Merge tag 'dm-3.12-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm

Pull device-mapper updates from Mike Snitzer:
"Add the ability to collect I/O statistics on user-defined regions of a
device-mapper device. This dm-stats code required the reintroduction
of a div64_u64_rem() helper, but as a separate method that doesn't
slow down div64_u64() -- especially on 32-bit systems.

Allow the error target to replace request-based DM devices (e.g.
multipath) in addition to bio-based DM devices.

Various other small code fixes and improvements to thin-provisioning,
DM cache and the DM ioctl interface"

* tag 'dm-3.12-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
dm stripe: silence a couple sparse warnings
dm: add statistics support
dm thin: always return -ENOSPC if no_free_space is set
dm ioctl: cleanup error handling in table_load
dm ioctl: increase granularity of type_lock when loading table
dm ioctl: prevent rename to empty name or uuid
dm thin: set pool read-only if breaking_sharing fails block allocation
dm thin: prefix pool error messages with pool device name
dm: allow error target to replace bio-based and request-based targets
math64: New separate div64_u64_rem helper
dm space map: optimise sm_ll_dec and sm_ll_inc
dm btree: prefetch child nodes when walking tree for a dm_btree_del
dm btree: use pop_frame in dm_btree_del to cleanup code
dm cache: eliminate holes in cache structure
dm cache: fix stacking of geometry limits
dm thin: fix stacking of geometry limits
dm thin: add data block size limits to Documentation
dm cache: add data block size limits to code and Documentation
dm cache: document metadata device is exclussive to a cache
dm: stop using WQ_NON_REENTRANT

Documentation/device-mapper/cache.txt

···
    which are dirty, and extra hints for use by the policy object.
    This information could be put on the cache device, but having it
    separate allows the volume manager to configure it differently,
-   e.g. as a mirror for extra robustness.
+   e.g. as a mirror for extra robustness.  This metadata device may only
+   be used by a single cache device.
 
 Fixed block size
 ----------------
 
 The origin is divided up into blocks of a fixed size.  This block size
 is configurable when you first create the cache.  Typically we've been
-using block sizes of 256k - 1024k.
+using block sizes of 256KB - 1024KB.  The block size must be between 64
+(32KB) and 2097152 (1GB) and a multiple of 64 (32KB).
 
 Having a fixed block size simplifies the target a lot.  But it is
 something of a compromise.  For instance, a small part of a block may be

Documentation/device-mapper/statistics.txt

···
+DM statistics
+=============
+
+Device Mapper supports the collection of I/O statistics on user-defined
+regions of a DM device.	 If no regions are defined no statistics are
+collected so there isn't any performance impact.  Only bio-based DM
+devices are currently supported.
+
+Each user-defined region specifies a starting sector, length and step.
+Individual statistics will be collected for each step-sized area within
+the range specified.
+
+The I/O statistics counters for each step-sized area of a region are
+in the same format as /sys/block/*/stat or /proc/diskstats (see:
+Documentation/iostats.txt).  But two extra counters (12 and 13) are
+provided: total time spent reading and writing in milliseconds.	 All
+these counters may be accessed by sending the @stats_print message to
+the appropriate DM device via dmsetup.
+
+Each region has a corresponding unique identifier, which we call a
+region_id, that is assigned when the region is created.	 The region_id
+must be supplied when querying statistics about the region, deleting the
+region, etc.  Unique region_ids enable multiple userspace programs to
+request and process statistics for the same DM device without stepping
+on each other's data.
+
+The creation of DM statistics will allocate memory via kmalloc or
+fallback to using vmalloc space.  At most, 1/4 of the overall system
+memory may be allocated by DM statistics.  The admin can see how much
+memory is used by reading
+/sys/module/dm_mod/parameters/stats_current_allocated_bytes
+
+Messages
+========
+
+    @stats_create <range> <step> [<program_id> [<aux_data>]]
+
+	Create a new region and return the region_id.
+
+	<range>
+	  "-" - whole device
+	  "<start_sector>+<length>" - a range of <length> 512-byte sectors
+				      starting with <start_sector>.
+
+	<step>
+	  "<area_size>" - the range is subdivided into areas each containing
+			  <area_size> sectors.
+	  "/<number_of_areas>" - the range is subdivided into the specified
+				 number of areas.
+
+	<program_id>
+	  An optional parameter.  A name that uniquely identifies
+	  the userspace owner of the range.  This groups ranges together
+	  so that userspace programs can identify the ranges they
+	  created and ignore those created by others.
+	  The kernel returns this string back in the output of
+	  @stats_list message, but it doesn't use it for anything else.
+
+	<aux_data>
+	  An optional parameter.  A word that provides auxiliary data
+	  that is useful to the client program that created the range.
+	  The kernel returns this string back in the output of
+	  @stats_list message, but it doesn't use this value for anything.
+
+    @stats_delete <region_id>
+
+	Delete the region with the specified id.
+
+	<region_id>
+	  region_id returned from @stats_create
+
+    @stats_clear <region_id>
+
+	Clear all the counters except the in-flight i/o counters.
+
+	<region_id>
+	  region_id returned from @stats_create
+
+    @stats_list [<program_id>]
+
+	List all regions registered with @stats_create.
+
+	<program_id>
+	  An optional parameter.
+	  If this parameter is specified, only matching regions
+	  are returned.
+	  If it is not specified, all regions are returned.
+
+	Output format:
+	  <region_id>: <start_sector>+<length> <step> <program_id> <aux_data>
+
+    @stats_print <region_id> [<starting_line> <number_of_lines>]
+
+	Print counters for each step-sized area of a region.
+
+	<region_id>
+	  region_id returned from @stats_create
+
+	<starting_line>
+	  The index of the starting line in the output.
+	  If omitted, all lines are returned.
+
+	<number_of_lines>
+	  The number of lines to include in the output.
+	  If omitted, all lines are returned.
+
+	Output format for each step-sized area of a region:
+
+	  <start_sector>+<length> counters
+
+	  The first 11 counters have the same meaning as
+	  /sys/block/*/stat or /proc/diskstats.
+
+	  Please refer to Documentation/iostats.txt for details.
+
+	  1. the number of reads completed
+	  2. the number of reads merged
+	  3. the number of sectors read
+	  4. the number of milliseconds spent reading
+	  5. the number of writes completed
+	  6. the number of writes merged
+	  7. the number of sectors written
+	  8. the number of milliseconds spent writing
+	  9. the number of I/Os currently in progress
+	  10. the number of milliseconds spent doing I/Os
+	  11. the weighted number of milliseconds spent doing I/Os
+
+	  Additional counters:
+	  12. the total time spent reading in milliseconds
+	  13. the total time spent writing in milliseconds
+
+    @stats_print_clear <region_id> [<starting_line> <number_of_lines>]
+
+	Atomically print and then clear all the counters except the
+	in-flight i/o counters.	 Useful when the client consuming the
+	statistics does not want to lose any statistics (those updated
+	between printing and clearing).
+
+	<region_id>
+	  region_id returned from @stats_create
+
+	<starting_line>
+	  The index of the starting line in the output.
+	  If omitted, all lines are printed and then cleared.
+
+	<number_of_lines>
+	  The number of lines to process.
+	  If omitted, all lines are printed and then cleared.
+
+    @stats_set_aux <region_id> <aux_data>
+
+	Store auxiliary data aux_data for the specified region.
+
+	<region_id>
+	  region_id returned from @stats_create
+
+	<aux_data>
+	  The string that identifies data which is useful to the client
+	  program that created the range.  The kernel returns this
+	  string back in the output of @stats_list message, but it
+	  doesn't use this value for anything.
+
+Examples
+========
+
+Subdivide the DM device 'vol' into 100 pieces and start collecting
+statistics on them:
+
+  dmsetup message vol 0 @stats_create - /100
+
+Set the auxillary data string to "foo bar baz" (the escape for each
+space must also be escaped, otherwise the shell will consume them):
+
+  dmsetup message vol 0 @stats_set_aux 0 foo\\ bar\\ baz
+
+List the statistics:
+
+  dmsetup message vol 0 @stats_list
+
+Print the statistics:
+
+  dmsetup message vol 0 @stats_print 0
+
+Delete the statistics:
+
+  dmsetup message vol 0 @stats_delete 0

Documentation/device-mapper/thin-provisioning.txt

···
 		 $data_block_size $low_water_mark"
 
 $data_block_size gives the smallest unit of disk space that can be
-allocated at a time expressed in units of 512-byte sectors.  People
-primarily interested in thin provisioning may want to use a value such
-as 1024 (512KB).  People doing lots of snapshotting may want a smaller value
-such as 128 (64KB).  If you are not zeroing newly-allocated data,
-a larger $data_block_size in the region of 256000 (128MB) is suggested.
-$data_block_size must be the same for the lifetime of the
-metadata device.
+allocated at a time expressed in units of 512-byte sectors.
+$data_block_size must be between 128 (64KB) and 2097152 (1GB) and a
+multiple of 128 (64KB).  $data_block_size cannot be changed after the
+thin-pool is created.  People primarily interested in thin provisioning
+may want to use a value such as 1024 (512KB).  People doing lots of
+snapshotting may want a smaller value such as 128 (64KB).  If you are
+not zeroing newly-allocated data, a larger $data_block_size in the
+region of 256000 (128MB) is suggested.
 
 $low_water_mark is expressed in blocks of size $data_block_size.  If
 free space on the data device drops below this level then a dm event

drivers/md/Makefile

···
 #
 
 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-ioctl.o dm-io.o dm-kcopyd.o dm-sysfs.o dm-stats.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-cache-target.c

···
 #define MIGRATION_COUNT_WINDOW 10
 
 /*
- * The block size of the device holding cache data must be >= 32KB
+ * The block size of the device holding cache data must be
+ * between 32KB and 1GB.
  */
 #define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (32 * 1024 >> SECTOR_SHIFT)
+#define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
 
 /*
  * FIXME: the cache is read/write for the time being.
···
 	struct dm_target *ti;
 	struct dm_target_callbacks callbacks;
 
+	struct dm_cache_metadata *cmd;
+
 	/*
 	 * Metadata is written to this device.
 	 */
···
 	 * The faster of the two data devices.  Typically an SSD.
 	 */
 	struct dm_dev *cache_dev;
-
-	/*
-	 * Cache features such as write-through.
-	 */
-	struct cache_features features;
 
 	/*
 	 * Size of the origin device in _complete_ blocks and native sectors.
···
 	uint32_t sectors_per_block;
 	int sectors_per_block_shift;
 
-	struct dm_cache_metadata *cmd;
-
 	spinlock_t lock;
 	struct bio_list deferred_bios;
 	struct bio_list deferred_flush_bios;
···
 	struct list_head completed_migrations;
 	struct list_head need_commit_migrations;
 	sector_t migration_threshold;
-	atomic_t nr_migrations;
 	wait_queue_head_t migration_wait;
+	atomic_t nr_migrations;
 
 	/*
 	 * cache_size entries, dirty if set
···
 	/*
 	 * origin_blocks entries, discarded if set.
 	 */
-	uint32_t discard_block_size; /* a power of 2 times sectors per block */
 	dm_dblock_t discard_nr_blocks;
 	unsigned long *discard_bitset;
+	uint32_t discard_block_size; /* a power of 2 times sectors per block */
+
+	/*
+	 * Rather than reconstructing the table line for the status we just
+	 * save it and regurgitate.
+	 */
+	unsigned nr_ctr_args;
+	const char **ctr_args;
 
 	struct dm_kcopyd_client *copier;
 	struct workqueue_struct *wq;
···
 	bool loaded_mappings:1;
 	bool loaded_discards:1;
 
-	struct cache_stats stats;
-
 	/*
-	 * Rather than reconstructing the table line for the status we just
-	 * save it and regurgitate.
+	 * Cache features such as write-through.
 	 */
-	unsigned nr_ctr_args;
-	const char **ctr_args;
+	struct cache_features features;
+
+	struct cache_stats stats;
 };
 
 struct per_bio_data {
···
 static int parse_block_size(struct cache_args *ca, struct dm_arg_set *as,
 			    char **error)
 {
-	unsigned long tmp;
+	unsigned long block_size;
 
 	if (!at_least_one_arg(as, error))
 		return -EINVAL;
 
-	if (kstrtoul(dm_shift_arg(as), 10, &tmp) || !tmp ||
-	    tmp < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
-	    tmp & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
+	if (kstrtoul(dm_shift_arg(as), 10, &block_size) || !block_size ||
+	    block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
+	    block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
+	    block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
 		*error = "Invalid data block size";
 		return -EINVAL;
 	}
 
-	if (tmp > ca->cache_sectors) {
+	if (block_size > ca->cache_sectors) {
 		*error = "Data block size is larger than the cache device";
 		return -EINVAL;
 	}
 
-	ca->block_size = tmp;
+	ca->block_size = block_size;
 
 	return 0;
 }
···
 static void cache_io_hints(struct dm_target *ti, struct queue_limits *limits)
 {
 	struct cache *cache = ti->private;
+	uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
 
-	blk_limits_io_min(limits, 0);
-	blk_limits_io_opt(limits, cache->sectors_per_block << SECTOR_SHIFT);
+	/*
+	 * If the system-determined stacked limits are compatible with the
+	 * cache's blocksize (io_opt is a factor) do not override them.
+	 */
+	if (io_opt_sectors < cache->sectors_per_block ||
+	    do_div(io_opt_sectors, cache->sectors_per_block)) {
+		blk_limits_io_min(limits, 0);
+		blk_limits_io_opt(limits, cache->sectors_per_block << SECTOR_SHIFT);
+	}
 	set_discard_limits(cache, limits);
 }
 

drivers/md/dm-crypt.c

···
 	}
 
 	ret = -ENOMEM;
-	cc->io_queue = alloc_workqueue("kcryptd_io",
-				       WQ_NON_REENTRANT|
-				       WQ_MEM_RECLAIM,
-				       1);
+	cc->io_queue = alloc_workqueue("kcryptd_io", WQ_MEM_RECLAIM, 1);
 	if (!cc->io_queue) {
 		ti->error = "Couldn't create kcryptd io queue";
 		goto bad;
 	}
 
 	cc->crypt_queue = alloc_workqueue("kcryptd",
-					  WQ_NON_REENTRANT|
-					  WQ_CPU_INTENSIVE|
-					  WQ_MEM_RECLAIM,
-					  1);
+					  WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM, 1);
 	if (!cc->crypt_queue) {
 		ti->error = "Couldn't create kcryptd queue";
 		goto bad;

drivers/md/dm-ioctl.c

···
 	unsigned change_uuid = (param->flags & DM_UUID_FLAG) ? 1 : 0;
 
 	if (new_data < param->data ||
-	    invalid_str(new_data, (void *) param + param_size) ||
+	    invalid_str(new_data, (void *) param + param_size) || !*new_data ||
 	    strlen(new_data) > (change_uuid ? DM_UUID_LEN - 1 : DM_NAME_LEN - 1)) {
 		DMWARN("Invalid new mapped device name or uuid string supplied.");
 		return -EINVAL;
···
 
 	r = dm_table_create(&t, get_mode(param), param->target_count, md);
 	if (r)
-		goto out;
+		goto err;
 
+	/* Protect md->type and md->queue against concurrent table loads. */
+	dm_lock_md_type(md);
 	r = populate_table(t, param, param_size);
-	if (r) {
-		dm_table_destroy(t);
-		goto out;
-	}
+	if (r)
+		goto err_unlock_md_type;
 
 	immutable_target_type = dm_get_immutable_target_type(md);
 	if (immutable_target_type &&
 	    (immutable_target_type != dm_table_get_immutable_target_type(t))) {
 		DMWARN("can't replace immutable target type %s",
 		       immutable_target_type->name);
-		dm_table_destroy(t);
 		r = -EINVAL;
-		goto out;
+		goto err_unlock_md_type;
 	}
 
-	/* Protect md->type and md->queue against concurrent table loads. */
-	dm_lock_md_type(md);
 	if (dm_get_md_type(md) == DM_TYPE_NONE)
 		/* Initial table load: acquire type of table. */
 		dm_set_md_type(md, dm_table_get_type(t));
 	else if (dm_get_md_type(md) != dm_table_get_type(t)) {
 		DMWARN("can't change device type after initial table load.");
-		dm_table_destroy(t);
-		dm_unlock_md_type(md);
 		r = -EINVAL;
-		goto out;
+		goto err_unlock_md_type;
 	}
 
 	/* setup md->queue to reflect md's type (may block) */
 	r = dm_setup_md_queue(md);
 	if (r) {
 		DMWARN("unable to set up device queue for new table.");
-		dm_table_destroy(t);
-		dm_unlock_md_type(md);
-		goto out;
+		goto err_unlock_md_type;
 	}
 	dm_unlock_md_type(md);
 
···
 	if (!hc || hc->md != md) {
 		DMWARN("device has been removed from the dev hash table.");
 		up_write(&_hash_lock);
-		dm_table_destroy(t);
 		r = -ENXIO;
-		goto out;
+		goto err_destroy_table;
 	}
 
 	if (hc->new_map)
···
 	param->flags |= DM_INACTIVE_PRESENT_FLAG;
 	__dev_status(md, param);
 
-out:
 	if (old_map) {
 		dm_sync_table(md);
 		dm_table_destroy(old_map);
 	}
 
+	dm_put(md);
+
+	return 0;
+
+err_unlock_md_type:
+	dm_unlock_md_type(md);
+err_destroy_table:
+	dm_table_destroy(t);
+err:
 	dm_put(md);
 
 	return r;
···
 	return 0;
 }
 
-static bool buffer_test_overflow(char *result, unsigned maxlen)
-{
-	return !maxlen || strlen(result) + 1 >= maxlen;
-}
-
 /*
- * Process device-mapper dependent messages.
+ * Process device-mapper dependent messages.  Messages prefixed with '@'
+ * are processed by the DM core.  All others are delivered to the target.
  * Returns a number <= 1 if message was processed by device mapper.
  * Returns 2 if message should be delivered to the target.
  */
 static int message_for_md(struct mapped_device *md, unsigned argc, char **argv,
 			  char *result, unsigned maxlen)
 {
-	return 2;
+	int r;
+
+	if (**argv != '@')
+		return 2; /* no '@' prefix, deliver to target */
+
+	r = dm_stats_message(md, argc, argv, result, maxlen);
+	if (r < 2)
+		return r;
+
+	DMERR("Unsupported message sent to DM core: %s", argv[0]);
+	return -EINVAL;
 }
 
 /*
···
 
 	if (r == 1) {
 		param->flags |= DM_DATA_OUT_FLAG;
-		if (buffer_test_overflow(result, maxlen))
+		if (dm_message_test_buffer_overflow(result, maxlen))
 			param->flags |= DM_BUFFER_FULL_FLAG;
 		else
 			param->data_size = param->data_start + strlen(result) + 1;

drivers/md/dm-kcopyd.c

···
 		goto bad_slab;
 
 	INIT_WORK(&kc->kcopyd_work, do_work);
-	kc->kcopyd_wq = alloc_workqueue("kcopyd",
-					WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
+	kc->kcopyd_wq = alloc_workqueue("kcopyd", WQ_MEM_RECLAIM, 0);
 	if (!kc->kcopyd_wq)
 		goto bad_workqueue;
 

drivers/md/dm-raid1.c

···
 	ti->per_bio_data_size = sizeof(struct dm_raid1_bio_record);
 	ti->discard_zeroes_data_unsupported = true;
 
-	ms->kmirrord_wq = alloc_workqueue("kmirrord",
-					  WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
+	ms->kmirrord_wq = alloc_workqueue("kmirrord", WQ_MEM_RECLAIM, 0);
 	if (!ms->kmirrord_wq) {
 		DMERR("couldn't start kmirrord");
 		r = -ENOMEM;

drivers/md/dm-stats.c

···
+#include <linux/errno.h>
+#include <linux/numa.h>
+#include <linux/slab.h>
+#include <linux/rculist.h>
+#include <linux/threads.h>
+#include <linux/preempt.h>
+#include <linux/irqflags.h>
+#include <linux/vmalloc.h>
+#include <linux/mm.h>
+#include <linux/module.h>
+#include <linux/device-mapper.h>
+
+#include "dm.h"
+#include "dm-stats.h"
+
+#define DM_MSG_PREFIX "stats"
+
+static int dm_stat_need_rcu_barrier;
+
+/*
+ * Using 64-bit values to avoid overflow (which is a
+ * problem that block/genhd.c's IO accounting has).
+ */
+struct dm_stat_percpu {
+	unsigned long long sectors[2];
+	unsigned long long ios[2];
+	unsigned long long merges[2];
+	unsigned long long ticks[2];
+	unsigned long long io_ticks[2];
+	unsigned long long io_ticks_total;
+	unsigned long long time_in_queue;
+};
+
+struct dm_stat_shared {
+	atomic_t in_flight[2];
+	unsigned long stamp;
+	struct dm_stat_percpu tmp;
+};
+
+struct dm_stat {
+	struct list_head list_entry;
+	int id;
+	size_t n_entries;
+	sector_t start;
+	sector_t end;
+	sector_t step;
+	const char *program_id;
+	const char *aux_data;
+	struct rcu_head rcu_head;
+	size_t shared_alloc_size;
+	size_t percpu_alloc_size;
+	struct dm_stat_percpu *stat_percpu[NR_CPUS];
+	struct dm_stat_shared stat_shared[0];
+};
+
+struct dm_stats_last_position {
+	sector_t last_sector;
+	unsigned last_rw;
+};
+
+/*
+ * A typo on the command line could possibly make the kernel run out of memory
+ * and crash. To prevent the crash we account all used memory. We fail if we
+ * exhaust 1/4 of all memory or 1/2 of vmalloc space.
+ */
+#define DM_STATS_MEMORY_FACTOR		4
+#define DM_STATS_VMALLOC_FACTOR		2
+
+static DEFINE_SPINLOCK(shared_memory_lock);
+
+static unsigned long shared_memory_amount;
+
+static bool __check_shared_memory(size_t alloc_size)
+{
+	size_t a;
+
+	a = shared_memory_amount + alloc_size;
+	if (a < shared_memory_amount)
+		return false;
+	if (a >> PAGE_SHIFT > totalram_pages / DM_STATS_MEMORY_FACTOR)
+		return false;
+#ifdef CONFIG_MMU
+	if (a > (VMALLOC_END - VMALLOC_START) / DM_STATS_VMALLOC_FACTOR)
+		return false;
+#endif
+	return true;
+}
+
+static bool check_shared_memory(size_t alloc_size)
+{
+	bool ret;
+
+	spin_lock_irq(&shared_memory_lock);
+
+	ret = __check_shared_memory(alloc_size);
+
+	spin_unlock_irq(&shared_memory_lock);
+
+	return ret;
+}
+
+static bool claim_shared_memory(size_t alloc_size)
+{
+	spin_lock_irq(&shared_memory_lock);
+
+	if (!__check_shared_memory(alloc_size)) {
+		spin_unlock_irq(&shared_memory_lock);
+		return false;
+	}
+
+	shared_memory_amount += alloc_size;
+
+	spin_unlock_irq(&shared_memory_lock);
+
+	return true;
+}
+
+static void free_shared_memory(size_t alloc_size)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&shared_memory_lock, flags);
+
+	if (WARN_ON_ONCE(shared_memory_amount < alloc_size)) {
+		spin_unlock_irqrestore(&shared_memory_lock, flags);
+		DMCRIT("Memory usage accounting bug.");
+		return;
+	}
+
+	shared_memory_amount -= alloc_size;
+
+	spin_unlock_irqrestore(&shared_memory_lock, flags);
+}
+
+static void *dm_kvzalloc(size_t alloc_size, int node)
+{
+	void *p;
+
+	if (!claim_shared_memory(alloc_size))
+		return NULL;
+
+	if (alloc_size <= KMALLOC_MAX_SIZE) {
+		p = kzalloc_node(alloc_size, GFP_KERNEL | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN, node);
+		if (p)
+			return p;
+	}
+	p = vzalloc_node(alloc_size, node);
+	if (p)
+		return p;
+
+	free_shared_memory(alloc_size);
+
+	return NULL;
+}
+
+static void dm_kvfree(void *ptr, size_t alloc_size)
+{
+	if (!ptr)
+		return;
+
+	free_shared_memory(alloc_size);
+
+	if (is_vmalloc_addr(ptr))
+		vfree(ptr);
+	else
+		kfree(ptr);
+}
+
+static void dm_stat_free(struct rcu_head *head)
+{
+	int cpu;
+	struct dm_stat *s = container_of(head, struct dm_stat, rcu_head);
+
+	kfree(s->program_id);
+	kfree(s->aux_data);
+	for_each_possible_cpu(cpu)
+		dm_kvfree(s->stat_percpu[cpu], s->percpu_alloc_size);
+	dm_kvfree(s, s->shared_alloc_size);
+}
+
+static int dm_stat_in_flight(struct dm_stat_shared *shared)
+{
+	return atomic_read(&shared->in_flight[READ]) +
+	       atomic_read(&shared->in_flight[WRITE]);
+}
+
+void dm_stats_init(struct dm_stats *stats)
+{
+	int cpu;
+	struct dm_stats_last_position *last;
+
+	mutex_init(&stats->mutex);
+	INIT_LIST_HEAD(&stats->list);
+	stats->last = alloc_percpu(struct dm_stats_last_position);
+	for_each_possible_cpu(cpu) {
+		last = per_cpu_ptr(stats->last, cpu);
+		last->last_sector = (sector_t)ULLONG_MAX;
+		last->last_rw = UINT_MAX;
+	}
+}
+
+void dm_stats_cleanup(struct dm_stats *stats)
+{
+	size_t ni;
+	struct dm_stat *s;
+	struct dm_stat_shared *shared;
+
+	while (!list_empty(&stats->list)) {
+		s = container_of(stats->list.next, struct dm_stat, list_entry);
+		list_del(&s->list_entry);
+		for (ni = 0; ni < s->n_entries; ni++) {
+			shared = &s->stat_shared[ni];
+			if (WARN_ON(dm_stat_in_flight(shared))) {
+				DMCRIT("leaked in-flight counter at index %lu "
+				       "(start %llu, end %llu, step %llu): reads %d, writes %d",
+				       (unsigned long)ni,
+				       (unsigned long long)s->start,
+				       (unsigned long long)s->end,
+				       (unsigned long long)s->step,
+				       atomic_read(&shared->in_flight[READ]),
+				       atomic_read(&shared->in_flight[WRITE]));
+			}
+		}
+		dm_stat_free(&s->rcu_head);
+	}
+	free_percpu(stats->last);
+}
+
+static int dm_stats_create(struct dm_stats *stats, sector_t start, sector_t end,
+			   sector_t step, const char *program_id, const char *aux_data,
+			   void (*suspend_callback)(struct mapped_device *),
+			   void (*resume_callback)(struct mapped_device *),
+			   struct mapped_device *md)
+{
+	struct list_head *l;
+	struct dm_stat *s, *tmp_s;
+	sector_t n_entries;
+	size_t ni;
+	size_t shared_alloc_size;
+	size_t percpu_alloc_size;
+	struct dm_stat_percpu *p;
+	int cpu;
+	int ret_id;
+	int r;
+
+	if (end < start || !step)
+		return -EINVAL;
+
+	n_entries = end - start;
+	if (dm_sector_div64(n_entries, step))
+		n_entries++;
+
+	if (n_entries != (size_t)n_entries || !(size_t)(n_entries + 1))
+		return -EOVERFLOW;
+
+	shared_alloc_size = sizeof(struct dm_stat) + (size_t)n_entries * sizeof(struct dm_stat_shared);
+	if ((shared_alloc_size - sizeof(struct dm_stat)) / sizeof(struct dm_stat_shared) != n_entries)
+		return -EOVERFLOW;
+
+	percpu_alloc_size = (size_t)n_entries * sizeof(struct dm_stat_percpu);
+	if (percpu_alloc_size / sizeof(struct dm_stat_percpu) != n_entries)
+		return -EOVERFLOW;
+
+	if (!check_shared_memory(shared_alloc_size + num_possible_cpus() * percpu_alloc_size))
+		return -ENOMEM;
+
+	s = dm_kvzalloc(shared_alloc_size, NUMA_NO_NODE);
+	if (!s)
+		return -ENOMEM;
+
+	s->n_entries = n_entries;
+	s->start = start;
+	s->end = end;
+	s->step = step;
+	s->shared_alloc_size = shared_alloc_size;
+	s->percpu_alloc_size = percpu_alloc_size;
+
+	s->program_id = kstrdup(program_id, GFP_KERNEL);
+	if (!s->program_id) {
+		r = -ENOMEM;
+		goto out;
+	}
+	s->aux_data = kstrdup(aux_data, GFP_KERNEL);
+	if (!s->aux_data) {
+		r = -ENOMEM;
+		goto out;
+	}
+
+	for (ni = 0; ni < n_entries; ni++) {
+		atomic_set(&s->stat_shared[ni].in_flight[READ], 0);
+		atomic_set(&s->stat_shared[ni].in_flight[WRITE], 0);
+	}
+
+	for_each_possible_cpu(cpu) {
+		p = dm_kvzalloc(percpu_alloc_size, cpu_to_node(cpu));
+		if (!p) {
+			r = -ENOMEM;
+			goto out;
+		}
+		s->stat_percpu[cpu] = p;
+	}
+
+	/*
+	 * Suspend/resume to make sure there is no i/o in flight,
+	 * so that newly created statistics will be exact.
+	 *
+	 * (note: we couldn't suspend earlier because we must not
+	 * allocate memory while suspended)
+	 */
+	suspend_callback(md);
+
+	mutex_lock(&stats->mutex);
+	s->id = 0;
+	list_for_each(l, &stats->list) {
+		tmp_s = container_of(l, struct dm_stat, list_entry);
+		if (WARN_ON(tmp_s->id < s->id)) {
+			r = -EINVAL;
+			goto out_unlock_resume;
+		}
+		if (tmp_s->id > s->id)
+			break;
+		if (unlikely(s->id == INT_MAX)) {
+			r = -ENFILE;
+			goto out_unlock_resume;
+		}
+		s->id++;
+	}
+	ret_id = s->id;
+	list_add_tail_rcu(&s->list_entry, l);
+	mutex_unlock(&stats->mutex);
+
+	resume_callback(md);
+
+	return ret_id;
+
+out_unlock_resume:
+	mutex_unlock(&stats->mutex);
+	resume_callback(md);
+out:
+	dm_stat_free(&s->rcu_head);
+	return r;
+}
+
+static struct dm_stat *__dm_stats_find(struct dm_stats *stats, int id)
+{
+	struct dm_stat *s;
+
+	list_for_each_entry(s, &stats->list, list_entry) {
+		if (s->id > id)
+			break;
+		if (s->id == id)
+			return s;
+	}
+
+	return NULL;
+}
+
+static int dm_stats_delete(struct dm_stats *stats, int id)
+{
+	struct dm_stat *s;
+	int cpu;
+
+	mutex_lock(&stats->mutex);
+
+	s = __dm_stats_find(stats, id);
+	if (!s) {
+		mutex_unlock(&stats->mutex);
+		return -ENOENT;
+	}
+
+	list_del_rcu(&s->list_entry);
+	mutex_unlock(&stats->mutex);
+
+	/*
+	 * vfree can't be called from RCU callback
+	 */
+	for_each_possible_cpu(cpu)
+		if (is_vmalloc_addr(s->stat_percpu))
+			goto do_sync_free;
+	if (is_vmalloc_addr(s)) {
+do_sync_free:
+		synchronize_rcu_expedited();
+		dm_stat_free(&s->rcu_head);
+	} else {
+		ACCESS_ONCE(dm_stat_need_rcu_barrier) = 1;
+		call_rcu(&s->rcu_head, dm_stat_free);
+	}
+	return 0;
+}
+
+static int dm_stats_list(struct dm_stats *stats, const char *program,
+			 char *result, unsigned maxlen)
+{
+	struct dm_stat *s;
+	sector_t len;
+	unsigned sz = 0;
+
+	/*
+	 * Output format:
+	 *   <region_id>: <start_sector>+<length> <step> <program_id> <aux_data>
+	 */
+
+	mutex_lock(&stats->mutex);
+	list_for_each_entry(s, &stats->list, list_entry) {
+		if (!program || !strcmp(program, s->program_id)) {
+			len = s->end - s->start;
+			DMEMIT("%d: %llu+%llu %llu %s %s\n", s->id,
+				(unsigned long long)s->start,
+				(unsigned long long)len,
+				(unsigned long long)s->step,
+				s->program_id,
+				s->aux_data);
+		}
+	}
+	mutex_unlock(&stats->mutex);
+
+	return 1;
+}
+
+static void dm_stat_round(struct dm_stat_shared *shared, struct dm_stat_percpu *p)
+{
+	/*
+	 * This is racy, but so is part_round_stats_single.
+	 */
+	unsigned long now = jiffies;
+	unsigned in_flight_read;
+	unsigned in_flight_write;
+	unsigned long difference = now - shared->stamp;
+
+	if (!difference)
+		return;
+	in_flight_read = (unsigned)atomic_read(&shared->in_flight[READ]);
+	in_flight_write = (unsigned)atomic_read(&shared->in_flight[WRITE]);
+	if (in_flight_read)
+		p->io_ticks[READ] += difference;
+	if (in_flight_write)
+		p->io_ticks[WRITE] += difference;
+	if (in_flight_read + in_flight_write) {
+		p->io_ticks_total += difference;
+		p->time_in_queue += (in_flight_read + in_flight_write) * difference;
+	}
+	shared->stamp = now;
+}
+
+static void dm_stat_for_entry(struct dm_stat *s, size_t entry,
+			      unsigned long bi_rw, sector_t len, bool merged,
+			      bool end, unsigned long duration)
+{
+	unsigned long idx = bi_rw & REQ_WRITE;
+	struct dm_stat_shared *shared = &s->stat_shared[entry];
+	struct dm_stat_percpu *p;
+
+	/*
+	 * For strict correctness we should use local_irq_disable/enable
+	 * instead of preempt_disable/enable.
+	 *
+	 * This is racy if the driver finishes bios from non-interrupt
+	 * context as well as from interrupt context or from more different
+	 * interrupts.
+	 *
+	 * However, the race only results in not counting some events,
+	 * so it is acceptable.
+	 *
+	 * part_stat_lock()/part_stat_unlock() have this race too.
+	 */
+	preempt_disable();
+	p = &s->stat_percpu[smp_processor_id()][entry];
+
+	if (!end) {
+		dm_stat_round(shared, p);
+		atomic_inc(&shared->in_flight[idx]);
+	} else {
+		dm_stat_round(shared, p);
+		atomic_dec(&shared->in_flight[idx]);
+		p->sectors[idx] += len;
+		p->ios[idx] += 1;
+		p->merges[idx] += merged;
+		p->ticks[idx] += duration;
+	}
+
+	preempt_enable();
+}
+
+static void __dm_stat_bio(struct dm_stat *s, unsigned long bi_rw,
+			  sector_t bi_sector, sector_t end_sector,
+			  bool end, unsigned long duration,
+			  struct dm_stats_aux *stats_aux)
+{
+	sector_t rel_sector, offset, todo, fragment_len;
+	size_t entry;
+
+	if (end_sector <= s->start || bi_sector >= s->end)
+		return;
+	if (unlikely(bi_sector < s->start)) {
+		rel_sector = 0;
+		todo = end_sector - s->start;
+	} else {
+		rel_sector = bi_sector - s->start;
+		todo = end_sector - bi_sector;
+	}
+	if (unlikely(end_sector > s->end))
+		todo -= (end_sector - s->end);
+
+	offset = dm_sector_div64(rel_sector, s->step);
+	entry = rel_sector;
+	do {
+		if (WARN_ON_ONCE(entry >= s->n_entries)) {
+			DMCRIT("Invalid area access in region id %d", s->id);
+			return;
+		}
+		fragment_len = todo;
+		if (fragment_len > s->step - offset)
+			fragment_len = s->step - offset;
+		dm_stat_for_entry(s, entry, bi_rw, fragment_len,
+				  stats_aux->merged, end, duration);
+		todo -= fragment_len;
+		entry++;
+		offset = 0;
+	} while (unlikely(todo != 0));
+}
+
+void dm_stats_account_io(struct dm_stats *stats, unsigned long bi_rw,
+			 sector_t bi_sector, unsigned bi_sectors, bool end,
+			 unsigned long duration, struct dm_stats_aux *stats_aux)
+{
+	struct dm_stat *s;
+	sector_t end_sector;
+	struct dm_stats_last_position *last;
+
+	if (unlikely(!bi_sectors))
+		return;
+
+	end_sector = bi_sector + bi_sectors;
+
+	if (!end) {
+		/*
+		 * A race condition can at worst result in the merged flag being
+		 * misrepresented, so we don't have to disable preemption here.
+		 */
+		last = __this_cpu_ptr(stats->last);
+		stats_aux->merged =
+			(bi_sector == (ACCESS_ONCE(last->last_sector) &&
+				       ((bi_rw & (REQ_WRITE | REQ_DISCARD)) ==
+					(ACCESS_ONCE(last->last_rw) & (REQ_WRITE | REQ_DISCARD)))
+				       ));
+		ACCESS_ONCE(last->last_sector) = end_sector;
+		ACCESS_ONCE(last->last_rw) = bi_rw;
+	}
+
+	rcu_read_lock();
+
+	list_for_each_entry_rcu(s, &stats->list, list_entry)
+		__dm_stat_bio(s, bi_rw, bi_sector, end_sector, end, duration, stats_aux);
+
+	rcu_read_unlock();
+}
+
+static void __dm_stat_init_temporary_percpu_totals(struct dm_stat_shared *shared,
+						   struct dm_stat *s, size_t x)
+{
+	int cpu;
+	struct dm_stat_percpu *p;
+
+	local_irq_disable();
+	p = &s->stat_percpu[smp_processor_id()][x];
+	dm_stat_round(shared, p);
+	local_irq_enable();
+
+	memset(&shared->tmp, 0, sizeof(shared->tmp));
+	for_each_possible_cpu(cpu) {
+		p = &s->stat_percpu[cpu][x];
+		shared->tmp.sectors[READ] += ACCESS_ONCE(p->sectors[READ]);
+		shared->tmp.sectors[WRITE] += ACCESS_ONCE(p->sectors[WRITE]);
+		shared->tmp.ios[READ] += ACCESS_ONCE(p->ios[READ]);
+		shared->tmp.ios[WRITE] += ACCESS_ONCE(p->ios[WRITE]);
+		shared->tmp.merges[READ] += ACCESS_ONCE(p->merges[READ]);
+		shared->tmp.merges[WRITE] += ACCESS_ONCE(p->merges[WRITE]);
+		shared->tmp.ticks[READ] += ACCESS_ONCE(p->ticks[READ]);
+		shared->tmp.ticks[WRITE] += ACCESS_ONCE(p->ticks[WRITE]);
+		shared->tmp.io_ticks[READ] += ACCESS_ONCE(p->io_ticks[READ]);
+		shared->tmp.io_ticks[WRITE] += ACCESS_ONCE(p->io_ticks[WRITE]);
+		shared->tmp.io_ticks_total += ACCESS_ONCE(p->io_ticks_total);
+		shared->tmp.time_in_queue += ACCESS_ONCE(p->time_in_queue);
+	}
+}
+
+static void __dm_stat_clear(struct dm_stat *s, size_t idx_start, size_t idx_end,
+			    bool init_tmp_percpu_totals)
+{
+	size_t x;
+	struct dm_stat_shared *shared;
+	struct dm_stat_percpu *p;
+
+	for (x = idx_start; x < idx_end; x++) {
+		shared = &s->stat_shared[x];
+		if (init_tmp_percpu_totals)
+			__dm_stat_init_temporary_percpu_totals(shared, s, x);
+		local_irq_disable();
+		p = &s->stat_percpu[smp_processor_id()][x];
+		p->sectors[READ] -= shared->tmp.sectors[READ];
+		p->sectors[WRITE] -= shared->tmp.sectors[WRITE];
+		p->ios[READ] -= shared->tmp.ios[READ];
+		p->ios[WRITE] -= shared->tmp.ios[WRITE];
+		p->merges[READ] -= shared->tmp.merges[READ];
+		p->merges[WRITE] -= shared->tmp.merges[WRITE];
+		p->ticks[READ] -= shared->tmp.ticks[READ];
+		p->ticks[WRITE] -= shared->tmp.ticks[WRITE];
+		p->io_ticks[READ] -= shared->tmp.io_ticks[READ];
+		p->io_ticks[WRITE] -= shared->tmp.io_ticks[WRITE];
+		p->io_ticks_total -= shared->tmp.io_ticks_total;
+		p->time_in_queue -= shared->tmp.time_in_queue;
+		local_irq_enable();
+	}
+}
+
+static int dm_stats_clear(struct dm_stats *stats, int id)
+{
+	struct dm_stat *s;
+
+	mutex_lock(&stats->mutex);
+
+	s = __dm_stats_find(stats, id);
+	if (!s) {
+		mutex_unlock(&stats->mutex);
+		return -ENOENT;
+	}
+
+	__dm_stat_clear(s, 0, s->n_entries, true);
+
+	mutex_unlock(&stats->mutex);
+
+	return 1;
+}
+
+/*
+ * This is like jiffies_to_msec, but works for 64-bit values.
+ */
+static unsigned long long dm_jiffies_to_msec64(unsigned long long j)
+{
+	unsigned long long result = 0;
+	unsigned mult;
+
+	if (j)
+		result = jiffies_to_msecs(j & 0x3fffff);
+	if (j >= 1 << 22) {
+		mult = jiffies_to_msecs(1 << 22);
+		result += (unsigned long long)mult * (unsigned long long)jiffies_to_msecs((j >> 22) & 0x3fffff);
+	}
+	if (j >= 1ULL << 44)
+		result += (unsigned long long)mult * (unsigned long long)mult * (unsigned long long)jiffies_to_msecs(j >> 44);
+
+	return result;
+}
+
+static int dm_stats_print(struct dm_stats *stats, int id,
+			  size_t idx_start, size_t idx_len,
+			  bool clear, char *result, unsigned maxlen)
+{
+	unsigned sz = 0;
+	struct dm_stat *s;
+	size_t x;
+	sector_t start, end, step;
+	size_t idx_end;
+	struct dm_stat_shared *shared;
+
+	/*
+	 * Output format:
+	 *   <start_sector>+<length> counters
+	 */
+
+	mutex_lock(&stats->mutex);
+
+	s = __dm_stats_find(stats, id);
+	if (!s) {
+		mutex_unlock(&stats->mutex);
+		return -ENOENT;
+	}
+
+	idx_end = idx_start + idx_len;
+	if (idx_end < idx_start ||
+	    idx_end > s->n_entries)
+		idx_end = s->n_entries;
+
+	if (idx_start > idx_end)
+		idx_start = idx_end;
+
+	step = s->step;
+	start = s->start + (step * idx_start);
+
+	for (x = idx_start; x < idx_end; x++, start = end) {
+		shared = &s->stat_shared[x];
+		end = start + step;
+		if (unlikely(end > s->end))
+			end = s->end;
+
+		__dm_stat_init_temporary_percpu_totals(shared, s, x);
+
+		DMEMIT("%llu+%llu %llu %llu %llu %llu %llu %llu %llu %llu %d %llu %llu %llu %llu\n",
+		       (unsigned long long)start,
+		       (unsigned long long)step,
+		       shared->tmp.ios[READ],
+		       shared->tmp.merges[READ],
+		       shared->tmp.sectors[READ],
+		       dm_jiffies_to_msec64(shared->tmp.ticks[READ]),
+		       shared->tmp.ios[WRITE],
+		       shared->tmp.merges[WRITE],
+		       shared->tmp.sectors[WRITE],
+		       dm_jiffies_to_msec64(shared->tmp.ticks[WRITE]),
+		       dm_stat_in_flight(shared),
+		       dm_jiffies_to_msec64(shared->tmp.io_ticks_total),
+		       dm_jiffies_to_msec64(shared->tmp.time_in_queue),
+		       dm_jiffies_to_msec64(shared->tmp.io_ticks[READ]),
+		       dm_jiffies_to_msec64(shared->tmp.io_ticks[WRITE]));
+
+		if (unlikely(sz + 1 >= maxlen))
+			goto buffer_overflow;
+	}
+
+	if (clear)
+		__dm_stat_clear(s, idx_start, idx_end, false);
+
+buffer_overflow:
+	mutex_unlock(&stats->mutex);
+
+	return 1;
+}
+
+static int dm_stats_set_aux(struct dm_stats *stats, int id, const char *aux_data)
+{
+	struct dm_stat *s;
+	const char *new_aux_data;
+
+	mutex_lock(&stats->mutex);
+
+	s = __dm_stats_find(stats, id);
+	if (!s) {
+		mutex_unlock(&stats->mutex);
+		return -ENOENT;
+	}
+
+	new_aux_data = kstrdup(aux_data, GFP_KERNEL);
+	if (!new_aux_data) {
+		mutex_unlock(&stats->mutex);
+		return -ENOMEM;
+	}
+
+	kfree(s->aux_data);
+	s->aux_data = new_aux_data;
+
+	mutex_unlock(&stats->mutex);
+
+	return 0;
+}
+
+static int message_stats_create(struct mapped_device *md,
+				unsigned argc, char **argv,
+				char *result, unsigned maxlen)
+{
+	int id;
+	char dummy;
+	unsigned long long start, end, len, step;
+	unsigned divisor;
+	const char *program_id, *aux_data;
+
+	/*
+	 * Input format:
+	 *   <range> <step> [<program_id> [<aux_data>]]
+	 */
+
+	if (argc < 3 || argc > 5)
+		return -EINVAL;
+
+	if (!strcmp(argv[1], "-")) {
+		start = 0;
+		len = dm_get_size(md);
+		if (!len)
+			len = 1;
+	} else if (sscanf(argv[1], "%llu+%llu%c", &start, &len, &dummy) != 2 ||
+		   start != (sector_t)start || len != (sector_t)len)
+		return -EINVAL;
+
+	end = start + len;
+	if (start >= end)
+		return -EINVAL;
+
+	if (sscanf(argv[2], "/%u%c", &divisor, &dummy) == 1) {
+		step = end - start;
+		if (do_div(step, divisor))
+			step++;
+		if (!step)
+			step = 1;
+	} else if (sscanf(argv[2], "%llu%c", &step, &dummy) != 1 ||
+		   step != (sector_t)step || !step)
+		return -EINVAL;
+
+	program_id = "-";
+	aux_data = "-";
+
+	if (argc > 3)
+		program_id = argv[3];
+
+	if (argc > 4)
+		aux_data = argv[4];
+
+	/*
+	 * If a buffer overflow happens after we created the region,
+	 * it's too late (the userspace would retry with a larger
+	 * buffer, but the region id that caused the overflow is already
+	 * leaked).  So we must detect buffer overflow in advance.
+	 */
+	snprintf(result, maxlen, "%d", INT_MAX);
+	if (dm_message_test_buffer_overflow(result, maxlen))
+		return 1;
+
+	id = dm_stats_create(dm_get_stats(md), start, end, step, program_id, aux_data,
+			     dm_internal_suspend, dm_internal_resume, md);
+	if (id < 0)
+		return id;
+
+	snprintf(result, maxlen, "%d", id);
+
+	return 1;
+}
+
+static int message_stats_delete(struct mapped_device *md,
+				unsigned argc, char **argv)
+{
+	int id;
+	char dummy;
+
+	if (argc != 2)
+		return -EINVAL;
+
+	if (sscanf(argv[1], "%d%c", &id, &dummy) != 1 || id < 0)
+		return -EINVAL;
+
+	return dm_stats_delete(dm_get_stats(md), id);
+}
+
+static int message_stats_clear(struct mapped_device *md,
+			       unsigned argc, char **argv)
+{
+	int id;
+	char dummy;
+
+	if (argc != 2)
+		return -EINVAL;
+
+	if (sscanf(argv[1], "%d%c", &id, &dummy) != 1 || id < 0)
+		return -EINVAL;
+
+	return dm_stats_clear(dm_get_stats(md), id);
+}
+
+static int message_stats_list(struct mapped_device *md,
+			      unsigned argc, char **argv,
+			      char *result, unsigned maxlen)
+{
+	int r;
+	const char *program = NULL;
+
+	if (argc < 1 || argc > 2)
+		return -EINVAL;
+
+	if (argc > 1) {
+		program = kstrdup(argv[1], GFP_KERNEL);
+		if (!program)
+			return -ENOMEM;
+	}
+
+	r = dm_stats_list(dm_get_stats(md), program, result, maxlen);
+
+	kfree(program);
+
+	return r;
+}
+
+static int message_stats_print(struct mapped_device *md,
+			       unsigned argc, char **argv, bool clear,
+			       char *result, unsigned maxlen)
+{
+	int id;
+	char dummy;
+	unsigned long idx_start = 0, idx_len = ULONG_MAX;
+
+	if (argc != 2 && argc != 4)
+		return -EINVAL;
+
+	if (sscanf(argv[1], "%d%c", &id, &dummy) != 1 || id < 0)
+		return -EINVAL;
+
+	if (argc > 3) {
+		if (strcmp(argv[2], "-") &&
+		    sscanf(argv[2], "%lu%c", &idx_start, &dummy) != 1)
+			return -EINVAL;
+		if (strcmp(argv[3], "-") &&
+		    sscanf(argv[3], "%lu%c", &idx_len, &dummy) != 1)
+			return -EINVAL;
+	}
+
+	return dm_stats_print(dm_get_stats(md), id, idx_start, idx_len, clear,
+			      result, maxlen);
+}
+
+static int message_stats_set_aux(struct mapped_device *md,
+				 unsigned argc, char **argv)
+{
+	int id;
+	char dummy;
+
+	if (argc != 3)
+		return -EINVAL;
+
+	if (sscanf(argv[1], "%d%c", &id, &dummy) != 1 || id < 0)
+		return -EINVAL;
+
+	return dm_stats_set_aux(dm_get_stats(md), id, argv[2]);
+}
+
+int dm_stats_message(struct mapped_device *md, unsigned argc, char **argv,
+		     char *result, unsigned maxlen)
+{
+	int r;
+
+	if (dm_request_based(md)) {
+		DMWARN("Statistics are only supported for bio-based devices");
+		return -EOPNOTSUPP;
+	}
+
+	/* All messages here must start with '@' */
+	if (!strcasecmp(argv[0], "@stats_create"))
+		r = message_stats_create(md, argc, argv, result, maxlen);
+	else if (!strcasecmp(argv[0], "@stats_delete"))
+		r = message_stats_delete(md, argc, argv);
+	else if (!strcasecmp(argv[0], "@stats_clear"))
+		r = message_stats_clear(md, argc, argv);
+	else if (!strcasecmp(argv[0], "@stats_list"))
+		r = message_stats_list(md, argc, argv, result, maxlen);
+	else if (!strcasecmp(argv[0], "@stats_print"))
+		r = message_stats_print(md, argc, argv, false, result, maxlen);
+	else if (!strcasecmp(argv[0], "@stats_print_clear"))
+		r = message_stats_print(md, argc, argv, true, result, maxlen);
+	else if (!strcasecmp(argv[0], "@stats_set_aux"))
+		r = message_stats_set_aux(md, argc, argv);
+	else
+		return 2; /* this wasn't a stats message */
+
+	if (r == -EINVAL)
+		DMWARN("Invalid parameters for message %s", argv[0]);
+
+	return r;
+}
+
+int __init dm_statistics_init(void)
+{
+	dm_stat_need_rcu_barrier = 0;
+	return 0;
+}
+
+void dm_statistics_exit(void)
+{
+	if (dm_stat_need_rcu_barrier)
+		rcu_barrier();
+	if (WARN_ON(shared_memory_amount))
+		DMCRIT("shared_memory_amount leaked: %lu", shared_memory_amount);
+}
+
+module_param_named(stats_current_allocated_bytes, shared_memory_amount, ulong, S_IRUGO);
+MODULE_PARM_DESC(stats_current_allocated_bytes, "Memory currently used by statistics");

drivers/md/dm-stats.h

···
+#ifndef DM_STATS_H
+#define DM_STATS_H
+
+#include <linux/types.h>
+#include <linux/mutex.h>
+#include <linux/list.h>
+
+int dm_statistics_init(void);
+void dm_statistics_exit(void);
+
+struct dm_stats {
+	struct mutex mutex;
+	struct list_head list;	/* list of struct dm_stat */
+	struct dm_stats_last_position __percpu *last;
+	sector_t last_sector;
+	unsigned last_rw;
+};
+
+struct dm_stats_aux {
+	bool merged;
+};
+
+void dm_stats_init(struct dm_stats *st);
+void dm_stats_cleanup(struct dm_stats *st);
+
+struct mapped_device;
+
+int dm_stats_message(struct mapped_device *md, unsigned argc, char **argv,
+		     char *result, unsigned maxlen);
+
+void dm_stats_account_io(struct dm_stats *stats, unsigned long bi_rw,
+			 sector_t bi_sector, unsigned bi_sectors, bool end,
+			 unsigned long duration, struct dm_stats_aux *aux);
+
+static inline bool dm_stats_used(struct dm_stats *st)
+{
+	return !list_empty(&st->list);
+}
+
+#endif

drivers/md/dm-stripe.c

···
  * This file is released under the GPL.
  */
 
+#include "dm.h"
 #include <linux/device-mapper.h>
 
 #include <linux/module.h>

drivers/md/dm-table.c

···
 static int dm_table_set_type(struct dm_table *t)
 {
 	unsigned i;
-	unsigned bio_based = 0, request_based = 0;
+	unsigned bio_based = 0, request_based = 0, hybrid = 0;
 	struct dm_target *tgt;
 	struct dm_dev_internal *dd;
 	struct list_head *devices;
+	unsigned live_md_type;
 
 	for (i = 0; i < t->num_targets; i++) {
 		tgt = t->targets + i;
-		if (dm_target_request_based(tgt))
+		if (dm_target_hybrid(tgt))
+			hybrid = 1;
+		else if (dm_target_request_based(tgt))
 			request_based = 1;
 		else
 			bio_based = 1;
···
 			       " can't be mixed up");
 			return -EINVAL;
 		}
+	}
+
+	if (hybrid && !bio_based && !request_based) {
+		/*
+		 * The targets can work either way.
+		 * Determine the type from the live device.
+		 * Default to bio-based if device is new.
+		 */
+		live_md_type = dm_get_md_type(t->md);
+		if (live_md_type == DM_TYPE_REQUEST_BASED)
+			request_based = 1;
+		else
+			bio_based = 1;
 	}
 
 	if (bio_based) {

drivers/md/dm-target.c

···
 	return -EIO;
 }
 
+static int io_err_map_rq(struct dm_target *ti, struct request *clone,
+			 union map_info *map_context)
+{
+	return -EIO;
+}
+
 static struct target_type error_target = {
 	.name = "error",
-	.version = {1, 1, 0},
+	.version = {1, 2, 0},
 	.ctr  = io_err_ctr,
 	.dtr  = io_err_dtr,
 	.map  = io_err_map,
+	.map_rq = io_err_map_rq,
 };
 
 int __init dm_target_init(void)

drivers/md/dm-thin.c

···
 
 	r = dm_pool_commit_metadata(pool->pmd);
 	if (r)
-		DMERR_LIMIT("commit failed: error = %d", r);
+		DMERR_LIMIT("%s: commit failed: error = %d",
+			    dm_device_name(pool->pool_md), r);
 
 	return r;
 }
···
 	unsigned long flags;
 	struct pool *pool = tc->pool;
 
+	/*
+	 * Once no_free_space is set we must not allow allocation to succeed.
+	 * Otherwise it is difficult to explain, debug, test and support.
+	 */
+	if (pool->no_free_space)
+		return -ENOSPC;
+
 	r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
 	if (r)
 		return r;
···
 	}
 
 	if (!free_blocks) {
-		if (pool->no_free_space)
+		/*
+		 * Try to commit to see if that will free up some
+		 * more space.
+		 */
+		(void) commit_or_fallback(pool);
+
+		r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
+		if (r)
+			return r;
+
+		/*
+		 * If we still have no space we set a flag to avoid
+		 * doing all this checking and return -ENOSPC.  This
+		 * flag serves as a latch that disallows allocations from
+		 * this pool until the admin takes action (e.g. resize or
+		 * table reload).
+		 */
+		if (!free_blocks) {
+			DMWARN("%s: no free space available.",
+			       dm_device_name(pool->pool_md));
+			spin_lock_irqsave(&pool->lock, flags);
+			pool->no_free_space = 1;
+			spin_unlock_irqrestore(&pool->lock, flags);
 			return -ENOSPC;
-		else {
-			/*
-			 * Try to commit to see if that will free up some
-			 * more space.
-			 */
-			(void) commit_or_fallback(pool);
-
-			r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
-			if (r)
-				return r;
-
-			/*
-			 * If we still have no space we set a flag to avoid
-			 * doing all this checking and return -ENOSPC.
-			 */
-			if (!free_blocks) {
-				DMWARN("%s: no free space available.",
-				       dm_device_name(pool->pool_md));
-				spin_lock_irqsave(&pool->lock, flags);
-				pool->no_free_space = 1;
-				spin_unlock_irqrestore(&pool->lock, flags);
-				return -ENOSPC;
-			}
 		}
 	}
 
···
 {
 	int r;
 	dm_block_t data_block;
+	struct pool *pool = tc->pool;
 
 	r = alloc_data_block(tc, &data_block);
 	switch (r) {
···
 		break;
 
 	case -ENOSPC:
-		no_space(tc->pool, cell);
+		no_space(pool, cell);
 		break;
 
 	default:
 		DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
 			    __func__, r);
-		cell_error(tc->pool, cell);
+		set_pool_mode(pool, PM_READ_ONLY);
+		cell_error(pool, cell);
 		break;
 	}
 }
···
 
 	switch (mode) {
 	case PM_FAIL:
-		DMERR("switching pool to failure mode");
+		DMERR("%s: switching pool to failure mode",
+		      dm_device_name(pool->pool_md));
 		pool->process_bio = process_bio_fail;
 		pool->process_discard = process_bio_fail;
 		pool->process_prepared_mapping = process_prepared_mapping_fail;
···
 		break;
 
 	case PM_READ_ONLY:
-		DMERR("switching pool to read-only mode");
+		DMERR("%s: switching pool to read-only mode",
+		      dm_device_name(pool->pool_md));
 		r = dm_pool_abort_metadata(pool->pmd);
 		if (r) {
-			DMERR("aborting transaction failed");
+			DMERR("%s: aborting transaction failed",
+			      dm_device_name(pool->pool_md));
 			set_pool_mode(pool, PM_FAIL);
 		} else {
 			dm_pool_metadata_read_only(pool->pmd);
···
 
 	r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
 	if (r) {
-		DMERR("failed to retrieve data device size");
+		DMERR("%s: failed to retrieve data device size",
+		      dm_device_name(pool->pool_md));
 		return r;
 	}
 
 	if (data_size < sb_data_size) {
-		DMERR("pool target (%llu blocks) too small: expected %llu",
+		DMERR("%s: pool target (%llu blocks) too small: expected %llu",
+		      dm_device_name(pool->pool_md),
 		      (unsigned long long)data_size, sb_data_size);
 		return -EINVAL;
 
 	} else if (data_size > sb_data_size) {
 		r = dm_pool_resize_data_dev(pool->pmd, data_size);
 		if (r) {
-			DMERR("failed to resize data device");
+			DMERR("%s: failed to resize data device",
+			      dm_device_name(pool->pool_md));
 			set_pool_mode(pool, PM_READ_ONLY);
 			return r;
 		}
···
 
 	r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
 	if (r) {
-		DMERR("failed to retrieve data device size");
+		DMERR("%s: failed to retrieve metadata device size",
+		      dm_device_name(pool->pool_md));
 		return r;
 	}
 
 	if (metadata_dev_size < sb_metadata_dev_size) {
-		DMERR("metadata device (%llu blocks) too small: expected %llu",
+		DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
+		      dm_device_name(pool->pool_md),
 		      metadata_dev_size, sb_metadata_dev_size);
 		return -EINVAL;
 
 	} else if (metadata_dev_size > sb_metadata_dev_size) {
 		r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
 		if (r) {
-			DMERR("failed to resize metadata device");
+			DMERR("%s: failed to resize metadata device",
+			      dm_device_name(pool->pool_md));
 			return r;
 		}
 
···
 
 		r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
 		if (r) {
-			DMERR("dm_pool_get_metadata_transaction_id returned %d", r);
+			DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
+			      dm_device_name(pool->pool_md), r);
 			goto err;
 		}
 
 		r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
 		if (r) {
-			DMERR("dm_pool_get_free_metadata_block_count returned %d", r);
+			DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
+			      dm_device_name(pool->pool_md), r);
 			goto err;
 		}
 
 		r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
 		if (r) {
-			DMERR("dm_pool_get_metadata_dev_size returned %d", r);
+			DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
+			      dm_device_name(pool->pool_md), r);
 			goto err;
 		}
 
 		r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
 		if (r) {
-			DMERR("dm_pool_get_free_block_count returned %d", r);
+			DMERR("%s: dm_pool_get_free_block_count returned %d",
+			      dm_device_name(pool->pool_md), r);
 			goto err;
 		}
 
 		r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
 		if (r) {
-			DMERR("dm_pool_get_data_dev_size returned %d", r);
+			DMERR("%s: dm_pool_get_data_dev_size returned %d",
+			      dm_device_name(pool->pool_md), r);
 			goto err;
 		}
 
 		r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
 		if (r) {
-			DMERR("dm_pool_get_metadata_snap returned %d", r);
+			DMERR("%s: dm_pool_get_metadata_snap returned %d",
+			      dm_device_name(pool->pool_md), r);
 			goto err;
 		}
 
···
 {
 	struct pool_c *pt = ti->private;
 	struct pool *pool = pt->pool;
+	uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
 
-	blk_limits_io_min(limits, 0);
-	blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
+	/*
+	 * If the system-determined stacked limits are compatible with the
+	 * pool's blocksize (io_opt is a factor) do not override them.
+	 */
+	if (io_opt_sectors < pool->sectors_per_block ||
+	    do_div(io_opt_sectors, pool->sectors_per_block)) {
+		blk_limits_io_min(limits, 0);
+		blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
+	}
 
 	/*
 	 * pt->adjusted_pf is a staging area for the actual features to use.
···
 	.name = "thin-pool",
 	.features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
 		    DM_TARGET_IMMUTABLE,
-	.version = {1, 8, 0},
+	.version = {1, 9, 0},
 	.module = THIS_MODULE,
 	.ctr = pool_ctr,
 	.dtr = pool_dtr,
···
 
 static struct target_type thin_target = {
 	.name = "thin",
-	.version = {1, 8, 0},
+	.version = {1, 9, 0},
 	.module	= THIS_MODULE,
 	.ctr = thin_ctr,
 	.dtr = thin_dtr,

drivers/md/dm.c

···
 	struct bio *bio;
 	unsigned long start_time;
 	spinlock_t endio_lock;
+	struct dm_stats_aux stats_aux;
 };
 
 /*
···
 
 	/* zero-length flush that will be cloned and submitted to targets */
 	struct bio flush_bio;
+
+	struct dm_stats stats;
 };
 
 /*
···
 	dm_io_init,
 	dm_kcopyd_init,
 	dm_interface_init,
+	dm_statistics_init,
 };
 
 static void (*_exits[])(void) = {
···
 	dm_io_exit,
 	dm_kcopyd_exit,
 	dm_interface_exit,
+	dm_statistics_exit,
 };
 
 static int __init dm_init(void)
···
 	return r;
 }
 
+sector_t dm_get_size(struct mapped_device *md)
+{
+	return get_capacity(md->disk);
+}
+
+struct dm_stats *dm_get_stats(struct mapped_device *md)
+{
+	return &md->stats;
+}
+
 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 {
 	struct mapped_device *md = bdev->bd_disk->private_data;
···
 static void start_io_acct(struct dm_io *io)
 {
 	struct mapped_device *md = io->md;
+	struct bio *bio = io->bio;
 	int cpu;
-	int rw = bio_data_dir(io->bio);
+	int rw = bio_data_dir(bio);
 
 	io->start_time = jiffies;
 
···
 	part_stat_unlock();
 	atomic_set(&dm_disk(md)->part0.in_flight[rw],
 		atomic_inc_return(&md->pending[rw]));
+
+	if (unlikely(dm_stats_used(&md->stats)))
+		dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_sector,
+				    bio_sectors(bio), false, 0, &io->stats_aux);
 }
 
 static void end_io_acct(struct dm_io *io)
···
 	part_round_stats(cpu, &dm_disk(md)->part0);
 	part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
 	part_stat_unlock();
+
+	if (unlikely(dm_stats_used(&md->stats)))
+		dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_sector,
+				    bio_sectors(bio), true, duration, &io->stats_aux);
 
 	/*
 	 * After this is decremented the bio must not be touched if it is
···
 	return;
 }
 
-static int dm_request_based(struct mapped_device *md)
+int dm_request_based(struct mapped_device *md)
 {
 	return blk_queue_stackable(md->queue);
 }
···
 	add_disk(md->disk);
 	format_dev_t(md->name, MKDEV(_major, minor));
 
-	md->wq = alloc_workqueue("kdmflush",
-				 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
+	md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
 	if (!md->wq)
 		goto bad_thread;
 
···
 	bio_init(&md->flush_bio);
 	md->flush_bio.bi_bdev = md->bdev;
 	md->flush_bio.bi_rw = WRITE_FLUSH;
+
+	dm_stats_init(&md->stats);
 
 	/* Populate the mapping, nobody knows we exist yet */
 	spin_lock(&_minor_lock);
···
 
 	put_disk(md->disk);
 	blk_cleanup_queue(md->queue);
+	dm_stats_cleanup(&md->stats);
 	module_put(THIS_MODULE);
 	kfree(md);
 }
···
 	/*
 	 * Wipe any geometry if the size of the table changed.
 	 */
-	if (size != get_capacity(md->disk))
+	if (size != dm_get_size(md))
 		memset(&md->geometry, 0, sizeof(md->geometry));
 
 	__set_size(md, size);
···
 
 void dm_set_md_type(struct mapped_device *md, unsigned type)
 {
+	BUG_ON(!mutex_is_locked(&md->type_lock));
 	md->type = type;
 }
 
 unsigned dm_get_md_type(struct mapped_device *md)
 {
+	BUG_ON(!mutex_is_locked(&md->type_lock));
 	return md->type;
 }
 
···
 	mutex_unlock(&md->suspend_lock);
 
 	return r;
+}
+
+/*
+ * Internal suspend/resume works like userspace-driven suspend. It waits
+ * until all bios finish and prevents issuing new bios to the target drivers.
+ * It may be used only from the kernel.
+ *
+ * Internal suspend holds md->suspend_lock, which prevents interaction with
+ * userspace-driven suspend.
+ */
+
+void dm_internal_suspend(struct mapped_device *md)
+{
+	mutex_lock(&md->suspend_lock);
+	if (dm_suspended_md(md))
+		return;
+
+	set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
+	synchronize_srcu(&md->io_barrier);
+	flush_workqueue(md->wq);
+	dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
+}
+
+void dm_internal_resume(struct mapped_device *md)
+{
+	if (dm_suspended_md(md))
+		goto done;
+
+	dm_queue_flush(md);
+
+done:
+	mutex_unlock(&md->suspend_lock);
 }
 
 /*-----------------------------------------------------------------

drivers/md/dm.h

···
 #include <linux/blkdev.h>
 #include <linux/hdreg.h>
 
+#include "dm-stats.h"
+
 /*
  * Suspend feature flags
  */
···
 #define dm_target_is_valid(t) ((t)->table)
 
 /*
+ * To check whether the target type is bio-based or not (request-based).
+ */
+#define dm_target_bio_based(t) ((t)->type->map != NULL)
+
+/*
  * To check whether the target type is request-based or not (bio-based).
  */
 #define dm_target_request_based(t) ((t)->type->map_rq != NULL)
+
+/*
+ * To check whether the target type is a hybrid (capable of being
+ * either request-based or bio-based).
+ */
+#define dm_target_hybrid(t) (dm_target_bio_based(t) && dm_target_request_based(t))
 
 /*-----------------------------------------------------------------
  * A registry of target types.
···
 void dm_destroy_immediate(struct mapped_device *md);
 int dm_open_count(struct mapped_device *md);
 int dm_lock_for_deletion(struct mapped_device *md);
+int dm_request_based(struct mapped_device *md);
+sector_t dm_get_size(struct mapped_device *md);
+struct dm_stats *dm_get_stats(struct mapped_device *md);
 
 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
 		      unsigned cookie);
+
+void dm_internal_suspend(struct mapped_device *md);
+void dm_internal_resume(struct mapped_device *md);
 
 int dm_io_init(void);
 void dm_io_exit(void);
···
  */
 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size);
 void dm_free_md_mempools(struct dm_md_mempools *pools);
+
+/*
+ * Helpers that are used by DM core
+ */
+static inline bool dm_message_test_buffer_overflow(char *result, unsigned maxlen)
+{
+	return !maxlen || strlen(result) + 1 >= maxlen;
+}
 
 #endif

drivers/md/persistent-data/dm-block-manager.c

···
 }
 EXPORT_SYMBOL_GPL(dm_bm_flush_and_unlock);
 
+void dm_bm_prefetch(struct dm_block_manager *bm, dm_block_t b)
+{
+	dm_bufio_prefetch(bm->bufio, b, 1);
+}
+
 void dm_bm_set_read_only(struct dm_block_manager *bm)
 {
 	bm->read_only = true;

drivers/md/persistent-data/dm-block-manager.h

···
 int dm_bm_flush_and_unlock(struct dm_block_manager *bm,
 			   struct dm_block *superblock);
 
+ /*
+  * Request data be prefetched into the cache.
+  */
+void dm_bm_prefetch(struct dm_block_manager *bm, dm_block_t b);
+
 /*
  * Switches the bm to a read only mode.  Once read-only mode
  * has been entered the following functions will return -EPERM.

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

···
 };
 
 struct del_stack {
+	struct dm_btree_info *info;
 	struct dm_transaction_manager *tm;
 	int top;
 	struct frame spine[MAX_SPINE_DEPTH];
···
 static int unprocessed_frames(struct del_stack *s)
 {
 	return s->top >= 0;
+}
+
+static void prefetch_children(struct del_stack *s, struct frame *f)
+{
+	unsigned i;
+	struct dm_block_manager *bm = dm_tm_get_bm(s->tm);
+
+	for (i = 0; i < f->nr_children; i++)
+		dm_bm_prefetch(bm, value64(f->n, i));
+}
+
+static bool is_internal_level(struct dm_btree_info *info, struct frame *f)
+{
+	return f->level < (info->levels - 1);
 }
 
 static int push_frame(struct del_stack *s, dm_block_t b, unsigned level)
···
 		dm_tm_dec(s->tm, b);
 
 	else {
+		uint32_t flags;
 		struct frame *f = s->spine + ++s->top;
 
 		r = dm_tm_read_lock(s->tm, b, &btree_node_validator, &f->b);
···
 		f->level = level;
 		f->nr_children = le32_to_cpu(f->n->header.nr_entries);
 		f->current_child = 0;
+
+		flags = le32_to_cpu(f->n->header.flags);
+		if (flags & INTERNAL_NODE || is_internal_level(s->info, f))
+			prefetch_children(s, f);
 	}
 
 	return 0;
···
 	dm_tm_unlock(s->tm, f->b);
 }
 
-static bool is_internal_level(struct dm_btree_info *info, struct frame *f)
-{
-	return f->level < (info->levels - 1);
-}
-
 int dm_btree_del(struct dm_btree_info *info, dm_block_t root)
 {
 	int r;
···
 	s = kmalloc(sizeof(*s), GFP_KERNEL);
 	if (!s)
 		return -ENOMEM;
+	s->info = info;
 	s->tm = info->tm;
 	s->top = -1;
 
···
 					info->value_type.dec(info->value_type.context,
 							     value_ptr(f->n, i));
 			}
-			f->current_child = f->nr_children;
+			pop_frame(s);
 		}
 	}
 

drivers/md/persistent-data/dm-space-map-common.c

···
 	return dm_tm_unlock(ll->tm, blk);
 }
 
-int sm_ll_lookup(struct ll_disk *ll, dm_block_t b, uint32_t *result)
+static int sm_ll_lookup_big_ref_count(struct ll_disk *ll, dm_block_t b,
+				      uint32_t *result)
 {
 	__le32 le_rc;
-	int r = sm_ll_lookup_bitmap(ll, b, result);
-
-	if (r)
-		return r;
-
-	if (*result != 3)
-		return r;
+	int r;
 
 	r = dm_btree_lookup(&ll->ref_count_info, ll->ref_count_root, &b, &le_rc);
 	if (r < 0)
···
 	*result = le32_to_cpu(le_rc);
 
 	return r;
+}
+
+int sm_ll_lookup(struct ll_disk *ll, dm_block_t b, uint32_t *result)
+{
+	int r = sm_ll_lookup_bitmap(ll, b, result);
+
+	if (r)
+		return r;
+
+	if (*result != 3)
+		return r;
+
+	return sm_ll_lookup_big_ref_count(ll, b, result);
 }
 
 int sm_ll_find_free_block(struct ll_disk *ll, dm_block_t begin,
···
 	return -ENOSPC;
 }
 
-int sm_ll_insert(struct ll_disk *ll, dm_block_t b,
-		 uint32_t ref_count, enum allocation_event *ev)
+static int sm_ll_mutate(struct ll_disk *ll, dm_block_t b,
+			uint32_t (*mutator)(void *context, uint32_t old),
+			void *context, enum allocation_event *ev)
 {
 	int r;
-	uint32_t bit, old;
+	uint32_t bit, old, ref_count;
 	struct dm_block *nb;
 	dm_block_t index = b;
 	struct disk_index_entry ie_disk;
···
 
 	bm_le = dm_bitmap_data(nb);
 	old = sm_lookup_bitmap(bm_le, bit);
+
+	if (old > 2) {
+		r = sm_ll_lookup_big_ref_count(ll, b, &old);
+		if (r < 0)
+			return r;
+	}
+
+	ref_count = mutator(context, old);
 
 	if (ref_count <= 2) {
 		sm_set_bitmap(bm_le, bit, ref_count);
···
 	return ll->save_ie(ll, index, &ie_disk);
 }
 
+static uint32_t set_ref_count(void *context, uint32_t old)
+{
+	return *((uint32_t *) context);
+}
+
+int sm_ll_insert(struct ll_disk *ll, dm_block_t b,
+		 uint32_t ref_count, enum allocation_event *ev)
+{
+	return sm_ll_mutate(ll, b, set_ref_count, &ref_count, ev);
+}
+
+static uint32_t inc_ref_count(void *context, uint32_t old)
+{
+	return old + 1;
+}
+
 int sm_ll_inc(struct ll_disk *ll, dm_block_t b, enum allocation_event *ev)
 {
-	int r;
-	uint32_t rc;
+	return sm_ll_mutate(ll, b, inc_ref_count, NULL, ev);
+}
 
-	r = sm_ll_lookup(ll, b, &rc);
-	if (r)
-		return r;
-
-	return sm_ll_insert(ll, b, rc + 1, ev);
+static uint32_t dec_ref_count(void *context, uint32_t old)
+{
+	return old - 1;
 }
 
 int sm_ll_dec(struct ll_disk *ll, dm_block_t b, enum allocation_event *ev)
 {
-	int r;
-	uint32_t rc;
-
-	r = sm_ll_lookup(ll, b, &rc);
-	if (r)
-		return r;
-
-	if (!rc)
-		return -EINVAL;
-
-	return sm_ll_insert(ll, b, rc - 1, ev);
+	return sm_ll_mutate(ll, b, dec_ref_count, NULL, ev);
 }
 
 int sm_ll_commit(struct ll_disk *ll)

include/linux/device-mapper.h

···
 
 #include <linux/bio.h>
 #include <linux/blkdev.h>
+#include <linux/math64.h>
 #include <linux/ratelimit.h>
 
 struct dm_dev;
···
 #define DM_MAPIO_SUBMITTED	0
 #define DM_MAPIO_REMAPPED	1
 #define DM_MAPIO_REQUEUE	DM_ENDIO_REQUEUE
+
+#define dm_sector_div64(x, y)( \
+{ \
+	u64 _res; \
+	(x) = div64_u64_rem(x, y, &_res); \
+	_res; \
+} \
+)
 
 /*
  * Ceiling(n / sz)

include/linux/math64.h

···
 }
 
 /**
+ * div64_u64_rem - unsigned 64bit divide with 64bit divisor and remainder
+ */
+static inline u64 div64_u64_rem(u64 dividend, u64 divisor, u64 *remainder)
+{
+	*remainder = dividend % divisor;
+	return dividend / divisor;
+}
+
+/**
  * div64_u64 - unsigned 64bit divide with 64bit divisor
  */
 static inline u64 div64_u64(u64 dividend, u64 divisor)
···
 
 #ifndef div_s64_rem
 extern s64 div_s64_rem(s64 dividend, s32 divisor, s32 *remainder);
+#endif
+
+#ifndef div64_u64_rem
+extern u64 div64_u64_rem(u64 dividend, u64 divisor, u64 *remainder);
 #endif
 
 #ifndef div64_u64

include/uapi/linux/dm-ioctl.h

···
 #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	25
+#define DM_VERSION_MINOR	26
 #define DM_VERSION_PATCHLEVEL	0
-#define DM_VERSION_EXTRA	"-ioctl (2013-06-26)"
+#define DM_VERSION_EXTRA	"-ioctl (2013-08-15)"
 
 /* Status bits */
 #define DM_READONLY_FLAG	(1 << 0) /* In/Out */

lib/div64.c

···
 #endif
 
 /**
+ * div64_u64_rem - unsigned 64bit divide with 64bit divisor and remainder
+ * @dividend:	64bit dividend
+ * @divisor:	64bit divisor
+ * @remainder:  64bit remainder
+ *
+ * This implementation is a comparable to algorithm used by div64_u64.
+ * But this operation, which includes math for calculating the remainder,
+ * is kept distinct to avoid slowing down the div64_u64 operation on 32bit
+ * systems.
+ */
+#ifndef div64_u64_rem
+u64 div64_u64_rem(u64 dividend, u64 divisor, u64 *remainder)
+{
+	u32 high = divisor >> 32;
+	u64 quot;
+
+	if (high == 0) {
+		u32 rem32;
+		quot = div_u64_rem(dividend, divisor, &rem32);
+		*remainder = rem32;
+	} else {
+		int n = 1 + fls(high);
+		quot = div_u64(dividend >> n, divisor >> n);
+
+		if (quot != 0)
+			quot--;
+
+		*remainder = dividend - quot * divisor;
+		if (*remainder >= divisor) {
+			quot++;
+			*remainder -= divisor;
+		}
+	}
+
+	return quot;
+}
+EXPORT_SYMBOL(div64_u64_rem);
+#endif
+
+/**
  * div64_u64 - unsigned 64bit divide with 64bit divisor
  * @dividend:	64bit dividend
  * @divisor:	64bit divisor