1 files changed, 245 insertions, 148 deletions

diff --git a/drivers/md/dm-raid1.c b/drivers/md/dm-raid1.c
index ce8868c768c..7dfdb5c746d 100644
--- a/drivers/md/dm-raid1.c
+++ b/drivers/md/dm-raid1.c
@@ -22,8 +22,6 @@
 #define DM_MSG_PREFIX "raid1"
 
 #define MAX_RECOVERY 1	/* Maximum number of regions recovered in parallel. */
-#define DM_IO_PAGES 64
-#define DM_KCOPYD_PAGES 64
 
 #define DM_RAID1_HANDLE_ERRORS 0x01
 #define errors_handled(p)	((p)->features & DM_RAID1_HANDLE_ERRORS)
@@ -35,6 +33,7 @@ static DECLARE_WAIT_QUEUE_HEAD(_kmirrord_recovery_stopped);
  *---------------------------------------------------------------*/
 enum dm_raid1_error {
 	DM_RAID1_WRITE_ERROR,
+	DM_RAID1_FLUSH_ERROR,
 	DM_RAID1_SYNC_ERROR,
 	DM_RAID1_READ_ERROR
 };
@@ -57,16 +56,17 @@ struct mirror_set {
 	struct bio_list reads;
 	struct bio_list writes;
 	struct bio_list failures;
+	struct bio_list holds;	/* bios are waiting until suspend */
 
 	struct dm_region_hash *rh;
 	struct dm_kcopyd_client *kcopyd_client;
 	struct dm_io_client *io_client;
-	mempool_t *read_record_pool;
 
 	/* recovery */
 	region_t nr_regions;
 	int in_sync;
 	int log_failure;
+	int leg_failure;
 	atomic_t suspend;
 
 	atomic_t default_mirror;	/* Default mirror */
@@ -82,6 +82,9 @@ struct mirror_set {
 	struct mirror mirror[0];
 };
 
+DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(raid1_resync_throttle,
+		"A percentage of time allocated for raid resynchronization");
+
 static void wakeup_mirrord(void *context)
 {
 	struct mirror_set *ms = context;
@@ -138,14 +141,13 @@ static void dispatch_bios(void *context, struct bio_list *bio_list)
 		queue_bio(ms, bio, WRITE);
 }
 
-#define MIN_READ_RECORDS 20
-struct dm_raid1_read_record {
+struct dm_raid1_bio_record {
 	struct mirror *m;
+	/* if details->bi_bdev == NULL, details were not saved */
 	struct dm_bio_details details;
+	region_t write_region;
 };
 
-static struct kmem_cache *_dm_raid1_read_record_cache;
-
 /*
  * Every mirror should look like this one.
  */
@@ -179,6 +181,17 @@ static void set_default_mirror(struct mirror *m)
 	atomic_set(&ms->default_mirror, m - m0);
 }
 
+static struct mirror *get_valid_mirror(struct mirror_set *ms)
+{
+	struct mirror *m;
+
+	for (m = ms->mirror; m < ms->mirror + ms->nr_mirrors; m++)
+		if (!atomic_read(&m->error_count))
+			return m;
+
+	return NULL;
+}
+
 /* fail_mirror
  * @m: mirror device to fail
  * @error_type: one of the enum's, DM_RAID1_*_ERROR
@@ -198,6 +211,8 @@ static void fail_mirror(struct mirror *m, enum dm_raid1_error error_type)
 	struct mirror_set *ms = m->ms;
 	struct mirror *new;
 
+	ms->leg_failure = 1;
+
 	/*
 	 * error_count is used for nothing more than a
 	 * simple way to tell if a device has encountered
@@ -224,19 +239,50 @@ static void fail_mirror(struct mirror *m, enum dm_raid1_error error_type)
 		goto out;
 	}
 
-	for (new = ms->mirror; new < ms->mirror + ms->nr_mirrors; new++)
-		if (!atomic_read(&new->error_count)) {
-			set_default_mirror(new);
-			break;
-		}
-
-	if (unlikely(new == ms->mirror + ms->nr_mirrors))
+	new = get_valid_mirror(ms);
+	if (new)
+		set_default_mirror(new);
+	else
 		DMWARN("All sides of mirror have failed.");
 
 out:
 	schedule_work(&ms->trigger_event);
 }
 
+static int mirror_flush(struct dm_target *ti)
+{
+	struct mirror_set *ms = ti->private;
+	unsigned long error_bits;
+
+	unsigned int i;
+	struct dm_io_region io[ms->nr_mirrors];
+	struct mirror *m;
+	struct dm_io_request io_req = {
+		.bi_rw = WRITE_FLUSH,
+		.mem.type = DM_IO_KMEM,
+		.mem.ptr.addr = NULL,
+		.client = ms->io_client,
+	};
+
+	for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++) {
+		io[i].bdev = m->dev->bdev;
+		io[i].sector = 0;
+		io[i].count = 0;
+	}
+
+	error_bits = -1;
+	dm_io(&io_req, ms->nr_mirrors, io, &error_bits);
+	if (unlikely(error_bits != 0)) {
+		for (i = 0; i < ms->nr_mirrors; i++)
+			if (test_bit(i, &error_bits))
+				fail_mirror(ms->mirror + i,
+					    DM_RAID1_FLUSH_ERROR);
+		return -EIO;
+	}
+
+	return 0;
+}
+
 /*-----------------------------------------------------------------
  * Recovery.
  *
@@ -386,7 +432,7 @@ static int mirror_available(struct mirror_set *ms, struct bio *bio)
 	region_t region = dm_rh_bio_to_region(ms->rh, bio);
 
 	if (log->type->in_sync(log, region, 0))
-		return choose_mirror(ms,  bio->bi_sector) ? 1 : 0;
+		return choose_mirror(ms,  bio->bi_iter.bi_sector) ? 1 : 0;
 
 	return 0;
 }
@@ -396,13 +442,15 @@ static int mirror_available(struct mirror_set *ms, struct bio *bio)
  */
 static sector_t map_sector(struct mirror *m, struct bio *bio)
 {
-	return m->offset + (bio->bi_sector - m->ms->ti->begin);
+	if (unlikely(!bio->bi_iter.bi_size))
+		return 0;
+	return m->offset + dm_target_offset(m->ms->ti, bio->bi_iter.bi_sector);
 }
 
 static void map_bio(struct mirror *m, struct bio *bio)
 {
 	bio->bi_bdev = m->dev->bdev;
-	bio->bi_sector = map_sector(m, bio);
+	bio->bi_iter.bi_sector = map_sector(m, bio);
 }
 
 static void map_region(struct dm_io_region *io, struct mirror *m,
@@ -410,7 +458,35 @@ static void map_region(struct dm_io_region *io, struct mirror *m,
 {
 	io->bdev = m->dev->bdev;
 	io->sector = map_sector(m, bio);
-	io->count = bio->bi_size >> 9;
+	io->count = bio_sectors(bio);
+}
+
+static void hold_bio(struct mirror_set *ms, struct bio *bio)
+{
+	/*
+	 * Lock is required to avoid race condition during suspend
+	 * process.
+	 */
+	spin_lock_irq(&ms->lock);
+
+	if (atomic_read(&ms->suspend)) {
+		spin_unlock_irq(&ms->lock);
+
+		/*
+		 * If device is suspended, complete the bio.
+		 */
+		if (dm_noflush_suspending(ms->ti))
+			bio_endio(bio, DM_ENDIO_REQUEUE);
+		else
+			bio_endio(bio, -EIO);
+		return;
+	}
+
+	/*
+	 * Hold bio until the suspend is complete.
+	 */
+	bio_list_add(&ms->holds, bio);
+	spin_unlock_irq(&ms->lock);
 }
 
 /*-----------------------------------------------------------------
@@ -450,8 +526,8 @@ static void read_async_bio(struct mirror *m, struct bio *bio)
 	struct dm_io_region io;
 	struct dm_io_request io_req = {
 		.bi_rw = READ,
-		.mem.type = DM_IO_BVEC,
-		.mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx,
+		.mem.type = DM_IO_BIO,
+		.mem.ptr.bio = bio,
 		.notify.fn = read_callback,
 		.notify.context = bio,
 		.client = m->ms->io_client,
@@ -483,7 +559,7 @@ static void do_reads(struct mirror_set *ms, struct bio_list *reads)
 		 * We can only read balance if the region is in sync.
 		 */
 		if (likely(region_in_sync(ms, region, 1)))
-			m = choose_mirror(ms, bio->bi_sector);
+			m = choose_mirror(ms, bio->bi_iter.bi_sector);
 		else if (m && atomic_read(&m->error_count))
 			m = NULL;
 
@@ -511,7 +587,6 @@ static void write_callback(unsigned long error, void *context)
 	unsigned i, ret = 0;
 	struct bio *bio = (struct bio *) context;
 	struct mirror_set *ms;
-	int uptodate = 0;
 	int should_wake = 0;
 	unsigned long flags;
 
@@ -524,36 +599,27 @@ static void write_callback(unsigned long error, void *context)
 	 * This way we handle both writes to SYNC and NOSYNC
 	 * regions with the same code.
 	 */
-	if (likely(!error))
-		goto out;
+	if (likely(!error)) {
+		bio_endio(bio, ret);
+		return;
+	}
 
 	for (i = 0; i < ms->nr_mirrors; i++)
 		if (test_bit(i, &error))
 			fail_mirror(ms->mirror + i, DM_RAID1_WRITE_ERROR);
-		else
-			uptodate = 1;
 
-	if (unlikely(!uptodate)) {
-		DMERR("All replicated volumes dead, failing I/O");
-		/* None of the writes succeeded, fail the I/O. */
-		ret = -EIO;
-	} else if (errors_handled(ms)) {
-		/*
-		 * Need to raise event.  Since raising
-		 * events can block, we need to do it in
-		 * the main thread.
-		 */
-		spin_lock_irqsave(&ms->lock, flags);
-		if (!ms->failures.head)
-			should_wake = 1;
-		bio_list_add(&ms->failures, bio);
-		spin_unlock_irqrestore(&ms->lock, flags);
-		if (should_wake)
-			wakeup_mirrord(ms);
-		return;
-	}
-out:
-	bio_endio(bio, ret);
+	/*
+	 * Need to raise event.  Since raising
+	 * events can block, we need to do it in
+	 * the main thread.
+	 */
+	spin_lock_irqsave(&ms->lock, flags);
+	if (!ms->failures.head)
+		should_wake = 1;
+	bio_list_add(&ms->failures, bio);
+	spin_unlock_irqrestore(&ms->lock, flags);
+	if (should_wake)
+		wakeup_mirrord(ms);
 }
 
 static void do_write(struct mirror_set *ms, struct bio *bio)
@@ -562,14 +628,20 @@ static void do_write(struct mirror_set *ms, struct bio *bio)
 	struct dm_io_region io[ms->nr_mirrors], *dest = io;
 	struct mirror *m;
 	struct dm_io_request io_req = {
-		.bi_rw = WRITE,
-		.mem.type = DM_IO_BVEC,
-		.mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx,
+		.bi_rw = WRITE | (bio->bi_rw & WRITE_FLUSH_FUA),
+		.mem.type = DM_IO_BIO,
+		.mem.ptr.bio = bio,
 		.notify.fn = write_callback,
 		.notify.context = bio,
 		.client = ms->io_client,
 	};
 
+	if (bio->bi_rw & REQ_DISCARD) {
+		io_req.bi_rw |= REQ_DISCARD;
+		io_req.mem.type = DM_IO_KMEM;
+		io_req.mem.ptr.addr = NULL;
+	}
+
 	for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++)
 		map_region(dest++, m, bio);
 
@@ -603,6 +675,12 @@ static void do_writes(struct mirror_set *ms, struct bio_list *writes)
 	bio_list_init(&requeue);
 
 	while ((bio = bio_list_pop(writes))) {
+		if ((bio->bi_rw & REQ_FLUSH) ||
+		    (bio->bi_rw & REQ_DISCARD)) {
+			bio_list_add(&sync, bio);
+			continue;
+		}
+
 		region = dm_rh_bio_to_region(ms->rh, bio);
 
 		if (log->type->is_remote_recovering &&
@@ -638,6 +716,7 @@ static void do_writes(struct mirror_set *ms, struct bio_list *writes)
 		spin_lock_irq(&ms->lock);
 		bio_list_merge(&ms->writes, &requeue);
 		spin_unlock_irq(&ms->lock);
+		delayed_wake(ms);
 	}
 
 	/*
@@ -647,12 +726,18 @@ static void do_writes(struct mirror_set *ms, struct bio_list *writes)
 	 */
 	dm_rh_inc_pending(ms->rh, &sync);
 	dm_rh_inc_pending(ms->rh, &nosync);
-	ms->log_failure = dm_rh_flush(ms->rh) ? 1 : 0;
+
+	/*
+	 * If the flush fails on a previous call and succeeds here,
+	 * we must not reset the log_failure variable.  We need
+	 * userspace interaction to do that.
+	 */
+	ms->log_failure = dm_rh_flush(ms->rh) ? 1 : ms->log_failure;
 
 	/*
 	 * Dispatch io.
 	 */
-	if (unlikely(ms->log_failure)) {
+	if (unlikely(ms->log_failure) && errors_handled(ms)) {
 		spin_lock_irq(&ms->lock);
 		bio_list_merge(&ms->failures, &sync);
 		spin_unlock_irq(&ms->lock);
@@ -665,8 +750,15 @@ static void do_writes(struct mirror_set *ms, struct bio_list *writes)
 		dm_rh_delay(ms->rh, bio);
 
 	while ((bio = bio_list_pop(&nosync))) {
-		map_bio(get_default_mirror(ms), bio);
-		generic_make_request(bio);
+		if (unlikely(ms->leg_failure) && errors_handled(ms)) {
+			spin_lock_irq(&ms->lock);
+			bio_list_add(&ms->failures, bio);
+			spin_unlock_irq(&ms->lock);
+			wakeup_mirrord(ms);
+		} else {
+			map_bio(get_default_mirror(ms), bio);
+			generic_make_request(bio);
+		}
 	}
 }
 
@@ -674,20 +766,12 @@ static void do_failures(struct mirror_set *ms, struct bio_list *failures)
 {
 	struct bio *bio;
 
-	if (!failures->head)
+	if (likely(!failures->head))
 		return;
 
-	if (!ms->log_failure) {
-		while ((bio = bio_list_pop(failures))) {
-			ms->in_sync = 0;
-			dm_rh_mark_nosync(ms->rh, bio, bio->bi_size, 0);
-		}
-		return;
-	}
-
 	/*
 	 * If the log has failed, unattempted writes are being
-	 * put on the failures list.  We can't issue those writes
+	 * put on the holds list.  We can't issue those writes
 	 * until a log has been marked, so we must store them.
 	 *
 	 * If a 'noflush' suspend is in progress, we can requeue
@@ -702,23 +786,27 @@ static void do_failures(struct mirror_set *ms, struct bio_list *failures)
 	 * for us to treat them the same and requeue them
 	 * as well.
 	 */
-	if (dm_noflush_suspending(ms->ti)) {
-		while ((bio = bio_list_pop(failures)))
-			bio_endio(bio, DM_ENDIO_REQUEUE);
-		return;
-	}
+	while ((bio = bio_list_pop(failures))) {
+		if (!ms->log_failure) {
+			ms->in_sync = 0;
+			dm_rh_mark_nosync(ms->rh, bio);
+		}
 
-	if (atomic_read(&ms->suspend)) {
-		while ((bio = bio_list_pop(failures)))
+		/*
+		 * If all the legs are dead, fail the I/O.
+		 * If we have been told to handle errors, hold the bio
+		 * and wait for userspace to deal with the problem.
+		 * Otherwise pretend that the I/O succeeded. (This would
+		 * be wrong if the failed leg returned after reboot and
+		 * got replicated back to the good legs.)
+		 */
+		if (!get_valid_mirror(ms))
 			bio_endio(bio, -EIO);
-		return;
+		else if (errors_handled(ms))
+			hold_bio(ms, bio);
+		else
+			bio_endio(bio, 0);
 	}
-
-	spin_lock_irq(&ms->lock);
-	bio_list_merge(&ms->failures, failures);
-	spin_unlock_irq(&ms->lock);
-
-	delayed_wake(ms);
 }
 
 static void trigger_event(struct work_struct *work)
@@ -753,8 +841,6 @@ static void do_mirror(struct work_struct *work)
 	do_reads(ms, &reads);
 	do_writes(ms, &writes);
 	do_failures(ms, &failures);
-
-	dm_table_unplug_all(ms->ti->table);
 }
 
 /*-----------------------------------------------------------------
@@ -777,28 +863,23 @@ static struct mirror_set *alloc_context(unsigned int nr_mirrors,
 	}
 
 	spin_lock_init(&ms->lock);
+	bio_list_init(&ms->reads);
+	bio_list_init(&ms->writes);
+	bio_list_init(&ms->failures);
+	bio_list_init(&ms->holds);
 
 	ms->ti = ti;
 	ms->nr_mirrors = nr_mirrors;
 	ms->nr_regions = dm_sector_div_up(ti->len, region_size);
 	ms->in_sync = 0;
 	ms->log_failure = 0;
+	ms->leg_failure = 0;
 	atomic_set(&ms->suspend, 0);
 	atomic_set(&ms->default_mirror, DEFAULT_MIRROR);
 
-	ms->read_record_pool = mempool_create_slab_pool(MIN_READ_RECORDS,
-						_dm_raid1_read_record_cache);
-
-	if (!ms->read_record_pool) {
-		ti->error = "Error creating mirror read_record_pool";
-		kfree(ms);
-		return NULL;
-	}
-
-	ms->io_client = dm_io_client_create(DM_IO_PAGES);
+	ms->io_client = dm_io_client_create();
 	if (IS_ERR(ms->io_client)) {
 		ti->error = "Error creating dm_io client";
-		mempool_destroy(ms->read_record_pool);
 		kfree(ms);
  		return NULL;
 	}
@@ -810,7 +891,6 @@ static struct mirror_set *alloc_context(unsigned int nr_mirrors,
 	if (IS_ERR(ms->rh)) {
 		ti->error = "Error creating dirty region hash";
 		dm_io_client_destroy(ms->io_client);
-		mempool_destroy(ms->read_record_pool);
 		kfree(ms);
 		return NULL;
 	}
@@ -826,7 +906,6 @@ static void free_context(struct mirror_set *ms, struct dm_target *ti,
 
 	dm_io_client_destroy(ms->io_client);
 	dm_region_hash_destroy(ms->rh);
-	mempool_destroy(ms->read_record_pool);
 	kfree(ms);
 }
 
@@ -834,14 +913,14 @@ static int get_mirror(struct mirror_set *ms, struct dm_target *ti,
 		      unsigned int mirror, char **argv)
 {
 	unsigned long long offset;
+	char dummy;
 
-	if (sscanf(argv[1], "%llu", &offset) != 1) {
+	if (sscanf(argv[1], "%llu%c", &offset, &dummy) != 1) {
 		ti->error = "Invalid offset";
 		return -EINVAL;
 	}
 
-	if (dm_get_device(ti, argv[0], offset, ti->len,
-			  dm_table_get_mode(ti->table),
+	if (dm_get_device(ti, argv[0], dm_table_get_mode(ti->table),
 			  &ms->mirror[mirror].dev)) {
 		ti->error = "Device lookup failure";
 		return -ENXIO;
@@ -864,13 +943,14 @@ static struct dm_dirty_log *create_dirty_log(struct dm_target *ti,
 {
 	unsigned param_count;
 	struct dm_dirty_log *dl;
+	char dummy;
 
 	if (argc < 2) {
 		ti->error = "Insufficient mirror log arguments";
 		return NULL;
 	}
 
-	if (sscanf(argv[1], "%u", &param_count) != 1) {
+	if (sscanf(argv[1], "%u%c", &param_count, &dummy) != 1) {
 		ti->error = "Invalid mirror log argument count";
 		return NULL;
 	}
@@ -882,7 +962,8 @@ static struct dm_dirty_log *create_dirty_log(struct dm_target *ti,
 		return NULL;
 	}
 
-	dl = dm_dirty_log_create(argv[0], ti, param_count, argv + 2);
+	dl = dm_dirty_log_create(argv[0], ti, mirror_flush, param_count,
+				 argv + 2);
 	if (!dl) {
 		ti->error = "Error creating mirror dirty log";
 		return NULL;
@@ -896,13 +977,14 @@ static int parse_features(struct mirror_set *ms, unsigned argc, char **argv,
 {
 	unsigned num_features;
 	struct dm_target *ti = ms->ti;
+	char dummy;
 
 	*args_used = 0;
 
 	if (!argc)
 		return 0;
 
-	if (sscanf(argv[0], "%u", &num_features) != 1) {
+	if (sscanf(argv[0], "%u%c", &num_features, &dummy) != 1) {
 		ti->error = "Invalid number of features";
 		return -EINVAL;
 	}
@@ -946,6 +1028,7 @@ static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 	unsigned int nr_mirrors, m, args_used;
 	struct mirror_set *ms;
 	struct dm_dirty_log *dl;
+	char dummy;
 
 	dl = create_dirty_log(ti, argc, argv, &args_used);
 	if (!dl)
@@ -954,7 +1037,7 @@ static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 	argv += args_used;
 	argc -= args_used;
 
-	if (!argc || sscanf(argv[0], "%u", &nr_mirrors) != 1 ||
+	if (!argc || sscanf(argv[0], "%u%c", &nr_mirrors, &dummy) != 1 ||
 	    nr_mirrors < 2 || nr_mirrors > DM_KCOPYD_MAX_REGIONS + 1) {
 		ti->error = "Invalid number of mirrors";
 		dm_dirty_log_destroy(dl);
@@ -987,9 +1070,17 @@ static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 	}
 
 	ti->private = ms;
-	ti->split_io = dm_rh_get_region_size(ms->rh);
 
-	ms->kmirrord_wq = create_singlethread_workqueue("kmirrord");
+	r = dm_set_target_max_io_len(ti, dm_rh_get_region_size(ms->rh));
+	if (r)
+		goto err_free_context;
+
+	ti->num_flush_bios = 1;
+	ti->num_discard_bios = 1;
+	ti->per_bio_data_size = sizeof(struct dm_raid1_bio_record);
+	ti->discard_zeroes_data_unsupported = true;
+
+	ms->kmirrord_wq = alloc_workqueue("kmirrord", WQ_MEM_RECLAIM, 0);
 	if (!ms->kmirrord_wq) {
 		DMERR("couldn't start kmirrord");
 		r = -ENOMEM;
@@ -1022,9 +1113,11 @@ static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 		goto err_destroy_wq;
 	}
 
-	r = dm_kcopyd_client_create(DM_KCOPYD_PAGES, &ms->kcopyd_client);
-	if (r)
+	ms->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle);
+	if (IS_ERR(ms->kcopyd_client)) {
+		r = PTR_ERR(ms->kcopyd_client);
 		goto err_destroy_wq;
+	}
 
 	wakeup_mirrord(ms);
 	return 0;
@@ -1042,7 +1135,7 @@ static void mirror_dtr(struct dm_target *ti)
 
 	del_timer_sync(&ms->timer);
 	flush_workqueue(ms->kmirrord_wq);
-	flush_scheduled_work();
+	flush_work(&ms->trigger_event);
 	dm_kcopyd_client_destroy(ms->kcopyd_client);
 	destroy_workqueue(ms->kmirrord_wq);
 	free_context(ms, ti, ms->nr_mirrors);
@@ -1051,18 +1144,20 @@ static void mirror_dtr(struct dm_target *ti)
 /*
  * Mirror mapping function
  */
-static int mirror_map(struct dm_target *ti, struct bio *bio,
-		      union map_info *map_context)
+static int mirror_map(struct dm_target *ti, struct bio *bio)
 {
 	int r, rw = bio_rw(bio);
 	struct mirror *m;
 	struct mirror_set *ms = ti->private;
-	struct dm_raid1_read_record *read_record = NULL;
 	struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
+	struct dm_raid1_bio_record *bio_record =
+	  dm_per_bio_data(bio, sizeof(struct dm_raid1_bio_record));
+
+	bio_record->details.bi_bdev = NULL;
 
 	if (rw == WRITE) {
 		/* Save region for mirror_end_io() handler */
-		map_context->ll = dm_rh_bio_to_region(ms->rh, bio);
+		bio_record->write_region = dm_rh_bio_to_region(ms->rh, bio);
 		queue_bio(ms, bio, rw);
 		return DM_MAPIO_SUBMITTED;
 	}
@@ -1086,47 +1181,44 @@ static int mirror_map(struct dm_target *ti, struct bio *bio,
 	 * The region is in-sync and we can perform reads directly.
 	 * Store enough information so we can retry if it fails.
 	 */
-	m = choose_mirror(ms, bio->bi_sector);
+	m = choose_mirror(ms, bio->bi_iter.bi_sector);
 	if (unlikely(!m))
 		return -EIO;
 
-	read_record = mempool_alloc(ms->read_record_pool, GFP_NOIO);
-	if (likely(read_record)) {
-		dm_bio_record(&read_record->details, bio);
-		map_context->ptr = read_record;
-		read_record->m = m;
-	}
+	dm_bio_record(&bio_record->details, bio);
+	bio_record->m = m;
 
 	map_bio(m, bio);
 
 	return DM_MAPIO_REMAPPED;
 }
 
-static int mirror_end_io(struct dm_target *ti, struct bio *bio,
-			 int error, union map_info *map_context)
+static int mirror_end_io(struct dm_target *ti, struct bio *bio, int error)
 {
 	int rw = bio_rw(bio);
 	struct mirror_set *ms = (struct mirror_set *) ti->private;
 	struct mirror *m = NULL;
 	struct dm_bio_details *bd = NULL;
-	struct dm_raid1_read_record *read_record = map_context->ptr;
+	struct dm_raid1_bio_record *bio_record =
+	  dm_per_bio_data(bio, sizeof(struct dm_raid1_bio_record));
 
 	/*
 	 * We need to dec pending if this was a write.
 	 */
 	if (rw == WRITE) {
-		dm_rh_dec(ms->rh, map_context->ll);
+		if (!(bio->bi_rw & (REQ_FLUSH | REQ_DISCARD)))
+			dm_rh_dec(ms->rh, bio_record->write_region);
 		return error;
 	}
 
 	if (error == -EOPNOTSUPP)
 		goto out;
 
-	if ((error == -EWOULDBLOCK) && bio_rw_ahead(bio))
+	if ((error == -EWOULDBLOCK) && (bio->bi_rw & REQ_RAHEAD))
 		goto out;
 
 	if (unlikely(error)) {
-		if (!read_record) {
+		if (!bio_record->details.bi_bdev) {
 			/*
 			 * There wasn't enough memory to record necessary
 			 * information for a retry or there was no other
@@ -1136,7 +1228,7 @@ static int mirror_end_io(struct dm_target *ti, struct bio *bio,
 			return -EIO;
 		}
 
-		m = read_record->m;
+		m = bio_record->m;
 
 		DMERR("Mirror read failed from %s. Trying alternative device.",
 		      m->dev->name);
@@ -1148,22 +1240,21 @@ static int mirror_end_io(struct dm_target *ti, struct bio *bio,
 		 * mirror.
 		 */
 		if (default_ok(m) || mirror_available(ms, bio)) {
-			bd = &read_record->details;
+			bd = &bio_record->details;
 
 			dm_bio_restore(bd, bio);
-			mempool_free(read_record, ms->read_record_pool);
-			map_context->ptr = NULL;
+			bio_record->details.bi_bdev = NULL;
+
+			atomic_inc(&bio->bi_remaining);
+
 			queue_bio(ms, bio, rw);
-			return 1;
+			return DM_ENDIO_INCOMPLETE;
 		}
 		DMERR("All replicated volumes dead, failing I/O");
 	}
 
 out:
-	if (read_record) {
-		mempool_free(read_record, ms->read_record_pool);
-		map_context->ptr = NULL;
-	}
+	bio_record->details.bi_bdev = NULL;
 
 	return error;
 }
@@ -1173,9 +1264,26 @@ static void mirror_presuspend(struct dm_target *ti)
 	struct mirror_set *ms = (struct mirror_set *) ti->private;
 	struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
 
+	struct bio_list holds;
+	struct bio *bio;
+
 	atomic_set(&ms->suspend, 1);
 
 	/*
+	 * Process bios in the hold list to start recovery waiting
+	 * for bios in the hold list. After the process, no bio has
+	 * a chance to be added in the hold list because ms->suspend
+	 * is set.
+	 */
+	spin_lock_irq(&ms->lock);
+	holds = ms->holds;
+	bio_list_init(&ms->holds);
+	spin_unlock_irq(&ms->lock);
+
+	while ((bio = bio_list_pop(&holds)))
+		hold_bio(ms, bio);
+
+	/*
 	 * We must finish up all the work that we've
 	 * generated (i.e. recovery work).
 	 */
@@ -1237,14 +1345,15 @@ static char device_status_char(struct mirror *m)
 	if (!atomic_read(&(m->error_count)))
 		return 'A';
 
-	return (test_bit(DM_RAID1_WRITE_ERROR, &(m->error_type))) ? 'D' :
+	return (test_bit(DM_RAID1_FLUSH_ERROR, &(m->error_type))) ? 'F' :
+		(test_bit(DM_RAID1_WRITE_ERROR, &(m->error_type))) ? 'D' :
 		(test_bit(DM_RAID1_SYNC_ERROR, &(m->error_type))) ? 'S' :
 		(test_bit(DM_RAID1_READ_ERROR, &(m->error_type))) ? 'R' : 'U';
 }
 
 
-static int mirror_status(struct dm_target *ti, status_type_t type,
-			 char *result, unsigned int maxlen)
+static void mirror_status(struct dm_target *ti, status_type_t type,
+			  unsigned status_flags, char *result, unsigned maxlen)
 {
 	unsigned int m, sz = 0;
 	struct mirror_set *ms = (struct mirror_set *) ti->private;
@@ -1279,8 +1388,6 @@ static int mirror_status(struct dm_target *ti, status_type_t type,
 		if (ms->features & DM_RAID1_HANDLE_ERRORS)
 			DMEMIT(" 1 handle_errors");
 	}
-
-	return 0;
 }
 
 static int mirror_iterate_devices(struct dm_target *ti,
@@ -1292,14 +1399,14 @@ static int mirror_iterate_devices(struct dm_target *ti,
 
 	for (i = 0; !ret && i < ms->nr_mirrors; i++)
 		ret = fn(ti, ms->mirror[i].dev,
-			 ms->mirror[i].offset, data);
+			 ms->mirror[i].offset, ti->len, data);
 
 	return ret;
 }
 
 static struct target_type mirror_target = {
 	.name	 = "mirror",
-	.version = {1, 12, 0},
+	.version = {1, 13, 2},
 	.module	 = THIS_MODULE,
 	.ctr	 = mirror_ctr,
 	.dtr	 = mirror_dtr,
@@ -1316,13 +1423,6 @@ static int __init dm_mirror_init(void)
 {
 	int r;
 
-	_dm_raid1_read_record_cache = KMEM_CACHE(dm_raid1_read_record, 0);
-	if (!_dm_raid1_read_record_cache) {
-		DMERR("Can't allocate dm_raid1_read_record cache");
-		r = -ENOMEM;
-		goto bad_cache;
-	}
-
 	r = dm_register_target(&mirror_target);
 	if (r < 0) {
 		DMERR("Failed to register mirror target");
@@ -1332,15 +1432,12 @@ static int __init dm_mirror_init(void)
 	return 0;
 
 bad_target:
-	kmem_cache_destroy(_dm_raid1_read_record_cache);
-bad_cache:
 	return r;
 }
 
 static void __exit dm_mirror_exit(void)
 {
 	dm_unregister_target(&mirror_target);
-	kmem_cache_destroy(_dm_raid1_read_record_cache);
 }
 
 /* Module hooks */