[Cluster-devel] cluster/cmirror-kernel/src dm-cmirror-client.c

[Cluster-devel] cluster/cmirror-kernel/src dm-cmirror-client.c

[jbrassow]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4
Changes by:	jbrassow at sourceware.org	2006-11-27 23:15:45

Modified files:
	cmirror-kernel/src: dm-cmirror-client.c 

Log message:
	Bug 217438: scrolling kernel requests to mark mirror regions
	
	It is sometimes possible for a mark request to be issued on a region
	that is being recovered.  The mark must be delayed until the recovery
	is complete.  This is what was happening, but a rather severe error
	was being reported and the (mark region) retries were happening too
	fast.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cmirror-kernel/src/dm-cmirror-client.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.1.2.28&r2=1.1.2.29

--- cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/11/27 22:36:48	1.1.2.28
+++ cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/11/27 23:15:43	1.1.2.29
@@ -941,9 +941,19 @@
 	spin_unlock(&region_state_lock);
 
 	while((error = consult_server(lc, region, LRT_MARK_REGION, NULL))){
+		if (error == -EBUSY) {
+			/* Remote recovering delay and try again */
+			DMDEBUG("Delaying mark to region %Lu, due to recovery",
+				region);
+			set_current_state(TASK_INTERRUPTIBLE);
+			schedule_timeout(HZ/2);
+			continue;
+		}
+
 		DMWARN("unable to get server (%u) to mark region (%Lu)",
 		       lc->server_id, region);
 		DMWARN("Reason :: %d", error);
+
 		if (error == -EIO) {
 			lc->log_dev_failed = 1;
 			break;

[Cluster-devel] cluster/cmirror-kernel/src dm-cmirror-client.c

[jbrassow]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4
Changes by:	jbrassow at sourceware.org	2007-08-23 16:51:39

Modified files:
	cmirror-kernel/src: dm-cmirror-client.c 

Log message:
	BUG 239856 Processed: failed server election due to suspended mirror...
	
	When two logs are loaded that represent the same entity, the second
	log should start out suspended.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cmirror-kernel/src/dm-cmirror-client.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.1.2.49&r2=1.1.2.50

--- cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2007/07/11 16:18:03	1.1.2.49
+++ cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2007/08/23 16:51:39	1.1.2.50
@@ -586,6 +586,11 @@
 		if(!strncmp(tmp_lc->uuid, lc->uuid, MAX_NAME_LEN)){
 			lc->uuid_ref = (lc->uuid_ref > tmp_lc->uuid_ref) ?
 				lc->uuid_ref : tmp_lc->uuid_ref + 1;
+			/*
+			 * A second instance of the same log
+			 * should start out suspended.
+			 */
+			atomic_set(&lc->suspended, 1);
 		}
 	}
 
@@ -601,7 +606,6 @@
 	INIT_LIST_HEAD(&lc->mark_logged);
 	spin_lock_init(&lc->state_lock);
 
-	atomic_set(&lc->suspended, 1);
 	lc->server_valid = 0;
 	lc->server_id = 0xDEAD;
 

[Cluster-devel] cluster/cmirror-kernel/src dm-cmirror-client.c

[jbrassow]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4
Changes by:	jbrassow at sourceware.org	2007-05-09 21:44:34

Modified files:
	cmirror-kernel/src: dm-cmirror-client.c 

Log message:
	Bug 235040 is fixed in the 4.6 kernel, no need to work around it now.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cmirror-kernel/src/dm-cmirror-client.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.1.2.47&r2=1.1.2.48

--- cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2007/04/26 16:54:49	1.1.2.47
+++ cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2007/05/09 21:44:34	1.1.2.48
@@ -756,13 +756,8 @@
 		DMERR("Unable to disconnect from cluster infrastructure.\n");
 }
 
-static int cluster_flush(struct dirty_log *log);
 static int cluster_presuspend(struct dirty_log *log)
 {
-	/* FIXME: flush is work-around for bug 235040 */
-	DMDEBUG("Performing flush to work around bug 235040");
-	cluster_flush(log);
-	DMDEBUG("Log flush complete");
 	return 0;
 }
 

[Cluster-devel] cluster/cmirror-kernel/src dm-cmirror-client.c

[jbrassow]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4
Changes by:	jbrassow at sourceware.org	2007-03-02 22:31:14

Modified files:
	cmirror-kernel/src: dm-cmirror-client.c 

Log message:
	- When a cluster log resumes, the server must re-read the log device.
	However, it doesn't _need_ to happen until a request comes in which
	forces an election.
	
	If a resume/suspend loop is tight enough, it is possible for a
	log read to happen after the log device has been suspended.
	
	The fix is to force the log read during the resume call.
	
	Bug 227398: cmirror request to LRT_CLEAR_REGION fails...

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cmirror-kernel/src/dm-cmirror-client.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.1.2.39&r2=1.1.2.40

--- cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2007/02/26 17:38:06	1.1.2.39
+++ cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2007/03/02 22:31:14	1.1.2.40
@@ -305,7 +305,8 @@
 		lc->server_id = lr.u.lr_coordinator;
 	} else {
 		/* ATTENTION -- what do we do with this ? */
-		DMWARN("Failed to receive election results from server: %d", len);
+		DMWARN("Failed to receive election results from server: (%s,%d)",
+		       lc->uuid + (strlen(lc->uuid) - 8), len);
 		error = len;
 	}
 
@@ -464,6 +465,8 @@
 			       (type == LRT_ELECTION)? "LRT_ELECTION":
 			       (type == LRT_SELECTION)? "LRT_SELECTION": "UNKNOWN");
 			DMDEBUG(" - error   :: %d", error);
+			DMINFO("Too many retries, attempting to re-establish server connection.");
+			lc->server_id = 0xDEAD;
 		}
 	}
 
@@ -764,11 +767,6 @@
 	list_del_init(&lc->log_list);
 	spin_unlock(&log_list_lock);
 
-	if ((lc->server_id == my_id) && !atomic_read(&lc->suspended))
-		consult_server(lc, 0, LRT_MASTER_LEAVING, NULL);
-
-	sock_release(lc->client_sock);
-
 	spin_lock(&region_state_lock);
 
 	list_for_each_entry_safe(rs, tmp_rs, &clear_region_list, rs_list) {
@@ -783,6 +781,10 @@
 
 	spin_unlock(&region_state_lock);
 
+	if ((lc->server_id == my_id) && !atomic_read(&lc->suspended))
+		consult_server(lc, 0, LRT_MASTER_LEAVING, NULL);
+
+	sock_release(lc->client_sock);
 
 	if (lc->log_dev)
 		disk_dtr(log);
@@ -842,6 +844,7 @@
 	lc->sync_search = 0;
 	resume_server_requests();
 	atomic_set(&lc->suspended, 0);
+	consult_server(lc, 0, LRT_IN_SYNC, NULL);
 
 	return 0;
 }

[Cluster-devel] cluster/cmirror-kernel/src dm-cmirror-client.c

[jbrassow]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4
Changes by:	jbrassow at sourceware.org	2006-12-05 17:49:08

Modified files:
	cmirror-kernel/src: dm-cmirror-client.c 

Log message:
	Just a comment update, so I don't forget why I'm checking for
	duplicate mark region requests.
	
	<       /* ATTENTION -- this check should not be necessary.   **
	<       ** Why are regions being marked again before a clear? */
	---
	>       /*
	>        * In the mirroring code, it is possible for a write
	>        * to complete and call rh_dec - putting the region on
	>        * the clear_region list.  However, before the actual
	>        * clear request is issued to the log (rh_update_states)
	>        * another mark happens.  So, we check for and remove
	>        * duplicates.
	>        */

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cmirror-kernel/src/dm-cmirror-client.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.1.2.31&r2=1.1.2.32

--- cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/12/05 15:08:17	1.1.2.31
+++ cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/12/05 17:49:08	1.1.2.32
@@ -918,8 +918,14 @@
 			return;
 		}
 	}
-	/* ATTENTION -- this check should not be necessary.   **
-	** Why are regions being marked again before a clear? */
+	/*
+	 * In the mirroring code, it is possible for a write
+	 * to complete and call rh_dec - putting the region on
+	 * the clear_region list.  However, before the actual
+	 * clear request is issued to the log (rh_update_states)
+	 * another mark happens.  So, we check for and remove
+	 * duplicates.
+	 */
 	list_for_each_entry(rs, &marked_region_list, rs_list){
 		if(lc == rs->rs_lc && region == rs->rs_region){
 #ifdef DEBUG

[Cluster-devel] cluster/cmirror-kernel/src dm-cmirror-client.c

[jbrassow]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4
Changes by:	jbrassow at sourceware.org	2006-12-05 15:08:18

Modified files:
	cmirror-kernel/src: dm-cmirror-client.c 

Log message:
	Bug 214487: "Attempt to mark a already marked region" messages when ...
	
	Similar to the last checkin.  When a log server moved, all entries in
	the clear region list were removed rather than just those that where
	associated with the log that moved.  Again, this would cause the server
	to report that the regions where already marked when the next mark
	request would come in.
	
	I believe that this action was harmless (but annoying).

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cmirror-kernel/src/dm-cmirror-client.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.1.2.30&r2=1.1.2.31

--- cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/12/05 03:13:50	1.1.2.30
+++ cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/12/05 15:08:17	1.1.2.31
@@ -503,11 +503,14 @@
 			DMINFO(" - Wiping clear region list");
 			list_for_each_entry_safe(rs, tmp_rs,
 						 &clear_region_list, rs_list){
+				/* Remove only those associated with referenced log */
+				if (rs->rs_lc != lc)
+					continue;
 				i++;
 				list_del_init(&rs->rs_list);
 				mempool_free(rs, region_state_pool);
 			}
-			clear_region_count=0;
+			clear_region_count -= i;
 			DMINFO(" - %d clear region requests wiped", i);
 
 			DMINFO(" - Resending all mark region requests");

[Cluster-devel] cluster/cmirror-kernel/src dm-cmirror-client.c

[jbrassow]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4
Changes by:	jbrassow at sourceware.org	2006-12-05 03:13:51

Modified files:
	cmirror-kernel/src: dm-cmirror-client.c 

Log message:
	Bug 214487: "Attempt to mark a already marked region" messages when ...
	
	When a log server moves (like durring a failure or the mirror being
	shutdown on that node), the clients send the new server the regions
	that they have marked.
	
	However, the clients where sending _all_ of the regions they had
	marked for every mirror that was active - even for the ones whose
	log server had not moved.  This was causing those servers to report
	that they already had those regions marked.
	
	Other than the annoying messages, I don't believe there were any
	side affects of this bug.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cmirror-kernel/src/dm-cmirror-client.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.1.2.29&r2=1.1.2.30

--- cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/11/27 23:15:43	1.1.2.29
+++ cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/12/05 03:13:50	1.1.2.30
@@ -512,6 +512,9 @@
 
 			DMINFO(" - Resending all mark region requests");
 			list_for_each_entry(rs, &marked_region_list, rs_list){
+				/* Resend only those associated with referenced log */
+				if (rs->rs_lc != lc)
+					continue;
 				do {
 					retry = 0;
 					DMINFO("   - " SECTOR_FORMAT, rs->rs_region);

[Cluster-devel] cluster/cmirror-kernel/src dm-cmirror-client.c

[jbrassow]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4
Changes by:	jbrassow at sourceware.org	2006-11-27 22:36:48

Modified files:
	cmirror-kernel/src: dm-cmirror-client.c 

Log message:
	Bug 214517: hung cmirror operations due to looping mirror region requests
	
	Once we've completed handling a new server, we must reset the 'new_server'
	variable, so that if we need to retry a request we don't go through the
	whole process again.
	
	Also, added more error reporting for when a request must be retried.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cmirror-kernel/src/dm-cmirror-client.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.1.2.27&r2=1.1.2.28

--- cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/11/07 20:48:16	1.1.2.27
+++ cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/11/27 22:36:48	1.1.2.28
@@ -406,6 +406,7 @@
 
 	if(len <= 0){
 		/* ATTENTION -- what do we do with this ? */
+		DMWARN("Error while listening for server response: %d", len);
 		error = len;
 		*retry = 1;
 		goto fail;
@@ -420,6 +421,8 @@
 	}
 
 	if (lr->u.lr_int_rtn == -ENXIO) {
+		DMDEBUG("Server (%u) says it no longer controls this log (%s)",
+			lc->server_id, lc->uuid + (strlen(lc->uuid) - 8));
 		lc->server_id = 0xDEAD;
 		*retry = 1;
 		goto fail;
@@ -540,6 +543,7 @@
 				      (type == LRT_SELECTION)? "LRT_SELECTION": "UNKNOWN"
 					);
 			}
+			new_server = 0;
 		}
 
 		rs = NULL;

[Cluster-devel] cluster/cmirror-kernel/src dm-cmirror-client.c

[jbrassow]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4
Changes by:	jbrassow at sourceware.org	2006-11-07 20:48:16

Modified files:
	cmirror-kernel/src: dm-cmirror-client.c 

Log message:
	- Rather than just printing out "UNKNOWN", we print out the value
	and then "unknown".  This way, we can look up the value and see
	if it is valid.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cmirror-kernel/src/dm-cmirror-client.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.1.2.26&r2=1.1.2.27

--- cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/10/24 21:04:31	1.1.2.26
+++ cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/11/07 20:48:16	1.1.2.27
@@ -527,7 +527,7 @@
 				spin_unlock(&region_state_lock);
 				goto out;
 			} else {
-				DMINFO("Continuing request:: %s",
+				DMINFO("Continuing request type, %d (%s)", type,
 				      (type == LRT_IS_CLEAN)? "LRT_IS_CLEAN":
 				      (type == LRT_IN_SYNC)? "LRT_IN_SYNC":
 				      (type == LRT_MARK_REGION)? "LRT_MARK_REGION":

[Cluster-devel] cluster/cmirror-kernel/src dm-cmirror-client.c

[jbrassow]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4
Changes by:	jbrassow at sourceware.org	2006-10-24 21:04:31

Modified files:
	cmirror-kernel/src: dm-cmirror-client.c 

Log message:
	- Bug 194131 - cluster mirror copy status can give false (0.00) percent
	
	An error during communication can lead to a false reporting of 0%
	in-sync regions.  This can't be completely eliminated until rhel5,
	but we can at least retry a couple times in rhel4.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cmirror-kernel/src/dm-cmirror-client.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.1.2.25&r2=1.1.2.26

--- cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/09/08 15:53:56	1.1.2.25
+++ cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/10/24 21:04:31	1.1.2.26
@@ -1040,6 +1040,7 @@
 
 static region_t cluster_get_sync_count(struct dirty_log *log)
 {
+	int i;
 	region_t rtn;
 	struct log_c *lc = (struct log_c *) log->context;
 /* take out optimization
@@ -1047,7 +1048,9 @@
 		return lc->region_count;
 	}
 */
-	if(consult_server(lc, 0, LRT_GET_SYNC_COUNT, &rtn)){
+	/* Try to get sync count up to five times */
+	for (i = 0; i < 5 && consult_server(lc, 0, LRT_GET_SYNC_COUNT, &rtn); i++);
+	if(i >= 5){
 		return 0;
 	}
 

[Cluster-devel] cluster/cmirror-kernel/src dm-cmirror-client.c

[jbrassow]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4
Changes by:	jbrassow at sourceware.org	2006-09-08 15:53:57

Modified files:
	cmirror-kernel/src: dm-cmirror-client.c 

Log message:
	- remove speed-bump that was put in for debugging
	- delete 'marked region requests' when destructor is called
	The regions have been marked on disk, but the memory struct
	tracking that mark needs to be freed if a dtr comes along
	before a 'clear region request'.  This can happen when a
	mirror image fails.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cmirror-kernel/src/dm-cmirror-client.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.1.2.24&r2=1.1.2.25

--- cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/09/05 17:50:11	1.1.2.24
+++ cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/09/08 15:53:56	1.1.2.25
@@ -528,7 +528,7 @@
 				goto out;
 			} else {
 				DMINFO("Continuing request:: %s",
-				      (type == LRT_IS_CLEAN)? "LRT_IS_C	LEAN":
+				      (type == LRT_IS_CLEAN)? "LRT_IS_CLEAN":
 				      (type == LRT_IN_SYNC)? "LRT_IN_SYNC":
 				      (type == LRT_MARK_REGION)? "LRT_MARK_REGION":
 				      (type == LRT_GET_RESYNC_WORK)? "LRT_GET_RESYNC_WORK":
@@ -732,6 +732,7 @@
 static void cluster_dtr(struct dirty_log *log)
 {
 	struct log_c *lc = (struct log_c *) log->context;
+	struct region_state *rs, *tmp_rs;
 
 	if (!list_empty(&clear_region_list))
 		DMINFO("Leaving while clear region requests remain.");
@@ -742,6 +743,20 @@
 
 	sock_release(lc->client_sock);
 
+	spin_lock(&region_state_lock);
+
+	list_for_each_entry_safe(rs, tmp_rs, &clear_region_list, rs_list) {
+		if (lc == rs->rs_lc)
+			list_del_init(&rs->rs_list);
+	}
+
+	list_for_each_entry_safe(rs, tmp_rs, &marked_region_list, rs_list) {
+		if (lc == rs->rs_lc)
+			list_del_init(&rs->rs_list);
+	}
+
+	spin_unlock(&region_state_lock);
+
 	if (lc->log_dev)
 		disk_dtr(log);
 	else
@@ -758,7 +773,6 @@
 
 static int cluster_postsuspend(struct dirty_log *log)
 {
-	int r;
 	struct log_c *lc = (struct log_c *) log->context;
 
 	while (1) {
@@ -876,6 +890,8 @@
 
 	rs_new = mempool_alloc(region_state_pool, GFP_KERNEL);
 
+	memset(rs_new, 0, sizeof(struct region_state));
+
 	spin_lock(&region_state_lock);
 	list_for_each_entry_safe(rs, tmp_rs, &clear_region_list, rs_list){
 		if(lc == rs->rs_lc && region == rs->rs_region){
@@ -950,6 +966,8 @@
 
 	rs_new = mempool_alloc(region_state_pool, GFP_ATOMIC);
 
+	memset(rs_new, 0, sizeof(struct region_state));
+
 	spin_lock(&region_state_lock);
 
 	list_for_each_entry_safe(rs, tmp_rs, &clear_region_list, rs_list){
@@ -968,9 +986,6 @@
 			list_del_init(&rs->rs_list);
 			list_add(&rs->rs_list, &clear_region_list);
 			clear_region_count++;
-			if(!(clear_region_count & 0x7F)){
-				DMINFO("clear_region_count :: %d", clear_region_count);
-			}
 			spin_unlock(&region_state_lock);
 			if (rs_new)
 				mempool_free(rs_new, region_state_pool);
@@ -1017,11 +1032,8 @@
 	while(consult_server(lc, region, LRT_COMPLETE_RESYNC_WORK, &success_tmp)){
 		DMWARN("unable to notify server of completed resync work");
 	}
-	if (!success) {
+	if (!success)
 		DMERR("Attempting to revert sync status of region #%llu", region);
-		set_current_state(TASK_INTERRUPTIBLE);
-		schedule_timeout(HZ/5);
-	}
 
 	return;
 }

[Cluster-devel] cluster/cmirror-kernel/src dm-cmirror-client.c

[jbrassow]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4U4
Changes by:	jbrassow at sourceware.org	2006-09-08 15:52:41

Modified files:
	cmirror-kernel/src: dm-cmirror-client.c 

Log message:
	- remove speed-bump that was put in for debugging
	- delete 'marked region requests' when destructor is called
	The regions have been marked on disk, but the memory struct
	tracking that mark needs to be freed if a dtr comes along
	before a 'clear region request'.  This can happen when a
	mirror image fails.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cmirror-kernel/src/dm-cmirror-client.c.diff?cvsroot=cluster&only_with_tag=RHEL4U4&r1=1.1.2.19.2.5&r2=1.1.2.19.2.6

--- cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/09/05 17:48:02	1.1.2.19.2.5
+++ cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/09/08 15:52:40	1.1.2.19.2.6
@@ -528,7 +528,7 @@
 				goto out;
 			} else {
 				DMINFO("Continuing request:: %s",
-				      (type == LRT_IS_CLEAN)? "LRT_IS_C	LEAN":
+				      (type == LRT_IS_CLEAN)? "LRT_IS_CLEAN":
 				      (type == LRT_IN_SYNC)? "LRT_IN_SYNC":
 				      (type == LRT_MARK_REGION)? "LRT_MARK_REGION":
 				      (type == LRT_GET_RESYNC_WORK)? "LRT_GET_RESYNC_WORK":
@@ -732,6 +732,7 @@
 static void cluster_dtr(struct dirty_log *log)
 {
 	struct log_c *lc = (struct log_c *) log->context;
+	struct region_state *rs, *tmp_rs;
 
 	if (!list_empty(&clear_region_list))
 		DMINFO("Leaving while clear region requests remain.");
@@ -742,6 +743,20 @@
 
 	sock_release(lc->client_sock);
 
+	spin_lock(&region_state_lock);
+
+	list_for_each_entry_safe(rs, tmp_rs, &clear_region_list, rs_list) {
+		if (lc == rs->rs_lc)
+			list_del_init(&rs->rs_list);
+	}
+
+	list_for_each_entry_safe(rs, tmp_rs, &marked_region_list, rs_list) {
+		if (lc == rs->rs_lc)
+			list_del_init(&rs->rs_list);
+	}
+
+	spin_unlock(&region_state_lock);
+
 	if (lc->log_dev)
 		disk_dtr(log);
 	else
@@ -758,7 +773,6 @@
 
 static int cluster_postsuspend(struct dirty_log *log)
 {
-	int r;
 	struct log_c *lc = (struct log_c *) log->context;
 
 	while (1) {
@@ -876,6 +890,8 @@
 
 	rs_new = mempool_alloc(region_state_pool, GFP_KERNEL);
 
+	memset(rs_new, 0, sizeof(struct region_state));
+
 	spin_lock(&region_state_lock);
 	list_for_each_entry_safe(rs, tmp_rs, &clear_region_list, rs_list){
 		if(lc == rs->rs_lc && region == rs->rs_region){
@@ -950,6 +966,8 @@
 
 	rs_new = mempool_alloc(region_state_pool, GFP_ATOMIC);
 
+	memset(rs_new, 0, sizeof(struct region_state));
+
 	spin_lock(&region_state_lock);
 
 	list_for_each_entry_safe(rs, tmp_rs, &clear_region_list, rs_list){
@@ -968,9 +986,6 @@
 			list_del_init(&rs->rs_list);
 			list_add(&rs->rs_list, &clear_region_list);
 			clear_region_count++;
-			if(!(clear_region_count & 0x7F)){
-				DMINFO("clear_region_count :: %d", clear_region_count);
-			}
 			spin_unlock(&region_state_lock);
 			if (rs_new)
 				mempool_free(rs_new, region_state_pool);
@@ -1017,11 +1032,8 @@
 	while(consult_server(lc, region, LRT_COMPLETE_RESYNC_WORK, &success_tmp)){
 		DMWARN("unable to notify server of completed resync work");
 	}
-	if (!success) {
+	if (!success)
 		DMERR("Attempting to revert sync status of region #%llu", region);
-		set_current_state(TASK_INTERRUPTIBLE);
-		schedule_timeout(HZ/5);
-	}
 
 	return;
 }

[Cluster-devel] cluster/cmirror-kernel/src dm-cmirror-client.c

[jbrassow]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4U4
Changes by:	jbrassow at sourceware.org	2006-07-07 17:12:22

Modified files:
	cmirror-kernel/src: dm-cmirror-client.c 

Log message:
	- If a cluster mirror was removed, and a new one created fast enough, the
	processes could overlap such that the creation would fail to register
	with cman because the remove hadn't unregistered yet (-EEXIST)
	
	Fix for bug 197952 - cluster mirror creation fails due to an inablity to connect to cluster infrastructure

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cmirror-kernel/src/dm-cmirror-client.c.diff?cvsroot=cluster&only_with_tag=RHEL4U4&r1=1.1.2.19.2.1&r2=1.1.2.19.2.2

--- cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/06/29 19:46:37	1.1.2.19.2.1
+++ cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/07/07 17:12:22	1.1.2.19.2.2
@@ -1206,26 +1206,31 @@
 };
 
 static int mirror_set_count = 0; /* used to prevent multiple cluster [dis]connects */
+static DECLARE_MUTEX(cmirror_register_lock);
 
 static int cluster_connect(void)
 {
-	int r;
+	int r = 0;
 
-	if (mirror_set_count++)
-		return 0;
+	down(&cmirror_register_lock);
+
+	if (mirror_set_count++) {
+		up(&cmirror_register_lock);
+		goto out;
+	}
 
 	r = kcl_register_service("clustered_log", 13, SERVICE_LEVEL_GDLM, &clog_ops,
 				 1, NULL, &local_id);
 	if (r) {
-		DMWARN("Couldn't register clustered_log service");
-		return r;
+		DMWARN("Couldn't register clustered_log service.  Reason: %d", r);
+		goto out;
 	}
 
 	r = start_server();
 	if(r){
 		DMWARN("Unable to start clustered log server daemon");
 		kcl_unregister_service(local_id);
-		return r;
+		goto out;
 	}
 
 	r = kcl_join_service(local_id);
@@ -1236,13 +1241,19 @@
 		kcl_unregister_service(local_id);
 	}
 
+out:
+	up(&cmirror_register_lock);
 	return r;
 }
 
 static int cluster_disconnect(void)
 {
-	if (--mirror_set_count)
+	down(&cmirror_register_lock);
+
+	if (--mirror_set_count) {
+		up(&cmirror_register_lock);
 		return 0;
+	}
 
 	/* By setting 'shutting_down', the server will not be suspended **
 	** when a stop is received */
@@ -1251,6 +1262,7 @@
 	stop_server();
 	kcl_unregister_service(local_id);
 
+	up(&cmirror_register_lock);
 	return 0;
 }
 

[Cluster-devel] cluster/cmirror-kernel/src dm-cmirror-client.c

[jbrassow]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	STABLE
Changes by:	jbrassow at sourceware.org	2006-07-07 17:09:54

Modified files:
	cmirror-kernel/src: dm-cmirror-client.c 

Log message:
	- If a cluster mirror was removed, and a new one created fast enough, the
	processes could overlap such that the creation would fail to register
	with cman because the remove hadn't unregistered yet (-EEXIST)

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cmirror-kernel/src/dm-cmirror-client.c.diff?cvsroot=cluster&only_with_tag=STABLE&r1=1.1.4.3&r2=1.1.4.4

--- cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/06/29 19:49:32	1.1.4.3
+++ cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/07/07 17:09:54	1.1.4.4
@@ -1222,26 +1222,31 @@
 };
 
 static int mirror_set_count = 0; /* used to prevent multiple cluster [dis]connects */
+static DECLARE_MUTEX(cmirror_register_lock);
 
 static int cluster_connect(void)
 {
-	int r;
+	int r = 0;
 
-	if (mirror_set_count++)
-		return 0;
+	down(&cmirror_register_lock);
+
+	if (mirror_set_count++) {
+		up(&cmirror_register_lock);
+		goto out;
+	}
 
 	r = kcl_register_service("clustered_log", 13, SERVICE_LEVEL_GDLM, &clog_ops,
 				 1, NULL, &local_id);
 	if (r) {
-		DMWARN("Couldn't register clustered_log service");
-		return r;
+		DMWARN("Couldn't register clustered_log service.  Reason: %d", r);
+		goto out;
 	}
 
 	r = start_server();
 	if(r){
 		DMWARN("Unable to start clustered log server daemon");
 		kcl_unregister_service(local_id);
-		return r;
+		goto out;
 	}
 
 	r = kcl_join_service(local_id);
@@ -1252,13 +1257,19 @@
 		kcl_unregister_service(local_id);
 	}
 
+out:
+	up(&cmirror_register_lock);
 	return r;
 }
 
 static int cluster_disconnect(void)
 {
-	if (--mirror_set_count)
+	down(&cmirror_register_lock);
+
+	if (--mirror_set_count) {
+		up(&cmirror_register_lock);
 		return 0;
+	}
 
 	/* By setting 'shutting_down', the server will not be suspended **
 	** when a stop is received */
@@ -1267,6 +1278,7 @@
 	stop_server();
 	kcl_unregister_service(local_id);
 
+	up(&cmirror_register_lock);
 	return 0;
 }
 

[Cluster-devel] cluster/cmirror-kernel/src dm-cmirror-client.c

[jbrassow]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4
Changes by:	jbrassow at sourceware.org	2006-07-07 17:08:56

Modified files:
	cmirror-kernel/src: dm-cmirror-client.c 

Log message:
	- If a cluster mirror was removed, and a new one created fast enough, the
	processes could overlap such that the creation would fail to register
	with cman because the remove hadn't unregistered yet (-EEXIST)
	
	This should fix bug 197952
	
	Waiting to commit this to rhel4u4 until the bug becomes a blocker.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cmirror-kernel/src/dm-cmirror-client.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.1.2.20&r2=1.1.2.21

--- cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/06/29 19:48:01	1.1.2.20
+++ cluster/cmirror-kernel/src/Attic/dm-cmirror-client.c	2006/07/07 17:08:56	1.1.2.21
@@ -1206,26 +1206,31 @@
 };
 
 static int mirror_set_count = 0; /* used to prevent multiple cluster [dis]connects */
+static DECLARE_MUTEX(cmirror_register_lock);
 
 static int cluster_connect(void)
 {
-	int r;
+	int r = 0;
 
-	if (mirror_set_count++)
-		return 0;
+	down(&cmirror_register_lock);
+
+	if (mirror_set_count++) {
+		up(&cmirror_register_lock);
+		goto out;
+	}
 
 	r = kcl_register_service("clustered_log", 13, SERVICE_LEVEL_GDLM, &clog_ops,
 				 1, NULL, &local_id);
 	if (r) {
-		DMWARN("Couldn't register clustered_log service");
-		return r;
+		DMWARN("Couldn't register clustered_log service.  Reason: %d", r);
+		goto out;
 	}
 
 	r = start_server();
 	if(r){
 		DMWARN("Unable to start clustered log server daemon");
 		kcl_unregister_service(local_id);
-		return r;
+		goto out;
 	}
 
 	r = kcl_join_service(local_id);
@@ -1236,13 +1241,19 @@
 		kcl_unregister_service(local_id);
 	}
 
+out:
+	up(&cmirror_register_lock);
 	return r;
 }
 
 static int cluster_disconnect(void)
 {
-	if (--mirror_set_count)
+	down(&cmirror_register_lock);
+
+	if (--mirror_set_count) {
+		up(&cmirror_register_lock);
 		return 0;
+	}
 
 	/* By setting 'shutting_down', the server will not be suspended **
 	** when a stop is received */
@@ -1251,6 +1262,7 @@
 	stop_server();
 	kcl_unregister_service(local_id);
 
+	up(&cmirror_register_lock);
 	return 0;
 }
 

[Cluster-devel] cluster/cmirror-kernel/src dm-cmirror-client.c

[jbrassow]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4
Changes by:	jbrassow at sourceware.org	2006-06-15 17:53:15

Modified files:
	cmirror-kernel/src: dm-cmirror-client.c 

Log message:
	- bah!  typo.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cmirror-kernel/src/dm-cmirror-client.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.1.2.16&r2=1.1.2.17

[Cluster-devel] cluster/cmirror-kernel/src dm-cmirror-client.c

[jbrassow]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4
Changes by:	jbrassow at sourceware.org	2006-06-15 17:44:56

Modified files:
	cmirror-kernel/src: dm-cmirror-client.c 

Log message:
	- it was possible under heavy I/O and machine joining/leaving to run into
	a situation where the log server leaves and we have to rerun elections.
	Failing to do so would result in never making progress.
	
	This bug only occurs if the log server fails/leaves, we need to mark a
	region, and the newly appointed server also fails/leaves.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/cmirror-kernel/src/dm-cmirror-client.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.1.2.15&r2=1.1.2.16