq_recv);
 		/* Still need to ACK it though, in case it was the ACK that got
 		 * lost */
 		cl_sendack(csock, header->seq, addrlen, addr, header->tgtport, 0);
--- cluster/cman-kernel/src/Attic/membership.c	2007/09/19 15:01:07	1.44.2.28
+++ cluster/cman-kernel/src/Attic/membership.c	2007/11/20 11:02:46	1.44.2.29
@@ -1398,6 +1398,7 @@
 		cluster_members--;
 	node->state = NODESTATE_DEAD;
 	node->last_seq_recv = 0;
+	node->last_ackneeded_seq_recv = 0;
 	up(&cluster_members_lock);
 
 	send_nodedown(node->node_id, node->leave_reason);
@@ -1683,6 +1684,7 @@
 		newnode->us = 0;
 		newnode->leave_reason = 0;
 		newnode->last_seq_recv = 0;
+		newnode->last_ackneeded_seq_recv = 0;
 		newnode->last_seq_acked = 0;
 		newnode->last_seq_sent = 0;
 		newnode->incarnation++;
@@ -1717,6 +1719,7 @@
 	newnode->us = 0;
 	newnode->leave_reason = 0;
 	newnode->last_seq_recv = 0;
+	newnode->last_ackneeded_seq_recv = 0;
 	newnode->last_seq_acked = 0;
 	newnode->last_seq_sent = 0;
 	newnode->incarnation = 0;

[Cluster-devel] cluster/cman-kernel/src cnxman.c membership.c

[pcaulfield]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL45
Changes by:	pcaulfield at sourceware.org	2008-02-05 16:21:14

Modified files:
	cman-kernel/src: cnxman.c membership.c 

Log message:
	Fix for cman throwing away messages that arrive out of order.
	bz#430284

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cman-kernel/src/cnxman.c.diff?cvsroot=cluster&only_with_tag=RHEL45&r1=1.42.2.28.2.1&r2=1.42.2.28.2.2
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cman-kernel/src/membership.c.diff?cvsroot=cluster&only_with_tag=RHEL45&r1=1.44.2.26&r2=1.44.2.26.2.1

--- cluster/cman-kernel/src/Attic/cnxman.c	2007/04/02 08:03:13	1.42.2.28.2.1
+++ cluster/cman-kernel/src/Attic/cnxman.c	2008/02/05 16:21:14	1.42.2.28.2.2
@@ -867,8 +867,8 @@
         /* Have we received this message before ? If so just ignore it, it's a
 	 * resend for someone else's benefit */
 	if (!(flags & MSG_NOACK) &&
-	    rem_node && rem_node->last_seq_recv &&
-	    (short)((short)le16_to_cpu(header->seq) - (short)rem_node->last_seq_recv) <= 0) {
+	    rem_node && rem_node->last_ackneeded_seq_recv &&
+	    (short)((short)le16_to_cpu(header->seq) - (short)rem_node->last_ackneeded_seq_recv) <= 0) {
 		P_COMMS("Discarding message -  seq = %d, last_seen = %d\n",
 			header->seq, rem_node->last_seq_recv);
 		/* Still need to ACK it though, in case it was the ACK that got
--- cluster/cman-kernel/src/Attic/membership.c	2007/01/19 16:20:14	1.44.2.26
+++ cluster/cman-kernel/src/Attic/membership.c	2008/02/05 16:21:14	1.44.2.26.2.1
@@ -1395,6 +1395,7 @@
 		cluster_members--;
 	node->state = NODESTATE_DEAD;
 	node->last_seq_recv = 0;
+	node->last_ackneeded_seq_recv = 0;
 	up(&cluster_members_lock);
 
 	send_nodedown(node->node_id, node->leave_reason);
@@ -1680,6 +1681,7 @@
 		newnode->us = 0;
 		newnode->leave_reason = 0;
 		newnode->last_seq_recv = 0;
+		newnode->last_ackneeded_seq_recv = 0;
 		newnode->last_seq_acked = 0;
 		newnode->last_seq_sent = 0;
 		newnode->incarnation++;
@@ -1714,6 +1716,7 @@
 	newnode->us = 0;
 	newnode->leave_reason = 0;
 	newnode->last_seq_recv = 0;
+	newnode->last_ackneeded_seq_recv = 0;
 	newnode->last_seq_acked = 0;
 	newnode->last_seq_sent = 0;
 	newnode->incarnation = 0;