[dm-devel] mirroring: [patch 6 of 8] device failure tolerance

[Jonathan E Brassow <jbrassow@xxxxxxxxxx>]

This patch adds the necessary components to do cluster mirroring 
properly, _if_ the log type supports it.

 brassow

diff -urN linux-2.6.12-006/drivers/md/dm-log.c 
linux-2.6.12-007/drivers/md/dm-log.c
--- linux-2.6.12-006/drivers/md/dm-log.c	2005-06-29 09:54:33.001960796 
-0500
+++ linux-2.6.12-007/drivers/md/dm-log.c	2005-06-29 11:57:48.224957146 
-0500
@@ -705,6 +705,7 @@
 static struct dirty_log_type _core_type = {
 	.name = "core",
 	.module = THIS_MODULE,
+	.multi_node = 0,
 	.ctr = core_ctr,
 	.dtr = core_dtr,
 	.get_region_size = core_get_region_size,
@@ -722,6 +723,7 @@
 static struct dirty_log_type _disk_type = {
 	.name = "disk",
 	.module = THIS_MODULE,
+	.multi_node = 0,
 	.ctr = disk_ctr,
 	.dtr = disk_dtr,
 	.suspend = disk_flush,
diff -urN linux-2.6.12-006/drivers/md/dm-log.h 
linux-2.6.12-007/drivers/md/dm-log.h
--- linux-2.6.12-006/drivers/md/dm-log.h	2005-06-29 09:54:24.957955783 
-0500
+++ linux-2.6.12-007/drivers/md/dm-log.h	2005-06-29 11:58:43.001589620 
-0500
@@ -29,6 +29,7 @@
 	const char *name;
 	struct module *module;
 	unsigned int use_count;
+	unsigned int multi_node;

 	int (*ctr)(struct dirty_log *log, struct dm_target *ti,
 		   unsigned int argc, char **argv);
diff -urN linux-2.6.12-006/drivers/md/dm-raid1.c 
linux-2.6.12-007/drivers/md/dm-raid1.c
--- linux-2.6.12-006/drivers/md/dm-raid1.c	2005-06-29 
11:44:57.552713175 -0500
+++ linux-2.6.12-007/drivers/md/dm-raid1.c	2005-06-29 
12:08:09.838784593 -0500
@@ -573,6 +573,7 @@
 	struct bio_list writes;
 	struct bio_list failures;
 	struct work_struct failure_work;
+	struct completion failure_completion;

 	/* recovery */
 	atomic_t suspended;
@@ -861,7 +862,7 @@
 				      0) == RH_CLEAN))
 			m = choose_mirror(ms, NULL);
 		else {
-			m = ms->default_mirror;;
+			m = ms->default_mirror;

 			/* If the default fails, we give up .*/
 			if (unlikely(m && atomic_read(&m->error_count)))
@@ -890,9 +891,21 @@
 	struct bio *bio;
 	struct bio_list failed_writes;
 	struct mirror_set *ms = (struct mirror_set *)data;
+	struct dirty_log *log = ms->rh.log;

 	dm_table_event(ms->ti->table);

+	if (log->type->multi_node) {
+		DMERR("Event signaled.  Waiting to start failure handling.");
+		wait_for_completion(&ms->failure_completion);
+		DMINFO("Wait complete");
+	}
+
+	/*
+	 * Device must be suspended to prevent corruption in
+	 * cluster context.
+	 */
+
 	/* Take list out to handle endios. */
 	spin_lock(&ms->lock);
 	failed_writes = ms->failures;
@@ -902,6 +915,10 @@
 	while ((bio = bio_list_pop(&failed_writes))) {
 		bio_endio(bio, bio->bi_size, 0);
 	}
+
+	if (log->type->multi_node) {
+		DMERR("Failure handling complete.");
+	}
 }

 static void write_callback(unsigned long error, void *context)
@@ -966,7 +983,7 @@
 		}
 	}

-	bio_endio(bio, bio->bi_size, 0);
+	bio_endio(bio, bio->bi_size, ret);
 }

 static void do_write(struct mirror_set *ms, struct bio *bio)
@@ -974,9 +991,11 @@
 	unsigned int i;
 	struct io_region io[ms->nr_mirrors], *dest = io;
 	struct mirror *m;
+	struct dirty_log *log = ms->rh.log;

 	for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++) {
-		if (likely(!atomic_read(&m->error_count)))
+		if (likely(!atomic_read(&m->error_count)) ||
+		    log->type->multi_node)
 			map_region(dest++, m, bio);
 	}

@@ -999,6 +1018,7 @@
 	int state;
 	struct bio *bio;
 	struct bio_list sync, nosync, recover, *this_list = NULL;
+	struct bio_list tmp;

 	if (!writes->head)
 		return;
@@ -1009,6 +1029,7 @@
 	bio_list_init(&sync);
 	bio_list_init(&nosync);
 	bio_list_init(&recover);
+	bio_list_init(&tmp);

 	while ((bio = bio_list_pop(writes))) {
 		state = rh_state(&ms->rh, bio_to_region(&ms->rh, bio), 1);
@@ -1025,10 +1046,15 @@
 		case RH_RECOVERING:
 			this_list = &recover;
 			break;
+
+		case RH_REMOTE_RECOVERING:
+			this_list = &tmp;
+			break;
 		}

 		bio_list_add(this_list, bio);
 	}
+	bio_list_merge(writes, &tmp);

 	/*
 	 * Increment the pending counts for any regions that will
@@ -1132,6 +1158,8 @@
 	bio_list_init(&ms->failures);
 	INIT_WORK(&ms->failure_work, write_failure_handler, ms);
 	
+	init_completion(&ms->failure_completion);
+
 	return ms;
 }

@@ -1451,6 +1479,7 @@
 	struct mirror_set *ms = (struct mirror_set *)ti->private;

 	atomic_set(&ms->suspended, 1);
+	complete(&ms->failure_completion);
 }