dm raid1: support flush

Flush support for dm-raid1. When it receives an empty barrier, submit it to all the devices via dm-io. Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>

dm raid1: support flush
Flush support for dm-raid1. When it receives an empty barrier, submit it to all the devices via dm-io. Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Mikulas Patocka · Alasdair G Kergon
1 parent f1e5398746
Showing 2 changed files with 34 additions and 4 deletions Side-by-side Diff
drivers/md/dm-raid1.c
drivers/md/dm-region-hash.c
@@ -396,6 +396,8 @@
  */
 static sector_t map_sector(struct mirror *m, struct bio *bio)
 {
+	if (unlikely(!bio->bi_size))
+		return 0;
 	return m->offset + (bio->bi_sector - m->ms->ti->begin);
 }
  
@@ -562,7 +564,7 @@
 	struct dm_io_region io[ms->nr_mirrors], *dest = io;
 	struct mirror *m;
 	struct dm_io_request io_req = {
-		.bi_rw = WRITE,
+		.bi_rw = WRITE | (bio->bi_rw & WRITE_BARRIER),
 		.mem.type = DM_IO_BVEC,
 		.mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx,
 		.notify.fn = write_callback,
@@ -603,6 +605,11 @@
 	bio_list_init(&requeue);
  
 	while ((bio = bio_list_pop(writes))) {
+		if (unlikely(bio_empty_barrier(bio))) {
+			bio_list_add(&sync, bio);
+			continue;
+		}
+
 		region = dm_rh_bio_to_region(ms->rh, bio);
  
 		if (log->type->is_remote_recovering &&
@@ -995,6 +1002,7 @@
  
 	ti->private = ms;
 	ti->split_io = dm_rh_get_region_size(ms->rh);
+	ti->num_flush_requests = 1;
  
 	ms->kmirrord_wq = create_singlethread_workqueue("kmirrord");
 	if (!ms->kmirrord_wq) {
@@ -1122,7 +1130,8 @@
 	 * We need to dec pending if this was a write.
 	 */
 	if (rw == WRITE) {
-		dm_rh_dec(ms->rh, map_context->ll);
+		if (likely(!bio_empty_barrier(bio)))
+			dm_rh_dec(ms->rh, map_context->ll);
 		return error;
 	}
  
@@ -79,6 +79,11 @@
 	struct list_head recovered_regions;
 	struct list_head failed_recovered_regions;
  
+	/*
+	 * If there was a barrier failure no regions can be marked clean.
+	 */
+	int barrier_failure;
+
 	void *context;
 	sector_t target_begin;
  
@@ -211,6 +216,7 @@
 	INIT_LIST_HEAD(&rh->quiesced_regions);
 	INIT_LIST_HEAD(&rh->recovered_regions);
 	INIT_LIST_HEAD(&rh->failed_recovered_regions);
+	rh->barrier_failure = 0;
  
 	rh->region_pool = mempool_create_kmalloc_pool(MIN_REGIONS,
 						      sizeof(struct dm_region));
@@ -395,6 +401,11 @@
 	region_t region = dm_rh_bio_to_region(rh, bio);
 	int recovering = 0;
  
+	if (bio_empty_barrier(bio)) {
+		rh->barrier_failure = 1;
+		return;
+	}
+
 	/* We must inform the log that the sync count has changed. */
 	log->type->set_region_sync(log, region, 0);
  
  
@@ -515,8 +526,11 @@
 {
 	struct bio *bio;
  
-	for (bio = bios->head; bio; bio = bio->bi_next)
+	for (bio = bios->head; bio; bio = bio->bi_next) {
+		if (bio_empty_barrier(bio))
+			continue;
 		rh_inc(rh, dm_rh_bio_to_region(rh, bio));
+	}
 }
 EXPORT_SYMBOL_GPL(dm_rh_inc_pending);
  
@@ -544,7 +558,14 @@
 		 */
  
 		/* do nothing for DM_RH_NOSYNC */
-		if (reg->state == DM_RH_RECOVERING) {
+		if (unlikely(rh->barrier_failure)) {
+			/*
+			 * If a write barrier failed some time ago, we
+			 * don't know whether or not this write made it
+			 * to the disk, so we must resync the device.
+			 */
+			reg->state = DM_RH_NOSYNC;
+		} else if (reg->state == DM_RH_RECOVERING) {
 			list_add_tail(&reg->list, &rh->quiesced_regions);
 		} else if (reg->state == DM_RH_DIRTY) {
 			reg->state = DM_RH_CLEAN;
...	...	@@ -396,6 +396,8 @@
396	396	*/
397	397	static sector_t map_sector(struct mirror m, struct bio bio)
398	398	{
	399	+ if (unlikely(!bio->bi_size))
	400	+ return 0;
399	401	return m->offset + (bio->bi_sector - m->ms->ti->begin);
400	402	}
401	403
...	...	@@ -562,7 +564,7 @@
562	564	struct dm_io_region io[ms->nr_mirrors], *dest = io;
563	565	struct mirror *m;
564	566	struct dm_io_request io_req = {
565		- .bi_rw = WRITE,
	567	+ .bi_rw = WRITE \| (bio->bi_rw & WRITE_BARRIER),
566	568	.mem.type = DM_IO_BVEC,
567	569	.mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx,
568	570	.notify.fn = write_callback,
...	...	@@ -603,6 +605,11 @@
603	605	bio_list_init(&requeue);
604	606
605	607	while ((bio = bio_list_pop(writes))) {
	608	+ if (unlikely(bio_empty_barrier(bio))) {
	609	+ bio_list_add(&sync, bio);
	610	+ continue;
	611	+ }
	612	+
606	613	region = dm_rh_bio_to_region(ms->rh, bio);
607	614
608	615	if (log->type->is_remote_recovering &&
...	...	@@ -995,6 +1002,7 @@
995	1002
996	1003	ti->private = ms;
997	1004	ti->split_io = dm_rh_get_region_size(ms->rh);
	1005	+ ti->num_flush_requests = 1;
998	1006
999	1007	ms->kmirrord_wq = create_singlethread_workqueue("kmirrord");
1000	1008	if (!ms->kmirrord_wq) {
...	...	@@ -1122,7 +1130,8 @@
1122	1130	* We need to dec pending if this was a write.
1123	1131	*/
1124	1132	if (rw == WRITE) {
1125		- dm_rh_dec(ms->rh, map_context->ll);
	1133	+ if (likely(!bio_empty_barrier(bio)))
	1134	+ dm_rh_dec(ms->rh, map_context->ll);
1126	1135	return error;
1127	1136	}
1128	1137
...	...	@@ -79,6 +79,11 @@
79	79	struct list_head recovered_regions;
80	80	struct list_head failed_recovered_regions;
81	81
	82	+ /*
	83	+ * If there was a barrier failure no regions can be marked clean.
	84	+ */
	85	+ int barrier_failure;
	86	+
82	87	void *context;
83	88	sector_t target_begin;
84	89
...	...	@@ -211,6 +216,7 @@
211	216	INIT_LIST_HEAD(&rh->quiesced_regions);
212	217	INIT_LIST_HEAD(&rh->recovered_regions);
213	218	INIT_LIST_HEAD(&rh->failed_recovered_regions);
	219	+ rh->barrier_failure = 0;
214	220
215	221	rh->region_pool = mempool_create_kmalloc_pool(MIN_REGIONS,
216	222	sizeof(struct dm_region));
...	...	@@ -395,6 +401,11 @@
395	401	region_t region = dm_rh_bio_to_region(rh, bio);
396	402	int recovering = 0;
397	403
	404	+ if (bio_empty_barrier(bio)) {
	405	+ rh->barrier_failure = 1;
	406	+ return;
	407	+ }
	408	+
398	409	/* We must inform the log that the sync count has changed. */
399	410	log->type->set_region_sync(log, region, 0);
400	411
401	412
...	...	@@ -515,8 +526,11 @@
515	526	{
516	527	struct bio *bio;
517	528
518		- for (bio = bios->head; bio; bio = bio->bi_next)
	529	+ for (bio = bios->head; bio; bio = bio->bi_next) {
	530	+ if (bio_empty_barrier(bio))
	531	+ continue;
519	532	rh_inc(rh, dm_rh_bio_to_region(rh, bio));
	533	+ }
520	534	}
521	535	EXPORT_SYMBOL_GPL(dm_rh_inc_pending);
522	536
...	...	@@ -544,7 +558,14 @@
544	558	*/
545	559
546	560	/* do nothing for DM_RH_NOSYNC */
547		- if (reg->state == DM_RH_RECOVERING) {
	561	+ if (unlikely(rh->barrier_failure)) {
	562	+ /*
	563	+ * If a write barrier failed some time ago, we
	564	+ * don't know whether or not this write made it
	565	+ * to the disk, so we must resync the device.
	566	+ */
	567	+ reg->state = DM_RH_NOSYNC;
	568	+ } else if (reg->state == DM_RH_RECOVERING) {
548	569	list_add_tail(&reg->list, &rh->quiesced_regions);
549	570	} else if (reg->state == DM_RH_DIRTY) {
550	571	reg->state = DM_RH_CLEAN;