Merge tag 'for-4.16/dm-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm

Pull device mapper updates from Mike Snitzer:

 - DM core fixes to ensure that bio submission follows a depth-first
   tree walk; this is critical to allow forward progress without the
   need to use the bioset's BIOSET_NEED_RESCUER.

 - Remove DM core's BIOSET_NEED_RESCUER based dm_offload infrastructure.

 - DM core cleanups and improvements to make bio-based DM more efficient
   (e.g. reduced memory footprint as well leveraging per-bio-data more).

 - Introduce new bio-based mode (DM_TYPE_NVME_BIO_BASED) that leverages
   the more direct IO submission path in the block layer; this mode is
   used by DM multipath and also optimizes targets like DM thin-pool
   that stack directly on NVMe data device.

 - DM multipath improvements to factor out legacy SCSI-only (e.g.
   scsi_dh) code paths to allow for more optimized support for NVMe
   multipath.

 - A fix for DM multipath path selectors (service-time and queue-length)
   to select paths in a more balanced way; largely academic but doesn't
   hurt.

 - Numerous DM raid target fixes and improvements.

 - Add a new DM "unstriped" target that enables Intel to workaround
   firmware limitations in some NVMe drives that are striped internally
   (this target also works when stacked above the DM "striped" target).

 - Various Documentation fixes and improvements.

 - Misc cleanups and fixes across various DM infrastructure and targets
   (e.g. bufio, flakey, log-writes, snapshot).

* tag 'for-4.16/dm-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm: (69 commits)
  dm cache: Documentation: update default migration_throttling value
  dm mpath selector: more evenly distribute ties
  dm unstripe: fix target length versus number of stripes size check
  dm thin: fix trailing semicolon in __remap_and_issue_shared_cell
  dm table: fix NVMe bio-based dm_table_determine_type() validation
  dm: various cleanups to md->queue initialization code
  dm mpath: delay the retry of a request if the target responded as busy
  dm mpath: return DM_MAPIO_DELAY_REQUEUE if QUEUE_IO or PG_INIT_REQUIRED
  dm mpath: return DM_MAPIO_REQUEUE on blk-mq rq allocation failure
  dm log writes: fix max length used for kstrndup
  dm: backfill missing calls to mutex_destroy()
  dm snapshot: use mutex instead of rw_semaphore
  dm flakey: check for null arg_name in parse_features()
  dm thin: extend thinpool status format string with omitted fields
  dm thin: fixes in thin-provisioning.txt
  dm thin: document representation of <highest mapped sector> when there is none
  dm thin: fix documentation relative to low water mark threshold
  dm cache: be consistent in specifying sectors and SI units in cache.txt
  dm cache: delete obsoleted paragraph in cache.txt
  dm cache: fix grammar in cache-policies.txt
  ...

24 files changed, 1252 insertions, 612 deletions

diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig
index 83b9362be09c..2c8ac3688815 100644
--- a/drivers/md/Kconfig
+++ b/drivers/md/Kconfig
@@ -269,6 +269,13 @@ config DM_BIO_PRISON
 
 source "drivers/md/persistent-data/Kconfig"
 
+config DM_UNSTRIPED
+       tristate "Unstriped target"
+       depends on BLK_DEV_DM
+       ---help---
+	  Unstripes I/O so it is issued solely on a single drive in a HW
+	  RAID0 or dm-striped target.
+
 config DM_CRYPT
 	tristate "Crypt target support"
 	depends on BLK_DEV_DM
diff --git a/drivers/md/Makefile b/drivers/md/Makefile
index f701bb211783..63255f3ebd97 100644
--- a/drivers/md/Makefile
+++ b/drivers/md/Makefile
@@ -43,6 +43,7 @@ obj-$(CONFIG_BCACHE)		+= bcache/
 obj-$(CONFIG_BLK_DEV_MD)	+= md-mod.o
 obj-$(CONFIG_BLK_DEV_DM)	+= dm-mod.o
 obj-$(CONFIG_BLK_DEV_DM_BUILTIN) += dm-builtin.o
+obj-$(CONFIG_DM_UNSTRIPED)	+= dm-unstripe.o
 obj-$(CONFIG_DM_BUFIO)		+= dm-bufio.o
 obj-$(CONFIG_DM_BIO_PRISON)	+= dm-bio-prison.o
 obj-$(CONFIG_DM_CRYPT)		+= dm-crypt.o
diff --git a/drivers/md/dm-bufio.c b/drivers/md/dm-bufio.c
index c546b567f3b5..414c9af54ded 100644
--- a/drivers/md/dm-bufio.c
+++ b/drivers/md/dm-bufio.c
@@ -662,7 +662,7 @@ static void submit_io(struct dm_buffer *b, int rw, bio_end_io_t *end_io)
 
 	sector = (b->block << b->c->sectors_per_block_bits) + b->c->start;
 
-	if (rw != WRITE) {
+	if (rw != REQ_OP_WRITE) {
 		n_sectors = 1 << b->c->sectors_per_block_bits;
 		offset = 0;
 	} else {
@@ -740,7 +740,7 @@ static void __write_dirty_buffer(struct dm_buffer *b,
 	b->write_end = b->dirty_end;
 
 	if (!write_list)
-		submit_io(b, WRITE, write_endio);
+		submit_io(b, REQ_OP_WRITE, write_endio);
 	else
 		list_add_tail(&b->write_list, write_list);
 }
@@ -753,7 +753,7 @@ static void __flush_write_list(struct list_head *write_list)
 		struct dm_buffer *b =
 			list_entry(write_list->next, struct dm_buffer, write_list);
 		list_del(&b->write_list);
-		submit_io(b, WRITE, write_endio);
+		submit_io(b, REQ_OP_WRITE, write_endio);
 		cond_resched();
 	}
 	blk_finish_plug(&plug);
@@ -1123,7 +1123,7 @@ static void *new_read(struct dm_bufio_client *c, sector_t block,
 		return NULL;
 
 	if (need_submit)
-		submit_io(b, READ, read_endio);
+		submit_io(b, REQ_OP_READ, read_endio);
 
 	wait_on_bit_io(&b->state, B_READING, TASK_UNINTERRUPTIBLE);
 
@@ -1193,7 +1193,7 @@ void dm_bufio_prefetch(struct dm_bufio_client *c,
 			dm_bufio_unlock(c);
 
 			if (need_submit)
-				submit_io(b, READ, read_endio);
+				submit_io(b, REQ_OP_READ, read_endio);
 			dm_bufio_release(b);
 
 			cond_resched();
@@ -1454,7 +1454,7 @@ retry:
 		old_block = b->block;
 		__unlink_buffer(b);
 		__link_buffer(b, new_block, b->list_mode);
-		submit_io(b, WRITE, write_endio);
+		submit_io(b, REQ_OP_WRITE, write_endio);
 		wait_on_bit_io(&b->state, B_WRITING,
 			       TASK_UNINTERRUPTIBLE);
 		__unlink_buffer(b);
@@ -1716,7 +1716,7 @@ struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsign
 			if (!DM_BUFIO_CACHE_NAME(c)) {
 				r = -ENOMEM;
 				mutex_unlock(&dm_bufio_clients_lock);
-				goto bad_cache;
+				goto bad;
 			}
 		}
 
@@ -1727,7 +1727,7 @@ struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsign
 			if (!DM_BUFIO_CACHE(c)) {
 				r = -ENOMEM;
 				mutex_unlock(&dm_bufio_clients_lock);
-				goto bad_cache;
+				goto bad;
 			}
 		}
 	}
@@ -1738,27 +1738,28 @@ struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsign
 
 		if (!b) {
 			r = -ENOMEM;
-			goto bad_buffer;
+			goto bad;
 		}
 		__free_buffer_wake(b);
 	}
 
+	c->shrinker.count_objects = dm_bufio_shrink_count;
+	c->shrinker.scan_objects = dm_bufio_shrink_scan;
+	c->shrinker.seeks = 1;
+	c->shrinker.batch = 0;
+	r = register_shrinker(&c->shrinker);
+	if (r)
+		goto bad;
+
 	mutex_lock(&dm_bufio_clients_lock);
 	dm_bufio_client_count++;
 	list_add(&c->client_list, &dm_bufio_all_clients);
 	__cache_size_refresh();
 	mutex_unlock(&dm_bufio_clients_lock);
 
-	c->shrinker.count_objects = dm_bufio_shrink_count;
-	c->shrinker.scan_objects = dm_bufio_shrink_scan;
-	c->shrinker.seeks = 1;
-	c->shrinker.batch = 0;
-	register_shrinker(&c->shrinker);
-
 	return c;
 
-bad_buffer:
-bad_cache:
+bad:
 	while (!list_empty(&c->reserved_buffers)) {
 		struct dm_buffer *b = list_entry(c->reserved_buffers.next,
 						 struct dm_buffer, lru_list);
@@ -1767,6 +1768,7 @@ bad_cache:
 	}
 	dm_io_client_destroy(c->dm_io);
 bad_dm_io:
+	mutex_destroy(&c->lock);
 	kfree(c);
 bad_client:
 	return ERR_PTR(r);
@@ -1811,6 +1813,7 @@ void dm_bufio_client_destroy(struct dm_bufio_client *c)
 		BUG_ON(c->n_buffers[i]);
 
 	dm_io_client_destroy(c->dm_io);
+	mutex_destroy(&c->lock);
 	kfree(c);
 }
 EXPORT_SYMBOL_GPL(dm_bufio_client_destroy);
diff --git a/drivers/md/dm-core.h b/drivers/md/dm-core.h
index 6a14f945783c..3222e21cbbf8 100644
--- a/drivers/md/dm-core.h
+++ b/drivers/md/dm-core.h
@@ -91,8 +91,7 @@ struct mapped_device {
 	/*
 	 * io objects are allocated from here.
 	 */
-	mempool_t *io_pool;
-
+	struct bio_set *io_bs;
 	struct bio_set *bs;
 
 	/*
@@ -130,8 +129,6 @@ struct mapped_device {
 	struct srcu_struct io_barrier;
 };
 
-void dm_init_md_queue(struct mapped_device *md);
-void dm_init_normal_md_queue(struct mapped_device *md);
 int md_in_flight(struct mapped_device *md);
 void disable_write_same(struct mapped_device *md);
 void disable_write_zeroes(struct mapped_device *md);
diff --git a/drivers/md/dm-crypt.c b/drivers/md/dm-crypt.c
index 2ad429100d25..8168f737590e 100644
--- a/drivers/md/dm-crypt.c
+++ b/drivers/md/dm-crypt.c
@@ -2193,6 +2193,8 @@ static void crypt_dtr(struct dm_target *ti)
 	kzfree(cc->cipher_auth);
 	kzfree(cc->authenc_key);
 
+	mutex_destroy(&cc->bio_alloc_lock);
+
 	/* Must zero key material before freeing */
 	kzfree(cc);
 }
@@ -2702,8 +2704,7 @@ static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 		goto bad;
 	}
 
-	cc->bs = bioset_create(MIN_IOS, 0, (BIOSET_NEED_BVECS |
-					    BIOSET_NEED_RESCUER));
+	cc->bs = bioset_create(MIN_IOS, 0, BIOSET_NEED_BVECS);
 	if (!cc->bs) {
 		ti->error = "Cannot allocate crypt bioset";
 		goto bad;
diff --git a/drivers/md/dm-delay.c b/drivers/md/dm-delay.c
index 288386bfbfb5..1783d80c9cad 100644
--- a/drivers/md/dm-delay.c
+++ b/drivers/md/dm-delay.c
@@ -229,6 +229,8 @@ static void delay_dtr(struct dm_target *ti)
 	if (dc->dev_write)
 		dm_put_device(ti, dc->dev_write);
 
+	mutex_destroy(&dc->timer_lock);
+
 	kfree(dc);
 }
 
diff --git a/drivers/md/dm-flakey.c b/drivers/md/dm-flakey.c
index b82cb1ab1eaa..1b907b15f5c3 100644
--- a/drivers/md/dm-flakey.c
+++ b/drivers/md/dm-flakey.c
@@ -70,6 +70,11 @@ static int parse_features(struct dm_arg_set *as, struct flakey_c *fc,
 		arg_name = dm_shift_arg(as);
 		argc--;
 
+		if (!arg_name) {
+			ti->error = "Insufficient feature arguments";
+			return -EINVAL;
+		}
+
 		/*
 		 * drop_writes
 		 */
diff --git a/drivers/md/dm-io.c b/drivers/md/dm-io.c
index b4357ed4d541..a8d914d5abbe 100644
--- a/drivers/md/dm-io.c
+++ b/drivers/md/dm-io.c
@@ -58,8 +58,7 @@ struct dm_io_client *dm_io_client_create(void)
 	if (!client->pool)
 		goto bad;
 
-	client->bios = bioset_create(min_ios, 0, (BIOSET_NEED_BVECS |
-						  BIOSET_NEED_RESCUER));
+	client->bios = bioset_create(min_ios, 0, BIOSET_NEED_BVECS);
 	if (!client->bios)
 		goto bad;
 
diff --git a/drivers/md/dm-kcopyd.c b/drivers/md/dm-kcopyd.c
index eb45cc3df31d..e6e7c686646d 100644
--- a/drivers/md/dm-kcopyd.c
+++ b/drivers/md/dm-kcopyd.c
@@ -477,8 +477,10 @@ static int run_complete_job(struct kcopyd_job *job)
 	 * If this is the master job, the sub jobs have already
 	 * completed so we can free everything.
 	 */
-	if (job->master_job == job)
+	if (job->master_job == job) {
+		mutex_destroy(&job->lock);
 		mempool_free(job, kc->job_pool);
+	}
 	fn(read_err, write_err, context);
 
 	if (atomic_dec_and_test(&kc->nr_jobs))
@@ -750,6 +752,7 @@ int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
 	 * followed by SPLIT_COUNT sub jobs.
 	 */
 	job = mempool_alloc(kc->job_pool, GFP_NOIO);
+	mutex_init(&job->lock);
 
 	/*
 	 * set up for the read.
@@ -811,7 +814,6 @@ int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
 	if (job->source.count <= SUB_JOB_SIZE)
 		dispatch_job(job);
 	else {
-		mutex_init(&job->lock);
 		job->progress = 0;
 		split_job(job);
 	}
diff --git a/drivers/md/dm-log-writes.c b/drivers/md/dm-log-writes.c
index 189badbeddaf..3362d866793b 100644
--- a/drivers/md/dm-log-writes.c
+++ b/drivers/md/dm-log-writes.c
@@ -594,7 +594,7 @@ static int log_mark(struct log_writes_c *lc, char *data)
 		return -ENOMEM;
 	}
 
-	block->data = kstrndup(data, maxsize, GFP_KERNEL);
+	block->data = kstrndup(data, maxsize - 1, GFP_KERNEL);
 	if (!block->data) {
 		DMERR("Error copying mark data");
 		kfree(block);
diff --git a/drivers/md/dm-mpath.c b/drivers/md/dm-mpath.c
index ef57c6d1c887..7d3e572072f5 100644
--- a/drivers/md/dm-mpath.c
+++ b/drivers/md/dm-mpath.c
@@ -64,36 +64,30 @@ struct priority_group {
 
 /* Multipath context */
 struct multipath {
-	struct list_head list;
-	struct dm_target *ti;
-
-	const char *hw_handler_name;
-	char *hw_handler_params;
+	unsigned long flags;		/* Multipath state flags */
 
 	spinlock_t lock;
-
-	unsigned nr_priority_groups;
-	struct list_head priority_groups;
-
-	wait_queue_head_t pg_init_wait;	/* Wait for pg_init completion */
+	enum dm_queue_mode queue_mode;
 
 	struct pgpath *current_pgpath;
 	struct priority_group *current_pg;
 	struct priority_group *next_pg;	/* Switch to this PG if set */
 
-	unsigned long flags;		/* Multipath state flags */
+	atomic_t nr_valid_paths;	/* Total number of usable paths */
+	unsigned nr_priority_groups;
+	struct list_head priority_groups;
 
+	const char *hw_handler_name;
+	char *hw_handler_params;
+	wait_queue_head_t pg_init_wait;	/* Wait for pg_init completion */
 	unsigned pg_init_retries;	/* Number of times to retry pg_init */
 	unsigned pg_init_delay_msecs;	/* Number of msecs before pg_init retry */
-
-	atomic_t nr_valid_paths;	/* Total number of usable paths */
 	atomic_t pg_init_in_progress;	/* Only one pg_init allowed at once */
 	atomic_t pg_init_count;		/* Number of times pg_init called */
 
-	enum dm_queue_mode queue_mode;
-
 	struct mutex work_mutex;
 	struct work_struct trigger_event;
+	struct dm_target *ti;
 
 	struct work_struct process_queued_bios;
 	struct bio_list queued_bios;
@@ -135,10 +129,10 @@ static struct pgpath *alloc_pgpath(void)
 {
 	struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL);
 
-	if (pgpath) {
-		pgpath->is_active = true;
-		INIT_DELAYED_WORK(&pgpath->activate_path, activate_path_work);
-	}
+	if (!pgpath)
+		return NULL;
+
+	pgpath->is_active = true;
 
 	return pgpath;
 }
@@ -193,13 +187,8 @@ static struct multipath *alloc_multipath(struct dm_target *ti)
 	if (m) {
 		INIT_LIST_HEAD(&m->priority_groups);
 		spin_lock_init(&m->lock);
-		set_bit(MPATHF_QUEUE_IO, &m->flags);
 		atomic_set(&m->nr_valid_paths, 0);
-		atomic_set(&m->pg_init_in_progress, 0);
-		atomic_set(&m->pg_init_count, 0);
-		m->pg_init_delay_msecs = DM_PG_INIT_DELAY_DEFAULT;
 		INIT_WORK(&m->trigger_event, trigger_event);
-		init_waitqueue_head(&m->pg_init_wait);
 		mutex_init(&m->work_mutex);
 
 		m->queue_mode = DM_TYPE_NONE;
@@ -221,13 +210,26 @@ static int alloc_multipath_stage2(struct dm_target *ti, struct multipath *m)
 			m->queue_mode = DM_TYPE_MQ_REQUEST_BASED;
 		else
 			m->queue_mode = DM_TYPE_REQUEST_BASED;
-	} else if (m->queue_mode == DM_TYPE_BIO_BASED) {
+
+	} else if (m->queue_mode == DM_TYPE_BIO_BASED ||
+		   m->queue_mode == DM_TYPE_NVME_BIO_BASED) {
 		INIT_WORK(&m->process_queued_bios, process_queued_bios);
-		/*
-		 * bio-based doesn't support any direct scsi_dh management;
-		 * it just discovers if a scsi_dh is attached.
-		 */
-		set_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags);
+
+		if (m->queue_mode == DM_TYPE_BIO_BASED) {
+			/*
+			 * bio-based doesn't support any direct scsi_dh management;
+			 * it just discovers if a scsi_dh is attached.
+			 */
+			set_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags);
+		}
+	}
+
+	if (m->queue_mode != DM_TYPE_NVME_BIO_BASED) {
+		set_bit(MPATHF_QUEUE_IO, &m->flags);
+		atomic_set(&m->pg_init_in_progress, 0);
+		atomic_set(&m->pg_init_count, 0);
+		m->pg_init_delay_msecs = DM_PG_INIT_DELAY_DEFAULT;
+		init_waitqueue_head(&m->pg_init_wait);
 	}
 
 	dm_table_set_type(ti->table, m->queue_mode);
@@ -246,6 +248,7 @@ static void free_multipath(struct multipath *m)
 
 	kfree(m->hw_handler_name);
 	kfree(m->hw_handler_params);
+	mutex_destroy(&m->work_mutex);
 	kfree(m);
 }
 
@@ -264,29 +267,23 @@ static struct dm_mpath_io *get_mpio_from_bio(struct bio *bio)
 	return dm_per_bio_data(bio, multipath_per_bio_data_size());
 }
 
-static struct dm_bio_details *get_bio_details_from_bio(struct bio *bio)
+static struct dm_bio_details *get_bio_details_from_mpio(struct dm_mpath_io *mpio)
 {
 	/* dm_bio_details is immediately after the dm_mpath_io in bio's per-bio-data */
-	struct dm_mpath_io *mpio = get_mpio_from_bio(bio);
 	void *bio_details = mpio + 1;
-
 	return bio_details;
 }
 
-static void multipath_init_per_bio_data(struct bio *bio, struct dm_mpath_io **mpio_p,
-					struct dm_bio_details **bio_details_p)
+static void multipath_init_per_bio_data(struct bio *bio, struct dm_mpath_io **mpio_p)
 {
 	struct dm_mpath_io *mpio = get_mpio_from_bio(bio);
-	struct dm_bio_details *bio_details = get_bio_details_from_bio(bio);
+	struct dm_bio_details *bio_details = get_bio_details_from_mpio(mpio);
 
-	memset(mpio, 0, sizeof(*mpio));
-	memset(bio_details, 0, sizeof(*bio_details));
-	dm_bio_record(bio_details, bio);
+	mpio->nr_bytes = bio->bi_iter.bi_size;
+	mpio->pgpath = NULL;
+	*mpio_p = mpio;
 
-	if (mpio_p)
-		*mpio_p = mpio;
-	if (bio_details_p)
-		*bio_details_p = bio_details;
+	dm_bio_record(bio_details, bio);
 }
 
 /*-----------------------------------------------
@@ -340,6 +337,9 @@ static void __switch_pg(struct multipath *m, struct priority_group *pg)
 {
 	m->current_pg = pg;
 
+	if (m->queue_mode == DM_TYPE_NVME_BIO_BASED)
+		return;
+
 	/* Must we initialise the PG first, and queue I/O till it's ready? */
 	if (m->hw_handler_name) {
 		set_bit(MPATHF_PG_INIT_REQUIRED, &m->flags);
@@ -385,7 +385,8 @@ static struct pgpath *choose_pgpath(struct multipath *m, size_t nr_bytes)
 	unsigned bypassed = 1;
 
 	if (!atomic_read(&m->nr_valid_paths)) {
-		clear_bit(MPATHF_QUEUE_IO, &m->flags);
+		if (m->queue_mode != DM_TYPE_NVME_BIO_BASED)
+			clear_bit(MPATHF_QUEUE_IO, &m->flags);
 		goto failed;
 	}
 
@@ -516,12 +517,10 @@ static int multipath_clone_and_map(struct dm_target *ti, struct request *rq,
 		return DM_MAPIO_KILL;
 	} else if (test_bit(MPATHF_QUEUE_IO, &m->flags) ||
 		   test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) {
-		if (pg_init_all_paths(m))
-			return DM_MAPIO_DELAY_REQUEUE;
-		return DM_MAPIO_REQUEUE;
+		pg_init_all_paths(m);
+		return DM_MAPIO_DELAY_REQUEUE;
 	}
 
-	memset(mpio, 0, sizeof(*mpio));
 	mpio->pgpath = pgpath;
 	mpio->nr_bytes = nr_bytes;
 
@@ -530,12 +529,23 @@ static int multipath_clone_and_map(struct dm_target *ti, struct request *rq,
 	clone = blk_get_request(q, rq->cmd_flags | REQ_NOMERGE, GFP_ATOMIC);
 	if (IS_ERR(clone)) {
 		/* EBUSY, ENODEV or EWOULDBLOCK: requeue */
-		bool queue_dying = blk_queue_dying(q);
-		if (queue_dying) {
+		if (blk_queue_dying(q)) {
 			atomic_inc(&m->pg_init_in_progress);
 			activate_or_offline_path(pgpath);
+			return DM_MAPIO_DELAY_REQUEUE;
 		}
-		return DM_MAPIO_DELAY_REQUEUE;
+
+		/*
+		 * blk-mq's SCHED_RESTART can cover this requeue, so we
+		 * needn't deal with it by DELAY_REQUEUE. More importantly,
+		 * we have to return DM_MAPIO_REQUEUE so that blk-mq can
+		 * get the queue busy feedback (via BLK_STS_RESOURCE),
+		 * otherwise I/O merging can suffer.
+		 */
+		if (q->mq_ops)
+			return DM_MAPIO_REQUEUE;
+		else
+			return DM_MAPIO_DELAY_REQUEUE;
 	}
 	clone->bio = clone->biotail = NULL;
 	clone->rq_disk = bdev->bd_disk;
@@ -557,9 +567,9 @@ static void multipath_release_clone(struct request *clone)
 /*
  * Map cloned bios (bio-based multipath)
  */
-static int __multipath_map_bio(struct multipath *m, struct bio *bio, struct dm_mpath_io *mpio)
+
+static struct pgpath *__map_bio(struct multipath *m, struct bio *bio)
 {
-	size_t nr_bytes = bio->bi_iter.bi_size;
 	struct pgpath *pgpath;
 	unsigned long flags;
 	bool queue_io;
@@ -568,7 +578,7 @@ static int __multipath_map_bio(struct multipath *m, struct bio *bio, struct dm_m
 	pgpath = READ_ONCE(m->current_pgpath);
 	queue_io = test_bit(MPATHF_QUEUE_IO, &m->flags);
 	if (!pgpath || !queue_io)
-		pgpath = choose_pgpath(m, nr_bytes);
+		pgpath = choose_pgpath(m, bio->bi_iter.bi_size);
 
 	if ((pgpath && queue_io) ||
 	    (!pgpath && test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))) {
@@ -576,14 +586,62 @@ static int __multipath_map_bio(struct multipath *m, struct bio *bio, struct dm_m
 		spin_lock_irqsave(&m->lock, flags);
 		bio_list_add(&m->queued_bios, bio);
 		spin_unlock_irqrestore(&m->lock, flags);
+
 		/* PG_INIT_REQUIRED cannot be set without QUEUE_IO */
 		if (queue_io || test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags))
 			pg_init_all_paths(m);
 		else if (!queue_io)
 			queue_work(kmultipathd, &m->process_queued_bios);
-		return DM_MAPIO_SUBMITTED;
+
+		return ERR_PTR(-EAGAIN);
 	}
 
+	return pgpath;
+}
+
+static struct pgpath *__map_bio_nvme(struct multipath *m, struct bio *bio)
+{
+	struct pgpath *pgpath;
+	unsigned long flags;
+
+	/* Do we need to select a new pgpath? */
+	/*
+	 * FIXME: currently only switching path if no path (due to failure, etc)
+	 * - which negates the point of using a path selector
+	 */
+	pgpath = READ_ONCE(m->current_pgpath);
+	if (!pgpath)
+		pgpath = choose_pgpath(m, bio->bi_iter.bi_size);
+
+	if (!pgpath) {
+		if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) {
+			/* Queue for the daemon to resubmit */
+			spin_lock_irqsave(&m->lock, flags);
+			bio_list_add(&m->queued_bios, bio);
+			spin_unlock_irqrestore(&m->lock, flags);
+			queue_work(kmultipathd, &m->process_queued_bios);
+
+			return ERR_PTR(-EAGAIN);
+		}
+		return NULL;
+	}
+
+	return pgpath;
+}
+
+static int __multipath_map_bio(struct multipath *m, struct bio *bio,
+			       struct dm_mpath_io *mpio)
+{
+	struct pgpath *pgpath;
+
+	if (m->queue_mode == DM_TYPE_NVME_BIO_BASED)
+		pgpath = __map_bio_nvme(m, bio);
+	else
+		pgpath = __map_bio(m, bio);
+
+	if (IS_ERR(pgpath))
+		return DM_MAPIO_SUBMITTED;
+
 	if (!pgpath) {
 		if (must_push_back_bio(m))
 			return DM_MAPIO_REQUEUE;
@@ -592,7 +650,6 @@ static int __multipath_map_bio(struct multipath *m, struct bio *bio, struct dm_m
 	}
 
 	mpio->pgpath = pgpath;
-	mpio->nr_bytes = nr_bytes;
 
 	bio->bi_status = 0;
 	bio_set_dev(bio, pgpath->path.dev->bdev);
@@ -601,7 +658,7 @@ static int __multipath_map_bio(struct multipath *m, struct bio *bio, struct dm_m
 	if (pgpath->pg->ps.type->start_io)
 		pgpath->pg->ps.type->start_io(&pgpath->pg->ps,
 					      &pgpath->path,
-					      nr_bytes);
+					      mpio->nr_bytes);
 	return DM_MAPIO_REMAPPED;
 }
 
@@ -610,8 +667,7 @@ static int multipath_map_bio(struct dm_target *ti, struct bio *bio)
 	struct multipath *m = ti->private;
 	struct dm_mpath_io *mpio = NULL;
 
-	multipath_init_per_bio_data(bio, &mpio, NULL);
-
+	multipath_init_per_bio_data(bio, &mpio);
 	return __multipath_map_bio(m, bio, mpio);
 }
 
@@ -619,7 +675,8 @@ static void process_queued_io_list(struct multipath *m)
 {
 	if (m->queue_mode == DM_TYPE_MQ_REQUEST_BASED)
 		dm_mq_kick_requeue_list(dm_table_get_md(m->ti->table));
-	else if (m->queue_mode == DM_TYPE_BIO_BASED)
+	else if (m->queue_mode == DM_TYPE_BIO_BASED ||
+		 m->queue_mode == DM_TYPE_NVME_BIO_BASED)
 		queue_work(kmultipathd, &m->process_queued_bios);
 }
 
@@ -649,7 +706,9 @@ static void process_queued_bios(struct work_struct *work)
 
 	blk_start_plug(&plug);
 	while ((bio = bio_list_pop(&bios))) {
-		r = __multipath_map_bio(m, bio, get_mpio_from_bio(bio));
+		struct dm_mpath_io *mpio = get_mpio_from_bio(bio);
+		dm_bio_restore(get_bio_details_from_mpio(mpio), bio);
+		r = __multipath_map_bio(m, bio, mpio);
 		switch (r) {
 		case DM_MAPIO_KILL:
 			bio->bi_status = BLK_STS_IOERR;
@@ -752,34 +811,11 @@ static int parse_path_selector(struct dm_arg_set *as, struct priority_group *pg,
 	return 0;
 }
 
-static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps,
-			       struct dm_target *ti)
+static int setup_scsi_dh(struct block_device *bdev, struct multipath *m, char **error)
 {
-	int r;
-	struct pgpath *p;
-	struct multipath *m = ti->private;
-	struct request_queue *q = NULL;
+	struct request_queue *q = bdev_get_queue(bdev);
 	const char *attached_handler_name;
-
-	/* we need at least a path arg */
-	if (as->argc < 1) {
-		ti->error = "no device given";
-		return ERR_PTR(-EINVAL);
-	}
-
-	p = alloc_pgpath();
-	if (!p)
-		return ERR_PTR(-ENOMEM);
-
-	r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table),
-			  &p->path.dev);
-	if (r) {
-		ti->error = "error getting device";
-		goto bad;
-	}
-
-	if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags) || m->hw_handler_name)
-		q = bdev_get_queue(p->path.dev->bdev);
+	int r;
 
 	if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags)) {
 retain:
@@ -811,26 +847,59 @@ retain:
 			char b[BDEVNAME_SIZE];
 
 			printk(KERN_INFO "dm-mpath: retaining handler on device %s\n",
-				bdevname(p->path.dev->bdev, b));
+			       bdevname(bdev, b));
 			goto retain;
 		}
 		if (r < 0) {
-			ti->error = "error attaching hardware handler";
-			dm_put_device(ti, p->path.dev);
-			goto bad;
+			*error = "error attaching hardware handler";
+			return r;
 		}
 
 		if (m->hw_handler_params) {
 			r = scsi_dh_set_params(q, m->hw_handler_params);
 			if (r < 0) {
-				ti->error = "unable to set hardware "
-							"handler parameters";
-				dm_put_device(ti, p->path.dev);
-				goto bad;
+				*error = "unable to set hardware handler parameters";
+				return r;
 			}
 		}
 	}
 
+	return 0;
+}
+
+static struct pgpath *parse_path(struct dm_arg_set *as, struct path_selector *ps,
+				 struct dm_target *ti)
+{
+	int r;
+	struct pgpath *p;
+	struct multipath *m = ti->private;
+
+	/* we need at least a path arg */
+	if (as->argc < 1) {
+		ti->error = "no device given";
+		return ERR_PTR(-EINVAL);
+	}
+
+	p = alloc_pgpath();
+	if (!p)
+		return ERR_PTR(-ENOMEM);
+
+	r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table),
+			  &p->path.dev);
+	if (r) {
+		ti->error = "error getting device";
+		goto bad;
+	}
+
+	if (m->queue_mode != DM_TYPE_NVME_BIO_BASED) {
+		INIT_DELAYED_WORK(&p->activate_path, activate_path_work);
+		r = setup_scsi_dh(p->path.dev->bdev, m, &ti->error);
+		if (r) {
+			dm_put_device(ti, p->path.dev);
+			goto bad;
+		}
+	}
+
 	r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error);
 	if (r) {
 		dm_put_device(ti, p->path.dev);
@@ -838,7 +907,6 @@ retain:
 	}
 
 	return p;
-
  bad:
 	free_pgpath(p);
 	return ERR_PTR(r);
@@ -933,7 +1001,8 @@ static int parse_hw_handler(struct dm_arg_set *as, struct multipath *m)
 	if (!hw_argc)
 		return 0;
 
-	if (m->queue_mode == DM_TYPE_BIO_BASED) {
+	if (m->queue_mode == DM_TYPE_BIO_BASED ||
+	    m->queue_mode == DM_TYPE_NVME_BIO_BASED) {
 		dm_consume_args(as, hw_argc);
 		DMERR("bio-based multipath doesn't allow hardware handler args");
 		return 0;
@@ -1022,6 +1091,8 @@ static int parse_features(struct dm_arg_set *as, struct multipath *m)
 
 			if (!strcasecmp(queue_mode_name, "bio"))
 				m->queue_mode = DM_TYPE_BIO_BASED;
+			else if (!strcasecmp(queue_mode_name, "nvme"))
+				m->queue_mode = DM_TYPE_NVME_BIO_BASED;
 			else if (!strcasecmp(queue_mode_name, "rq"))
 				m->queue_mode = DM_TYPE_REQUEST_BASED;
 			else if (!strcasecmp(queue_mode_name, "mq"))
@@ -1122,7 +1193,7 @@ static int multipath_ctr(struct dm_target *ti, unsigned argc, char **argv)
 	ti->num_discard_bios = 1;
 	ti->num_write_same_bios = 1;
 	ti->num_write_zeroes_bios = 1;
-	if (m->queue_mode == DM_TYPE_BIO_BASED)
+	if (m->queue_mode == DM_TYPE_BIO_BASED || m->queue_mode == DM_TYPE_NVME_BIO_BASED)
 		ti->per_io_data_size = multipath_per_bio_data_size();
 	else
 		ti->per_io_data_size = sizeof(struct dm_mpath_io);
@@ -1151,16 +1222,19 @@ static void multipath_wait_for_pg_init_completion(struct multipath *m)
 
 static void flush_multipath_work(struct multipath *m)
 {
-	set_bit(MPATHF_PG_INIT_DISABLED, &m->flags);
-	smp_mb__after_atomic();
+	if (m->hw_handler_name) {
+		set_bit(MPATHF_PG_INIT_DISABLED, &m->flags);
+		smp_mb__after_atomic();