Merge tag 'dm-4.8-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm

Pull device mapper updates from Mike Snitzer:

 - initially based on Jens' 'for-4.8/core' (given all the flag churn)
   and later merged with 'for-4.8/core' to pickup the QUEUE_FLAG_DAX
   commits that DM depends on to provide its DAX support

 - clean up the bio-based vs request-based DM core code by moving the
   request-based DM core code out to dm-rq.[hc]

 - reinstate bio-based support in the DM multipath target (done with the
   idea that fast storage like NVMe over Fabrics could benefit) -- while
   preserving support for request_fn and blk-mq request-based DM mpath

 - SCSI and DM multipath persistent reservation fixes that were
   coordinated with Martin Petersen.

 - the DM raid target saw the most extensive change this cycle; it now
   provides reshape and takeover support (by layering ontop of the
   corresponding MD capabilities)

 - DAX support for DM core and the linear, stripe and error targets

 - a DM thin-provisioning block discard vs allocation race fix that
   addresses potential for corruption

 - a stable fix for DM verity-fec's block calculation during decode

 - a few cleanups and fixes to DM core and various targets

* tag 'dm-4.8-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm: (73 commits)
  dm: allow bio-based table to be upgraded to bio-based with DAX support
  dm snap: add fake origin_direct_access
  dm stripe: add DAX support
  dm error: add DAX support
  dm linear: add DAX support
  dm: add infrastructure for DAX support
  dm thin: fix a race condition between discarding and provisioning a block
  dm btree: fix a bug in dm_btree_find_next_single()
  dm raid: fix random optimal_io_size for raid0
  dm raid: address checkpatch.pl complaints
  dm: call PR reserve/unreserve on each underlying device
  sd: don't use the ALL_TG_PT bit for reservations
  dm: fix second blk_delay_queue() parameter to be in msec units not jiffies
  dm raid: change logical functions to actually return bool
  dm raid: use rdev_for_each in status
  dm raid: use rs->raid_disks to avoid memory leaks on free
  dm raid: support delta_disks for raid1, fix table output
  dm raid: enhance reshape check and factor out reshape setup
  dm raid: allow resize during recovery
  dm raid: fix rs_is_recovering() to allow for lvextend
  ...

Documentation/device-mapper/dm-raid.txt

@@ -14,8 +14,12 @@ The target is named "raid" and it accepts the following parameters:
     <#raid_devs> <metadata_dev0> <dev0> [.. <metadata_devN> <devN>]
 
 <raid_type>:
+  raid0		RAID0 striping (no resilience)
   raid1		RAID1 mirroring
-  raid4		RAID4 dedicated parity disk
+  raid4		RAID4 with dedicated last parity disk
+  raid5_n 	RAID5 with dedicated last parity disk suporting takeover
+		Same as raid4
+		-Transitory layout
   raid5_la	RAID5 left asymmetric
 		- rotating parity 0 with data continuation
   raid5_ra	RAID5 right asymmetric
@@ -30,7 +34,19 @@ The target is named "raid" and it accepts the following parameters:
 		- rotating parity N (right-to-left) with data restart
   raid6_nc	RAID6 N continue
 		- rotating parity N (right-to-left) with data continuation
+  raid6_n_6	RAID6 with dedicate parity disks
+		- parity and Q-syndrome on the last 2 disks;
+		  laylout for takeover from/to raid4/raid5_n
+  raid6_la_6	Same as "raid_la" plus dedicated last Q-syndrome disk
+		- layout for takeover from raid5_la from/to raid6
+  raid6_ra_6	Same as "raid5_ra" dedicated last Q-syndrome disk
+		- layout for takeover from raid5_ra from/to raid6
+  raid6_ls_6	Same as "raid5_ls" dedicated last Q-syndrome disk
+		- layout for takeover from raid5_ls from/to raid6
+  raid6_rs_6	Same as "raid5_rs" dedicated last Q-syndrome disk
+		- layout for takeover from raid5_rs from/to raid6
   raid10        Various RAID10 inspired algorithms chosen by additional params
+		(see raid10_format and raid10_copies below)
 		- RAID10: Striped Mirrors (aka 'Striping on top of mirrors')
 		- RAID1E: Integrated Adjacent Stripe Mirroring
 		- RAID1E: Integrated Offset Stripe Mirroring
@@ -116,10 +132,41 @@ The target is named "raid" and it accepts the following parameters:
 		Here we see layouts closely akin to 'RAID1E - Integrated
 		Offset Stripe Mirroring'.
 
+        [delta_disks <N>]
+		The delta_disks option value (-251 < N < +251) triggers
+		device removal (negative value) or device addition (positive
+		value) to any reshape supporting raid levels 4/5/6 and 10.
+		RAID levels 4/5/6 allow for addition of devices (metadata
+		and data device tupel), raid10_near and raid10_offset only
+		allow for device addtion. raid10_far does not support any
+		reshaping at all.
+		A minimum of devices have to be kept to enforce resilience,
+		which is 3 devices for raid4/5 and 4 devices for raid6.
+
+        [data_offset <sectors>]
+		This option value defines the offset into each data device
+		where the data starts. This is used to provide out-of-place
+		reshaping space to avoid writing over data whilst
+		changing the layout of stripes, hence an interruption/crash
+		may happen at any time without the risk of losing data.
+		E.g. when adding devices to an existing raid set during
+		forward reshaping, the out-of-place space will be allocated
+		at the beginning of each raid device. The kernel raid4/5/6/10
+		MD personalities supporting such device addition will read the data from
+		the existing first stripes (those with smaller number of stripes)
+		starting at data_offset to fill up a new stripe with the larger
+		number of stripes, calculate the redundancy blocks (CRC/Q-syndrome)
+		and write that new stripe to offset 0. Same will be applied to all
+		N-1 other new stripes. This out-of-place scheme is used to change
+		the RAID type (i.e. the allocation algorithm) as well, e.g.
+		changing from raid5_ls to raid5_n.
+
 <#raid_devs>: The number of devices composing the array.
 	Each device consists of two entries.  The first is the device
 	containing the metadata (if any); the second is the one containing the
-	data.
+	data. A Maximum of 64 metadata/data device entries are supported
+	up to target version 1.8.0.
+	1.9.0 supports up to 253 which is enforced by the used MD kernel runtime.
 
 	If a drive has failed or is missing at creation time, a '-' can be
 	given for both the metadata and data drives for a given position.
@@ -207,7 +254,6 @@ include:
 	"recover"- Initiate/continue a recover process.
 	"check"  - Initiate a check (i.e. a "scrub") of the array.
 	"repair" - Initiate a repair of the array.
-	"reshape"- Currently unsupported (-EINVAL).
 
 
 Discard Support
@@ -257,3 +303,9 @@ Version History
 1.5.2   'mismatch_cnt' is zero unless [last_]sync_action is "check".
 1.6.0   Add discard support (and devices_handle_discard_safely module param).
 1.7.0   Add support for MD RAID0 mappings.
+1.8.0   Explictely check for compatible flags in the superblock metadata
+	and reject to start the raid set if any are set by a newer
+	target version, thus avoiding data corruption on a raid set
+	with a reshape in progress.
+1.9.0   Add support for RAID level takeover/reshape/region size
+	and set size reduction.

drivers/md/Makefile

@@ -3,7 +3,8 @@
 #
 
 dm-mod-y	+= dm.o dm-table.o dm-target.o dm-linear.o dm-stripe.o \
-		   dm-ioctl.o dm-io.o dm-kcopyd.o dm-sysfs.o dm-stats.o
+		   dm-ioctl.o dm-io.o dm-kcopyd.o dm-sysfs.o dm-stats.o \
+		   dm-rq.o
 dm-multipath-y	+= dm-path-selector.o dm-mpath.o
 dm-snapshot-y	+= dm-snap.o dm-exception-store.o dm-snap-transient.o \
 		    dm-snap-persistent.o

drivers/md/dm-builtin.c

-#include "dm.h"
+#include "dm-core.h"
 
 /*
  * The kobject release method must not be placed in the module itself,

drivers/md/dm-core.h

+/*
+ * Internal header file _only_ for device mapper core
+ *
+ * Copyright (C) 2016 Red Hat, Inc. All rights reserved.
+ *
+ * This file is released under the LGPL.
+ */
+
+#ifndef DM_CORE_INTERNAL_H
+#define DM_CORE_INTERNAL_H
+
+#include <linux/kthread.h>
+#include <linux/ktime.h>
+#include <linux/blk-mq.h>
+
+#include <trace/events/block.h>
+
+#include "dm.h"
+
+#define DM_RESERVED_MAX_IOS		1024
+
+struct dm_kobject_holder {
+	struct kobject kobj;
+	struct completion completion;
+};
+
+/*
+ * DM core internal structure that used directly by dm.c and dm-rq.c
+ * DM targets must _not_ deference a mapped_device to directly access its members!
+ */
+struct mapped_device {
+	struct srcu_struct io_barrier;
+	struct mutex suspend_lock;
+
+	/*
+	 * The current mapping (struct dm_table *).
+	 * Use dm_get_live_table{_fast} or take suspend_lock for
+	 * dereference.
+	 */
+	void __rcu *map;
+
+	struct list_head table_devices;
+	struct mutex table_devices_lock;
+
+	unsigned long flags;
+
+	struct request_queue *queue;
+	int numa_node_id;
+
+	unsigned type;
+	/* Protect queue and type against concurrent access. */
+	struct mutex type_lock;
+
+	atomic_t holders;
+	atomic_t open_count;
+
+	struct dm_target *immutable_target;
+	struct target_type *immutable_target_type;
+
+	struct gendisk *disk;
+	char name[16];
+
+	void *interface_ptr;
+
+	/*
+	 * A list of ios that arrived while we were suspended.
+	 */
+	atomic_t pending[2];
+	wait_queue_head_t wait;
+	struct work_struct work;
+	spinlock_t deferred_lock;
+	struct bio_list deferred;
+
+	/*
+	 * Event handling.
+	 */
+	wait_queue_head_t eventq;
+	atomic_t event_nr;
+	atomic_t uevent_seq;
+	struct list_head uevent_list;
+	spinlock_t uevent_lock; /* Protect access to uevent_list */
+
+	/* the number of internal suspends */
+	unsigned internal_suspend_count;
+
+	/*
+	 * Processing queue (flush)
+	 */
+	struct workqueue_struct *wq;
+
+	/*
+	 * io objects are allocated from here.
+	 */
+	mempool_t *io_pool;
+	mempool_t *rq_pool;
+
+	struct bio_set *bs;
+
+	/*
+	 * freeze/thaw support require holding onto a super block
+	 */
+	struct super_block *frozen_sb;
+
+	/* forced geometry settings */
+	struct hd_geometry geometry;
+
+	struct block_device *bdev;
+
+	/* kobject and completion */
+	struct dm_kobject_holder kobj_holder;
+
+	/* zero-length flush that will be cloned and submitted to targets */
+	struct bio flush_bio;
+
+	struct dm_stats stats;
+
+	struct kthread_worker kworker;
+	struct task_struct *kworker_task;
+
+	/* for request-based merge heuristic in dm_request_fn() */
+	unsigned seq_rq_merge_deadline_usecs;
+	int last_rq_rw;
+	sector_t last_rq_pos;
+	ktime_t last_rq_start_time;
+
+	/* for blk-mq request-based DM support */
+	struct blk_mq_tag_set *tag_set;
+	bool use_blk_mq:1;
+	bool init_tio_pdu:1;
+};
+
+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);
+
+static inline struct completion *dm_get_completion_from_kobject(struct kobject *kobj)
+{
+	return &container_of(kobj, struct dm_kobject_holder, kobj)->completion;
+}
+
+unsigned __dm_get_module_param(unsigned *module_param, unsigned def, unsigned max);
+
+static inline bool dm_message_test_buffer_overflow(char *result, unsigned maxlen)
+{
+	return !maxlen || strlen(result) + 1 >= maxlen;
+}
+
+#endif

drivers/md/dm-crypt.c

@@ -683,7 +683,7 @@ static int crypt_iv_tcw_whitening(struct crypt_config *cc,
 				  u8 *data)
 {
 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
-	u64 sector = cpu_to_le64((u64)dmreq->iv_sector);
+	__le64 sector = cpu_to_le64(dmreq->iv_sector);
 	u8 buf[TCW_WHITENING_SIZE];
 	SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
 	int i, r;
@@ -722,7 +722,7 @@ static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
 			    struct dm_crypt_request *dmreq)
 {
 	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
-	u64 sector = cpu_to_le64((u64)dmreq->iv_sector);
+	__le64 sector = cpu_to_le64(dmreq->iv_sector);
 	u8 *src;
 	int r = 0;
 

drivers/md/dm-io.c

@@ -5,7 +5,7 @@
  * This file is released under the GPL.
  */
 
-#include "dm.h"
+#include "dm-core.h"
 
 #include <linux/device-mapper.h>
 

drivers/md/dm-ioctl.c

@@ -5,7 +5,7 @@
  * This file is released under the GPL.
  */
 
-#include "dm.h"
+#include "dm-core.h"
 
 #include <linux/module.h>
 #include <linux/vmalloc.h>
@@ -1267,6 +1267,15 @@ static int populate_table(struct dm_table *table,
 	return dm_table_complete(table);
 }
 
+static bool is_valid_type(unsigned cur, unsigned new)
+{
+	if (cur == new ||
+	    (cur == DM_TYPE_BIO_BASED && new == DM_TYPE_DAX_BIO_BASED))
+		return true;
+
+	return false;
+}
+
 static int table_load(struct dm_ioctl *param, size_t param_size)
 {
 	int r;
@@ -1309,7 +1318,7 @@ static int table_load(struct dm_ioctl *param, size_t param_size)
 			DMWARN("unable to set up device queue for new table.");
 			goto err_unlock_md_type;
 		}
-	} else if (dm_get_md_type(md) != dm_table_get_type(t)) {
+	} else if (!is_valid_type(dm_get_md_type(md), dm_table_get_type(t))) {
 		DMWARN("can't change device type after initial table load.");
 		r = -EINVAL;
 		goto err_unlock_md_type;
@@ -1670,8 +1679,7 @@ static int check_version(unsigned int cmd, struct dm_ioctl __user *user)
 	return r;
 }
 
-#define DM_PARAMS_KMALLOC	0x0001	/* Params alloced with kmalloc */
-#define DM_PARAMS_VMALLOC	0x0002	/* Params alloced with vmalloc */
+#define DM_PARAMS_MALLOC	0x0001	/* Params allocated with kvmalloc() */
 #define DM_WIPE_BUFFER		0x0010	/* Wipe input buffer before returning from ioctl */
 
 static void free_params(struct dm_ioctl *param, size_t param_size, int param_flags)
@@ -1679,10 +1687,8 @@ static void free_params(struct dm_ioctl *param, size_t param_size, int param_fla
 	if (param_flags & DM_WIPE_BUFFER)
 		memset(param, 0, param_size);
 
-	if (param_flags & DM_PARAMS_KMALLOC)
-		kfree(param);
-	if (param_flags & DM_PARAMS_VMALLOC)
-		vfree(param);
+	if (param_flags & DM_PARAMS_MALLOC)
+		kvfree(param);
 }
 
 static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl *param_kernel,
@@ -1714,19 +1720,14 @@ static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl *param_kern
 	 * Use kmalloc() rather than vmalloc() when we can.
 	 */
 	dmi = NULL;
-	if (param_kernel->data_size <= KMALLOC_MAX_SIZE) {
+	if (param_kernel->data_size <= KMALLOC_MAX_SIZE)
 		dmi = kmalloc(param_kernel->data_size, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
-		if (dmi)
-			*param_flags |= DM_PARAMS_KMALLOC;
-	}
 
 	if (!dmi) {
 		unsigned noio_flag;
 		noio_flag = memalloc_noio_save();
 		dmi = __vmalloc(param_kernel->data_size, GFP_NOIO | __GFP_HIGH | __GFP_HIGHMEM, PAGE_KERNEL);
 		memalloc_noio_restore(noio_flag);
-		if (dmi)
-			*param_flags |= DM_PARAMS_VMALLOC;
 	}
 
 	if (!dmi) {
@@ -1735,6 +1736,8 @@ static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl *param_kern
 		return -ENOMEM;
 	}
 
+	*param_flags |= DM_PARAMS_MALLOC;
+
 	if (copy_from_user(dmi, user, param_kernel->data_size))
 		goto bad;
 

drivers/md/dm-kcopyd.c

@@ -26,7 +26,7 @@
 #include <linux/device-mapper.h>
 #include <linux/dm-kcopyd.h>
 
-#include "dm.h"
+#include "dm-core.h"
 
 #define SUB_JOB_SIZE	128
 #define SPLIT_COUNT	8

drivers/md/dm-linear.c

@@ -141,9 +141,27 @@ static int linear_iterate_devices(struct dm_target *ti,
 	return fn(ti, lc->dev, lc->start, ti->len, data);
 }
 
+static long linear_direct_access(struct dm_target *ti, sector_t sector,
+				 void __pmem **kaddr, pfn_t *pfn, long size)
+{
+	struct linear_c *lc = ti->private;
+	struct block_device *bdev = lc->dev->bdev;
+	struct blk_dax_ctl dax = {
+		.sector = linear_map_sector(ti, sector),
+		.size = size,
+	};
+	long ret;
+
+	ret = bdev_direct_access(bdev, &dax);
+	*kaddr = dax.addr;
+	*pfn = dax.pfn;
+
+	return ret;
+}
+
 static struct target_type linear_target = {
 	.name   = "linear",
-	.version = {1, 2, 1},
+	.version = {1, 3, 0},
 	.module = THIS_MODULE,
 	.ctr    = linear_ctr,
 	.dtr    = linear_dtr,
@@ -151,6 +169,7 @@ static struct target_type linear_target = {
 	.status = linear_status,
 	.prepare_ioctl = linear_prepare_ioctl,
 	.iterate_devices = linear_iterate_devices,
+	.direct_access = linear_direct_access,
 };
 
 int __init dm_linear_init(void)

drivers/md/dm-mpath.c

@@ -7,7 +7,8 @@
 
 #include <linux/device-mapper.h>
 
-#include "dm.h"
+#include "dm-rq.h"
+#include "dm-bio-record.h"
 #include "dm-path-selector.h"
 #include "dm-uevent.h"
 
@@ -89,6 +90,8 @@ struct multipath {
 	atomic_t pg_init_in_progress;	/* Only one pg_init allowed at once */
 	atomic_t pg_init_count;		/* Number of times pg_init called */
 
+	unsigned queue_mode;
+
 	/*
 	 * We must use a mempool of dm_mpath_io structs so that we
 	 * can resubmit bios on error.
@@ -97,10 +100,13 @@ struct multipath {
 
 	struct mutex work_mutex;
 	struct work_struct trigger_event;
+
+	struct work_struct process_queued_bios;
+	struct bio_list queued_bios;
 };
 
 /*
- * Context information attached to each bio we process.
+ * Context information attached to each io we process.
  */
 struct dm_mpath_io {
 	struct pgpath *pgpath;
@@ -114,6 +120,7 @@ static struct kmem_cache *_mpio_cache;
 static struct workqueue_struct *kmultipathd, *kmpath_handlerd;
 static void trigger_event(struct work_struct *work);
 static void activate_path(struct work_struct *work);
+static void process_queued_bios(struct work_struct *work);
 
 /*-----------------------------------------------
  * Multipath state flags.
@@ -185,7 +192,7 @@ static void free_priority_group(struct priority_group *pg,
 	kfree(pg);
 }
 
-static struct multipath *alloc_multipath(struct dm_target *ti, bool use_blk_mq)
+static struct multipath *alloc_multipath(struct dm_target *ti)
 {
 	struct multipath *m;
 
@@ -203,21 +210,46 @@ static struct multipath *alloc_multipath(struct dm_target *ti, bool use_blk_mq)
 		mutex_init(&m->work_mutex);
 
 		m->mpio_pool = NULL;
-		if (!use_blk_mq) {
+		m->queue_mode = DM_TYPE_NONE;
+
+		m->ti = ti;
+		ti->private = m;
+	}
+
+	return m;
+}
+
+static int alloc_multipath_stage2(struct dm_target *ti, struct multipath *m)
+{
+	if (m->queue_mode == DM_TYPE_NONE) {
+		/*
+		 * Default to request-based.
+		 */
+		if (dm_use_blk_mq(dm_table_get_md(ti->table)))
+			m->queue_mode = DM_TYPE_MQ_REQUEST_BASED;
+		else
+			m->queue_mode = DM_TYPE_REQUEST_BASED;
+	}
+
+	if (m->queue_mode == DM_TYPE_REQUEST_BASED) {
 		unsigned min_ios = dm_get_reserved_rq_based_ios();
 
 		m->mpio_pool = mempool_create_slab_pool(min_ios, _mpio_cache);
-			if (!m->mpio_pool) {
-				kfree(m);
-				return NULL;
+		if (!m->mpio_pool)
+			return -ENOMEM;
 	}
+	else if (m->queue_mode == DM_TYPE_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);
 	}
 
-		m->ti = ti;
-		ti->private = m;
-	}
+	dm_table_set_type(ti->table, m->queue_mode);
 
-	return m;
+	return 0;
 }
 
 static void free_multipath(struct multipath *m)
@@ -272,6 +304,41 @@ static void clear_request_fn_mpio(struct multipath *m, union map_info *info)
 	}
 }
 
+static size_t multipath_per_bio_data_size(void)
+{
+	return sizeof(struct dm_mpath_io) + sizeof(struct dm_bio_details);
+}
+
+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)
+{
+	/* 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)
+{
+	struct dm_mpath_io *mpio = get_mpio_from_bio(bio);
+	struct dm_bio_details *bio_details = get_bio_details_from_bio(bio);
+
+	memset(mpio, 0, sizeof(*mpio));
+	memset(bio_details, 0, sizeof(*bio_details));
+	dm_bio_record(bio_details, bio);
+
+	if (mpio_p)
+		*mpio_p = mpio;
+	if (bio_details_p)
+		*bio_details_p = bio_details;
+}
+
 /*-----------------------------------------------
  * Path selection
  *-----------------------------------------------*/
@@ -431,16 +498,26 @@ static struct pgpath *choose_pgpath(struct multipath *m, size_t nr_bytes)
  * and multipath_resume() calls and we have no need to check
  * for the DMF_NOFLUSH_SUSPENDING flag.
  */
-static int must_push_back(struct multipath *m)
+static bool __must_push_back(struct multipath *m)
 {
-	return (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) ||
-		((test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) !=
+	return ((test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) !=
 		 test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &m->flags)) &&
-		 dm_noflush_suspending(m->ti)));
+		dm_noflush_suspending(m->ti));
+}
+
+static bool must_push_back_rq(struct multipath *m)
+{
+	return (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) ||
+		__must_push_back(m));
+}
+
+static bool must_push_back_bio(struct multipath *m)
+{
+	return __must_push_back(m);
 }
 
 /*
- * Map cloned requests
+ * Map cloned requests (request-based multipath)
  */
 static int __multipath_map(struct dm_target *ti, struct request *clone,
 			   union map_info *map_context,
@@ -459,7 +536,7 @@ static int __multipath_map(struct dm_target *ti, struct request *clone,
 		pgpath = choose_pgpath(m, nr_bytes);
 
 	if (!pgpath) {
-		if (!must_push_back(m))
+		if (!must_push_back_rq(m))
 			r = -EIO;	/* Failed */
 		return r;
 	} else if (test_bit(MPATHF_QUEUE_IO, &m->flags) ||
@@ -529,6 +606,108 @@ static void multipath_release_clone(struct request *clone)
 	blk_mq_free_request(clone);
 }
 
+/*
+ * Map cloned bios (bio-based multipath)
+ */
+static int __multipath_map_bio(struct multipath *m, struct bio *bio, struct dm_mpath_io *mpio)
+{
+	size_t nr_bytes = bio->bi_iter.bi_size;
+	struct pgpath *pgpath;
+	unsigned long flags;
+	bool queue_io;
+
+	/* Do we need to select a new pgpath? */
+	pgpath = lockless_dereference(m->current_pgpath);
+	queue_io = test_bit(MPATHF_QUEUE_IO, &m->flags);
+	if (!pgpath || !queue_io)
+		pgpath = choose_pgpath(m, nr_bytes);
+
+	if ((pgpath && queue_io) ||
+	    (!pgpath && 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);
+		/* 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;
+	}
+
+	if (!pgpath) {
+		if (!must_push_back_bio(m))
+			return -EIO;
+		return DM_MAPIO_REQUEUE;
+	}
+
+	mpio->pgpath = pgpath;
+	mpio->nr_bytes = nr_bytes;
+
+	bio->bi_error = 0;
+	bio->bi_bdev = pgpath->path.dev->bdev;
+	bio->bi_rw |= REQ_FAILFAST_TRANSPORT;
+
+	if (pgpath->pg->ps.type->start_io)
+		pgpath->pg->ps.type->start_io(&pgpath->pg->ps,
+					      &pgpath->path,
+					      nr_bytes);
+	return DM_MAPIO_REMAPPED;
+}
+
+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);
+
+	return __multipath_map_bio(m, bio, mpio);
+}
+
+static void process_queued_bios_list(struct multipath *m)
+{
+	if (m->queue_mode == DM_TYPE_BIO_BASED)
+		queue_work(kmultipathd, &m->process_queued_bios);
+}
+
+static void process_queued_bios(struct work_struct *work)
+{
+	int r;
+	unsigned long flags;
+	struct bio *bio;
+	struct bio_list bios;
+	struct blk_plug plug;
+	struct multipath *m =
+		container_of(work, struct multipath, process_queued_bios);
+
+	bio_list_init(&bios);
+
+	spin_lock_irqsave(&m->lock, flags);
+
+	if (bio_list_empty(&m->queued_bios)) {
+		spin_unlock_irqrestore(&m->lock, flags);
+		return;
+	}
+
+	bio_list_merge(&bios, &m->queued_bios);
+	bio_list_init(&m->queued_bios);
+
+	spin_unlock_irqrestore(&m->lock, flags);
+
+	blk_start_plug(&plug);
+	while ((bio = bio_list_pop(&bios))) {
+		r = __multipath_map_bio(m, bio, get_mpio_from_bio(bio));
+		if (r < 0 || r == DM_MAPIO_REQUEUE) {
+			bio->bi_error = r;
+			bio_endio(bio);
+		} else if (r == DM_MAPIO_REMAPPED)
+			generic_make_request(bio);
+	}
+	blk_finish_plug(&plug);
+}
+
 /*
  * If we run out of usable paths, should we queue I/O or error it?
  */
@@ -557,8 +736,10 @@ static int queue_if_no_path(struct multipath *m, bool queue_if_no_path,
 
 	spin_unlock_irqrestore(&m->lock, flags);
 
-	if (!queue_if_no_path)
+	if (!queue_if_no_path) {
 		dm_table_run_md_queue_async(m->ti->table);
+		process_queued_bios_list(m);
+	}
 
 	return 0;
 }
@@ -798,6 +979,12 @@ 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) {
+		dm_consume_args(as, hw_argc);
+		DMERR("bio-based multipath doesn't allow hardware handler args");
+		return 0;
+	}
+
 	m->hw_handler_name = kstrdup(dm_shift_arg(as), GFP_KERNEL);
 
 	if (hw_argc > 1) {
@@ -833,7 +1020,7 @@ static int parse_features(struct dm_arg_set *as, struct multipath *m)
 	const char *arg_name;
 
 	static struct dm_arg _args[] = {
-		{0, 6, "invalid number of feature args"},
+		{0, 8, "invalid number of feature args"},
 		{1, 50, "pg_init_retries must be between 1 and 50"},
 		{0, 60000, "pg_init_delay_msecs must be between 0 and 60000"},
 	};
@@ -873,6 +1060,24 @@ static int parse_features(struct dm_arg_set *as, struct multipath *m)
 			continue;
 		}
 
+		if (!strcasecmp(arg_name, "queue_mode") &&
+		    (argc >= 1)) {
+			const char *queue_mode_name = dm_shift_arg(as);
+
+			if (!strcasecmp(queue_mode_name, "bio"))
+				m->queue_mode = DM_TYPE_BIO_BASED;
+			else if (!strcasecmp(queue_mode_name, "rq"))
+				m->queue_mode = DM_TYPE_REQUEST_BASED;
+			else if (!strcasecmp(queue_mode_name, "mq"))
+				m->queue_mode = DM_TYPE_MQ_REQUEST_BASED;
+			else {
+				ti->error = "Unknown 'queue_mode' requested";
+				r = -EINVAL;
+			}
+			argc--;
+			continue;
+		}
+
 		ti->error = "Unrecognised multipath feature request";
 		r = -EINVAL;
 	} while (argc && !r);
@@ -880,8 +1085,7 @@ static int parse_features(struct dm_arg_set *as, struct multipath *m)
 	return r;
 }
 
-static int multipath_ctr(struct dm_target *ti, unsigned int argc,
-			 char **argv)
+static int multipath_ctr(struct dm_target *ti, unsigned argc, char **argv)
 {
 	/* target arguments */
 	static struct dm_arg _args[] = {
@@ -894,12 +1098,11 @@ static int multipath_ctr(struct dm_target *ti, unsigned int argc,
 	struct dm_arg_set as;
 	unsigned pg_count = 0;
 	unsigned next_pg_num;
-	bool use_blk_mq = dm_use_blk_mq(dm_table_get_md(ti->table));
 
 	as.argc = argc;
 	as.argv = argv;
 
-	m = alloc_multipath(ti, use_blk_mq);
+	m = alloc_multipath(ti);
 	if (!m) {
 		ti->error = "can't allocate multipath";
 		return -EINVAL;
@@ -909,6 +1112,10 @@ static int multipath_ctr(struct dm_target *ti, unsigned int argc,
 	if (r)
 		goto bad;
 
+	r = alloc_multipath_stage2(ti, m);
+	if (r)
+		goto bad;
+
 	r = parse_hw_handler(&as, m);
 	if (r)
 		goto bad;
@@ -958,7 +1165,9 @@ static int multipath_ctr(struct dm_target *ti, unsigned int argc,
 	ti->num_flush_bios = 1;
 	ti->num_discard_bios = 1;
 	ti->num_write_same_bios = 1;
-	if (use_blk_mq)
+	if (m->queue_mode == DM_TYPE_BIO_BASED)
+		ti->per_io_data_size = multipath_per_bio_data_size();
+	else if (m->queue_mode == DM_TYPE_MQ_REQUEST_BASED)
 		ti->per_io_data_size = sizeof(struct dm_mpath_io);
 
 	return 0;
@@ -1083,8 +1292,10 @@ static int reinstate_path(struct pgpath *pgpath)
 
 out:
 	spin_unlock_irqrestore(&m->lock, flags);
-	if (run_queue)
+	if (run_queue) {
 		dm_table_run_md_queue_async(m->ti->table);
+		process_queued_bios_list(m);
+	}
 
 	return r;
 }
@@ -1281,6 +1492,8 @@ static void pg_init_done(void *data, int errors)
 	}
 	clear_bit(MPATHF_QUEUE_IO, &m->flags);
 
+	process_queued_bios_list(m);
+
 	/*
 	 * Wake up any thread waiting to suspend.
 	 */
@@ -1328,7 +1541,7 @@ static int do_end_io(struct multipath *m, struct request *clone,
 	 * during end I/O handling, since those clone requests don't have
 	 * bio clones.  If we queue them inside the multipath target,
 	 * we need to make bio clones, that requires memory allocation.
-	 * (See drivers/md/dm.c:end_clone_bio() about why the clone requests
+	 * (See drivers/md/dm-rq.c:end_clone_bio() about why the clone requests
 	 *  don't have bio clones.)
 	 * Instead of queueing the clone request here, we queue the original
 	 * request into dm core, which will remake a clone request and
@@ -1347,7 +1560,7 @@ static int do_end_io(struct multipath *m, struct request *clone,
 
 	if (!atomic_read(&m->nr_valid_paths)) {
 		if (!test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) {
-			if (!must_push_back(m))
+			if (!must_push_back_rq(m))
 				r = -EIO;
 		} else {
 			if (error == -EBADE)
@@ -1381,6 +1594,64 @@ static int multipath_end_io(struct dm_target *ti, struct request *clone,
 	return r;
 }
 
+static int do_end_io_bio(struct multipath *m, struct bio *clone,
+			 int error, struct dm_mpath_io *mpio)
+{
+	unsigned long flags;
+
+	if (!error)
+		return 0;	/* I/O complete */
+
+	if (noretry_error(error))
+		return error;
+
+	if (mpio->pgpath)
+		fail_path(mpio->pgpath);
+
+	if (!atomic_read(&m->nr_valid_paths)) {
+		if (!test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) {
+			if (!must_push_back_bio(m))
+				return -EIO;
+			return DM_ENDIO_REQUEUE;
+		} else {
+			if (error == -EBADE)
+				return error;
+		}
+	}
+
+	/* Queue for the daemon to resubmit */
+	dm_bio_restore(get_bio_details_from_bio(clone), clone);
+
+	spin_lock_irqsave(&m->lock, flags);
+	bio_list_add(&m->queued_bios, clone);
+	spin_unlock_irqrestore(&m->lock, flags);
+	if (!test_bit(MPATHF_QUEUE_IO, &m->flags))
+		queue_work(kmultipathd, &m->process_queued_bios);
+
+	return DM_ENDIO_INCOMPLETE;
+}
+
+static int multipath_end_io_bio(struct dm_target *ti, struct bio *clone, int error)
+{
+	struct multipath *m = ti->private;
+	struct dm_mpath_io *mpio = get_mpio_from_bio(clone);
+	struct pgpath *pgpath;
+	struct path_selector *ps;
+	int r;
+
+	BUG_ON(!mpio);
+
+	r = do_end_io_bio(m, clone, error, mpio);
+	pgpath = mpio->pgpath;
+	if (pgpath) {
+		ps = &pgpath->pg->ps;
+		if (ps->type->end_io)
+			ps->type->end_io(ps, &pgpath->path, mpio->nr_bytes);
+	}
+
+	return r;
+}
+
 /*
  * Suspend can't complete until all the I/O is processed so if
  * the last path fails we must error any remaining I/O.
@@ -1454,7 +1725,9 @@ static void multipath_status(struct dm_target *ti, status_type_t type,
 		DMEMIT("%u ", test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags) +
 			      (m->pg_init_retries > 0) * 2 +
 			      (m->pg_init_delay_msecs != DM_PG_INIT_DELAY_DEFAULT) * 2 +
-			      test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags));
+			      test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags) +
+			      (m->queue_mode != DM_TYPE_REQUEST_BASED) * 2);
+
 		if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))
 			DMEMIT("queue_if_no_path ");
 		if (m->pg_init_retries)
@@ -1463,6 +1736,16 @@ static void multipath_status(struct dm_target *ti, status_type_t type,
 			DMEMIT("pg_init_delay_msecs %u ", m->pg_init_delay_msecs);
 		if (test_bit(MPATHF_RETAIN_ATTACHED_HW_HANDLER, &m->flags))
 			DMEMIT("retain_attached_hw_handler ");
+		if (m->queue_mode != DM_TYPE_REQUEST_BASED) {
+			switch(m->queue_mode) {
+			case DM_TYPE_BIO_BASED:
+				DMEMIT("queue_mode bio ");
+				break;
+			case DM_TYPE_MQ_REQUEST_BASED:
+				DMEMIT("queue_mode mq ");
+				break;
+			}
+		}
 	}
 
 	if (!m->hw_handler_name || type == STATUSTYPE_INFO)
@@ -1642,6 +1925,7 @@ static int multipath_prepare_ioctl(struct dm_target *ti,
 		if (test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags))
 			pg_init_all_paths(m);
 		dm_table_run_md_queue_async(m->ti->table);
+		process_queued_bios_list(m);
 	}
 
 	/*
@@ -1748,7 +2032,7 @@ static int multipath_busy(struct dm_target *ti)
  *---------------------------------------------------------------*/
 static struct target_type multipath_target = {
 	.name = "multipath",
-	.version = {1, 11, 0},
+	.version = {1, 12, 0},
 	.features = DM_TARGET_SINGLETON | DM_TARGET_IMMUTABLE,
 	.module = THIS_MODULE,
 	.ctr = multipath_ctr,
@@ -1757,6 +2041,8 @@ static struct target_type multipath_target = {
 	.clone_and_map_rq = multipath_clone_and_map,
 	.release_clone_rq = multipath_release_clone,
 	.rq_end_io = multipath_end_io,
+	.map = multipath_map_bio,
+	.end_io = multipath_end_io_bio,
 	.presuspend = multipath_presuspend,
 	.postsuspend = multipath_postsuspend,
 	.resume = multipath_resume,
@@ -1771,14 +2057,14 @@ static int __init dm_multipath_init(void)
 {
 	int r;
 
-	/* allocate a slab for the dm_ios */
+	/* allocate a slab for the dm_mpath_ios */
 	_mpio_cache = KMEM_CACHE(dm_mpath_io, 0);
 	if (!_mpio_cache)
 		return -ENOMEM;
 
 	r = dm_register_target(&multipath_target);
 	if (r < 0) {
-		DMERR("register failed %d", r);
+		DMERR("request-based register failed %d", r);
 		r = -EINVAL;
 		goto bad_register_target;
 	}
@@ -1804,10 +2090,6 @@ static int __init dm_multipath_init(void)
 		goto bad_alloc_kmpath_handlerd;
 	}
 
-	DMINFO("version %u.%u.%u loaded",
-	       multipath_target.version[0], multipath_target.version[1],
-	       multipath_target.version[2]);
-
 	return 0;
 
 bad_alloc_kmpath_handlerd:

drivers/md/dm-rq.c

+/*
+ * Copyright (C) 2016 Red Hat, Inc. All rights reserved.
+ *
+ * This file is released under the GPL.
+ */
+
+#include "dm-core.h"
+#include "dm-rq.h"
+
+#include <linux/elevator.h> /* for rq_end_sector() */
+#include <linux/blk-mq.h>
+
+#define DM_MSG_PREFIX "core-rq"
+
+#define DM_MQ_NR_HW_QUEUES 1
+#define DM_MQ_QUEUE_DEPTH 2048
+static unsigned dm_mq_nr_hw_queues = DM_MQ_NR_HW_QUEUES;
+static unsigned dm_mq_queue_depth = DM_MQ_QUEUE_DEPTH;
+
+/*
+ * Request-based DM's mempools' reserved IOs set by the user.
+ */
+#define RESERVED_REQUEST_BASED_IOS	256
+static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS;
+
+#ifdef CONFIG_DM_MQ_DEFAULT
+static bool use_blk_mq = true;
+#else
+static bool use_blk_mq = false;
+#endif
+
+bool dm_use_blk_mq_default(void)
+{
+	return use_blk_mq;
+}
+
+bool dm_use_blk_mq(struct mapped_device *md)
+{
+	return md->use_blk_mq;
+}
+EXPORT_SYMBOL_GPL(dm_use_blk_mq);
+
+unsigned dm_get_reserved_rq_based_ios(void)
+{
+	return __dm_get_module_param(&reserved_rq_based_ios,
+				     RESERVED_REQUEST_BASED_IOS, DM_RESERVED_MAX_IOS);
+}
+EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios);
+
+static unsigned dm_get_blk_mq_nr_hw_queues(void)
+{
+	return __dm_get_module_param(&dm_mq_nr_hw_queues, 1, 32);
+}
+
+static unsigned dm_get_blk_mq_queue_depth(void)
+{
+	return __dm_get_module_param(&dm_mq_queue_depth,
+				     DM_MQ_QUEUE_DEPTH, BLK_MQ_MAX_DEPTH);
+}
+
+int dm_request_based(struct mapped_device *md)
+{
+	return blk_queue_stackable(md->queue);
+}
+
+static void dm_old_start_queue(struct request_queue *q)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(q->queue_lock, flags);
+	if (blk_queue_stopped(q))
+		blk_start_queue(q);
+	spin_unlock_irqrestore(q->queue_lock, flags);
+}
+
+void dm_start_queue(struct request_queue *q)
+{
+	if (!q->mq_ops)
+		dm_old_start_queue(q);
+	else {
+		blk_mq_start_stopped_hw_queues(q, true);
+		blk_mq_kick_requeue_list(q);
+	}
+}
+
+static void dm_old_stop_queue(struct request_queue *q)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(q->queue_lock, flags);
+	if (blk_queue_stopped(q)) {
+		spin_unlock_irqrestore(q->queue_lock, flags);
+		return;
+	}
+
+	blk_stop_queue(q);
+	spin_unlock_irqrestore(q->queue_lock, flags);
+}
+
+void dm_stop_queue(struct request_queue *q)
+{
+	if (!q->mq_ops)
+		dm_old_stop_queue(q);
+	else
+		blk_mq_stop_hw_queues(q);
+}
+
+static struct dm_rq_target_io *alloc_old_rq_tio(struct mapped_device *md,
+						gfp_t gfp_mask)
+{
+	return mempool_alloc(md->io_pool, gfp_mask);
+}
+
+static void free_old_rq_tio(struct dm_rq_target_io *tio)
+{
+	mempool_free(tio, tio->md->io_pool);
+}
+
+static struct request *alloc_old_clone_request(struct mapped_device *md,
+					       gfp_t gfp_mask)
+{
+	return mempool_alloc(md->rq_pool, gfp_mask);
+}
+
+static void free_old_clone_request(struct mapped_device *md, struct request *rq)
+{
+	mempool_free(rq, md->rq_pool);
+}
+
+/*
+ * Partial completion handling for request-based dm
+ */
+static void end_clone_bio(struct bio *clone)
+{
+	struct dm_rq_clone_bio_info *info =
+		container_of(clone, struct dm_rq_clone_bio_info, clone);
+	struct dm_rq_target_io *tio = info->tio;
+	struct bio *bio = info->orig;
+	unsigned int nr_bytes = info->orig->bi_iter.bi_size;
+	int error = clone->bi_error;
+
+	bio_put(clone);
+
+	if (tio->error)
+		/*
+		 * An error has already been detected on the request.
+		 * Once error occurred, just let clone->end_io() handle
+		 * the remainder.
+		 */
+		return;
+	else if (error) {
+		/*
+		 * Don't notice the error to the upper layer yet.
+		 * The error handling decision is made by the target driver,
+		 * when the request is completed.
+		 */
+		tio->error = error;
+		return;
+	}
+
+	/*
+	 * I/O for the bio successfully completed.
+	 * Notice the data completion to the upper layer.
+	 */
+
+	/*
+	 * bios are processed from the head of the list.
+	 * So the completing bio should always be rq->bio.
+	 * If it's not, something wrong is happening.
+	 */
+	if (tio->orig->bio != bio)
+		DMERR("bio completion is going in the middle of the request");
+
+	/*
+	 * Update the original request.
+	 * Do not use blk_end_request() here, because it may complete
+	 * the original request before the clone, and break the ordering.
+	 */
+	blk_update_request(tio->orig, 0, nr_bytes);
+}
+
+static struct dm_rq_target_io *tio_from_request(struct request *rq)
+{
+	return (rq->q->mq_ops ? blk_mq_rq_to_pdu(rq) : rq->special);
+}
+
+static void rq_end_stats(struct mapped_device *md, struct request *orig)
+{
+	if (unlikely(dm_stats_used(&md->stats))) {
+		struct dm_rq_target_io *tio = tio_from_request(orig);
+		tio->duration_jiffies = jiffies - tio->duration_jiffies;
+		dm_stats_account_io(&md->stats, rq_data_dir(orig),
+				    blk_rq_pos(orig), tio->n_sectors, true,
+				    tio->duration_jiffies, &tio->stats_aux);
+	}
+}
+
+/*
+ * Don't touch any member of the md after calling this function because
+ * the md may be freed in dm_put() at the end of this function.
+ * Or do dm_get() before calling this function and dm_put() later.
+ */
+static void rq_completed(struct mapped_device *md, int rw, bool run_queue)
+{
+	atomic_dec(&md->pending[rw]);
+
+	/* nudge anyone waiting on suspend queue */
+	if (!md_in_flight(md))
+		wake_up(&md->wait);
+
+	/*
+	 * Run this off this callpath, as drivers could invoke end_io while
+	 * inside their request_fn (and holding the queue lock). Calling
+	 * back into ->request_fn() could deadlock attempting to grab the
+	 * queue lock again.
+	 */
+	if (!md->queue->mq_ops && run_queue)
+		blk_run_queue_async(md->queue);
+
+	/*
+	 * dm_put() must be at the end of this function. See the comment above
+	 */
+	dm_put(md);
+}
+
+static void free_rq_clone(struct request *clone)
+{
+	struct dm_rq_target_io *tio = clone->end_io_data;
+	struct mapped_device *md = tio->md;
+
+	blk_rq_unprep_clone(clone);
+
+	/*
+	 * It is possible for a clone_old_rq() allocated clone to
+	 * get passed in -- it may not yet have a request_queue.
+	 * This is known to occur if the error target replaces
+	 * a multipath target that has a request_fn queue stacked
+	 * on blk-mq queue(s).
+	 */
+	if (clone->q && clone->q->mq_ops)
+		/* stacked on blk-mq queue(s) */
+		tio->ti->type->release_clone_rq(clone);
+	else if (!md->queue->mq_ops)
+		/* request_fn queue stacked on request_fn queue(s) */
+		free_old_clone_request(md, clone);
+
+	if (!md->queue->mq_ops)
+		free_old_rq_tio(tio);
+}
+
+/*
+ * Complete the clone and the original request.
+ * Must be called without clone's queue lock held,
+ * see end_clone_request() for more details.
+ */
+static void dm_end_request(struct request *clone, int error)
+{
+	int rw = rq_data_dir(clone);
+	struct dm_rq_target_io *tio = clone->end_io_data;
+	struct mapped_device *md = tio->md;
+	struct request *rq = tio->orig;
+
+	if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
+		rq->errors = clone->errors;
+		rq->resid_len = clone->resid_len;
+
+		if (rq->sense)
+			/*
+			 * We are using the sense buffer of the original
+			 * request.
+			 * So setting the length of the sense data is enough.
+			 */
+			rq->sense_len = clone->sense_len;
+	}
+
+	free_rq_clone(clone);
+	rq_end_stats(md, rq);
+	if (!rq->q->mq_ops)
+		blk_end_request_all(rq, error);
+	else
+		blk_mq_end_request(rq, error);
+	rq_completed(md, rw, true);
+}
+
+static void dm_unprep_request(struct request *rq)
+{
+	struct dm_rq_target_io *tio = tio_from_request(rq);
+	struct request *clone = tio->clone;
+
+	if (!rq->q->mq_ops) {
+		rq->special = NULL;
+		rq->cmd_flags &= ~REQ_DONTPREP;
+	}
+
+	if (clone)
+		free_rq_clone(clone);
+	else if (!tio->md->queue->mq_ops)
+		free_old_rq_tio(tio);
+}
+
+/*
+ * Requeue the original request of a clone.
+ */
+static void dm_old_requeue_request(struct request *rq)
+{
+	struct request_queue *q = rq->q;
+	unsigned long flags;
+
+	spin_lock_irqsave(q->queue_lock, flags);
+	blk_requeue_request(q, rq);
+	blk_run_queue_async(q);
+	spin_unlock_irqrestore(q->queue_lock, flags);
+}
+
+static void dm_mq_requeue_request(struct request *rq)
+{
+	struct request_queue *q = rq->q;
+	unsigned long flags;
+
+	blk_mq_requeue_request(rq);
+	spin_lock_irqsave(q->queue_lock, flags);
+	if (!blk_queue_stopped(q))
+		blk_mq_kick_requeue_list(q);
+	spin_unlock_irqrestore(q->queue_lock, flags);
+}
+
+static void dm_requeue_original_request(struct mapped_device *md,
+					struct request *rq)
+{
+	int rw = rq_data_dir(rq);
+
+	rq_end_stats(md, rq);
+	dm_unprep_request(rq);
+
+	if (!rq->q->mq_ops)
+		dm_old_requeue_request(rq);
+	else
+		dm_mq_requeue_request(rq);
+
+	rq_completed(md, rw, false);
+}
+
+static void dm_done(struct request *clone, int error, bool mapped)
+{
+	int r = error;
+	struct dm_rq_target_io *tio = clone->end_io_data;
+	dm_request_endio_fn rq_end_io = NULL;
+
+	if (tio->ti) {
+		rq_end_io = tio->ti->type->rq_end_io;
+
+		if (mapped && rq_end_io)
+			r = rq_end_io(tio->ti, clone, error, &tio->info);
+	}
+
+	if (unlikely(r == -EREMOTEIO && (req_op(clone) == REQ_OP_WRITE_SAME) &&
+		     !clone->q->limits.max_write_same_sectors))
+		disable_write_same(tio->md);
+
+	if (r <= 0)
+		/* The target wants to complete the I/O */
+		dm_end_request(clone, r);
+	else if (r == DM_ENDIO_INCOMPLETE)
+		/* The target will handle the I/O */
+		return;
+	else if (r == DM_ENDIO_REQUEUE)
+		/* The target wants to requeue the I/O */
+		dm_requeue_original_request(tio->md, tio->orig);
+	else {
+		DMWARN("unimplemented target endio return value: %d", r);
+		BUG();
+	}
+}
+
+/*
+ * Request completion handler for request-based dm
+ */
+static void dm_softirq_done(struct request *rq)
+{
+	bool mapped = true;
+	struct dm_rq_target_io *tio = tio_from_request(rq);
+	struct request *clone = tio->clone;
+	int rw;
+
+	if (!clone) {
+		rq_end_stats(tio->md, rq);
+		rw = rq_data_dir(rq);
+		if (!rq->q->mq_ops) {
+			blk_end_request_all(rq, tio->error);
+			rq_completed(tio->md, rw, false);
+			free_old_rq_tio(tio);
+		} else {
+			blk_mq_end_request(rq, tio->error);
+			rq_completed(tio->md, rw, false);
+		}
+		return;
+	}
+
+	if (rq->cmd_flags & REQ_FAILED)
+		mapped = false;
+
+	dm_done(clone, tio->error, mapped);
+}
+
+/*
+ * Complete the clone and the original request with the error status
+ * through softirq context.
+ */
+static void dm_complete_request(struct request *rq, int error)
+{
+	struct dm_rq_target_io *tio = tio_from_request(rq);
+
+	tio->error = error;
+	if (!rq->q->mq_ops)
+		blk_complete_request(rq);
+	else
+		blk_mq_complete_request(rq, error);
+}
+
+/*
+ * Complete the not-mapped clone and the original request with the error status
+ * through softirq context.
+ * Target's rq_end_io() function isn't called.
+ * This may be used when the target's map_rq() or clone_and_map_rq() functions fail.
+ */
+static void dm_kill_unmapped_request(struct request *rq, int error)
+{
+	rq->cmd_flags |= REQ_FAILED;
+	dm_complete_request(rq, error);
+}
+
+/*
+ * Called with the clone's queue lock held (in the case of .request_fn)
+ */
+static void end_clone_request(struct request *clone, int error)
+{
+	struct dm_rq_target_io *tio = clone->end_io_data;
+
+	if (!clone->q->mq_ops) {
+		/*
+		 * For just cleaning up the information of the queue in which
+		 * the clone was dispatched.
+		 * The clone is *NOT* freed actually here because it is alloced
+		 * from dm own mempool (REQ_ALLOCED isn't set).
+		 */
+		__blk_put_request(clone->q, clone);
+	}
+
+	/*
+	 * Actual request completion is done in a softirq context which doesn't
+	 * hold the clone's queue lock.  Otherwise, deadlock could occur because:
+	 *     - another request may be submitted by the upper level driver
+	 *       of the stacking during the completion
+	 *     - the submission which requires queue lock may be done
+	 *       against this clone's queue
+	 */
+	dm_complete_request(tio->orig, error);
+}
+
+static void dm_dispatch_clone_request(struct request *clone, struct request *rq)
+{
+	int r;
+
+	if (blk_queue_io_stat(clone->q))
+		clone->cmd_flags |= REQ_IO_STAT;
+
+	clone->start_time = jiffies;
+	r = blk_insert_cloned_request(clone->q, clone);
+	if (r)
+		/* must complete clone in terms of original request */
+		dm_complete_request(rq, r);
+}
+
+static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
+				 void *data)
+{
+	struct dm_rq_target_io *tio = data;
+	struct dm_rq_clone_bio_info *info =
+		container_of(bio, struct dm_rq_clone_bio_info, clone);
+
+	info->orig = bio_orig;
+	info->tio = tio;
+	bio->bi_end_io = end_clone_bio;
+
+	return 0;
+}
+
+static int setup_clone(struct request *clone, struct request *rq,
+		       struct dm_rq_target_io *tio, gfp_t gfp_mask)
+{
+	int r;
+
+	r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask,
+			      dm_rq_bio_constructor, tio);
+	if (r)
+		return r;
+
+	clone->cmd = rq->cmd;
+	clone->cmd_len = rq->cmd_len;
+	clone->sense = rq->sense;
+	clone->end_io = end_clone_request;
+	clone->end_io_data = tio;
+
+	tio->clone = clone;
+
+	return 0;
+}
+
+static struct request *clone_old_rq(struct request *rq, struct mapped_device *md,
+				    struct dm_rq_target_io *tio, gfp_t gfp_mask)
+{
+	/*
+	 * Create clone for use with .request_fn request_queue
+	 */
+	struct request *clone;
+
+	clone = alloc_old_clone_request(md, gfp_mask);
+	if (!clone)
+		return NULL;
+
+	blk_rq_init(NULL, clone);
+	if (setup_clone(clone, rq, tio, gfp_mask)) {
+		/* -ENOMEM */
+		free_old_clone_request(md, clone);
+		return NULL;
+	}
+
+	return clone;
+}
+
+static void map_tio_request(struct kthread_work *work);
+
+static void init_tio(struct dm_rq_target_io *tio, struct request *rq,
+		     struct mapped_device *md)
+{
+	tio->md = md;
+	tio->ti = NULL;
+	tio->clone = NULL;
+	tio->orig = rq;
+	tio->error = 0;
+	/*
+	 * Avoid initializing info for blk-mq; it passes
+	 * target-specific data through info.ptr
+	 * (see: dm_mq_init_request)
+	 */
+	if (!md->init_tio_pdu)
+		memset(&tio->info, 0, sizeof(tio->info));
+	if (md->kworker_task)
+		init_kthread_work(&tio->work, map_tio_request);
+}
+
+static struct dm_rq_target_io *dm_old_prep_tio(struct request *rq,
+					       struct mapped_device *md,
+					       gfp_t gfp_mask)
+{
+	struct dm_rq_target_io *tio;
+	int srcu_idx;
+	struct dm_table *table;
+
+	tio = alloc_old_rq_tio(md, gfp_mask);
+	if (!tio)
+		return NULL;
+
+	init_tio(tio, rq, md);
+
+	table = dm_get_live_table(md, &srcu_idx);
+	/*
+	 * Must clone a request if this .request_fn DM device
+	 * is stacked on .request_fn device(s).
+	 */
+	if (!dm_table_all_blk_mq_devices(table)) {
+		if (!clone_old_rq(rq, md, tio, gfp_mask)) {
+			dm_put_live_table(md, srcu_idx);
+			free_old_rq_tio(tio);
+			return NULL;
+		}
+	}
+	dm_put_live_table(md, srcu_idx);
+
+	return tio;
+}
+
+/*
+ * Called with the queue lock held.
+ */
+static int dm_old_prep_fn(struct request_queue *q, struct request *rq)
+{
+	struct mapped_device *md = q->queuedata;
+	struct dm_rq_target_io *tio;
+
+	if (unlikely(rq->special)) {
+		DMWARN("Already has something in rq->special.");
+		return BLKPREP_KILL;
+	}
+
+	tio = dm_old_prep_tio(rq, md, GFP_ATOMIC);
+	if (!tio)
+		return BLKPREP_DEFER;
+
+	rq->special = tio;
+	rq->cmd_flags |= REQ_DONTPREP;
+
+	return BLKPREP_OK;
+}
+
+/*
+ * Returns:
+ * 0                : the request has been processed
+ * DM_MAPIO_REQUEUE : the original request needs to be requeued
+ * < 0              : the request was completed due to failure
+ */
+static int map_request(struct dm_rq_target_io *tio, struct request *rq,
+		       struct mapped_device *md)
+{
+	int r;
+	struct dm_target *ti = tio->ti;
+	struct request *clone = NULL;
+
+	if (tio->clone) {
+		clone = tio->clone;
+		r = ti->type->map_rq(ti, clone, &tio->info);
+	} else {
+		r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone);
+		if (r < 0) {
+			/* The target wants to complete the I/O */
+			dm_kill_unmapped_request(rq, r);
+			return r;
+		}
+		if (r != DM_MAPIO_REMAPPED)
+			return r;
+		if (setup_clone(clone, rq, tio, GFP_ATOMIC)) {
+			/* -ENOMEM */
+			ti->type->release_clone_rq(clone);
+			return DM_MAPIO_REQUEUE;
+		}
+	}
+
+	switch (r) {
+	case DM_MAPIO_SUBMITTED:
+		/* The target has taken the I/O to submit by itself later */
+		break;
+	case DM_MAPIO_REMAPPED:
+		/* The target has remapped the I/O so dispatch it */
+		trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
+				     blk_rq_pos(rq));
+		dm_dispatch_clone_request(clone, rq);
+		break;
+	case DM_MAPIO_REQUEUE:
+		/* The target wants to requeue the I/O */
+		dm_requeue_original_request(md, tio->orig);
+		break;
+	default:
+		if (r > 0) {
+			DMWARN("unimplemented target map return value: %d", r);
+			BUG();
+		}
+
+		/* The target wants to complete the I/O */
+		dm_kill_unmapped_request(rq, r);
+		return r;
+	}
+
+	return 0;
+}
+
+static void dm_start_request(struct mapped_device *md, struct request *orig)
+{
+	if (!orig->q->mq_ops)
+		blk_start_request(orig);
+	else
+		blk_mq_start_request(orig);
+	atomic_inc(&md->pending[rq_data_dir(orig)]);
+
+	if (md->seq_rq_merge_deadline_usecs) {
+		md->last_rq_pos = rq_end_sector(orig);
+		md->last_rq_rw = rq_data_dir(orig);
+		md->last_rq_start_time = ktime_get();
+	}
+
+	if (unlikely(dm_stats_used(&md->stats))) {
+		struct dm_rq_target_io *tio = tio_from_request(orig);
+		tio->duration_jiffies = jiffies;
+		tio->n_sectors = blk_rq_sectors(orig);
+		dm_stats_account_io(&md->stats, rq_data_dir(orig),
+				    blk_rq_pos(orig), tio->n_sectors, false, 0,
+				    &tio->stats_aux);
+	}
+
+	/*
+	 * Hold the md reference here for the in-flight I/O.
+	 * We can't rely on the reference count by device opener,
+	 * because the device may be closed during the request completion
+	 * when all bios are completed.
+	 * See the comment in rq_completed() too.
+	 */
+	dm_get(md);
+}
+
+static void map_tio_request(struct kthread_work *work)
+{
+	struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work);
+	struct request *rq = tio->orig;
+	struct mapped_device *md = tio->md;
+
+	if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE)
+		dm_requeue_original_request(md, rq);
+}
+
+ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf)
+{
+	return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs);
+}
+
+#define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000
+
+ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md,
+						     const char *buf, size_t count)
+{
+	unsigned deadline;
+
+	if (dm_get_md_type(md) != DM_TYPE_REQUEST_BASED)
+		return count;
+
+	if (kstrtouint(buf, 10, &deadline))
+		return -EINVAL;
+
+	if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS)
+		deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS;
+
+	md->seq_rq_merge_deadline_usecs = deadline;
+
+	return count;
+}
+
+static bool dm_old_request_peeked_before_merge_deadline(struct mapped_device *md)
+{
+	ktime_t kt_deadline;
+
+	if (!md->seq_rq_merge_deadline_usecs)
+		return false;
+
+	kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC);
+	kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline);
+
+	return !ktime_after(ktime_get(), kt_deadline);
+}
+
+/*
+ * q->request_fn for old request-based dm.
+ * Called with the queue lock held.
+ */
+static void dm_old_request_fn(struct request_queue *q)
+{
+	struct mapped_device *md = q->queuedata;
+	struct dm_target *ti = md->immutable_target;
+	struct request *rq;
+	struct dm_rq_target_io *tio;
+	sector_t pos = 0;
+
+	if (unlikely(!ti)) {
+		int srcu_idx;
+		struct dm_table *map = dm_get_live_table(md, &srcu_idx);
+
+		ti = dm_table_find_target(map, pos);
+		dm_put_live_table(md, srcu_idx);
+	}
+
+	/*
+	 * For suspend, check blk_queue_stopped() and increment
+	 * ->pending within a single queue_lock not to increment the
+	 * number of in-flight I/Os after the queue is stopped in
+	 * dm_suspend().
+	 */
+	while (!blk_queue_stopped(q)) {
+		rq = blk_peek_request(q);
+		if (!rq)
+			return;
+
+		/* always use block 0 to find the target for flushes for now */
+		pos = 0;
+		if (req_op(rq) != REQ_OP_FLUSH)
+			pos = blk_rq_pos(rq);
+
+		if ((dm_old_request_peeked_before_merge_deadline(md) &&
+		     md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 &&
+		     md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq)) ||
+		    (ti->type->busy && ti->type->busy(ti))) {
+			blk_delay_queue(q, 10);
+			return;
+		}
+
+		dm_start_request(md, rq);
+
+		tio = tio_from_request(rq);
+		/* Establish tio->ti before queuing work (map_tio_request) */
+		tio->ti = ti;
+		queue_kthread_work(&md->kworker, &tio->work);
+		BUG_ON(!irqs_disabled());
+	}
+}
+
+/*
+ * Fully initialize a .request_fn request-based queue.
+ */
+int dm_old_init_request_queue(struct mapped_device *md)
+{
+	/* Fully initialize the queue */
+	if (!blk_init_allocated_queue(md->queue, dm_old_request_fn, NULL))
+		return -EINVAL;
+
+	/* disable dm_old_request_fn's merge heuristic by default */
+	md->seq_rq_merge_deadline_usecs = 0;
+
+	dm_init_normal_md_queue(md);
+	blk_queue_softirq_done(md->queue, dm_softirq_done);
+	blk_queue_prep_rq(md->queue, dm_old_prep_fn);
+
+	/* Initialize the request-based DM worker thread */
+	init_kthread_worker(&md->kworker);
+	md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker,
+				       "kdmwork-%s", dm_device_name(md));
+	if (IS_ERR(md->kworker_task))
+		return PTR_ERR(md->kworker_task);
+
+	elv_register_queue(md->queue);
+
+	return 0;
+}
+
+static int dm_mq_init_request(void *data, struct request *rq,
+		       unsigned int hctx_idx, unsigned int request_idx,
+		       unsigned int numa_node)
+{
+	struct mapped_device *md = data;
+	struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
+
+	/*
+	 * Must initialize md member of tio, otherwise it won't
+	 * be available in dm_mq_queue_rq.
+	 */
+	tio->md = md;
+
+	if (md->init_tio_pdu) {
+		/* target-specific per-io data is immediately after the tio */
+		tio->info.ptr = tio + 1;
+	}
+
+	return 0;
+}
+
+static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx,
+			  const struct blk_mq_queue_data *bd)
+{
+	struct request *rq = bd->rq;
+	struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
+	struct mapped_device *md = tio->md;
+	struct dm_target *ti = md->immutable_target;
+
+	if (unlikely(!ti)) {
+		int srcu_idx;
+		struct dm_table *map = dm_get_live_table(md, &srcu_idx);
+
+		ti = dm_table_find_target(map, 0);
+		dm_put_live_table(md, srcu_idx);
+	}
+
+	if (ti->type->busy && ti->type->busy(ti))
+		return BLK_MQ_RQ_QUEUE_BUSY;
+
+	dm_start_request(md, rq);
+
+	/* Init tio using md established in .init_request */
+	init_tio(tio, rq, md);
+
+	/*
+	 * Establish tio->ti before calling map_request().
+	 */
+	tio->ti = ti;
+
+	/* Direct call is fine since .queue_rq allows allocations */
+	if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) {
+		/* Undo dm_start_request() before requeuing */
+		rq_end_stats(md, rq);
+		rq_completed(md, rq_data_dir(rq), false);
+		return BLK_MQ_RQ_QUEUE_BUSY;
+	}
+
+	return BLK_MQ_RQ_QUEUE_OK;
+}
+
+static struct blk_mq_ops dm_mq_ops = {
+	.queue_rq = dm_mq_queue_rq,
+	.map_queue = blk_mq_map_queue,
+	.complete = dm_softirq_done,
+	.init_request = dm_mq_init_request,
+};
+
+int dm_mq_init_request_queue(struct mapped_device *md, struct dm_table *t)
+{
+	struct request_queue *q;
+	struct dm_target *immutable_tgt;
+	int err;
+
+	if (!dm_table_all_blk_mq_devices(t)) {
+		DMERR("request-based dm-mq may only be stacked on blk-mq device(s)");
+		return -EINVAL;
+	}
+
+	md->tag_set = kzalloc_node(sizeof(struct blk_mq_tag_set), GFP_KERNEL, md->numa_node_id);
+	if (!md->tag_set)
+		return -ENOMEM;
+
+	md->tag_set->ops = &dm_mq_ops;
+	md->tag_set->queue_depth = dm_get_blk_mq_queue_depth();
+	md->tag_set->numa_node = md->numa_node_id;
+	md->tag_set->flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
+	md->tag_set->nr_hw_queues = dm_get_blk_mq_nr_hw_queues();
+	md->tag_set->driver_data = md;
+
+	md->tag_set->cmd_size = sizeof(struct dm_rq_target_io);
+	immutable_tgt = dm_table_get_immutable_target(t);
+	if (immutable_tgt && immutable_tgt->per_io_data_size) {
+		/* any target-specific per-io data is immediately after the tio */
+		md->tag_set->cmd_size += immutable_tgt->per_io_data_size;
+		md->init_tio_pdu = true;
+	}
+
+	err = blk_mq_alloc_tag_set(md->tag_set);
+	if (err)
+		goto out_kfree_tag_set;
+
+	q = blk_mq_init_allocated_queue(md->tag_set, md->queue);
+	if (IS_ERR(q)) {
+		err = PTR_ERR(q);
+		goto out_tag_set;
+	}
+	dm_init_md_queue(md);
+
+	/* backfill 'mq' sysfs registration normally done in blk_register_queue */
+	blk_mq_register_disk(md->disk);
+
+	return 0;
+
+out_tag_set:
+	blk_mq_free_tag_set(md->tag_set);
+out_kfree_tag_set:
+	kfree(md->tag_set);
+
+	return err;
+}
+
+void dm_mq_cleanup_mapped_device(struct mapped_device *md)
+{
+	if (md->tag_set) {
+		blk_mq_free_tag_set(md->tag_set);
+		kfree(md->tag_set);
+	}
+}
+
+module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
+
+module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices");
+
+module_param(dm_mq_nr_hw_queues, uint, S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(dm_mq_nr_hw_queues, "Number of hardware queues for request-based dm-mq devices");
+
+module_param(dm_mq_queue_depth, uint, S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(dm_mq_queue_depth, "Queue depth for request-based dm-mq devices");

drivers/md/dm-rq.h

+/*
+ * Internal header file for device mapper
+ *
+ * Copyright (C) 2016 Red Hat, Inc. All rights reserved.
+ *
+ * This file is released under the LGPL.
+ */
+
+#ifndef DM_RQ_INTERNAL_H
+#define DM_RQ_INTERNAL_H
+
+#include <linux/bio.h>
+#include <linux/kthread.h>
+
+#include "dm-stats.h"
+
+struct mapped_device;
+
+/*
+ * One of these is allocated per request.
+ */
+struct dm_rq_target_io {
+	struct mapped_device *md;
+	struct dm_target *ti;
+	struct request *orig, *clone;
+	struct kthread_work work;
+	int error;
+	union map_info info;
+	struct dm_stats_aux stats_aux;
+	unsigned long duration_jiffies;
+	unsigned n_sectors;
+};
+
+/*
+ * For request-based dm - the bio clones we allocate are embedded in these
+ * structs.
+ *
+ * We allocate these with bio_alloc_bioset, using the front_pad parameter when
+ * the bioset is created - this means the bio has to come at the end of the
+ * struct.
+ */
+struct dm_rq_clone_bio_info {
+	struct bio *orig;
+	struct dm_rq_target_io *tio;
+	struct bio clone;
+};
+
+bool dm_use_blk_mq_default(void);
+bool dm_use_blk_mq(struct mapped_device *md);
+
+int dm_old_init_request_queue(struct mapped_device *md);
+int dm_mq_init_request_queue(struct mapped_device *md, struct dm_table *t);
+void dm_mq_cleanup_mapped_device(struct mapped_device *md);
+
+void dm_start_queue(struct request_queue *q);
+void dm_stop_queue(struct request_queue *q);
+
+unsigned dm_get_reserved_rq_based_ios(void);
+
+ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf);
+ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md,
+						     const char *buf, size_t count);
+
+#endif

drivers/md/dm-snap.c

@@ -2302,6 +2302,13 @@ static int origin_map(struct dm_target *ti, struct bio *bio)
 	return do_origin(o->dev, bio);
 }
 
+static long origin_direct_access(struct dm_target *ti, sector_t sector,
+		void __pmem **kaddr, pfn_t *pfn, long size)
+{
+	DMWARN("device does not support dax.");
+	return -EIO;
+}
+
 /*
  * Set the target "max_io_len" field to the minimum of all the snapshots'
  * chunk sizes.
@@ -2361,6 +2368,7 @@ static struct target_type origin_target = {
 	.postsuspend = origin_postsuspend,
 	.status  = origin_status,
 	.iterate_devices = origin_iterate_devices,
+	.direct_access = origin_direct_access,
 };
 
 static struct target_type snapshot_target = {

drivers/md/dm-stats.c

@@ -10,7 +10,7 @@
 #include <linux/module.h>
 #include <linux/device-mapper.h>
 
-#include "dm.h"
+#include "dm-core.h"
 #include "dm-stats.h"
 
 #define DM_MSG_PREFIX "stats"

drivers/md/dm-stripe.c

@@ -308,6 +308,29 @@ static int stripe_map(struct dm_target *ti, struct bio *bio)
 	return DM_MAPIO_REMAPPED;
 }
 
+static long stripe_direct_access(struct dm_target *ti, sector_t sector,
+				 void __pmem **kaddr, pfn_t *pfn, long size)
+{
+	struct stripe_c *sc = ti->private;
+	uint32_t stripe;
+	struct block_device *bdev;
+	struct blk_dax_ctl dax = {
+		.size = size,
+	};
+	long ret;
+
+	stripe_map_sector(sc, sector, &stripe, &dax.sector);
+
+	dax.sector += sc->stripe[stripe].physical_start;
+	bdev = sc->stripe[stripe].dev->bdev;
+
+	ret = bdev_direct_access(bdev, &dax);
+	*kaddr = dax.addr;
+	*pfn = dax.pfn;
+
+	return ret;
+}
+
 /*
  * Stripe status:
  *
@@ -416,7 +439,7 @@ static void stripe_io_hints(struct dm_target *ti,
 
 static struct target_type stripe_target = {
 	.name   = "striped",
-	.version = {1, 5, 1},
+	.version = {1, 6, 0},
 	.module = THIS_MODULE,
 	.ctr    = stripe_ctr,
 	.dtr    = stripe_dtr,
@@ -425,6 +448,7 @@ static struct target_type stripe_target = {
 	.status = stripe_status,
 	.iterate_devices = stripe_iterate_devices,
 	.io_hints = stripe_io_hints,
+	.direct_access = stripe_direct_access,
 };
 
 int __init dm_stripe_init(void)

drivers/md/dm-sysfs.c

@@ -6,7 +6,8 @@
 
 #include <linux/sysfs.h>
 #include <linux/dm-ioctl.h>
-#include "dm.h"
+#include "dm-core.h"
+#include "dm-rq.h"
 
 struct dm_sysfs_attr {
 	struct attribute attr;

drivers/md/dm-table.c

@@ -5,7 +5,7 @@
  * This file is released under the GPL.
  */
 
-#include "dm.h"
+#include "dm-core.h"
 
 #include <linux/module.h>
 #include <linux/vmalloc.h>
@@ -43,8 +43,10 @@ struct dm_table {
 	struct dm_target *targets;
 
 	struct target_type *immutable_target_type;
-	unsigned integrity_supported:1;
-	unsigned singleton:1;
+
+	bool integrity_supported:1;
+	bool singleton:1;
+	bool all_blk_mq:1;
 
 	/*
 	 * Indicates the rw permissions for the new logical
@@ -206,6 +208,7 @@ int dm_table_create(struct dm_table **result, fmode_t mode,
 		return -ENOMEM;
 	}
 
+	t->type = DM_TYPE_NONE;
 	t->mode = mode;
 	t->md = md;
 	*result = t;
@@ -703,7 +706,7 @@ int dm_table_add_target(struct dm_table *t, const char *type,
 			      dm_device_name(t->md), type);
 			return -EINVAL;
 		}
-		t->singleton = 1;
+		t->singleton = true;
 	}
 
 	if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
@@ -824,22 +827,70 @@ void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
 }
 EXPORT_SYMBOL(dm_consume_args);
 
+static bool __table_type_bio_based(unsigned table_type)
+{
+	return (table_type == DM_TYPE_BIO_BASED ||
+		table_type == DM_TYPE_DAX_BIO_BASED);
+}
+
 static bool __table_type_request_based(unsigned table_type)
 {
 	return (table_type == DM_TYPE_REQUEST_BASED ||
 		table_type == DM_TYPE_MQ_REQUEST_BASED);
 }
 
-static int dm_table_set_type(struct dm_table *t)
+void dm_table_set_type(struct dm_table *t, unsigned type)
+{
+	t->type = type;
+}
+EXPORT_SYMBOL_GPL(dm_table_set_type);
+
+static int device_supports_dax(struct dm_target *ti, struct dm_dev *dev,
+			       sector_t start, sector_t len, void *data)
+{
+	struct request_queue *q = bdev_get_queue(dev->bdev);
+
+	return q && blk_queue_dax(q);
+}
+
+static bool dm_table_supports_dax(struct dm_table *t)
+{
+	struct dm_target *ti;
+	unsigned i = 0;
+
+	/* Ensure that all targets support DAX. */
+	while (i < dm_table_get_num_targets(t)) {
+		ti = dm_table_get_target(t, i++);
+
+		if (!ti->type->direct_access)
+			return false;
+
+		if (!ti->type->iterate_devices ||
+		    !ti->type->iterate_devices(ti, device_supports_dax, NULL))
+			return false;
+	}
+
+	return true;
+}
+
+static int dm_table_determine_type(struct dm_table *t)
 {
 	unsigned i;
 	unsigned bio_based = 0, request_based = 0, hybrid = 0;
-	bool use_blk_mq = false;
+	bool verify_blk_mq = false;
 	struct dm_target *tgt;
 	struct dm_dev_internal *dd;
-	struct list_head *devices;
+	struct list_head *devices = dm_table_get_devices(t);
 	unsigned live_md_type = dm_get_md_type(t->md);
 
+	if (t->type != DM_TYPE_NONE) {
+		/* target already set the table's type */
+		if (t->type == DM_TYPE_BIO_BASED)
+			return 0;
+		BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
+		goto verify_rq_based;
+	}
+
 	for (i = 0; i < t->num_targets; i++) {
 		tgt = t->targets + i;
 		if (dm_target_hybrid(tgt))
@@ -871,11 +922,27 @@ static int dm_table_set_type(struct dm_table *t)
 	if (bio_based) {
 		/* We must use this table as bio-based */
 		t->type = DM_TYPE_BIO_BASED;
+		if (dm_table_supports_dax(t) ||
+		    (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED))
+			t->type = DM_TYPE_DAX_BIO_BASED;
 		return 0;
 	}
 
 	BUG_ON(!request_based); /* No targets in this table */
 
+	if (list_empty(devices) && __table_type_request_based(live_md_type)) {
+		/* inherit live MD type */
+		t->type = live_md_type;
+		return 0;
+	}
+
+	/*
+	 * The only way to establish DM_TYPE_MQ_REQUEST_BASED is by
+	 * having a compatible target use dm_table_set_type.
+	 */
+	t->type = DM_TYPE_REQUEST_BASED;
+
+verify_rq_based:
 	/*
 	 * Request-based dm supports only tables that have a single target now.
 	 * To support multiple targets, request splitting support is needed,
@@ -888,7 +955,6 @@ static int dm_table_set_type(struct dm_table *t)
 	}
 
 	/* Non-request-stackable devices can't be used for request-based dm */
-	devices = dm_table_get_devices(t);
 	list_for_each_entry(dd, devices, list) {
 		struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
 
@@ -899,10 +965,10 @@ static int dm_table_set_type(struct dm_table *t)
 		}
 
 		if (q->mq_ops)
-			use_blk_mq = true;
+			verify_blk_mq = true;
 	}
 
-	if (use_blk_mq) {
+	if (verify_blk_mq) {
 		/* verify _all_ devices in the table are blk-mq devices */
 		list_for_each_entry(dd, devices, list)
 			if (!bdev_get_queue(dd->dm_dev->bdev)->mq_ops) {
@@ -910,14 +976,9 @@ static int dm_table_set_type(struct dm_table *t)
 				      " are blk-mq request-stackable");
 				return -EINVAL;
 			}
-		t->type = DM_TYPE_MQ_REQUEST_BASED;
-
-	} else if (list_empty(devices) && __table_type_request_based(live_md_type)) {
-		/* inherit live MD type */
-		t->type = live_md_type;
 
-	} else
-		t->type = DM_TYPE_REQUEST_BASED;
+		t->all_blk_mq = true;
+	}
 
 	return 0;
 }
@@ -956,14 +1017,19 @@ struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
 	return NULL;
 }
 
+bool dm_table_bio_based(struct dm_table *t)
+{
+	return __table_type_bio_based(dm_table_get_type(t));
+}
+
 bool dm_table_request_based(struct dm_table *t)
 {
 	return __table_type_request_based(dm_table_get_type(t));
 }
 
-bool dm_table_mq_request_based(struct dm_table *t)
+bool dm_table_all_blk_mq_devices(struct dm_table *t)
 {
-	return dm_table_get_type(t) == DM_TYPE_MQ_REQUEST_BASED;
+	return t->all_blk_mq;
 }
 
 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
@@ -978,7 +1044,7 @@ static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *
 		return -EINVAL;
 	}
 
-	if (type == DM_TYPE_BIO_BASED)
+	if (__table_type_bio_based(type))
 		for (i = 0; i < t->num_targets; i++) {
 			tgt = t->targets + i;
 			per_io_data_size = max(per_io_data_size, tgt->per_io_data_size);
@@ -1106,7 +1172,7 @@ static int dm_table_register_integrity(struct dm_table *t)
 		return 0;
 
 	if (!integrity_profile_exists(dm_disk(md))) {
-		t->integrity_supported = 1;
+		t->integrity_supported = true;
 		/*
 		 * Register integrity profile during table load; we can do
 		 * this because the final profile must match during resume.
@@ -1129,7 +1195,7 @@ static int dm_table_register_integrity(struct dm_table *t)
 	}
 
 	/* Preserve existing integrity profile */
-	t->integrity_supported = 1;
+	t->integrity_supported = true;
 	return 0;
 }
 
@@ -1141,9 +1207,9 @@ int dm_table_complete(struct dm_table *t)
 {
 	int r;
 
-	r = dm_table_set_type(t);
+	r = dm_table_determine_type(t);
 	if (r) {
-		DMERR("unable to set table type");
+		DMERR("unable to determine table type");
 		return r;
 	}
 

drivers/md/dm-target.c

@@ -4,7 +4,7 @@
  * This file is released under the GPL.
  */
 
-#include "dm.h"
+#include "dm-core.h"
 
 #include <linux/module.h>
 #include <linux/init.h>
@@ -148,9 +148,15 @@ static void io_err_release_clone_rq(struct request *clone)
 {
 }
 
+static long io_err_direct_access(struct dm_target *ti, sector_t sector,
+				 void __pmem **kaddr, pfn_t *pfn, long size)
+{
+	return -EIO;
+}
+
 static struct target_type error_target = {
 	.name = "error",
-	.version = {1, 4, 0},
+	.version = {1, 5, 0},
 	.features = DM_TARGET_WILDCARD,
 	.ctr  = io_err_ctr,
 	.dtr  = io_err_dtr,
@@ -158,6 +164,7 @@ static struct target_type error_target = {
 	.map_rq = io_err_map_rq,
 	.clone_and_map_rq = io_err_clone_and_map_rq,
 	.release_clone_rq = io_err_release_clone_rq,
+	.direct_access = io_err_direct_access,
 };
 
 int __init dm_target_init(void)

drivers/md/dm-thin-metadata.c

@@ -1677,6 +1677,36 @@ int dm_pool_block_is_used(struct dm_pool_metadata *pmd, dm_block_t b, bool *resu
 	return r;
 }
 
+int dm_pool_inc_data_range(struct dm_pool_metadata *pmd, dm_block_t b, dm_block_t e)
+{
+	int r = 0;
+
+	down_write(&pmd->root_lock);
+	for (; b != e; b++) {
+		r = dm_sm_inc_block(pmd->data_sm, b);
+		if (r)
+			break;
+	}
+	up_write(&pmd->root_lock);
+
+	return r;
+}
+
+int dm_pool_dec_data_range(struct dm_pool_metadata *pmd, dm_block_t b, dm_block_t e)
+{
+	int r = 0;
+
+	down_write(&pmd->root_lock);
+	for (; b != e; b++) {
+		r = dm_sm_dec_block(pmd->data_sm, b);
+		if (r)
+			break;
+	}
+	up_write(&pmd->root_lock);
+
+	return r;
+}
+
 bool dm_thin_changed_this_transaction(struct dm_thin_device *td)
 {
 	int r;

drivers/md/dm-thin-metadata.h

@@ -197,6 +197,9 @@ int dm_pool_get_data_dev_size(struct dm_pool_metadata *pmd, dm_block_t *result);
 
 int dm_pool_block_is_used(struct dm_pool_metadata *pmd, dm_block_t b, bool *result);
 
+int dm_pool_inc_data_range(struct dm_pool_metadata *pmd, dm_block_t b, dm_block_t e);
+int dm_pool_dec_data_range(struct dm_pool_metadata *pmd, dm_block_t b, dm_block_t e);
+
 /*
  * Returns -ENOSPC if the new size is too small and already allocated
  * blocks would be lost.

drivers/md/dm-thin.c

@@ -253,6 +253,7 @@ struct pool {
 	struct bio_list deferred_flush_bios;
 	struct list_head prepared_mappings;
 	struct list_head prepared_discards;
+	struct list_head prepared_discards_pt2;
 	struct list_head active_thins;
 
 	struct dm_deferred_set *shared_read_ds;
@@ -269,6 +270,7 @@ struct pool {
 
 	process_mapping_fn process_prepared_mapping;
 	process_mapping_fn process_prepared_discard;
+	process_mapping_fn process_prepared_discard_pt2;
 
 	struct dm_bio_prison_cell **cell_sort_array;
 };
@@ -1001,7 +1003,8 @@ static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping *m)
 
 /*----------------------------------------------------------------*/
 
-static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m)
+static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m,
+						   struct bio *discard_parent)
 {
 	/*
 	 * We've already unmapped this range of blocks, but before we
@@ -1014,7 +1017,7 @@ static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m
 	dm_block_t b = m->data_block, e, end = m->data_block + m->virt_end - m->virt_begin;
 	struct discard_op op;
 
-	begin_discard(&op, tc, m->bio);
+	begin_discard(&op, tc, discard_parent);
 	while (b != end) {
 		/* find start of unmapped run */
 		for (; b < end; b++) {
@@ -1049,27 +1052,100 @@ static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m
 	end_discard(&op, r);
 }
 
-static void process_prepared_discard_passdown(struct dm_thin_new_mapping *m)
+static void queue_passdown_pt2(struct dm_thin_new_mapping *m)
+{
+	unsigned long flags;
+	struct pool *pool = m->tc->pool;
+
+	spin_lock_irqsave(&pool->lock, flags);
+	list_add_tail(&m->list, &pool->prepared_discards_pt2);
+	spin_unlock_irqrestore(&pool->lock, flags);
+	wake_worker(pool);
+}
+
+static void passdown_endio(struct bio *bio)
+{
+	/*
+	 * It doesn't matter if the passdown discard failed, we still want
+	 * to unmap (we ignore err).
+	 */
+	queue_passdown_pt2(bio->bi_private);
+}
+
+static void process_prepared_discard_passdown_pt1(struct dm_thin_new_mapping *m)
 {
 	int r;
 	struct thin_c *tc = m->tc;
 	struct pool *pool = tc->pool;
+	struct bio *discard_parent;
+	dm_block_t data_end = m->data_block + (m->virt_end - m->virt_begin);
 
+	/*
+	 * Only this thread allocates blocks, so we can be sure that the
+	 * newly unmapped blocks will not be allocated before the end of
+	 * the function.
+	 */
 	r = dm_thin_remove_range(tc->td, m->virt_begin, m->virt_end);
 	if (r) {
 		metadata_operation_failed(pool, "dm_thin_remove_range", r);
 		bio_io_error(m->bio);
+		cell_defer_no_holder(tc, m->cell);
+		mempool_free(m, pool->mapping_pool);
+		return;
+	}
 
-	} else if (m->maybe_shared) {
-		passdown_double_checking_shared_status(m);
+	discard_parent = bio_alloc(GFP_NOIO, 1);
+	if (!discard_parent) {
+		DMWARN("%s: unable to allocate top level discard bio for passdown. Skipping passdown.",
+		       dm_device_name(tc->pool->pool_md));
+		queue_passdown_pt2(m);
 
 	} else {
+		discard_parent->bi_end_io = passdown_endio;
+		discard_parent->bi_private = m;
+
+		if (m->maybe_shared)
+			passdown_double_checking_shared_status(m, discard_parent);
+		else {
 			struct discard_op op;
-		begin_discard(&op, tc, m->bio);
-		r = issue_discard(&op, m->data_block,
-				  m->data_block + (m->virt_end - m->virt_begin));
+
+			begin_discard(&op, tc, discard_parent);
+			r = issue_discard(&op, m->data_block, data_end);
 			end_discard(&op, r);
 		}
+	}
+
+	/*
+	 * Increment the unmapped blocks.  This prevents a race between the
+	 * passdown io and reallocation of freed blocks.
+	 */
+	r = dm_pool_inc_data_range(pool->pmd, m->data_block, data_end);
+	if (r) {
+		metadata_operation_failed(pool, "dm_pool_inc_data_range", r);
+		bio_io_error(m->bio);
+		cell_defer_no_holder(tc, m->cell);
+		mempool_free(m, pool->mapping_pool);
+		return;
+	}
+}
+
+static void process_prepared_discard_passdown_pt2(struct dm_thin_new_mapping *m)
+{
+	int r;
+	struct thin_c *tc = m->tc;
+	struct pool *pool = tc->pool;
+
+	/*
+	 * The passdown has completed, so now we can decrement all those
+	 * unmapped blocks.
+	 */
+	r = dm_pool_dec_data_range(pool->pmd, m->data_block,
+				   m->data_block + (m->virt_end - m->virt_begin));
+	if (r) {
+		metadata_operation_failed(pool, "dm_pool_dec_data_range", r);
+		bio_io_error(m->bio);
+	} else
+		bio_endio(m->bio);
 
 	cell_defer_no_holder(tc, m->cell);
 	mempool_free(m, pool->mapping_pool);
@@ -2215,6 +2291,8 @@ static void do_worker(struct work_struct *ws)
 	throttle_work_update(&pool->throttle);
 	process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
 	throttle_work_update(&pool->throttle);
+	process_prepared(pool, &pool->prepared_discards_pt2, &pool->process_prepared_discard_pt2);
+	throttle_work_update(&pool->throttle);
 	process_deferred_bios(pool);
 	throttle_work_complete(&pool->throttle);
 }
@@ -2343,7 +2421,8 @@ static void set_discard_callbacks(struct pool *pool)
 
 	if (passdown_enabled(pt)) {
 		pool->process_discard_cell = process_discard_cell_passdown;
-		pool->process_prepared_discard = process_prepared_discard_passdown;
+		pool->process_prepared_discard = process_prepared_discard_passdown_pt1;
+		pool->process_prepared_discard_pt2 = process_prepared_discard_passdown_pt2;
 	} else {
 		pool->process_discard_cell = process_discard_cell_no_passdown;
 		pool->process_prepared_discard = process_prepared_discard_no_passdown;
@@ -2830,6 +2909,7 @@ static struct pool *pool_create(struct mapped_device *pool_md,
 	bio_list_init(&pool->deferred_flush_bios);
 	INIT_LIST_HEAD(&pool->prepared_mappings);
 	INIT_LIST_HEAD(&pool->prepared_discards);
+	INIT_LIST_HEAD(&pool->prepared_discards_pt2);
 	INIT_LIST_HEAD(&pool->active_thins);
 	pool->low_water_triggered = false;
 	pool->suspended = true;

drivers/md/dm-verity-fec.c

@@ -453,9 +453,7 @@ int verity_fec_decode(struct dm_verity *v, struct dm_verity_io *io,
 	 */
 
 	offset = block << v->data_dev_block_bits;
-
-	res = offset;
-	div64_u64(res, v->fec->rounds << v->data_dev_block_bits);
+	res = div64_u64(offset, v->fec->rounds << v->data_dev_block_bits);
 
 	/*
 	 * The base RS block we can feed to the interleaver to find out all

drivers/md/dm.c

@@ -5,13 +5,13 @@
  * This file is released under the GPL.
  */
 
-#include "dm.h"
+#include "dm-core.h"
+#include "dm-rq.h"
 #include "dm-uevent.h"
 
 #include <linux/init.h>
 #include <linux/module.h>
 #include <linux/mutex.h>
-#include <linux/moduleparam.h>
 #include <linux/blkpg.h>
 #include <linux/bio.h>
 #include <linux/mempool.h>
@@ -20,14 +20,8 @@
 #include <linux/hdreg.h>
 #include <linux/delay.h>
 #include <linux/wait.h>
-#include <linux/kthread.h>
-#include <linux/ktime.h>
-#include <linux/elevator.h> /* for rq_end_sector() */
-#include <linux/blk-mq.h>
 #include <linux/pr.h>
 
-#include <trace/events/block.h>
-
 #define DM_MSG_PREFIX "core"
 
 #ifdef CONFIG_PRINTK
@@ -63,7 +57,6 @@ static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
 static struct workqueue_struct *deferred_remove_workqueue;
 
 /*
- * For bio-based dm.
  * One of these is allocated per bio.
  */
 struct dm_io {
@@ -76,36 +69,6 @@ struct dm_io {
 	struct dm_stats_aux stats_aux;
 };
 
-/*
- * For request-based dm.
- * One of these is allocated per request.
- */
-struct dm_rq_target_io {
-	struct mapped_device *md;
-	struct dm_target *ti;
-	struct request *orig, *clone;
-	struct kthread_work work;
-	int error;
-	union map_info info;
-	struct dm_stats_aux stats_aux;
-	unsigned long duration_jiffies;
-	unsigned n_sectors;
-};
-
-/*
- * For request-based dm - the bio clones we allocate are embedded in these
- * structs.
- *
- * We allocate these with bio_alloc_bioset, using the front_pad parameter when
- * the bioset is created - this means the bio has to come at the end of the
- * struct.
- */
-struct dm_rq_clone_bio_info {
-	struct bio *orig;
-	struct dm_rq_target_io *tio;
-	struct bio clone;
-};
-
 #define MINOR_ALLOCED ((void *)-1)
 
 /*
@@ -120,130 +83,9 @@ struct dm_rq_clone_bio_info {
 #define DMF_DEFERRED_REMOVE 6
 #define DMF_SUSPENDED_INTERNALLY 7
 
-/*
- * Work processed by per-device workqueue.
- */
-struct mapped_device {
-	struct srcu_struct io_barrier;
-	struct mutex suspend_lock;
-
-	/*
-	 * The current mapping (struct dm_table *).
-	 * Use dm_get_live_table{_fast} or take suspend_lock for
-	 * dereference.
-	 */
-	void __rcu *map;
-
-	struct list_head table_devices;
-	struct mutex table_devices_lock;
-
-	unsigned long flags;
-
-	struct request_queue *queue;
-	int numa_node_id;
-
-	unsigned type;
-	/* Protect queue and type against concurrent access. */
-	struct mutex type_lock;
-
-	atomic_t holders;
-	atomic_t open_count;
-
-	struct dm_target *immutable_target;
-	struct target_type *immutable_target_type;
-
-	struct gendisk *disk;
-	char name[16];
-
-	void *interface_ptr;
-
-	/*
-	 * A list of ios that arrived while we were suspended.
-	 */
-	atomic_t pending[2];
-	wait_queue_head_t wait;
-	struct work_struct work;
-	spinlock_t deferred_lock;
-	struct bio_list deferred;
-
-	/*
-	 * Event handling.
-	 */
-	wait_queue_head_t eventq;
-	atomic_t event_nr;
-	atomic_t uevent_seq;
-	struct list_head uevent_list;
-	spinlock_t uevent_lock; /* Protect access to uevent_list */
-
-	/* the number of internal suspends */
-	unsigned internal_suspend_count;
-
-	/*
-	 * Processing queue (flush)
-	 */
-	struct workqueue_struct *wq;
-
-	/*
-	 * io objects are allocated from here.
-	 */
-	mempool_t *io_pool;
-	mempool_t *rq_pool;
-
-	struct bio_set *bs;
-
-	/*
-	 * freeze/thaw support require holding onto a super block
-	 */
-	struct super_block *frozen_sb;
-
-	/* forced geometry settings */
-	struct hd_geometry geometry;
-
-	struct block_device *bdev;
-
-	/* kobject and completion */
-	struct dm_kobject_holder kobj_holder;
-
-	/* zero-length flush that will be cloned and submitted to targets */
-	struct bio flush_bio;
-
-	struct dm_stats stats;
-
-	struct kthread_worker kworker;
-	struct task_struct *kworker_task;
-
-	/* for request-based merge heuristic in dm_request_fn() */
-	unsigned seq_rq_merge_deadline_usecs;
-	int last_rq_rw;
-	sector_t last_rq_pos;
-	ktime_t last_rq_start_time;
-
-	/* for blk-mq request-based DM support */
-	struct blk_mq_tag_set *tag_set;
-	bool use_blk_mq:1;
-	bool init_tio_pdu:1;
-};
-
-#ifdef CONFIG_DM_MQ_DEFAULT
-static bool use_blk_mq = true;
-#else
-static bool use_blk_mq = false;
-#endif
-
-#define DM_MQ_NR_HW_QUEUES 1
-#define DM_MQ_QUEUE_DEPTH 2048
 #define DM_NUMA_NODE NUMA_NO_NODE
-
-static unsigned dm_mq_nr_hw_queues = DM_MQ_NR_HW_QUEUES;
-static unsigned dm_mq_queue_depth = DM_MQ_QUEUE_DEPTH;
 static int dm_numa_node = DM_NUMA_NODE;
 
-bool dm_use_blk_mq(struct mapped_device *md)
-{
-	return md->use_blk_mq;
-}
-EXPORT_SYMBOL_GPL(dm_use_blk_mq);
-
 /*
  * For mempools pre-allocation at the table loading time.
  */
@@ -259,9 +101,6 @@ struct table_device {
 	struct dm_dev dm_dev;
 };
 
-#define RESERVED_BIO_BASED_IOS		16
-#define RESERVED_REQUEST_BASED_IOS	256
-#define RESERVED_MAX_IOS		1024
 static struct kmem_cache *_io_cache;
 static struct kmem_cache *_rq_tio_cache;
 static struct kmem_cache *_rq_cache;
@@ -269,13 +108,9 @@ static struct kmem_cache *_rq_cache;
 /*
  * Bio-based DM's mempools' reserved IOs set by the user.
  */
+#define RESERVED_BIO_BASED_IOS		16
 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
 
-/*
- * Request-based DM's mempools' reserved IOs set by the user.
- */
-static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS;
-
 static int __dm_get_module_param_int(int *module_param, int min, int max)
 {
 	int param = ACCESS_ONCE(*module_param);
@@ -297,7 +132,7 @@ static int __dm_get_module_param_int(int *module_param, int min, int max)
 	return param;
 }
 
-static unsigned __dm_get_module_param(unsigned *module_param,
+unsigned __dm_get_module_param(unsigned *module_param,
 			       unsigned def, unsigned max)
 {
 	unsigned param = ACCESS_ONCE(*module_param);
@@ -319,28 +154,10 @@ static unsigned __dm_get_module_param(unsigned *module_param,
 unsigned dm_get_reserved_bio_based_ios(void)
 {
 	return __dm_get_module_param(&reserved_bio_based_ios,
-				     RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS);
+				     RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
 }
 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
 
-unsigned dm_get_reserved_rq_based_ios(void)
-{
-	return __dm_get_module_param(&reserved_rq_based_ios,
-				     RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS);
-}
-EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios);
-
-static unsigned dm_get_blk_mq_nr_hw_queues(void)
-{
-	return __dm_get_module_param(&dm_mq_nr_hw_queues, 1, 32);
-}
-
-static unsigned dm_get_blk_mq_queue_depth(void)
-{
-	return __dm_get_module_param(&dm_mq_queue_depth,
-				     DM_MQ_QUEUE_DEPTH, BLK_MQ_MAX_DEPTH);
-}
-
 static unsigned dm_get_numa_node(void)
 {
 	return __dm_get_module_param_int(&dm_numa_node,
@@ -679,29 +496,7 @@ static void free_tio(struct dm_target_io *tio)
 	bio_put(&tio->clone);
 }
 
-static struct dm_rq_target_io *alloc_old_rq_tio(struct mapped_device *md,
-						gfp_t gfp_mask)
-{
-	return mempool_alloc(md->io_pool, gfp_mask);
-}
-
-static void free_old_rq_tio(struct dm_rq_target_io *tio)
-{
-	mempool_free(tio, tio->md->io_pool);
-}
-
-static struct request *alloc_old_clone_request(struct mapped_device *md,
-					       gfp_t gfp_mask)
-{
-	return mempool_alloc(md->rq_pool, gfp_mask);
-}
-
-static void free_old_clone_request(struct mapped_device *md, struct request *rq)
-{
-	mempool_free(rq, md->rq_pool);
-}
-
-static int md_in_flight(struct mapped_device *md)
+int md_in_flight(struct mapped_device *md)
 {
 	return atomic_read(&md->pending[READ]) +
 	       atomic_read(&md->pending[WRITE]);
@@ -1019,7 +814,7 @@ static void dec_pending(struct dm_io *io, int error)
 	}
 }
 
-static void disable_write_same(struct mapped_device *md)
+void disable_write_same(struct mapped_device *md)
 {
 	struct queue_limits *limits = dm_get_queue_limits(md);
 
@@ -1061,371 +856,6 @@ static void clone_endio(struct bio *bio)
 	dec_pending(io, error);
 }
 
-/*
- * Partial completion handling for request-based dm
- */
-static void end_clone_bio(struct bio *clone)
-{
-	struct dm_rq_clone_bio_info *info =
-		container_of(clone, struct dm_rq_clone_bio_info, clone);
-	struct dm_rq_target_io *tio = info->tio;
-	struct bio *bio = info->orig;
-	unsigned int nr_bytes = info->orig->bi_iter.bi_size;
-	int error = clone->bi_error;
-
-	bio_put(clone);
-
-	if (tio->error)
-		/*
-		 * An error has already been detected on the request.
-		 * Once error occurred, just let clone->end_io() handle
-		 * the remainder.
-		 */
-		return;
-	else if (error) {
-		/*
-		 * Don't notice the error to the upper layer yet.
-		 * The error handling decision is made by the target driver,
-		 * when the request is completed.
-		 */
-		tio->error = error;
-		return;
-	}
-
-	/*
-	 * I/O for the bio successfully completed.
-	 * Notice the data completion to the upper layer.
-	 */
-
-	/*
-	 * bios are processed from the head of the list.
-	 * So the completing bio should always be rq->bio.
-	 * If it's not, something wrong is happening.
-	 */
-	if (tio->orig->bio != bio)
-		DMERR("bio completion is going in the middle of the request");
-
-	/*
-	 * Update the original request.
-	 * Do not use blk_end_request() here, because it may complete
-	 * the original request before the clone, and break the ordering.
-	 */
-	blk_update_request(tio->orig, 0, nr_bytes);
-}
-
-static struct dm_rq_target_io *tio_from_request(struct request *rq)
-{
-	return (rq->q->mq_ops ? blk_mq_rq_to_pdu(rq) : rq->special);
-}
-
-static void rq_end_stats(struct mapped_device *md, struct request *orig)
-{
-	if (unlikely(dm_stats_used(&md->stats))) {
-		struct dm_rq_target_io *tio = tio_from_request(orig);
-		tio->duration_jiffies = jiffies - tio->duration_jiffies;
-		dm_stats_account_io(&md->stats, rq_data_dir(orig),
-				    blk_rq_pos(orig), tio->n_sectors, true,
-				    tio->duration_jiffies, &tio->stats_aux);
-	}
-}
-
-/*
- * Don't touch any member of the md after calling this function because
- * the md may be freed in dm_put() at the end of this function.
- * Or do dm_get() before calling this function and dm_put() later.
- */
-static void rq_completed(struct mapped_device *md, int rw, bool run_queue)
-{
-	atomic_dec(&md->pending[rw]);
-
-	/* nudge anyone waiting on suspend queue */
-	if (!md_in_flight(md))
-		wake_up(&md->wait);
-
-	/*
-	 * Run this off this callpath, as drivers could invoke end_io while
-	 * inside their request_fn (and holding the queue lock). Calling
-	 * back into ->request_fn() could deadlock attempting to grab the
-	 * queue lock again.
-	 */
-	if (!md->queue->mq_ops && run_queue)
-		blk_run_queue_async(md->queue);
-
-	/*
-	 * dm_put() must be at the end of this function. See the comment above
-	 */
-	dm_put(md);
-}
-
-static void free_rq_clone(struct request *clone)
-{
-	struct dm_rq_target_io *tio = clone->end_io_data;
-	struct mapped_device *md = tio->md;
-
-	blk_rq_unprep_clone(clone);
-
-	if (md->type == DM_TYPE_MQ_REQUEST_BASED)
-		/* stacked on blk-mq queue(s) */
-		tio->ti->type->release_clone_rq(clone);
-	else if (!md->queue->mq_ops)
-		/* request_fn queue stacked on request_fn queue(s) */
-		free_old_clone_request(md, clone);
-
-	if (!md->queue->mq_ops)
-		free_old_rq_tio(tio);
-}
-
-/*
- * Complete the clone and the original request.
- * Must be called without clone's queue lock held,
- * see end_clone_request() for more details.
- */
-static void dm_end_request(struct request *clone, int error)
-{
-	int rw = rq_data_dir(clone);
-	struct dm_rq_target_io *tio = clone->end_io_data;
-	struct mapped_device *md = tio->md;
-	struct request *rq = tio->orig;
-
-	if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
-		rq->errors = clone->errors;
-		rq->resid_len = clone->resid_len;
-
-		if (rq->sense)
-			/*
-			 * We are using the sense buffer of the original
-			 * request.
-			 * So setting the length of the sense data is enough.
-			 */
-			rq->sense_len = clone->sense_len;
-	}
-
-	free_rq_clone(clone);
-	rq_end_stats(md, rq);
-	if (!rq->q->mq_ops)
-		blk_end_request_all(rq, error);
-	else
-		blk_mq_end_request(rq, error);
-	rq_completed(md, rw, true);
-}
-
-static void dm_unprep_request(struct request *rq)
-{
-	struct dm_rq_target_io *tio = tio_from_request(rq);
-	struct request *clone = tio->clone;
-
-	if (!rq->q->mq_ops) {
-		rq->special = NULL;
-		rq->cmd_flags &= ~REQ_DONTPREP;
-	}
-
-	if (clone)
-		free_rq_clone(clone);
-	else if (!tio->md->queue->mq_ops)
-		free_old_rq_tio(tio);
-}
-
-/*
- * Requeue the original request of a clone.
- */
-static void dm_old_requeue_request(struct request *rq)
-{
-	struct request_queue *q = rq->q;
-	unsigned long flags;
-
-	spin_lock_irqsave(q->queue_lock, flags);
-	blk_requeue_request(q, rq);
-	blk_run_queue_async(q);
-	spin_unlock_irqrestore(q->queue_lock, flags);
-}
-
-static void dm_mq_requeue_request(struct request *rq)
-{
-	struct request_queue *q = rq->q;
-	unsigned long flags;
-
-	blk_mq_requeue_request(rq);
-	spin_lock_irqsave(q->queue_lock, flags);
-	if (!blk_queue_stopped(q))
-		blk_mq_kick_requeue_list(q);
-	spin_unlock_irqrestore(q->queue_lock, flags);
-}
-
-static void dm_requeue_original_request(struct mapped_device *md,
-					struct request *rq)
-{
-	int rw = rq_data_dir(rq);
-
-	rq_end_stats(md, rq);
-	dm_unprep_request(rq);
-
-	if (!rq->q->mq_ops)
-		dm_old_requeue_request(rq);
-	else
-		dm_mq_requeue_request(rq);
-
-	rq_completed(md, rw, false);
-}
-
-static void dm_old_stop_queue(struct request_queue *q)
-{
-	unsigned long flags;
-
-	spin_lock_irqsave(q->queue_lock, flags);
-	if (blk_queue_stopped(q)) {
-		spin_unlock_irqrestore(q->queue_lock, flags);
-		return;
-	}
-
-	blk_stop_queue(q);
-	spin_unlock_irqrestore(q->queue_lock, flags);
-}
-
-static void dm_stop_queue(struct request_queue *q)
-{
-	if (!q->mq_ops)
-		dm_old_stop_queue(q);
-	else
-		blk_mq_stop_hw_queues(q);
-}
-
-static void dm_old_start_queue(struct request_queue *q)
-{
-	unsigned long flags;
-
-	spin_lock_irqsave(q->queue_lock, flags);
-	if (blk_queue_stopped(q))
-		blk_start_queue(q);
-	spin_unlock_irqrestore(q->queue_lock, flags);
-}
-
-static void dm_start_queue(struct request_queue *q)
-{
-	if (!q->mq_ops)
-		dm_old_start_queue(q);
-	else {
-		blk_mq_start_stopped_hw_queues(q, true);
-		blk_mq_kick_requeue_list(q);
-	}
-}
-
-static void dm_done(struct request *clone, int error, bool mapped)
-{
-	int r = error;
-	struct dm_rq_target_io *tio = clone->end_io_data;
-	dm_request_endio_fn rq_end_io = NULL;
-
-	if (tio->ti) {
-		rq_end_io = tio->ti->type->rq_end_io;
-
-		if (mapped && rq_end_io)
-			r = rq_end_io(tio->ti, clone, error, &tio->info);
-	}
-
-	if (unlikely(r == -EREMOTEIO && (req_op(clone) == REQ_OP_WRITE_SAME) &&
-		     !clone->q->limits.max_write_same_sectors))
-		disable_write_same(tio->md);
-
-	if (r <= 0)
-		/* The target wants to complete the I/O */
-		dm_end_request(clone, r);
-	else if (r == DM_ENDIO_INCOMPLETE)
-		/* The target will handle the I/O */
-		return;
-	else if (r == DM_ENDIO_REQUEUE)
-		/* The target wants to requeue the I/O */
-		dm_requeue_original_request(tio->md, tio->orig);
-	else {
-		DMWARN("unimplemented target endio return value: %d", r);
-		BUG();
-	}
-}
-
-/*
- * Request completion handler for request-based dm
- */
-static void dm_softirq_done(struct request *rq)
-{
-	bool mapped = true;
-	struct dm_rq_target_io *tio = tio_from_request(rq);
-	struct request *clone = tio->clone;
-	int rw;
-
-	if (!clone) {
-		rq_end_stats(tio->md, rq);
-		rw = rq_data_dir(rq);
-		if (!rq->q->mq_ops) {
-			blk_end_request_all(rq, tio->error);
-			rq_completed(tio->md, rw, false);
-			free_old_rq_tio(tio);
-		} else {
-			blk_mq_end_request(rq, tio->error);
-			rq_completed(tio->md, rw, false);
-		}
-		return;
-	}
-
-	if (rq->cmd_flags & REQ_FAILED)
-		mapped = false;
-
-	dm_done(clone, tio->error, mapped);
-}
-
-/*
- * Complete the clone and the original request with the error status
- * through softirq context.
- */
-static void dm_complete_request(struct request *rq, int error)
-{
-	struct dm_rq_target_io *tio = tio_from_request(rq);
-
-	tio->error = error;
-	if (!rq->q->mq_ops)
-		blk_complete_request(rq);
-	else
-		blk_mq_complete_request(rq, error);
-}
-
-/*
- * Complete the not-mapped clone and the original request with the error status
- * through softirq context.
- * Target's rq_end_io() function isn't called.
- * This may be used when the target's map_rq() or clone_and_map_rq() functions fail.
- */
-static void dm_kill_unmapped_request(struct request *rq, int error)
-{
-	rq->cmd_flags |= REQ_FAILED;
-	dm_complete_request(rq, error);
-}
-
-/*
- * Called with the clone's queue lock held (in the case of .request_fn)
- */
-static void end_clone_request(struct request *clone, int error)
-{
-	struct dm_rq_target_io *tio = clone->end_io_data;
-
-	if (!clone->q->mq_ops) {
-		/*
-		 * For just cleaning up the information of the queue in which
-		 * the clone was dispatched.
-		 * The clone is *NOT* freed actually here because it is alloced
-		 * from dm own mempool (REQ_ALLOCED isn't set).
-		 */
-		__blk_put_request(clone->q, clone);
-	}
-
-	/*
-	 * Actual request completion is done in a softirq context which doesn't
-	 * hold the clone's queue lock.  Otherwise, deadlock could occur because:
-	 *     - another request may be submitted by the upper level driver
-	 *       of the stacking during the completion
-	 *     - the submission which requires queue lock may be done
-	 *       against this clone's queue
-	 */
-	dm_complete_request(tio->orig, error);
-}
-
 /*
  * Return maximum size of I/O possible at the supplied sector up to the current
  * target boundary.
@@ -1475,13 +905,40 @@ int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
 }
 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
 
-/*
- * A target may call dm_accept_partial_bio only from the map routine.  It is
- * allowed for all bio types except REQ_PREFLUSH.
- *
- * dm_accept_partial_bio informs the dm that the target only wants to process
- * additional n_sectors sectors of the bio and the rest of the data should be
- * sent in a next bio.
+static long dm_blk_direct_access(struct block_device *bdev, sector_t sector,
+				 void __pmem **kaddr, pfn_t *pfn, long size)
+{
+	struct mapped_device *md = bdev->bd_disk->private_data;
+	struct dm_table *map;
+	struct dm_target *ti;
+	int srcu_idx;
+	long len, ret = -EIO;
+
+	map = dm_get_live_table(md, &srcu_idx);
+	if (!map)
+		goto out;
+
+	ti = dm_table_find_target(map, sector);
+	if (!dm_target_is_valid(ti))
+		goto out;
+
+	len = max_io_len(sector, ti) << SECTOR_SHIFT;
+	size = min(len, size);
+
+	if (ti->type->direct_access)
+		ret = ti->type->direct_access(ti, sector, kaddr, pfn, size);
+out:
+	dm_put_live_table(md, srcu_idx);
+	return min(ret, size);
+}
+
+/*
+ * A target may call dm_accept_partial_bio only from the map routine.  It is
+ * allowed for all bio types except REQ_PREFLUSH.
+ *
+ * dm_accept_partial_bio informs the dm that the target only wants to process
+ * additional n_sectors sectors of the bio and the rest of the data should be
+ * sent in a next bio.
  *
  * A diagram that explains the arithmetics:
  * +--------------------+---------------+-------+
@@ -1684,512 +1141,165 @@ static unsigned get_num_write_same_bios(struct dm_target *ti)
 	return ti->num_write_same_bios;
 }
 
-typedef bool (*is_split_required_fn)(struct dm_target *ti);
-
-static bool is_split_required_for_discard(struct dm_target *ti)
-{
-	return ti->split_discard_bios;
-}
-
-static int __send_changing_extent_only(struct clone_info *ci,
-				       get_num_bios_fn get_num_bios,
-				       is_split_required_fn is_split_required)
-{
-	struct dm_target *ti;
-	unsigned len;
-	unsigned num_bios;
-
-	do {
-		ti = dm_table_find_target(ci->map, ci->sector);
-		if (!dm_target_is_valid(ti))
-			return -EIO;
-
-		/*
-		 * Even though the device advertised support for this type of
-		 * request, that does not mean every target supports it, and
-		 * reconfiguration might also have changed that since the
-		 * check was performed.
-		 */
-		num_bios = get_num_bios ? get_num_bios(ti) : 0;
-		if (!num_bios)
-			return -EOPNOTSUPP;
-
-		if (is_split_required && !is_split_required(ti))
-			len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
-		else
-			len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
-
-		__send_duplicate_bios(ci, ti, num_bios, &len);
-
-		ci->sector += len;
-	} while (ci->sector_count -= len);
-
-	return 0;
-}
-
-static int __send_discard(struct clone_info *ci)
-{
-	return __send_changing_extent_only(ci, get_num_discard_bios,
-					   is_split_required_for_discard);
-}
-
-static int __send_write_same(struct clone_info *ci)
-{
-	return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
-}
-
-/*
- * Select the correct strategy for processing a non-flush bio.
- */
-static int __split_and_process_non_flush(struct clone_info *ci)
-{
-	struct bio *bio = ci->bio;
-	struct dm_target *ti;
-	unsigned len;
-	int r;
-
-	if (unlikely(bio_op(bio) == REQ_OP_DISCARD))
-		return __send_discard(ci);
-	else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
-		return __send_write_same(ci);
-
-	ti = dm_table_find_target(ci->map, ci->sector);
-	if (!dm_target_is_valid(ti))
-		return -EIO;
-
-	len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
-
-	r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
-	if (r < 0)
-		return r;
-
-	ci->sector += len;
-	ci->sector_count -= len;
-
-	return 0;
-}
-
-/*
- * Entry point to split a bio into clones and submit them to the targets.
- */
-static void __split_and_process_bio(struct mapped_device *md,
-				    struct dm_table *map, struct bio *bio)
-{
-	struct clone_info ci;
-	int error = 0;
-
-	if (unlikely(!map)) {
-		bio_io_error(bio);
-		return;
-	}
-
-	ci.map = map;
-	ci.md = md;
-	ci.io = alloc_io(md);
-	ci.io->error = 0;
-	atomic_set(&ci.io->io_count, 1);
-	ci.io->bio = bio;
-	ci.io->md = md;
-	spin_lock_init(&ci.io->endio_lock);
-	ci.sector = bio->bi_iter.bi_sector;
-
-	start_io_acct(ci.io);
-
-	if (bio->bi_rw & REQ_PREFLUSH) {
-		ci.bio = &ci.md->flush_bio;
-		ci.sector_count = 0;
-		error = __send_empty_flush(&ci);
-		/* dec_pending submits any data associated with flush */
-	} else {
-		ci.bio = bio;
-		ci.sector_count = bio_sectors(bio);
-		while (ci.sector_count && !error)
-			error = __split_and_process_non_flush(&ci);
-	}
-
-	/* drop the extra reference count */
-	dec_pending(ci.io, error);
-}
-/*-----------------------------------------------------------------
- * CRUD END
- *---------------------------------------------------------------*/
-
-/*
- * The request function that just remaps the bio built up by
- * dm_merge_bvec.
- */
-static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
-{
-	int rw = bio_data_dir(bio);
-	struct mapped_device *md = q->queuedata;
-	int srcu_idx;
-	struct dm_table *map;
-
-	map = dm_get_live_table(md, &srcu_idx);
-
-	generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
-
-	/* if we're suspended, we have to queue this io for later */
-	if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
-		dm_put_live_table(md, srcu_idx);
-
-		if (!(bio->bi_rw & REQ_RAHEAD))
-			queue_io(md, bio);
-		else
-			bio_io_error(bio);
-		return BLK_QC_T_NONE;
-	}
-
-	__split_and_process_bio(md, map, bio);
-	dm_put_live_table(md, srcu_idx);
-	return BLK_QC_T_NONE;
-}
-
-int dm_request_based(struct mapped_device *md)
-{
-	return blk_queue_stackable(md->queue);
-}
-
-static void dm_dispatch_clone_request(struct request *clone, struct request *rq)
-{
-	int r;
-
-	if (blk_queue_io_stat(clone->q))
-		clone->cmd_flags |= REQ_IO_STAT;
-
-	clone->start_time = jiffies;
-	r = blk_insert_cloned_request(clone->q, clone);
-	if (r)
-		/* must complete clone in terms of original request */
-		dm_complete_request(rq, r);
-}
-
-static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
-				 void *data)
-{
-	struct dm_rq_target_io *tio = data;
-	struct dm_rq_clone_bio_info *info =
-		container_of(bio, struct dm_rq_clone_bio_info, clone);
-
-	info->orig = bio_orig;
-	info->tio = tio;
-	bio->bi_end_io = end_clone_bio;
-
-	return 0;
-}
-
-static int setup_clone(struct request *clone, struct request *rq,
-		       struct dm_rq_target_io *tio, gfp_t gfp_mask)
-{
-	int r;
-
-	r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask,
-			      dm_rq_bio_constructor, tio);
-	if (r)
-		return r;
-
-	clone->cmd = rq->cmd;
-	clone->cmd_len = rq->cmd_len;
-	clone->sense = rq->sense;
-	clone->end_io = end_clone_request;
-	clone->end_io_data = tio;
-
-	tio->clone = clone;
-
-	return 0;
-}
-
-static struct request *clone_old_rq(struct request *rq, struct mapped_device *md,
-				    struct dm_rq_target_io *tio, gfp_t gfp_mask)
-{
-	/*
-	 * Create clone for use with .request_fn request_queue
-	 */
-	struct request *clone;
-
-	clone = alloc_old_clone_request(md, gfp_mask);
-	if (!clone)
-		return NULL;
-
-	blk_rq_init(NULL, clone);
-	if (setup_clone(clone, rq, tio, gfp_mask)) {
-		/* -ENOMEM */
-		free_old_clone_request(md, clone);
-		return NULL;
-	}
-
-	return clone;
-}
-
-static void map_tio_request(struct kthread_work *work);
-
-static void init_tio(struct dm_rq_target_io *tio, struct request *rq,
-		     struct mapped_device *md)
-{
-	tio->md = md;
-	tio->ti = NULL;
-	tio->clone = NULL;
-	tio->orig = rq;
-	tio->error = 0;
-	/*
-	 * Avoid initializing info for blk-mq; it passes
-	 * target-specific data through info.ptr
-	 * (see: dm_mq_init_request)
-	 */
-	if (!md->init_tio_pdu)
-		memset(&tio->info, 0, sizeof(tio->info));
-	if (md->kworker_task)
-		init_kthread_work(&tio->work, map_tio_request);
-}
-
-static struct dm_rq_target_io *dm_old_prep_tio(struct request *rq,
-					       struct mapped_device *md,
-					       gfp_t gfp_mask)
-{
-	struct dm_rq_target_io *tio;
-	int srcu_idx;
-	struct dm_table *table;
-
-	tio = alloc_old_rq_tio(md, gfp_mask);
-	if (!tio)
-		return NULL;
-
-	init_tio(tio, rq, md);
-
-	table = dm_get_live_table(md, &srcu_idx);
-	/*
-	 * Must clone a request if this .request_fn DM device
-	 * is stacked on .request_fn device(s).
-	 */
-	if (!dm_table_mq_request_based(table)) {
-		if (!clone_old_rq(rq, md, tio, gfp_mask)) {
-			dm_put_live_table(md, srcu_idx);
-			free_old_rq_tio(tio);
-			return NULL;
-		}
-	}
-	dm_put_live_table(md, srcu_idx);
-
-	return tio;
-}
-
-/*
- * Called with the queue lock held.
- */
-static int dm_old_prep_fn(struct request_queue *q, struct request *rq)
-{
-	struct mapped_device *md = q->queuedata;
-	struct dm_rq_target_io *tio;
-
-	if (unlikely(rq->special)) {
-		DMWARN("Already has something in rq->special.");
-		return BLKPREP_KILL;
-	}
-
-	tio = dm_old_prep_tio(rq, md, GFP_ATOMIC);
-	if (!tio)
-		return BLKPREP_DEFER;
-
-	rq->special = tio;
-	rq->cmd_flags |= REQ_DONTPREP;
-
-	return BLKPREP_OK;
-}
-
-/*
- * Returns:
- * 0                : the request has been processed
- * DM_MAPIO_REQUEUE : the original request needs to be requeued
- * < 0              : the request was completed due to failure
- */
-static int map_request(struct dm_rq_target_io *tio, struct request *rq,
-		       struct mapped_device *md)
-{
-	int r;
-	struct dm_target *ti = tio->ti;
-	struct request *clone = NULL;
-
-	if (tio->clone) {
-		clone = tio->clone;
-		r = ti->type->map_rq(ti, clone, &tio->info);
-	} else {
-		r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone);
-		if (r < 0) {
-			/* The target wants to complete the I/O */
-			dm_kill_unmapped_request(rq, r);
-			return r;
-		}
-		if (r != DM_MAPIO_REMAPPED)
-			return r;
-		if (setup_clone(clone, rq, tio, GFP_ATOMIC)) {
-			/* -ENOMEM */
-			ti->type->release_clone_rq(clone);
-			return DM_MAPIO_REQUEUE;
-		}
-	}
-
-	switch (r) {
-	case DM_MAPIO_SUBMITTED:
-		/* The target has taken the I/O to submit by itself later */
-		break;
-	case DM_MAPIO_REMAPPED:
-		/* The target has remapped the I/O so dispatch it */
-		trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
-				     blk_rq_pos(rq));
-		dm_dispatch_clone_request(clone, rq);
-		break;
-	case DM_MAPIO_REQUEUE:
-		/* The target wants to requeue the I/O */
-		dm_requeue_original_request(md, tio->orig);
-		break;
-	default:
-		if (r > 0) {
-			DMWARN("unimplemented target map return value: %d", r);
-			BUG();
-		}
-
-		/* The target wants to complete the I/O */
-		dm_kill_unmapped_request(rq, r);
-		return r;
-	}
-
-	return 0;
-}
-
-static void map_tio_request(struct kthread_work *work)
-{
-	struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work);
-	struct request *rq = tio->orig;
-	struct mapped_device *md = tio->md;
+typedef bool (*is_split_required_fn)(struct dm_target *ti);
 
-	if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE)
-		dm_requeue_original_request(md, rq);
+static bool is_split_required_for_discard(struct dm_target *ti)
+{
+	return ti->split_discard_bios;
 }
 
-static void dm_start_request(struct mapped_device *md, struct request *orig)
+static int __send_changing_extent_only(struct clone_info *ci,
+				       get_num_bios_fn get_num_bios,
+				       is_split_required_fn is_split_required)
 {
-	if (!orig->q->mq_ops)
-		blk_start_request(orig);
-	else
-		blk_mq_start_request(orig);
-	atomic_inc(&md->pending[rq_data_dir(orig)]);
-
-	if (md->seq_rq_merge_deadline_usecs) {
-		md->last_rq_pos = rq_end_sector(orig);
-		md->last_rq_rw = rq_data_dir(orig);
-		md->last_rq_start_time = ktime_get();
-	}
+	struct dm_target *ti;
+	unsigned len;
+	unsigned num_bios;
 
-	if (unlikely(dm_stats_used(&md->stats))) {
-		struct dm_rq_target_io *tio = tio_from_request(orig);
-		tio->duration_jiffies = jiffies;
-		tio->n_sectors = blk_rq_sectors(orig);
-		dm_stats_account_io(&md->stats, rq_data_dir(orig),
-				    blk_rq_pos(orig), tio->n_sectors, false, 0,
-				    &tio->stats_aux);
-	}
+	do {
+		ti = dm_table_find_target(ci->map, ci->sector);
+		if (!dm_target_is_valid(ti))
+			return -EIO;
 
 		/*
-	 * Hold the md reference here for the in-flight I/O.
-	 * We can't rely on the reference count by device opener,
-	 * because the device may be closed during the request completion
-	 * when all bios are completed.
-	 * See the comment in rq_completed() too.
+		 * Even though the device advertised support for this type of
+		 * request, that does not mean every target supports it, and
+		 * reconfiguration might also have changed that since the
+		 * check was performed.
 		 */
-	dm_get(md);
+		num_bios = get_num_bios ? get_num_bios(ti) : 0;
+		if (!num_bios)
+			return -EOPNOTSUPP;
+
+		if (is_split_required && !is_split_required(ti))
+			len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
+		else
+			len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
+
+		__send_duplicate_bios(ci, ti, num_bios, &len);
+
+		ci->sector += len;
+	} while (ci->sector_count -= len);
+
+	return 0;
 }
 
-#define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000
+static int __send_discard(struct clone_info *ci)
+{
+	return __send_changing_extent_only(ci, get_num_discard_bios,
+					   is_split_required_for_discard);
+}
 
-ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf)
+static int __send_write_same(struct clone_info *ci)
 {
-	return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs);
+	return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
 }
 
-ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md,
-						     const char *buf, size_t count)
+/*
+ * Select the correct strategy for processing a non-flush bio.
+ */
+static int __split_and_process_non_flush(struct clone_info *ci)
 {
-	unsigned deadline;
+	struct bio *bio = ci->bio;
+	struct dm_target *ti;
+	unsigned len;
+	int r;
+
+	if (unlikely(bio_op(bio) == REQ_OP_DISCARD))
+		return __send_discard(ci);
+	else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
+		return __send_write_same(ci);
 
-	if (!dm_request_based(md) || md->use_blk_mq)
-		return count;
+	ti = dm_table_find_target(ci->map, ci->sector);
+	if (!dm_target_is_valid(ti))
+		return -EIO;
 
-	if (kstrtouint(buf, 10, &deadline))
-		return -EINVAL;
+	len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
 
-	if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS)
-		deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS;
+	r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
+	if (r < 0)
+		return r;
 
-	md->seq_rq_merge_deadline_usecs = deadline;
+	ci->sector += len;
+	ci->sector_count -= len;
 
-	return count;
+	return 0;
 }
 
-static bool dm_request_peeked_before_merge_deadline(struct mapped_device *md)
+/*
+ * Entry point to split a bio into clones and submit them to the targets.
+ */
+static void __split_and_process_bio(struct mapped_device *md,
+				    struct dm_table *map, struct bio *bio)
 {
-	ktime_t kt_deadline;
+	struct clone_info ci;
+	int error = 0;
+
+	if (unlikely(!map)) {
+		bio_io_error(bio);
+		return;
+	}
+
+	ci.map = map;
+	ci.md = md;
+	ci.io = alloc_io(md);
+	ci.io->error = 0;
+	atomic_set(&ci.io->io_count, 1);
+	ci.io->bio = bio;
+	ci.io->md = md;
+	spin_lock_init(&ci.io->endio_lock);
+	ci.sector = bio->bi_iter.bi_sector;
 
-	if (!md->seq_rq_merge_deadline_usecs)
-		return false;
+	start_io_acct(ci.io);
 
-	kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC);
-	kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline);
+	if (bio->bi_rw & REQ_PREFLUSH) {
+		ci.bio = &ci.md->flush_bio;
+		ci.sector_count = 0;
+		error = __send_empty_flush(&ci);
+		/* dec_pending submits any data associated with flush */
+	} else {
+		ci.bio = bio;
+		ci.sector_count = bio_sectors(bio);
+		while (ci.sector_count && !error)
+			error = __split_and_process_non_flush(&ci);
+	}
 
-	return !ktime_after(ktime_get(), kt_deadline);
+	/* drop the extra reference count */
+	dec_pending(ci.io, error);
 }
+/*-----------------------------------------------------------------
+ * CRUD END
+ *---------------------------------------------------------------*/
 
 /*
- * q->request_fn for request-based dm.
- * Called with the queue lock held.
+ * The request function that just remaps the bio built up by
+ * dm_merge_bvec.
  */
-static void dm_request_fn(struct request_queue *q)
+static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
 {
+	int rw = bio_data_dir(bio);
 	struct mapped_device *md = q->queuedata;
-	struct dm_target *ti = md->immutable_target;
-	struct request *rq;
-	struct dm_rq_target_io *tio;
-	sector_t pos = 0;
-
-	if (unlikely(!ti)) {
 	int srcu_idx;
-		struct dm_table *map = dm_get_live_table(md, &srcu_idx);
+	struct dm_table *map;
 
-		ti = dm_table_find_target(map, pos);
-		dm_put_live_table(md, srcu_idx);
-	}
+	map = dm_get_live_table(md, &srcu_idx);
 
-	/*
-	 * For suspend, check blk_queue_stopped() and increment
-	 * ->pending within a single queue_lock not to increment the
-	 * number of in-flight I/Os after the queue is stopped in
-	 * dm_suspend().
-	 */
-	while (!blk_queue_stopped(q)) {
-		rq = blk_peek_request(q);
-		if (!rq)
-			return;
+	generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
 
-		/* always use block 0 to find the target for flushes for now */
-		pos = 0;
-		if (req_op(rq) != REQ_OP_FLUSH)
-			pos = blk_rq_pos(rq);
+	/* if we're suspended, we have to queue this io for later */
+	if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
+		dm_put_live_table(md, srcu_idx);
 
-		if ((dm_request_peeked_before_merge_deadline(md) &&
-		     md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 &&
-		     md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq)) ||
-		    (ti->type->busy && ti->type->busy(ti))) {
-			blk_delay_queue(q, HZ / 100);
-			return;
+		if (!(bio->bi_rw & REQ_RAHEAD))
+			queue_io(md, bio);
+		else
+			bio_io_error(bio);
+		return BLK_QC_T_NONE;
 	}
 
-		dm_start_request(md, rq);
-
-		tio = tio_from_request(rq);
-		/* Establish tio->ti before queuing work (map_tio_request) */
-		tio->ti = ti;
-		queue_kthread_work(&md->kworker, &tio->work);
-		BUG_ON(!irqs_disabled());
-	}
+	__split_and_process_bio(md, map, bio);
+	dm_put_live_table(md, srcu_idx);
+	return BLK_QC_T_NONE;
 }
 
 static int dm_any_congested(void *congested_data, int bdi_bits)
@@ -2269,7 +1379,7 @@ static const struct block_device_operations dm_blk_dops;
 
 static void dm_wq_work(struct work_struct *work);
 
-static void dm_init_md_queue(struct mapped_device *md)
+void dm_init_md_queue(struct mapped_device *md)
 {
 	/*
 	 * Request-based dm devices cannot be stacked on top of bio-based dm
@@ -2290,7 +1400,7 @@ static void dm_init_md_queue(struct mapped_device *md)
 	md->queue->backing_dev_info.congested_data = md;
 }
 
-static void dm_init_normal_md_queue(struct mapped_device *md)
+void dm_init_normal_md_queue(struct mapped_device *md)
 {
 	md->use_blk_mq = false;
 	dm_init_md_queue(md);
@@ -2330,6 +1440,8 @@ static void cleanup_mapped_device(struct mapped_device *md)
 		bdput(md->bdev);
 		md->bdev = NULL;
 	}
+
+	dm_mq_cleanup_mapped_device(md);
 }
 
 /*
@@ -2363,7 +1475,7 @@ static struct mapped_device *alloc_dev(int minor)
 		goto bad_io_barrier;
 
 	md->numa_node_id = numa_node_id;
-	md->use_blk_mq = use_blk_mq;
+	md->use_blk_mq = dm_use_blk_mq_default();
 	md->init_tio_pdu = false;
 	md->type = DM_TYPE_NONE;
 	mutex_init(&md->suspend_lock);
@@ -2448,10 +1560,6 @@ static void free_dev(struct mapped_device *md)
 	unlock_fs(md);
 
 	cleanup_mapped_device(md);
-	if (md->tag_set) {
-		blk_mq_free_tag_set(md->tag_set);
-		kfree(md->tag_set);
-	}
 
 	free_table_devices(&md->table_devices);
 	dm_stats_cleanup(&md->stats);
@@ -2467,7 +1575,7 @@ static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
 
 	if (md->bs) {
 		/* The md already has necessary mempools. */
-		if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
+		if (dm_table_bio_based(t)) {
 			/*
 			 * Reload bioset because front_pad may have changed
 			 * because a different table was loaded.
@@ -2657,176 +1765,15 @@ struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
 }
 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
 
-static void dm_old_init_rq_based_worker_thread(struct mapped_device *md)
-{
-	/* Initialize the request-based DM worker thread */
-	init_kthread_worker(&md->kworker);
-	md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker,
-				       "kdmwork-%s", dm_device_name(md));
-}
-
-/*
- * Fully initialize a .request_fn request-based queue.
- */
-static int dm_old_init_request_queue(struct mapped_device *md)
-{
-	/* Fully initialize the queue */
-	if (!blk_init_allocated_queue(md->queue, dm_request_fn, NULL))
-		return -EINVAL;
-
-	/* disable dm_request_fn's merge heuristic by default */
-	md->seq_rq_merge_deadline_usecs = 0;
-
-	dm_init_normal_md_queue(md);
-	blk_queue_softirq_done(md->queue, dm_softirq_done);
-	blk_queue_prep_rq(md->queue, dm_old_prep_fn);
-
-	dm_old_init_rq_based_worker_thread(md);
-
-	elv_register_queue(md->queue);
-
-	return 0;
-}
-
-static int dm_mq_init_request(void *data, struct request *rq,
-			      unsigned int hctx_idx, unsigned int request_idx,
-			      unsigned int numa_node)
-{
-	struct mapped_device *md = data;
-	struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
-
-	/*
-	 * Must initialize md member of tio, otherwise it won't
-	 * be available in dm_mq_queue_rq.
-	 */
-	tio->md = md;
-
-	if (md->init_tio_pdu) {
-		/* target-specific per-io data is immediately after the tio */
-		tio->info.ptr = tio + 1;
-	}
-
-	return 0;
-}
-
-static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx,
-			  const struct blk_mq_queue_data *bd)
-{
-	struct request *rq = bd->rq;
-	struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
-	struct mapped_device *md = tio->md;
-	struct dm_target *ti = md->immutable_target;
-
-	if (unlikely(!ti)) {
-		int srcu_idx;
-		struct dm_table *map = dm_get_live_table(md, &srcu_idx);
-
-		ti = dm_table_find_target(map, 0);
-		dm_put_live_table(md, srcu_idx);
-	}
-
-	if (ti->type->busy && ti->type->busy(ti))
-		return BLK_MQ_RQ_QUEUE_BUSY;
-
-	dm_start_request(md, rq);
-
-	/* Init tio using md established in .init_request */
-	init_tio(tio, rq, md);
-
-	/*
-	 * Establish tio->ti before queuing work (map_tio_request)
-	 * or making direct call to map_request().
-	 */
-	tio->ti = ti;
-
-	/* Direct call is fine since .queue_rq allows allocations */
-	if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) {
-		/* Undo dm_start_request() before requeuing */
-		rq_end_stats(md, rq);
-		rq_completed(md, rq_data_dir(rq), false);
-		return BLK_MQ_RQ_QUEUE_BUSY;
-	}
-
-	return BLK_MQ_RQ_QUEUE_OK;
-}
-
-static struct blk_mq_ops dm_mq_ops = {
-	.queue_rq = dm_mq_queue_rq,
-	.map_queue = blk_mq_map_queue,
-	.complete = dm_softirq_done,
-	.init_request = dm_mq_init_request,
-};
-
-static int dm_mq_init_request_queue(struct mapped_device *md,
-				    struct dm_target *immutable_tgt)
-{
-	struct request_queue *q;
-	int err;
-
-	if (dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) {
-		DMERR("request-based dm-mq may only be stacked on blk-mq device(s)");
-		return -EINVAL;
-	}
-
-	md->tag_set = kzalloc_node(sizeof(struct blk_mq_tag_set), GFP_KERNEL, md->numa_node_id);
-	if (!md->tag_set)
-		return -ENOMEM;
-
-	md->tag_set->ops = &dm_mq_ops;
-	md->tag_set->queue_depth = dm_get_blk_mq_queue_depth();
-	md->tag_set->numa_node = md->numa_node_id;
-	md->tag_set->flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
-	md->tag_set->nr_hw_queues = dm_get_blk_mq_nr_hw_queues();
-	md->tag_set->driver_data = md;
-
-	md->tag_set->cmd_size = sizeof(struct dm_rq_target_io);
-	if (immutable_tgt && immutable_tgt->per_io_data_size) {
-		/* any target-specific per-io data is immediately after the tio */
-		md->tag_set->cmd_size += immutable_tgt->per_io_data_size;
-		md->init_tio_pdu = true;
-	}
-
-	err = blk_mq_alloc_tag_set(md->tag_set);
-	if (err)
-		goto out_kfree_tag_set;
-
-	q = blk_mq_init_allocated_queue(md->tag_set, md->queue);
-	if (IS_ERR(q)) {
-		err = PTR_ERR(q);
-		goto out_tag_set;
-	}
-	dm_init_md_queue(md);
-
-	/* backfill 'mq' sysfs registration normally done in blk_register_queue */
-	blk_mq_register_disk(md->disk);
-
-	return 0;
-
-out_tag_set:
-	blk_mq_free_tag_set(md->tag_set);
-out_kfree_tag_set:
-	kfree(md->tag_set);
-
-	return err;
-}
-
-static unsigned filter_md_type(unsigned type, struct mapped_device *md)
-{
-	if (type == DM_TYPE_BIO_BASED)
-		return type;
-
-	return !md->use_blk_mq ? DM_TYPE_REQUEST_BASED : DM_TYPE_MQ_REQUEST_BASED;
-}
-
 /*
  * Setup the DM device's queue based on md's type
  */
 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
 {
 	int r;
-	unsigned md_type = filter_md_type(dm_get_md_type(md), md);
+	unsigned type = dm_get_md_type(md);
 
-	switch (md_type) {
+	switch (type) {
 	case DM_TYPE_REQUEST_BASED:
 		r = dm_old_init_request_queue(md);
 		if (r) {
@@ -2835,13 +1782,14 @@ int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
 		}
 		break;
 	case DM_TYPE_MQ_REQUEST_BASED:
-		r = dm_mq_init_request_queue(md, dm_table_get_immutable_target(t));
+		r = dm_mq_init_request_queue(md, t);
 		if (r) {
 			DMERR("Cannot initialize queue for request-based dm-mq mapped device");
 			return r;
 		}
 		break;
 	case DM_TYPE_BIO_BASED:
+	case DM_TYPE_DAX_BIO_BASED:
 		dm_init_normal_md_queue(md);
 		blk_queue_make_request(md->queue, dm_make_request);
 		/*
@@ -2850,6 +1798,9 @@ int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
 		 */
 		bioset_free(md->queue->bio_split);
 		md->queue->bio_split = NULL;
+
+		if (type == DM_TYPE_DAX_BIO_BASED)
+			queue_flag_set_unlocked(QUEUE_FLAG_DAX, md->queue);
 		break;
 	}
 
@@ -3544,10 +2495,9 @@ struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned t
 	if (!pools)
 		return NULL;
 
-	type = filter_md_type(type, md);
-
 	switch (type) {
 	case DM_TYPE_BIO_BASED:
+	case DM_TYPE_DAX_BIO_BASED:
 		cachep = _io_cache;
 		pool_size = dm_get_reserved_bio_based_ios();
 		front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
@@ -3604,26 +2554,76 @@ void dm_free_md_mempools(struct dm_md_mempools *pools)
 	kfree(pools);
 }
 
-static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
-			  u32 flags)
+struct dm_pr {
+	u64	old_key;
+	u64	new_key;
+	u32	flags;
+	bool	fail_early;
+};
+
+static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
+		      void *data)
 {
 	struct mapped_device *md = bdev->bd_disk->private_data;
-	const struct pr_ops *ops;
-	fmode_t mode;
-	int r;
+	struct dm_table *table;
+	struct dm_target *ti;
+	int ret = -ENOTTY, srcu_idx;
 
-	r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
-	if (r < 0)
-		return r;
+	table = dm_get_live_table(md, &srcu_idx);
+	if (!table || !dm_table_get_size(table))
+		goto out;
 
-	ops = bdev->bd_disk->fops->pr_ops;
-	if (ops && ops->pr_register)
-		r = ops->pr_register(bdev, old_key, new_key, flags);
-	else
-		r = -EOPNOTSUPP;
+	/* We only support devices that have a single target */
+	if (dm_table_get_num_targets(table) != 1)
+		goto out;
+	ti = dm_table_get_target(table, 0);
 
-	bdput(bdev);
-	return r;
+	ret = -EINVAL;
+	if (!ti->type->iterate_devices)
+		goto out;
+
+	ret = ti->type->iterate_devices(ti, fn, data);
+out:
+	dm_put_live_table(md, srcu_idx);
+	return ret;
+}
+
+/*
+ * For register / unregister we need to manually call out to every path.
+ */
+static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
+			    sector_t start, sector_t len, void *data)
+{
+	struct dm_pr *pr = data;
+	const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
+
+	if (!ops || !ops->pr_register)
+		return -EOPNOTSUPP;
+	return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
+}
+
+static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
+			  u32 flags)
+{
+	struct dm_pr pr = {
+		.old_key	= old_key,
+		.new_key	= new_key,
+		.flags		= flags,
+		.fail_early	= true,
+	};
+	int ret;
+
+	ret = dm_call_pr(bdev, __dm_pr_register, &pr);
+	if (ret && new_key) {
+		/* unregister all paths if we failed to register any path */
+		pr.old_key = new_key;
+		pr.new_key = 0;
+		pr.flags = 0;
+		pr.fail_early = false;
+		dm_call_pr(bdev, __dm_pr_register, &pr);
+	}
+
+	return ret;
 }
 
 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
@@ -3724,6 +2724,7 @@ static const struct block_device_operations dm_blk_dops = {
 	.open = dm_blk_open,
 	.release = dm_blk_close,
 	.ioctl = dm_blk_ioctl,
+	.direct_access = dm_blk_direct_access,
 	.getgeo = dm_blk_getgeo,
 	.pr_ops = &dm_pr_ops,
 	.owner = THIS_MODULE
@@ -3741,18 +2742,6 @@ MODULE_PARM_DESC(major, "The major number of the device mapper");
 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
 
-module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
-MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
-
-module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR);
-MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices");
-
-module_param(dm_mq_nr_hw_queues, uint, S_IRUGO | S_IWUSR);
-MODULE_PARM_DESC(dm_mq_nr_hw_queues, "Number of hardware queues for request-based dm-mq devices");
-
-module_param(dm_mq_queue_depth, uint, S_IRUGO | S_IWUSR);
-MODULE_PARM_DESC(dm_mq_queue_depth, "Queue depth for request-based dm-mq devices");
-
 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
 

drivers/md/dm.h

@@ -13,6 +13,7 @@
 #include <linux/fs.h>
 #include <linux/device-mapper.h>
 #include <linux/list.h>
+#include <linux/moduleparam.h>
 #include <linux/blkdev.h>
 #include <linux/backing-dev.h>
 #include <linux/hdreg.h>
@@ -32,14 +33,6 @@
  */
 #define DM_STATUS_NOFLUSH_FLAG		(1 << 0)
 
-/*
- * Type of table and mapped_device's mempool
- */
-#define DM_TYPE_NONE			0
-#define DM_TYPE_BIO_BASED		1
-#define DM_TYPE_REQUEST_BASED		2
-#define DM_TYPE_MQ_REQUEST_BASED	3
-
 /*
  * List of devices that a metadevice uses and should open/close.
  */
@@ -75,8 +68,9 @@ unsigned dm_table_get_type(struct dm_table *t);
 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t);
 struct dm_target *dm_table_get_immutable_target(struct dm_table *t);
 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t);
+bool dm_table_bio_based(struct dm_table *t);
 bool dm_table_request_based(struct dm_table *t);
-bool dm_table_mq_request_based(struct dm_table *t);
+bool dm_table_all_blk_mq_devices(struct dm_table *t);
 void dm_table_free_md_mempools(struct dm_table *t);
 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t);
 
@@ -161,16 +155,6 @@ void dm_interface_exit(void);
 /*
  * sysfs interface
  */
-struct dm_kobject_holder {
-	struct kobject kobj;
-	struct completion completion;
-};
-
-static inline struct completion *dm_get_completion_from_kobject(struct kobject *kobj)
-{
-	return &container_of(kobj, struct dm_kobject_holder, kobj)->completion;
-}
-
 int dm_sysfs_init(struct mapped_device *md);
 void dm_sysfs_exit(struct mapped_device *md);
 struct kobject *dm_kobject(struct mapped_device *md);
@@ -212,8 +196,6 @@ int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
 void dm_internal_suspend(struct mapped_device *md);
 void dm_internal_resume(struct mapped_device *md);
 
-bool dm_use_blk_mq(struct mapped_device *md);
-
 int dm_io_init(void);
 void dm_io_exit(void);
 
@@ -228,18 +210,8 @@ struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned t
 void dm_free_md_mempools(struct dm_md_mempools *pools);
 
 /*
- * Helpers that are used by DM core
+ * Various helpers
  */
 unsigned dm_get_reserved_bio_based_ios(void);
-unsigned dm_get_reserved_rq_based_ios(void);
-
-static inline bool dm_message_test_buffer_overflow(char *result, unsigned maxlen)
-{
-	return !maxlen || strlen(result) + 1 >= maxlen;
-}
-
-ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf);
-ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md,
-						     const char *buf, size_t count);
 
 #endif

drivers/md/persistent-data/dm-btree.c

@@ -429,7 +429,14 @@ static int dm_btree_lookup_next_single(struct dm_btree_info *info, dm_block_t ro
 
 	if (flags & INTERNAL_NODE) {
 		i = lower_bound(n, key);
-		if (i < 0 || i >= nr_entries) {
+		if (i < 0) {
+			/*
+			 * avoid early -ENODATA return when all entries are
+			 * higher than the search @key.
+			 */
+			i = 0;
+		}
+		if (i >= nr_entries) {
 			r = -ENODATA;
 			goto out;
 		}

drivers/scsi/sd.c

@@ -1619,8 +1619,7 @@ static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
 		return -EOPNOTSUPP;
 	return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
 			old_key, new_key, 0,
-			(1 << 0) /* APTPL */ |
-			(1 << 2) /* ALL_TG_PT */);
+			(1 << 0) /* APTPL */);
 }
 
 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,

include/linux/device-mapper.h

@@ -19,6 +19,15 @@ struct dm_table;
 struct mapped_device;
 struct bio_vec;
 
+/*
+ * Type of table, mapped_device's mempool and request_queue
+ */
+#define DM_TYPE_NONE			0
+#define DM_TYPE_BIO_BASED		1
+#define DM_TYPE_REQUEST_BASED		2
+#define DM_TYPE_MQ_REQUEST_BASED	3
+#define DM_TYPE_DAX_BIO_BASED		4
+
 typedef enum { STATUSTYPE_INFO, STATUSTYPE_TABLE } status_type_t;
 
 union map_info {
@@ -116,6 +125,14 @@ typedef void (*dm_io_hints_fn) (struct dm_target *ti,
  */
 typedef int (*dm_busy_fn) (struct dm_target *ti);
 
+/*
+ * Returns:
+ *  < 0 : error
+ * >= 0 : the number of bytes accessible at the address
+ */
+typedef long (*dm_direct_access_fn) (struct dm_target *ti, sector_t sector,
+				     void __pmem **kaddr, pfn_t *pfn, long size);
+
 void dm_error(const char *message);
 
 struct dm_dev {
@@ -162,6 +179,7 @@ struct target_type {
 	dm_busy_fn busy;
 	dm_iterate_devices_fn iterate_devices;
 	dm_io_hints_fn io_hints;
+	dm_direct_access_fn direct_access;
 
 	/* For internal device-mapper use. */
 	struct list_head list;
@@ -443,6 +461,14 @@ int dm_table_add_target(struct dm_table *t, const char *type,
  */
 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb);
 
+/*
+ * Target can use this to set the table's type.
+ * Can only ever be called from a target's ctr.
+ * Useful for "hybrid" target (supports both bio-based
+ * and request-based).
+ */
+void dm_table_set_type(struct dm_table *t, unsigned type);
+
 /*
  * Finally call this to make the table ready for use.
  */

include/uapi/linux/dm-ioctl.h

@@ -267,9 +267,9 @@ enum {
 #define DM_DEV_SET_GEOMETRY	_IOWR(DM_IOCTL, DM_DEV_SET_GEOMETRY_CMD, struct dm_ioctl)
 
 #define DM_VERSION_MAJOR	4
-#define DM_VERSION_MINOR	34
+#define DM_VERSION_MINOR	35
 #define DM_VERSION_PATCHLEVEL	0
-#define DM_VERSION_EXTRA	"-ioctl (2015-10-28)"
+#define DM_VERSION_EXTRA	"-ioctl (2016-06-23)"
 
 /* Status bits */
 #define DM_READONLY_FLAG	(1 << 0) /* In/Out */