dm: add thin provisioning target

Initial EXPERIMENTAL implementation of device-mapper thin provisioning
with snapshot support.  The 'thin' target is used to create instances of
the virtual devices that are hosted in the 'thin-pool' target.  The
thin-pool target provides data sharing among devices.  This sharing is
made possible using the persistent-data library in the previous patch.

The main highlight of this implementation, compared to the previous
implementation of snapshots, is that it allows many virtual devices to
be stored on the same data volume, simplifying administration and
allowing sharing of data between volumes (thus reducing disk usage).

Another big feature is support for arbitrary depth of recursive
snapshots (snapshots of snapshots of snapshots ...).  The previous
implementation of snapshots did this by chaining together lookup tables,
and so performance was O(depth).  This new implementation uses a single
data structure so we don't get this degradation with depth.

For further information and examples of how to use this, please read
Documentation/device-mapper/thin-provisioning.txt

Signed-off-by: Joe Thornber <thornber@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>

6 files changed, 4291 insertions, 0 deletions

diff --git a/Documentation/device-mapper/thin-provisioning.txt b/Documentation/device-mapper/thin-provisioning.txt
new file mode 100644
index 00000000000..801d9d1cf82
--- /dev/null
+++ b/Documentation/device-mapper/thin-provisioning.txt
@@ -0,0 +1,285 @@
+Introduction
+============
+
+This document descibes a collection of device-mapper targets that
+between them implement thin-provisioning and snapshots.
+
+The main highlight of this implementation, compared to the previous
+implementation of snapshots, is that it allows many virtual devices to
+be stored on the same data volume.  This simplifies administration and
+allows the sharing of data between volumes, thus reducing disk usage.
+
+Another significant feature is support for an arbitrary depth of
+recursive snapshots (snapshots of snapshots of snapshots ...).  The
+previous implementation of snapshots did this by chaining together
+lookup tables, and so performance was O(depth).  This new
+implementation uses a single data structure to avoid this degradation
+with depth.  Fragmentation may still be an issue, however, in some
+scenarios.
+
+Metadata is stored on a separate device from data, giving the
+administrator some freedom, for example to:
+
+- Improve metadata resilience by storing metadata on a mirrored volume
+  but data on a non-mirrored one.
+
+- Improve performance by storing the metadata on SSD.
+
+Status
+======
+
+These targets are very much still in the EXPERIMENTAL state.  Please
+do not yet rely on them in production.  But do experiment and offer us
+feedback.  Different use cases will have different performance
+characteristics, for example due to fragmentation of the data volume.
+
+If you find this software is not performing as expected please mail
+dm-devel@redhat.com with details and we'll try our best to improve
+things for you.
+
+Userspace tools for checking and repairing the metadata are under
+development.
+
+Cookbook
+========
+
+This section describes some quick recipes for using thin provisioning.
+They use the dmsetup program to control the device-mapper driver
+directly.  End users will be advised to use a higher-level volume
+manager such as LVM2 once support has been added.
+
+Pool device
+-----------
+
+The pool device ties together the metadata volume and the data volume.
+It maps I/O linearly to the data volume and updates the metadata via
+two mechanisms:
+
+- Function calls from the thin targets
+
+- Device-mapper 'messages' from userspace which control the creation of new
+  virtual devices amongst other things.
+
+Setting up a fresh pool device
+------------------------------
+
+Setting up a pool device requires a valid metadata device, and a
+data device.  If you do not have an existing metadata device you can
+make one by zeroing the first 4k to indicate empty metadata.
+
+    dd if=/dev/zero of=$metadata_dev bs=4096 count=1
+
+The amount of metadata you need will vary according to how many blocks
+are shared between thin devices (i.e. through snapshots).  If you have
+less sharing than average you'll need a larger-than-average metadata device.
+
+As a guide, we suggest you calculate the number of bytes to use in the
+metadata device as 48 * $data_dev_size / $data_block_size but round it up
+to 2MB if the answer is smaller.  The largest size supported is 16GB.
+
+If you're creating large numbers of snapshots which are recording large
+amounts of change, you may need find you need to increase this.
+
+Reloading a pool table
+----------------------
+
+You may reload a pool's table, indeed this is how the pool is resized
+if it runs out of space.  (N.B. While specifying a different metadata
+device when reloading is not forbidden at the moment, things will go
+wrong if it does not route I/O to exactly the same on-disk location as
+previously.)
+
+Using an existing pool device
+-----------------------------
+
+    dmsetup create pool \
+	--table "0 20971520 thin-pool $metadata_dev $data_dev \
+		 $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.
+
+$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
+will be triggered which a userspace daemon should catch allowing it to
+extend the pool device.  Only one such event will be sent.
+Resuming a device with a new table itself triggers an event so the
+userspace daemon can use this to detect a situation where a new table
+already exceeds the threshold.
+
+Thin provisioning
+-----------------
+
+i) Creating a new thinly-provisioned volume.
+
+  To create a new thinly- provisioned volume you must send a message to an
+  active pool device, /dev/mapper/pool in this example.
+
+    dmsetup message /dev/mapper/pool 0 "create_thin 0"
+
+  Here '0' is an identifier for the volume, a 24-bit number.  It's up
+  to the caller to allocate and manage these identifiers.  If the
+  identifier is already in use, the message will fail with -EEXIST.
+
+ii) Using a thinly-provisioned volume.
+
+  Thinly-provisioned volumes are activated using the 'thin' target:
+
+    dmsetup create thin --table "0 2097152 thin /dev/mapper/pool 0"
+
+  The last parameter is the identifier for the thinp device.
+
+Internal snapshots
+------------------
+
+i) Creating an internal snapshot.
+
+  Snapshots are created with another message to the pool.
+
+  N.B.  If the origin device that you wish to snapshot is active, you
+  must suspend it before creating the snapshot to avoid corruption.
+  This is NOT enforced at the moment, so please be careful!
+
+    dmsetup suspend /dev/mapper/thin
+    dmsetup message /dev/mapper/pool 0 "create_snap 1 0"
+    dmsetup resume /dev/mapper/thin
+
+  Here '1' is the identifier for the volume, a 24-bit number.  '0' is the
+  identifier for the origin device.
+
+ii) Using an internal snapshot.
+
+  Once created, the user doesn't have to worry about any connection
+  between the origin and the snapshot.  Indeed the snapshot is no
+  different from any other thinly-provisioned device and can be
+  snapshotted itself via the same method.  It's perfectly legal to
+  have only one of them active, and there's no ordering requirement on
+  activating or removing them both.  (This differs from conventional
+  device-mapper snapshots.)
+
+  Activate it exactly the same way as any other thinly-provisioned volume:
+
+    dmsetup create snap --table "0 2097152 thin /dev/mapper/pool 1"
+
+Deactivation
+------------
+
+All devices using a pool must be deactivated before the pool itself
+can be.
+
+    dmsetup remove thin
+    dmsetup remove snap
+    dmsetup remove pool
+
+Reference
+=========
+
+'thin-pool' target
+------------------
+
+i) Constructor
+
+    thin-pool <metadata dev> <data dev> <data block size (sectors)> \
+	      <low water mark (blocks)> [<number of feature args> [<arg>]*]
+
+    Optional feature arguments:
+    - 'skip_block_zeroing': skips the zeroing of newly-provisioned blocks.
+
+    Data block size must be between 64KB (128 sectors) and 1GB
+    (2097152 sectors) inclusive.
+
+
+ii) Status
+
+    <transaction id> <used metadata blocks>/<total metadata blocks>
+    <used data blocks>/<total data blocks> <held metadata root>
+
+
+    transaction id:
+	A 64-bit number used by userspace to help synchronise with metadata
+	from volume managers.
+
+    used data blocks / total data blocks
+	If the number of free blocks drops below the pool's low water mark a
+	dm event will be sent to userspace.  This event is edge-triggered and
+	it will occur only once after each resume so volume manager writers
+	should register for the event and then check the target's status.
+
+    held metadata root:
+	The location, in sectors, of the metadata root that has been
+	'held' for userspace read access.  '-' indicates there is no
+	held root.  This feature is not yet implemented so '-' is
+	always returned.
+
+iii) Messages
+
+    create_thin <dev id>
+
+	Create a new thinly-provisioned device.
+	<dev id> is an arbitrary unique 24-bit identifier chosen by
+	the caller.
+
+    create_snap <dev id> <origin id>
+
+	Create a new snapshot of another thinly-provisioned device.
+	<dev id> is an arbitrary unique 24-bit identifier chosen by
+	the caller.
+	<origin id> is the identifier of the thinly-provisioned device
+	of which the new device will be a snapshot.
+
+    delete <dev id>
+
+	Deletes a thin device.  Irreversible.
+
+    trim <dev id> <new size in sectors>
+
+	Delete mappings from the end of a thin device.  Irreversible.
+	You might want to use this if you're reducing the size of
+	your thinly-provisioned device.  In many cases, due to the
+	sharing of blocks between devices, it is not possible to
+	determine in advance how much space 'trim' will release.  (In
+	future a userspace tool might be able to perform this
+	calculation.)
+
+    set_transaction_id <current id> <new id>
+
+	Userland volume managers, such as LVM, need a way to
+	synchronise their external metadata with the internal metadata of the
+	pool target.  The thin-pool target offers to store an
+	arbitrary 64-bit transaction id and return it on the target's
+	status line.  To avoid races you must provide what you think
+	the current transaction id is when you change it with this
+	compare-and-swap message.
+
+'thin' target
+-------------
+
+i) Constructor
+
+    thin <pool dev> <dev id>
+
+    pool dev:
+	the thin-pool device, e.g. /dev/mapper/my_pool or 253:0
+
+    dev id:
+	the internal device identifier of the device to be
+	activated.
+
+The pool doesn't store any size against the thin devices.  If you
+load a thin target that is smaller than you've been using previously,
+then you'll have no access to blocks mapped beyond the end.  If you
+load a target that is bigger than before, then extra blocks will be
+provisioned as and when needed.
+
+If you wish to reduce the size of your thin device and potentially
+regain some space then send the 'trim' message to the pool.
+
+ii) Status
+
+     <nr mapped sectors> <highest mapped sector>
diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig
index b1a92149704..faa4741df6d 100644
--- a/drivers/md/Kconfig
+++ b/drivers/md/Kconfig
@@ -216,6 +216,8 @@ config DM_BUFIO
 	 as a cache, holding recently-read blocks in memory and performing
 	 delayed writes.
 
+source "drivers/md/persistent-data/Kconfig"
+
 config DM_CRYPT
 	tristate "Crypt target support"
 	depends on BLK_DEV_DM
@@ -241,6 +243,32 @@ config DM_SNAPSHOT
        ---help---
          Allow volume managers to take writable snapshots of a device.
 
+config DM_THIN_PROVISIONING
+       tristate "Thin provisioning target (EXPERIMENTAL)"
+       depends on BLK_DEV_DM && EXPERIMENTAL
+       select DM_PERSISTENT_DATA
+       ---help---
+         Provides thin provisioning and snapshots that share a data store.
+
+config DM_DEBUG_BLOCK_STACK_TRACING
+	boolean "Keep stack trace of thin provisioning block lock holders"
+	depends on STACKTRACE_SUPPORT && DM_THIN_PROVISIONING
+	select STACKTRACE
+	---help---
+	  Enable this for messages that may help debug problems with the
+	  block manager locking used by thin provisioning.
+
+	  If unsure, say N.
+
+config DM_DEBUG_SPACE_MAPS
+	boolean "Extra validation for thin provisioning space maps"
+	depends on DM_THIN_PROVISIONING
+	---help---
+	  Enable this for messages that may help debug problems with the
+	  space maps used by thin provisioning.
+
+          If unsure, say N.
+
 config DM_MIRROR
        tristate "Mirror target"
        depends on BLK_DEV_DM
diff --git a/drivers/md/Makefile b/drivers/md/Makefile
index 56661c4272f..046860c7a16 100644
--- a/drivers/md/Makefile
+++ b/drivers/md/Makefile
@@ -10,6 +10,7 @@ dm-snapshot-y	+= dm-snap.o dm-exception-store.o dm-snap-transient.o \
 dm-mirror-y	+= dm-raid1.o
 dm-log-userspace-y \
 		+= dm-log-userspace-base.o dm-log-userspace-transfer.o
+dm-thin-pool-y	+= dm-thin.o dm-thin-metadata.o
 md-mod-y	+= md.o bitmap.o
 raid456-y	+= raid5.o
 
@@ -35,10 +36,12 @@ obj-$(CONFIG_DM_MULTIPATH)	+= dm-multipath.o dm-round-robin.o
 obj-$(CONFIG_DM_MULTIPATH_QL)	+= dm-queue-length.o
 obj-$(CONFIG_DM_MULTIPATH_ST)	+= dm-service-time.o
 obj-$(CONFIG_DM_SNAPSHOT)	+= dm-snapshot.o
+obj-$(CONFIG_DM_PERSISTENT_DATA)	+= persistent-data/
 obj-$(CONFIG_DM_MIRROR)		+= dm-mirror.o dm-log.o dm-region-hash.o
 obj-$(CONFIG_DM_LOG_USERSPACE)	+= dm-log-userspace.o
 obj-$(CONFIG_DM_ZERO)		+= dm-zero.o
 obj-$(CONFIG_DM_RAID)	+= dm-raid.o
+obj-$(CONFIG_DM_THIN_PROVISIONING)	+= dm-thin-pool.o
 
 ifeq ($(CONFIG_DM_UEVENT),y)
 dm-mod-objs			+= dm-uevent.o
diff --git a/drivers/md/dm-thin-metadata.c b/drivers/md/dm-thin-metadata.c
new file mode 100644
index 00000000000..59c4f0446ff
--- /dev/null
+++ b/drivers/md/dm-thin-metadata.c
@@ -0,0 +1,1391 @@
+/*
+ * Copyright (C) 2011 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+
+#include "dm-thin-metadata.h"
+#include "persistent-data/dm-btree.h"
+#include "persistent-data/dm-space-map.h"
+#include "persistent-data/dm-space-map-disk.h"
+#include "persistent-data/dm-transaction-manager.h"
+
+#include <linux/list.h>
+#include <linux/device-mapper.h>
+#include <linux/workqueue.h>
+
+/*--------------------------------------------------------------------------
+ * As far as the metadata goes, there is:
+ *
+ * - A superblock in block zero, taking up fewer than 512 bytes for
+ *   atomic writes.
+ *
+ * - A space map managing the metadata blocks.
+ *
+ * - A space map managing the data blocks.
+ *
+ * - A btree mapping our internal thin dev ids onto struct disk_device_details.
+ *
+ * - A hierarchical btree, with 2 levels which effectively maps (thin
+ *   dev id, virtual block) -> block_time.  Block time is a 64-bit
+ *   field holding the time in the low 24 bits, and block in the top 48
+ *   bits.
+ *
+ * BTrees consist solely of btree_nodes, that fill a block.  Some are
+ * internal nodes, as such their values are a __le64 pointing to other
+ * nodes.  Leaf nodes can store data of any reasonable size (ie. much
+ * smaller than the block size).  The nodes consist of the header,
+ * followed by an array of keys, followed by an array of values.  We have
+ * to binary search on the keys so they're all held together to help the
+ * cpu cache.
+ *
+ * Space maps have 2 btrees:
+ *
+ * - One maps a uint64_t onto a struct index_entry.  Which points to a
+ *   bitmap block, and has some details about how many free entries there
+ *   are etc.
+ *
+ * - The bitmap blocks have a header (for the checksum).  Then the rest
+ *   of the block is pairs of bits.  With the meaning being:
+ *
+ *   0 - ref count is 0
+ *   1 - ref count is 1
+ *   2 - ref count is 2
+ *   3 - ref count is higher than 2
+ *
+ * - If the count is higher than 2 then the ref count is entered in a
+ *   second btree that directly maps the block_address to a uint32_t ref
+ *   count.
+ *
+ * The space map metadata variant doesn't have a bitmaps btree.  Instead
+ * it has one single blocks worth of index_entries.  This avoids
+ * recursive issues with the bitmap btree needing to allocate space in
+ * order to insert.  With a small data block size such as 64k the
+ * metadata support data devices that are hundreds of terrabytes.
+ *
+ * The space maps allocate space linearly from front to back.  Space that
+ * is freed in a transaction is never recycled within that transaction.
+ * To try and avoid fragmenting _free_ space the allocator always goes
+ * back and fills in gaps.
+ *
+ * All metadata io is in THIN_METADATA_BLOCK_SIZE sized/aligned chunks
+ * from the block manager.
+ *--------------------------------------------------------------------------*/
+
+#define DM_MSG_PREFIX   "thin metadata"
+
+#define THIN_SUPERBLOCK_MAGIC 27022010
+#define THIN_SUPERBLOCK_LOCATION 0
+#define THIN_VERSION 1
+#define THIN_METADATA_CACHE_SIZE 64
+#define SECTOR_TO_BLOCK_SHIFT 3
+
+/* This should be plenty */
+#define SPACE_MAP_ROOT_SIZE 128
+
+/*
+ * Little endian on-disk superblock and device details.
+ */
+struct thin_disk_superblock {
+	__le32 csum;	/* Checksum of superblock except for this field. */
+	__le32 flags;
+	__le64 blocknr;	/* This block number, dm_block_t. */
+
+	__u8 uuid[16];
+	__le64 magic;
+	__le32 version;
+	__le32 time;
+
+	__le64 trans_id;
+
+	/*
+	 * Root held by userspace transactions.
+	 */
+	__le64 held_root;
+
+	__u8 data_space_map_root[SPACE_MAP_ROOT_SIZE];
+	__u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
+
+	/*
+	 * 2-level btree mapping (dev_id, (dev block, time)) -> data block
+	 */
+	__le64 data_mapping_root;
+
+	/*
+	 * Device detail root mapping dev_id -> device_details
+	 */
+	__le64 device_details_root;
+
+	__le32 data_block_size;		/* In 512-byte sectors. */
+
+	__le32 metadata_block_size;	/* In 512-byte sectors. */
+	__le64 metadata_nr_blocks;
+
+	__le32 compat_flags;
+	__le32 compat_ro_flags;
+	__le32 incompat_flags;
+} __packed;
+
+struct disk_device_details {
+	__le64 mapped_blocks;
+	__le64 transaction_id;		/* When created. */
+	__le32 creation_time;
+	__le32 snapshotted_time;
+} __packed;
+
+struct dm_pool_metadata {
+	struct hlist_node hash;
+
+	struct block_device *bdev;
+	struct dm_block_manager *bm;
+	struct dm_space_map *metadata_sm;
+	struct dm_space_map *data_sm;
+	struct dm_transaction_manager *tm;
+	struct dm_transaction_manager *nb_tm;
+
+	/*
+	 * Two-level btree.
+	 * First level holds thin_dev_t.
+	 * Second level holds mappings.
+	 */
+	struct dm_btree_info info;
+
+	/*
+	 * Non-blocking version of the above.
+	 */
+	struct dm_btree_info nb_info;
+
+	/*
+	 * Just the top level for deleting whole devices.
+	 */
+	struct dm_btree_info tl_info;
+
+	/*
+	 * Just the bottom level for creating new devices.
+	 */
+	struct dm_btree_info bl_info;
+
+	/*
+	 * Describes the device details btree.
+	 */
+	struct dm_btree_info details_info;
+
+	struct rw_semaphore root_lock;
+	uint32_t time;
+	int need_commit;
+	dm_block_t root;
+	dm_block_t details_root;
+	struct list_head thin_devices;
+	uint64_t trans_id;
+	unsigned long flags;
+	sector_t data_block_size;
+};
+
+struct dm_thin_device {
+	struct list_head list;
+	struct dm_pool_metadata *pmd;
+	dm_thin_id id;
+
+	int open_count;
+	int changed;
+	uint64_t mapped_blocks;
+	uint64_t transaction_id;
+	uint32_t creation_time;
+	uint32_t snapshotted_time;
+};
+
+/*----------------------------------------------------------------
+ * superblock validator
+ *--------------------------------------------------------------*/
+
+#define SUPERBLOCK_CSUM_XOR 160774
+
+static void sb_prepare_for_write(struct dm_block_validator *v,
+				 struct dm_block *b,
+				 size_t block_size)
+{
+	struct thin_disk_superblock *disk_super = dm_block_data(b);
+
+	disk_super->blocknr = cpu_to_le64(dm_block_location(b));
+	disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
+						      block_size - sizeof(__le32),
+						      SUPERBLOCK_CSUM_XOR));
+}
+
+static int sb_check(struct dm_block_validator *v,
+		    struct dm_block *b,
+		    size_t block_size)
+{
+	struct thin_disk_superblock *disk_super = dm_block_data(b);
+	__le32 csum_le;
+
+	if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
+		DMERR("sb_check failed: blocknr %llu: "
+		      "wanted %llu", le64_to_cpu(disk_super->blocknr),
+		      (unsigned long long)dm_block_location(b));
+		return -ENOTBLK;
+	}
+
+	if (le64_to_cpu(disk_super->magic) != THIN_SUPERBLOCK_MAGIC) {
+		DMERR("sb_check failed: magic %llu: "
+		      "wanted %llu", le64_to_cpu(disk_super->magic),
+		      (unsigned long long)THIN_SUPERBLOCK_MAGIC);
+		return -EILSEQ;
+	}
+
+	csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
+					     block_size - sizeof(__le32),
+					     SUPERBLOCK_CSUM_XOR));
+	if (csum_le != disk_super->csum) {
+		DMERR("sb_check failed: csum %u: wanted %u",
+		      le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
+		return -EILSEQ;
+	}
+
+	return 0;
+}
+
+static struct dm_block_validator sb_validator = {
+	.name = "superblock",
+	.prepare_for_write = sb_prepare_for_write,
+	.check = sb_check
+};
+
+/*----------------------------------------------------------------
+ * Methods for the btree value types
+ *--------------------------------------------------------------*/
+
+static uint64_t pack_block_time(dm_block_t b, uint32_t t)
+{
+	return (b << 24) | t;
+}
+
+static void unpack_block_time(uint64_t v, dm_block_t *b, uint32_t *t)
+{
+	*b = v >> 24;
+	*t = v & ((1 << 24) - 1);
+}
+
+static void data_block_inc(void *context, void *value_le)
+{
+	struct dm_space_map *sm = context;
+	__le64 v_le;
+	uint64_t b;
+	uint32_t t;
+
+	memcpy(&v_le, value_le, sizeof(v_le));
+	unpack_block_time(le64_to_cpu(v_le), &b, &t);
+	dm_sm_inc_block(sm, b);
+}
+
+static void data_block_dec(void *context, void *value_le)
+{
+	struct dm_space_map *sm = context;
+	__le64 v_le;
+	uint64_t b;
+	uint32_t t;
+
+	memcpy(&v_le, value_le, sizeof(v_le));
+	unpack_block_time(le64_to_cpu(v_le), &b, &t);
+	dm_sm_dec_block(sm, b);
+}
+
+static int data_block_equal(void *context, void *value1_le, void *value2_le)
+{
+	__le64 v1_le, v2_le;
+	uint64_t b1, b2;
+	uint32_t t;
+
+	memcpy(&v1_le, value1_le, sizeof(v1_le));
+	memcpy(&v2_le, value2_le, sizeof(v2_le));
+	unpack_block_time(le64_to_cpu(v1_le), &b1, &t);
+	unpack_block_time(le64_to_cpu(v2_le), &b2, &t);
+
+	return b1 == b2;
+}
+
+static void subtree_inc(void *context, void *value)
+{
+	struct dm_btree_info *info = context;
+	__le64 root_le;
+	uint64_t root;
+
+	memcpy(&root_le, value, sizeof(root_le));
+	root = le64_to_cpu(root_le);
+	dm_tm_inc(info->tm, root);
+}
+
+static void subtree_dec(void *context, void *value)
+{
+	struct dm_btree_info *info = context;
+	__le64 root_le;
+	uint64_t root;
+
+	memcpy(&root_le, value, sizeof(root_le));
+	root = le64_to_cpu(root_le);
+	if (dm_btree_del(info, root))
+		DMERR("btree delete failed\n");
+}
+
+static int subtree_equal(void *context, void *value1_le, void *value2_le)
+{
+	__le64 v1_le, v2_le;
+	memcpy(&v1_le, value1_le, sizeof(v1_le));
+	memcpy(&v2_le, value2_le, sizeof(v2_le));
+
+	return v1_le == v2_le;
+}
+
+/*----------------------------------------------------------------*/
+
+static int superblock_all_zeroes(struct dm_block_manager *bm, int *result)
+{
+	int r;
+	unsigned i;
+	struct dm_block *b;
+	__le64 *data_le, zero = cpu_to_le64(0);
+	unsigned block_size = dm_bm_block_size(bm) / sizeof(__le64);
+
+	/*
+	 * We can't use a validator here - it may be all zeroes.
+	 */
+	r = dm_bm_read_lock(bm, THIN_SUPERBLOCK_LOCATION, NULL, &b);
+	if (r)
+		return r;
+
+	data_le = dm_block_data(b);
+	*result = 1;
+	for (i = 0; i < block_size; i++) {
+		if (data_le[i] != zero) {
+			*result = 0;
+			break;
+		}
+	}
+
+	return dm_bm_unlock(b);
+}
+
+static int init_pmd(struct dm_pool_metadata *pmd,
+		    struct dm_block_manager *bm,
+		    dm_block_t nr_blocks, int create)
+{
+	int r;
+	struct dm_space_map *sm, *data_sm;
+	struct dm_transaction_manager *tm;
+	struct dm_block *sblock;
+
+	if (create) {
+		r = dm_tm_create_with_sm(bm, THIN_SUPERBLOCK_LOCATION,
+					 &sb_validator, &tm, &sm, &sblock);
+		if (r < 0) {
+			DMERR("tm_create_with_sm failed");
+			return r;
+		}
+
+		data_sm = dm_sm_disk_create(tm, nr_blocks);
+		if (IS_ERR(data_sm)) {
+			DMERR("sm_disk_create failed");
+			r = PTR_ERR(data_sm);
+			goto bad;
+		}
+	} else {
+		struct thin_disk_superblock *disk_super = NULL;
+		size_t space_map_root_offset =
+			offsetof(struct thin_disk_superblock, metadata_space_map_root);
+
+		r = dm_tm_open_with_sm(bm, THIN_SUPERBLOCK_LOCATION,
+				       &sb_validator, space_map_root_offset,
+				       SPACE_MAP_ROOT_SIZE, &tm, &sm, &sblock);
+		if (r < 0) {
+			DMERR("tm_open_with_sm failed");
+			return r;
+		}
+
+		disk_super = dm_block_data(sblock);
+		data_sm = dm_sm_disk_open(tm, disk_super->data_space_map_root,
+					  sizeof(disk_super->data_space_map_root));
+		if (IS_ERR(data_sm)) {
+			DMERR("sm_disk_open failed");
+			r = PTR_ERR(data_sm);
+			goto bad;
+		}
+	}
+
+
+	r = dm_tm_unlock(tm, sblock);
+	if (r < 0) {
+		DMERR("couldn't unlock superblock");
+		goto bad_data_sm;
+	}
+
+	pmd->bm = bm;
+	pmd->metadata_sm = sm;
+	pmd->data_sm = data_sm;
+	pmd->tm = tm;
+	pmd->nb_tm = dm_tm_create_non_blocking_clone(tm);
+	if (!pmd->nb_tm) {
+		DMERR("could not create clone tm");
+		r = -ENOMEM;
+		goto bad_data_sm;
+	}
+
+	pmd->info.tm = tm;
+	pmd->info.levels = 2;
+	pmd->info.value_type.context = pmd->data_sm;
+	pmd->info.value_type.size = sizeof(__le64);
+	pmd->info.value_type.inc = data_block_inc;
+	pmd->info.value_type.dec = data_block_dec;
+	pmd->info.value_type.equal = data_block_equal;
+
+	memcpy(&pmd->nb_info, &pmd->info, sizeof(pmd->nb_info));
+	pmd->nb_info.tm = pmd->nb_tm;
+
+	pmd->tl_info.tm = tm;
+	pmd->tl_info.levels = 1;
+	pmd->tl_info.value_type.context = &pmd->info;
+	pmd->tl_info.value_type.size = sizeof(__le64);
+	pmd->tl_info.value_type.inc = subtree_inc;
+	pmd->tl_info.value_type.dec = subtree_dec;
+	pmd->tl_info.value_type.equal = subtree_equal;
+
+	pmd->bl_info.tm = tm;
+	pmd->bl_info.levels = 1;
+	pmd->bl_info.value_type.context = pmd->data_sm;
+	pmd->bl_info.value_type.size = sizeof(__le64);
+	pmd->bl_info.value_type.inc = data_block_inc;
+	pmd->bl_info.value_type.dec = data_block_dec;
+	pmd->bl_info.value_type.equal = data_block_equal;
+
+	pmd->details_info.tm = tm;
+	pmd->details_info.levels = 1;
+	pmd->details_info.value_type.context = NULL;
+	pmd->details_info.value_type.size = sizeof(struct disk_device_details);
+	pmd->details_info.value_type.inc = NULL;
+	pmd->details_info.value_type.dec = NULL;
+	pmd->details_info.value_type.equal = NULL;
+
+	pmd->root = 0;
+
+	init_rwsem(&pmd->root_lock);
+	pmd->time = 0;
+	pmd->need_commit = 0;
+	pmd->details_root = 0;
+	pmd->trans_id = 0;
+	pmd->flags = 0;
+	INIT_LIST_HEAD(&pmd->thin_devices);
+
+	return 0;
+
+bad_data_sm:
+	dm_sm_destroy(data_sm);
+bad:
+	dm_tm_destroy(tm);
+	dm_sm_destroy(sm);
+
+	return r;
+}
+
+static int __begin_transaction(struct dm_pool_metadata *pmd)
+{
+	int r;
+	u32 features;
+	struct thin_disk_superblock *disk_super;
+	struct dm_block *sblock;
+
+	/*
+	 * __maybe_commit_transaction() resets these
+	 */
+	WARN_ON(pmd->need_commit);
+
+	/*
+	 * We re-read the superblock every time.  Shouldn't need to do this
+	 * really.
+	 */
+	r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
+			    &sb_validator, &sblock);
+	if (r)
+		return r;
+
+	disk_super = dm_block_data(sblock);
+	pmd->time = le32_to_cpu(disk_super->time);
+	pmd->root = le64_to_cpu(disk_super->data_mapping_root);
+	pmd->details_root = le64_to_cpu(disk_super->device_details_root);
+	pmd->trans_id = le64_to_cpu(disk_super->trans_id);
+	pmd->flags = le32_to_cpu(disk_super->flags);
+	pmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
+
+	features = le32_to_cpu(disk_super->incompat_flags) & ~THIN_FEATURE_INCOMPAT_SUPP;
+	if (features) {
+		DMERR("could not access metadata due to "
+		      "unsupported optional features (%lx).",
+		      (unsigned long)features);
+		r = -EINVAL;
+		goto out;
+	}
+
+	/*
+	 * Check for read-only metadata to skip the following RDWR checks.
+	 */
+	if (get_disk_ro(pmd->bdev->bd_disk))
+		goto out;
+
+	features = le32_to_cpu(disk_super->compat_ro_flags) & ~THIN_FEATURE_COMPAT_RO_SUPP;
+	if (features) {
+		DMERR("could not access metadata RDWR due to "
+		      "unsupported optional features (%lx).",
+		      (unsigned long)features);
+		r = -EINVAL;
+	}
+
+out:
+	dm_bm_unlock(sblock);
+	return r;
+}
+
+static int __write_changed_details(struct dm_pool_metadata *pmd)
+{
+	int r;
+	struct dm_thin_device *td, *tmp;
+	struct disk_device_details details;
+	uint64_t key;
+
+	list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
+		if (!td->changed)
+			continue;
+
+		key = td->id;
+
+		details.mapped_blocks = cpu_to_le64(td->mapped_blocks);
+		details.transaction_id = cpu_to_le64(td->transaction_id);
+		details.creation_time = cpu_to_le32(td->creation_time);
+		details.snapshotted_time = cpu_to_le32(td->snapshotted_time);
+		__dm_bless_for_disk(&details);
+
+		r = dm_btree_insert(&pmd->details_info, pmd->details_root,
+				    &key, &details, &pmd->details_root);
+		if (r)
+			return r;
+
+		if (td->open_count)
+			td->changed = 0;
+		else {
+			list_del(&td->list);
+			kfree(td);
+		}
+
+		pmd->need_commit = 1;
+	}
+
+	return 0;
+}
+
+static int __commit_transaction(struct dm_pool_metadata *pmd)
+{
+	/*
+	 * FIXME: Associated pool should be made read-only on failure.
+	 */
+	int r;
+	size_t metadata_len, data_len;
+	struct thin_disk_superblock *disk_super;
+	struct dm_block *sblock;
+
+	/*
+	 * We need to know if the thin_disk_superblock exceeds a 512-byte sector.
+	 */
+	BUILD_BUG_ON(sizeof(struct thin_disk_superblock) > 512);
+
+	r = __write_changed_details(pmd);
+	if (r < 0)
+		goto out;
+
+	if (!pmd->need_commit)
+		goto out;
+
+	r = dm_sm_commit(pmd->data_sm);
+	if (r < 0)
+		goto out;
+
+	r = dm_tm_pre_commit(pmd->tm);
+	if (r < 0)
+		goto out;
+
+	r = dm_sm_root_size(pmd->metadata_sm, &metadata_len);
+	if (r < 0)
+		goto out;
+
+	r = dm_sm_root_size(pmd->metadata_sm, &data_len);
+	if (r < 0)
+		goto out;
+
+	r = dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
+			     &sb_validator, &sblock);
+	if (r)
+		goto out;
+
+	disk_super = dm_block_data(sblock);
+	disk_super->time = cpu_to_le32(pmd->time);
+	disk_super->data_mapping_root = cpu_to_le64(pmd->root);
+	disk_super->device_details_root = cpu_to_le64(pmd->details_root);
+	disk_super->trans_id = cpu_to_le64(pmd->trans_id);
+	disk_super->flags = cpu_to_le32(pmd->flags);
+
+	r = dm_sm_copy_root(pmd->metadata_sm, &disk_super->metadata_space_map_root,
+			    metadata_len);
+	if (r < 0)
+		goto out_locked;
+
+	r = dm_sm_copy_root(pmd->data_sm, &disk_super->data_space_map_root,
+			    data_len);
+	if (r < 0)
+		goto out_locked;
+
+	r = dm_tm_commit(pmd->tm, sblock);
+	if (!r)
+		pmd->need_commit = 0;
+
+out:
+	return r;
+
+out_locked:
+	dm_bm_unlock(sblock);
+	return r;
+}
+
+struct dm_pool_metadata *dm_pool_metadata_open(struct block_device *bdev,
+					       sector_t data_block_size)
+{
+	int r;
+	struct thin_disk_superblock *disk_super;
+	struct dm_pool_metadata *pmd;
+	sector_t bdev_size = i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
+	struct dm_block_manager *bm;
+	int create;
+	struct dm_block *sblock;
+
+	pmd = kmalloc(sizeof(*pmd), GFP_KERNEL);
+	if (!pmd) {
+		DMERR("could not allocate metadata struct");
+		return ERR_PTR(-ENOMEM);
+	}
+
+	/*
+	 * Max hex locks:
+	 *  3 for btree insert +
+	 *  2 for btree lookup used within space map
+	 */
+	bm = dm_block_manager_create(bdev, THIN_METADATA_BLOCK_SIZE,
+				     THIN_METADATA_CACHE_SIZE, 5);
+	if (!bm) {
+		DMERR("could not create block manager");
+		kfree(pmd);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	r = superblock_all_zeroes(bm, &create);
+	if (r) {
+		dm_block_manager_destroy(bm);
+		kfree(pmd);
+		return ERR_PTR(r);
+	}
+
+
+	r = init_pmd(pmd, bm, 0, create);
+	if (r) {
+		dm_block_manager_destroy(bm);
+		kfree(pmd);
+		return ERR_PTR(r);
+	}
+	pmd->bdev = bdev;
+
+	if (!create) {
+		r = __begin_transaction(pmd);
+		if (r < 0)
+			goto bad;
+		return pmd;
+	}
+
+	/*
+	 * Create.
+	 */
+	r = dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
+			     &sb_validator, &sblock);
+	if (r)
+		goto bad;
+
+	disk_super = dm_block_data(sblock);
+	disk_super->magic = cpu_to_le64(THIN_SUPERBLOCK_MAGIC);
+	disk_super->version = cpu_to_le32(THIN_VERSION);
+	disk_super->time = 0;
+	disk_super->metadata_block_size = cpu_to_le32(THIN_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
+	disk_super->metadata_nr_blocks = cpu_to_le64(bdev_size >> SECTOR_TO_BLOCK_SHIFT);
+	disk_super->data_block_size = cpu_to_le32(data_block_size);
+
+	r = dm_bm_unlock(sblock);
+	if (r < 0)
+		goto bad;
+
+	r = dm_btree_empty(&pmd->info, &pmd->root);
+	if (r < 0)
+		goto bad;
+
+	r = dm_btree_empty(&pmd->details_info, &pmd->details_root);
+	if (r < 0) {
+		DMERR("couldn't create devices root");
+		goto bad;
+	}
+
+	pmd->flags = 0;
+	pmd->need_commit = 1;
+	r = dm_pool_commit_metadata(pmd);
+	if (r < 0) {
+		DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
+		      __func__, r);
+		goto bad;
+	}
+
+	return pmd;
+
+bad:
+	if (dm_pool_metadata_close(pmd) < 0)
+		DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
+	return ERR_PTR(r);
+}
+
+int dm_pool_metadata_close(struct dm_pool_metadata *pmd)
+{
+	int r;
+	unsigned open_devices = 0;
+	struct dm_thin_device *td, *tmp;
+
+	down_read(&pmd->root_lock);
+	list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
+		if (td->open_count)
+			open_devices++;
+		else {
+			list_del(&td->list);
+			kfree(td);
+		}
+	}
+	up_read(&pmd->root_lock);
+
+	if (open_devices) {
+		DMERR("attempt to close pmd when %u device(s) are still open",
+		       open_devices);
+		return -EBUSY;
+	}
+
+	r = __commit_transaction(pmd);
+	if (r < 0)
+		DMWARN("%s: __commit_transaction() failed, error = %d",
+		       __func__, r);
+
+	dm_tm_destroy(pmd->tm);
+	dm_tm_destroy(pmd->nb_tm);
+	dm_block_manager_destroy(pmd->bm);
+	dm_sm_destroy(pmd->metadata_sm);
+	dm_sm_destroy(pmd->data_sm);
+	kfree(pmd);
+
+	return 0;
+}
+
+static int __open_device(struct dm_pool_metadata *pmd,
+			 dm_thin_id dev, int create,
+			 struct dm_thin_device **td)
+{
+	int r, changed = 0;
+	struct dm_thin_device *td2;
+	uint64_t key = dev;
+	struct disk_device_details details_le;
+
+	/*
+	 * Check the device isn't already open.
+	 */
+	list_for_each_entry(td2, &pmd->thin_devices, list)
+		if (td2->id == dev) {
+			td2->open_count++;
+			*td = td2;
+			return 0;
+		}
+
+	/*
+	 * Check the device exists.
+	 */
+	r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
+			    &key, &details_le);
+	if (r) {
+		if (r != -ENODATA || !create)
+			return r;
+
+		changed = 1;
+		details_le.mapped_blocks = 0;
+		details_le.transaction_id = cpu_to_le64(pmd->trans_id);
+		details_le.creation_time = cpu_to_le32(pmd->time);
+		details_le.snapshotted_time = cpu_to_le32(pmd->time);
+	}
+
+	*td = kmalloc(sizeof(**td), GFP_NOIO);
+	if (!*td)
+		return -ENOMEM;
+
+	(*td)->pmd = pmd;
+	(*td)->id = dev;
+	(*td)->open_count = 1;
+	(*td)->changed = changed;
+	(*td)->mapped_blocks = le64_to_cpu(details_le.mapped_blocks);
+	(*td)->transaction_id = le64_to_cpu(details_le.transaction_id);
+	(*td)->creation_time = le32_to_cpu(details_le.creation_time);
+	(*td)->snapshotted_time = le32_to_cpu(details_le.snapshotted_time);
+
+	list_add(&(*td)->list, &pmd->thin_devices);
+
+	return 0;
+}
+
+static void __close_device(struct dm_thin_device *td)
+{
+	--td->open_count;
+}
+
+static int __create_thin(struct dm_pool_metadata *pmd,
+			 dm_thin_id dev)
+{
+	int r;
+	dm_block_t dev_root;
+	uint64_t key = dev;
+	struct disk_device_details details_le;
+	struct dm_thin_device *td;
+	__le64 value;
+
+	r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
+			    &key, &details_le);
+	if (!r)
+		return -EEXIST;
+
+	/*
+	 * Create an empty btree for the mappings.
+	 */
+	r = dm_btree_empty(&pmd->bl_info, &dev_root);
+	if (r)
+		return r;
+
+	/*
+	 * Insert it into the main mapping tree.
+	 */
+	value = cpu_to_le64(dev_root);
+	__dm_bless_for_disk(&value);
+	r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
+	if (r) {
+		dm_btree_del(&pmd->bl_info, dev_root);
+		return r;
+	}
+
+	r = __open_device(pmd, dev, 1, &td);
+	if (r) {
+		__close_device(td);
+		dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
+		dm_btree_del(&pmd->bl_info, dev_root);
+		return r;
+	}
+	td->changed = 1;
+	__close_device(td);
+
+	return r;
+}
+
+int dm_pool_create_thin(struct dm_pool_metadata *pmd, dm_thin_id dev)
+{
+	int r;
+
+	down_write(&pmd->root_lock);
+	r = __create_thin(pmd, dev);
+	up_write(&pmd->root_lock);
+
+	return r;
+}
+
+static int __set_snapshot_details(struct dm_pool_metadata *pmd,
+				  struct dm_thin_device *snap,
+				  dm_thin_id origin, uint32_t time)
+{
+	int r;
+	struct dm_thin_device *td;
+
+	r = __open_device(pmd, origin, 0, &td);
+	if (r)
+		return r;
+
+	td->changed = 1;
+	td->snapshotted_time = time;
+
+	snap->mapped_blocks = td->mapped_blocks;
+	snap->snapshotted_time = time;
+	__close_device(td);
+
+	return 0;
+}
+
+static int __create_snap(struct dm_pool_metadata *pmd,
+			 dm_thin_id dev, dm_thin_id origin)
+{
+	int r;
+	dm_block_t origin_root;
+	uint64_t key = origin, dev_key = dev;
+	struct dm_thin_device *td;
+	struct disk_device_details details_le;
+	__le64 value;
+
+	/* check this device is unused */
+	r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
+			    &dev_key, &details_le);
+	if (!r)
+		return -EEXIST;
+
+	/* find the mapping tree for the origin */
+	r = dm_btree_lookup(&pmd->tl_info, pmd->root, &key, &value);
+	if (r)
+		return r;
+	origin_root = le64_to_cpu(value);
+
+	/* clone the origin, an inc will do */
+	dm_tm_inc(pmd->tm, origin_root);
+
+	/* insert into the main mapping tree */
+	value = cpu_to_le64(origin_root);
+	__dm_bless_for_disk(&value);
+	key = dev;
+	r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
+	if (r) {
+		dm_tm_dec(pmd->tm, origin_root);
+		return r;
+	}
+
+	pmd->time++;
+
+	r = __open_device(pmd, dev, 1, &td);
+	if (r)
+		goto bad;
+
+	r = __set_snapshot_details(pmd, td, origin, pmd->time);
+	if (r)
+		goto bad;
+
+	__close_device(td);
+	return 0;
+
+bad:
+	__close_device(td);
+	dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
+	dm_btree_remove(&pmd->details_info, pmd->details_root,
+			&key, &pmd->details_root);
+	return r;
+}
+
+int dm_pool_create_snap(struct dm_pool_metadata *pmd,
+				 dm_thin_id dev,
+				 dm_thin_id origin)
+{
+	int r;
+
+	down_write(&pmd->root_lock);
+	r = __create_snap(pmd, dev, origin);
+	up_write(&pmd->root_lock);
+
+	return r;
+}
+
+static int __delete_device(struct dm_pool_metadata *pmd, dm_thin_id dev)
+{
+	int r;
+	uint64_t key = dev;
+	struct dm_thin_device *td;
+
+	/* TODO: failure should mark the transaction invalid */
+	r = __open_device(pmd, dev, 0, &td);
+	if (r)
+		return r;
+
+	if (td->open_count > 1) {
+		__close_device(td);
+		return -EBUSY;
+	}
+
+	list_del(&td->list);
+	kfree(td);
+	r = dm_btree_remove(&pmd->details_info, pmd->details_root,
+			    &key, &pmd->details_root);
+	if (r)
+		return r;
+
+	r = dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
+	if (r)
+		return r;
+
+	pmd->need_commit = 1;
+
+	return 0;
+}
+
+int dm_pool_delete_thin_device(struct dm_pool_metadata *pmd,
+			       dm_thin_id dev)
+{
+	int r;
+
+	down_write(&pmd->root_lock);
+	r = __delete_device(pmd, dev);
+	up_write(&pmd->root_lock);
+
+	return r;
+}
+
+int dm_pool_set_metadata_transaction_id(struct dm_pool_metadata *pmd,
+					uint64_t current_id,
+					uint64_t new_id)
+{
+	down_write(&pmd->root_lock);
+	if (pmd->trans_id != current_id) {
+		up_write(&pmd->root_lock);
+		DMERR("mismatched transaction id");
+		return -EINVAL;
+	}
+
+	pmd->trans_id = new_id;
+	pmd->need_commit = 1;
+	up_write(&pmd->root_lock);
+
+	return 0;
+}
+
+int dm_pool_get_metadata_transaction_id(struct dm_pool_metadata *pmd,
+					uint64_t *result)
+{
+	down_read(&pmd->root_lock);
+	*result = pmd->trans_id;
+	up_read(&pmd->root_lock);
+
+	return 0;
+}
+
+static int __get_held_metadata_root(struct dm_pool_metadata *pmd,
+				    dm_block_t *result)
+{
+	int r;
+	struct thin_disk_superblock *disk_super;
+	struct dm_block *sblock;
+
+	r = dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
+			     &sb_validator, &sblock);
+	if (r)
+		return r;
+
+	disk_super = dm_block_data(sblock);
+	*result = le64_to_cpu(disk_super->held_root);
+
+	return dm_bm_unlock(sblock);
+}
+
+int dm_pool_get_held_metadata_root(struct dm_pool_metadata *pmd,
+				   dm_block_t *result)
+{
+	int r;
+
+	down_read(&pmd->root_lock);
+	r = __get_held_metadata_root(pmd, result);
+	up_read(&pmd->root_lock);
+
+	return r;
+}
+
+int dm_pool_open_thin_device(struct dm_pool_metadata *pmd, dm_thin_id dev,
+			     struct dm_thin_device **td)
+{
+	int r;
+
+	down_write(&pmd->root_lock);
+	r = __open_device(pmd, dev, 0, td);
+	up_write(&pmd->root_lock);
+
+	return r;
+}
+
+int dm_pool_close_thin_device(struct dm_thin_device *td)
+{
+	down_write(&td->pmd->root_lock);
+	__close_device(td);
+	up_write(&td->pmd->root_lock);
+
+	return 0;
+}
+
+dm_thin_id dm_thin_dev_id(struct dm_thin_device *td)
+{
+	return td->id;
+}
+
+static int __snapshotted_since(struct dm_thin_device *td, uint32_t time)
+{
+	return td->snapshotted_time > time;
+}
+
+int dm_thin_find_block(struct dm_thin_device *td, dm_block_t block,
+		       int can_block, struct dm_thin_lookup_result *result)
+{
+	int r;
+	uint64_t block_time = 0;
+	__le64 value;
+	struct dm_pool_metadata *pmd = td->pmd;
+	dm_block_t keys[2] = { td->id, block };
+
+	if (can_block) {
+		down_read(&pmd->root_lock);
+		r = dm_btree_lookup(&pmd->info, pmd->root, keys, &value);
+		if (!r)
+			block_time = le64_to_cpu(value);
+		up_read(&pmd->root_lock);
+
+	} else if (down_read_trylock(&pmd->root_lock)) {
+		r = dm_btree_lookup(&pmd->nb_info, pmd->root, keys, &value);
+		if (!r)
+			block_time = le64_to_cpu(value);
+		up_read(&pmd->root_lock);
+
+	} else
+		return -EWOULDBLOCK;
+
+	if (!r) {
+		dm_block_t exception_block;
+		uint32_t exception_time;
+		unpack_block_time(block_time, &exception_block,
+				  &exception_time);
+		result->block = exception_block;
+		result->shared = __snapshotted_since(td, exception_time);
+	}
+
+	return r;
+}
+
+static int __insert(struct dm_thin_device *td, dm_block_t block,
+		    dm_block_t data_block)
+{
+	int r, inserted;
+	__le64 value;
+	struct dm_pool_metadata *pmd = td->pmd;
+	dm_block_t keys[2] = { td->id, block };
+
+	pmd->need_commit = 1;
+	value = cpu_to_le64(pack_block_time(data_block, pmd->time));
+	__dm_bless_for_disk(&value);
+
+	r = dm_btree_insert_notify(&pmd->info, pmd->root, keys, &value,
+				   &pmd->root, &inserted);
+	if (r)
+		return r;
+
+	if (inserted) {
+		td->mapped_blocks++;
+		td->changed = 1;
+	}
+
+	return 0;
+}
+
+int dm_thin_insert_block(struct dm_thin_device *td, dm_block_t block,
+			 dm_block_t data_block)
+{
+	int r;
+
+	down_write(&td->pmd->root_lock);
+	r = __insert(td, block, data_block);
+	up_write(&td->pmd->root_lock);
+
+	return r;
+}
+
+static int __remove(struct dm_thin_device *td, dm_block_t block)
+{
+	int r;
+	struct dm_pool_metadata *pmd = td->pmd;
+	dm_block_t keys[2] = { td->id, block };
+
+	r = dm_btree_remove(&pmd->info, pmd->root, keys, &pmd->root);
+	if (r)
+		return r;
+
+	pmd->need_commit = 1;
+
+	return 0;
+}
+
+int dm_thin_remove_block(struct dm_thin_device *td, dm_block_t block)
+{
+	int r;
+
+	down_write(&td->pmd->root_lock);
+	r = __remove(td, block);
+	up_write(&td->pmd->root_lock);
+
+	return r;
+}
+
+int dm_pool_alloc_data_block(struct dm_pool_metadata *pmd, dm_block_t *result)
+{
+	int r;
+
+	down_write(&pmd->root_lock);
+
+	r = dm_sm_new_block(pmd->data_sm, result);
+	pmd->need_commit = 1;
+
+	up_write(&pmd->root_lock);
+
+	return r;
+}
+
+int dm_pool_commit_metadata(struct dm_pool_metadata *pmd)
+{
+	int r;
+
+	down_write(&pmd->root_lock);
+
+	r = __commit_transaction(pmd);
+	if (r <= 0)
+		goto out;
+
+	/*
+	 * Open the next transaction.
+	 */
+	r = __begin_transaction(pmd);
+out:
+	up_write(&pmd->root_lock);
+	return r;
+}
+
+int dm_pool_get_free_block_count(struct dm_pool_metadata *pmd, dm_block_t *result)
+{
+	int r;
+
+	down_read(&pmd->root_lock);
+	r = dm_sm_get_nr_free(pmd->data_sm, result);
+	up_read(&pmd->root_lock);
+
+	return r;
+}
+
+int dm_pool_get_free_metadata_block_count(struct dm_pool_metadata *pmd,
+					  dm_block_t *result)
+{
+	int r;
+
+	down_read(&pmd->root_lock);
+	r = dm_sm_get_nr_free(pmd->metadata_sm, result);
+	up_read(&pmd->root_lock);
+
+	return r;
+}
+
+int dm_pool_get_metadata_dev_size(struct dm_pool_metadata *pmd,
+				  dm_block_t *result)
+{
+	int r;
+
+	down_read(&pmd->root_lock);
+	r = dm_sm_get_nr_blocks(pmd->metadata_sm, result);
+	up_read(&pmd->root_lock);
+
+	return r;
+}
+
+int dm_pool_get_data_block_size(struct dm_pool_metadata *pmd, sector_t *result)
+{
+	down_read(&pmd->root_lock);
+	*result = pmd->data_block_size;
+	up_read(&pmd->root_lock);
+
+	return 0;
+}
+
+int dm_pool_get_data_dev_size(struct dm_pool_metadata *pmd, dm_block_t *result)
+{
+	int r;
+
+	down_read(&pmd->root_lock);
+	r = dm_sm_get_nr_blocks(pmd->data_sm, result);
+	up_read(&pmd->root_lock);
+
+	return r;
+}
+
+int dm_thin_get_mapped_count(struct dm_thin_device *td, dm_block_t *result)
+{
+	struct dm_pool_metadata *pmd = td->pmd;
+
+	down_read(&pmd->root_lock);
+	*result = td->mapped_blocks;
+	up_read(&pmd->root_lock);
+
+	return 0;
+}
+
+static int __highest_block(struct dm_thin_device *td, dm_block_t *result)
+{
+	int r;
+	__le64 value_le;
+	dm_block_t thin_root;
+	struct dm_pool_metadata *pmd = td->pmd;
+
+	r = dm_btree_lookup(&pmd->tl_info, pmd->root, &td->id, &value_le);
+	if (r)
+		return r;
+
+	thin_root = le64_to_cpu(value_le);
+
+	return dm_btree_find_highest_key(&pmd->bl_info, thin_root, result);
+}
+
+int dm_thin_get_highest_mapped_block(struct dm_thin_device *td,
+				     dm_block_t *result)
+{
+	int r;
+	struct dm_pool_metadata *pmd = td->pmd;
+
+	down_read(&pmd->root_lock);
+	r = __highest_block(td, result);
+	up_read(&pmd->root_lock);
+
+	return r;
+}
+
+static int __resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
+{
+	int r;
+	dm_block_t old_count;
+
+	r = dm_sm_get_nr_blocks(pmd->data_sm, &old_count);
+	if (r)
+		return r;
+
+	if (new_count == old_count)
+		return 0;
+
+	if (new_count < old_count) {
+		DMERR("cannot reduce size of data device");
+		return -EINVAL;
+	}
+
+	r = dm_sm_extend(pmd->data_sm, new_count - old_count);
+	if (!r)
+		pmd->need_commit = 1;
+
+	return r;
+}
+
+int dm_pool_resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
+{
+	int r;
+
+	down_write(&pmd->root_lock);
+	r = __resize_data_dev(pmd, new_count);
+	up_write(&pmd->root_lock);
+
+	return r;
+}
diff --git a/drivers/md/dm-thin-metadata.h b/drivers/md/dm-thin-metadata.h
new file mode 100644
index 00000000000..859c1689687
--- /dev/null
+++ b/drivers/md/dm-thin-metadata.h
@@ -0,0 +1,156 @@
+/*
+ * Copyright (C) 2010-2011 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+
+#ifndef DM_THIN_METADATA_H
+#define DM_THIN_METADATA_H
+
+#include "persistent-data/dm-block-manager.h"
+
+#define THIN_METADATA_BLOCK_SIZE 4096
+
+/*----------------------------------------------------------------*/
+
+struct dm_pool_metadata;
+struct dm_thin_device;
+
+/*
+ * Device identifier
+ */
+typedef uint64_t dm_thin_id;
+
+/*
+ * Reopens or creates a new, empty metadata volume.
+ */
+struct dm_pool_metadata *dm_pool_metadata_open(struct block_device *bdev,
+					       sector_t data_block_size);
+
+int dm_pool_metadata_close(struct dm_pool_metadata *pmd);
+
+/*
+ * Compat feature flags.  Any incompat flags beyond the ones
+ * specified below will prevent use of the thin metadata.
+ */
+#define THIN_FEATURE_COMPAT_SUPP	  0UL
+#define THIN_FEATURE_COMPAT_RO_SUPP	  0UL
+#define THIN_FEATURE_INCOMPAT_SUPP	  0UL
+
+/*
+ * Device creation/deletion.
+ */
+int dm_pool_create_thin(struct dm_pool_metadata *pmd, dm_thin_id dev);
+
+/*
+ * An internal snapshot.
+ *
+ * You can only snapshot a quiesced origin i.e. one that is either
+ * suspended or not instanced at all.
+ */
+int dm_pool_create_snap(struct dm_pool_metadata *pmd, dm_thin_id dev,
+			dm_thin_id origin);
+
+/*
+ * Deletes a virtual device from the metadata.  It _is_ safe to call this
+ * when that device is open.  Operations on that device will just start
+ * failing.  You still need to call close() on the device.
+ */
+int dm_pool_delete_thin_device(struct dm_pool_metadata *pmd,
+			       dm_thin_id dev);
+
+/*
+ * Commits _all_ metadata changes: device creation, deletion, mapping
+ * updates.
+ */
+int dm_pool_commit_metadata(struct dm_pool_metadata *pmd);
+
+/*
+ * Set/get userspace transaction id.
+ */
+int dm_pool_set_metadata_transaction_id(struct dm_pool_metadata *pmd,
+					uint64_t current_id,
+					uint64_t new_id);
+
+int dm_pool_get_metadata_transaction_id(struct dm_pool_metadata *pmd,
+					uint64_t *result);
+
+/*
+ * Hold/get root for userspace transaction.
+ */
+int dm_pool_hold_metadata_root(struct dm_pool_metadata *pmd);
+
+int dm_pool_get_held_metadata_root(struct dm_pool_metadata *pmd,
+				   dm_block_t *result);
+
+/*
+ * Actions on a single virtual device.
+ */
+
+/*
+ * Opening the same device more than once will fail with -EBUSY.
+ */
+int dm_pool_open_thin_device(struct dm_pool_metadata *pmd, dm_thin_id dev,
+			     struct dm_thin_device **td);
+
+int dm_pool_close_thin_device(struct dm_thin_device *td);
+
+dm_thin_id dm_thin_dev_id(struct dm_thin_device *td);
+
+struct dm_thin_lookup_result {
+	dm_block_t block;
+	int shared;
+};
+
+/*
+ * Returns:
+ *   -EWOULDBLOCK iff @can_block is set and would block.
+ *   -ENODATA iff that mapping is not present.
+ *   0 success
+ */
+int dm_thin_find_block(struct dm_thin_device *td, dm_block_t block,
+		       int can_block, struct dm_thin_lookup_result *result);
+
+/*
+ * Obtain an unused block.
+ */
+int dm_pool_alloc_data_block(struct dm_pool_metadata *pmd, dm_block_t *result);
+
+/*
+ * Insert or remove block.
+ */
+int dm_thin_insert_block(struct dm_thin_device *td, dm_block_t block,
+			 dm_block_t data_block);
+
+int dm_thin_remove_block(struct dm_thin_device *td, dm_block_t block);
+
+/*
+ * Queries.
+ */
+int dm_thin_get_highest_mapped_block(struct dm_thin_device *td,
+				     dm_block_t *highest_mapped);
+
+int dm_thin_get_mapped_count(struct dm_thin_device *td, dm_block_t *result);
+
+int dm_pool_get_free_block_count(struct dm_pool_metadata *pmd,
+				 dm_block_t *result);
+
+int dm_pool_get_free_metadata_block_count(struct dm_pool_metadata *pmd,
+					  dm_block_t *result);
+
+int dm_pool_get_metadata_dev_size(struct dm_pool_metadata *pmd,
+				  dm_block_t *result);
+
+int dm_pool_get_data_block_size(struct dm_pool_metadata *pmd, sector_t *result);
+
+int dm_pool_get_data_dev_size(struct dm_pool_metadata *pmd, dm_block_t *result);
+
+/*
+ * Returns -ENOSPC if the new size is too small and already allocated
+ * blocks would be lost.
+ */
+int dm_pool_resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_size);
+
+/*----------------------------------------------------------------*/
+
+#endif
diff --git a/drivers/md/dm-thin.c b/drivers/md/dm-thin.c
new file mode 100644
index 00000000000..c3087575fef
--- /dev/null
+++ b/drivers/md/dm-thin.c
@@ -0,0 +1,2428 @@
+/*
+ * Copyright (C) 2011 Red Hat UK.
+ *
+ * This file is released under the GPL.
+ */
+
+#include "dm-thin-metadata.h"
+
+#include <linux/device-mapper.h>
+#include <linux/dm-io.h>
+#include <linux/dm-kcopyd.h>
+#include <linux/list.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+
+#define	DM_MSG_PREFIX	"thin"
+
+/*
+ * Tunable constants
+ */
+#define ENDIO_HOOK_POOL_SIZE 10240
+#define DEFERRED_SET_SIZE 64
+#define MAPPING_POOL_SIZE 1024
+#define PRISON_CELLS 1024
+
+/*
+ * The block size of the device holding pool data must be
+ * between 64KB and 1GB.
+ */
+#define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
+#define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
+
+/*
+ * The metadata device is currently limited in size.  The limitation is
+ * checked lower down in dm-space-map-metadata, but we also check it here
+ * so we can fail early.
+ *
+ * We have one block of index, which can hold 255 index entries.  Each
+ * index entry contains allocation info about 16k metadata blocks.
+ */
+#define METADATA_DEV_MAX_SECTORS (255 * (1 << 14) * (THIN_METADATA_BLOCK_SIZE / (1 << SECTOR_SHIFT)))
+
+/*
+ * Device id is restricted to 24 bits.
+ */
+#define MAX_DEV_ID ((1 << 24) - 1)
+
+/*
+ * How do we handle breaking sharing of data blocks?
+ * =================================================
+ *
+ * We use a standard copy-on-write btree to store the mappings for the
+ * devices (note I'm talking about copy-on-write of the metadata here, not
+ * the data).  When you take an internal snapshot you clone the root node
+ * of the origin btree.  After this there is no concept of an origin or a
+ * snapshot.  They are just two device trees that happen to point to the
+ * same data blocks.
+ *
+ * When we get a write in we decide if it's to a shared data block using
+ * some timestamp magic.  If it is, we have to break sharing.
+ *
+ * Let's say we write to a shared block in what was the origin.  The
+ * steps are:
+ *
+ * i) plug io further to this physical block. (see bio_prison code).
+ *
+ * ii) quiesce any read io to that shared data block.  Obviously
+ * including all devices that share this block.  (see deferred_set code)
+ *
+ * iii) copy the data block to a newly allocate block.  This step can be
+ * missed out if the io covers the block. (schedule_copy).
+ *
+ * iv) insert the new mapping into the origin's btree
+ * (process_prepared_mappings).  This act of inserting breaks some
+ * sharing of btree nodes between the two devices.  Breaking sharing only
+ * effects the btree of that specific device.  Btrees for the other
+ * devices that share the block never change.  The btree for the origin
+ * device as it was after the last commit is untouched, ie. we're using
+ * persistent data structures in the functional programming sense.
+ *
+ * v) unplug io to this physical block, including the io that triggered
+ * the breaking of sharing.
+ *
+ * Steps (ii) and (iii) occur in parallel.
+ *
+ * The metadata _doesn't_ need to be committed before the io continues.  We
+ * get away with this because the io is always written to a _new_ block.
+ * If there's a crash, then:
+ *
+ * - The origin mapping will point to the old origin block (the shared
+ * one).  This will contain the data as it was before the io that triggered
+ * the breaking of sharing came in.
+ *
+ * - The snap mapping still points to the old block.  As it would after
+ * the commit.
+ *
+ * The downside of this scheme is the timestamp magic isn't perfect, and
+ * will continue to think that data block in the snapshot device is shared
+ * even after the write to the origin has broken sharing.  I suspect data
+ * blocks will typically be shared by many different devices, so we're
+ * breaking sharing n + 1 times, rather than n, where n is the number of
+ * devices that reference this data block.  At the moment I think the
+ * benefits far, far outweigh the disadvantages.
+ */
+
+/*----------------------------------------------------------------*/
+
+/*
+ * Sometimes we can't deal with a bio straight away.  We put them in prison
+ * where they can't cause any mischief.  Bios are put in a cell identified
+ * by a key, multiple bios can be in the same cell.  When the cell is
+ * subsequently unlocked the bios become available.
+ */
+struct bio_prison;
+
+struct cell_key {
+	int virtual;
+	dm_thin_id dev;
+	dm_block_t block;
+};
+
+struct cell {
+	struct hlist_node list;
+	struct bio_prison *prison;
+	struct cell_key key;
+	unsigned count;
+	struct bio_list bios;
+};
+
+struct bio_prison {
+	spinlock_t lock;
+	mempool_t *cell_pool;
+
+	unsigned nr_buckets;
+	unsigned hash_mask;
+	struct hlist_head *cells;
+};
+
+static uint32_t calc_nr_buckets(unsigned nr_cells)
+{
+	uint32_t n = 128;
+
+	nr_cells /= 4;
+	nr_cells = min(nr_cells, 8192u);
+
+	while (n < nr_cells)
+		n <<= 1;
+
+	return n;
+}
+
+/*
+ * @nr_cells should be the number of cells you want in use _concurrently_.
+ * Don't confuse it with the number of distinct keys.
+ */
+static struct bio_prison *prison_create(unsigned nr_cells)
+{
+	unsigned i;
+	uint32_t nr_buckets = calc_nr_buckets(nr_cells);
+	size_t len = sizeof(struct bio_prison) +
+		(sizeof(struct hlist_head) * nr_buckets);
+	struct bio_prison *prison = kmalloc(len, GFP_KERNEL);
+
+	if (!prison)
+		return NULL;
+
+	spin_lock_init(&prison->lock);
+	prison->cell_pool = mempool_create_kmalloc_pool(nr_cells,
+							sizeof(struct cell));
+	if (!prison->cell_pool) {
+		kfree(prison);
+		return NULL;
+	}
+
+	prison->nr_buckets = nr_buckets;
+	prison->hash_mask = nr_buckets - 1;
+	prison->cells = (struct hlist_head *) (prison + 1);
+	for (i = 0; i < nr_buckets; i++)
+		INIT_HLIST_HEAD(prison->cells + i);
+
+	return prison;
+}
+
+static void prison_destroy(struct bio_prison *prison)
+{
+	mempool_destroy(prison->cell_pool);
+	kfree(prison);
+}
+
+static uint32_t hash_key(struct bio_prison *prison, struct cell_key *key)
+{
+	const unsigned long BIG_PRIME = 4294967291UL;
+	uint64_t hash = key->block * BIG_PRIME;
+
+	return (uint32_t) (hash & prison->hash_mask);
+}
+
+static int keys_equal(struct cell_key *lhs, struct cell_key *rhs)
+{
+	       return (lhs->virtual == rhs->virtual) &&
+		       (lhs->dev == rhs->dev) &&
+		       (lhs->block == rhs->block);
+}
+
+static struct cell *__search_bucket(struct hlist_head *bucket,
+				    struct cell_key *key)
+{
+	struct cell *cell;
+	struct hlist_node *tmp;
+
+	hlist_for_each_entry(cell, tmp, bucket, list)
+		if (keys_equal(&cell->key, key))
+			return cell;
+
+	return NULL;
+}
+
+/*
+ * This may block if a new cell needs allocating.  You must ensure that
+ * cells will be unlocked even if the calling thread is blocked.
+ *
+ * Returns the number of entries in the cell prior to the new addition
+ * or < 0 on failure.
+ */
+static int bio_detain(struct bio_prison *prison, struct cell_key *key,
+		      struct bio *inmate, struct cell **ref)
+{
+	int r;
+	unsigned long flags;
+	uint32_t hash = hash_key(prison, key);
+	struct cell *uninitialized_var(cell), *cell2 = NULL;
+
+	BUG_ON(hash > prison->nr_buckets);
+
+	spin_lock_irqsave(&prison->lock, flags);
+	cell = __search_bucket(prison->cells + hash, key);
+
+	if (!cell) {
+		/*
+		 * Allocate a new cell
+		 */
+		spin_unlock_irqrestore(&prison->lock, flags);
+		cell2 = mempool_alloc(prison->cell_pool, GFP_NOIO);
+		spin_lock_irqsave(&prison->lock, flags);
+
+		/*
+		 * We've been unlocked, so we have to double check that
+		 * nobody else has inserted this cell in the meantime.
+		 */
+		cell = __search_bucket(prison->cells + hash, key);
+
+		if (!cell) {
+			cell = cell2;
+			cell2 = NULL;
+
+			cell->prison = prison;
+			memcpy(&cell->key, key, sizeof(cell->key));
+			cell->count = 0;
+			bio_list_init(&cell->bios);
+			hlist_add_head(&cell->list, prison->cells + hash);
+		}
+	}
+
+	r = cell->count++;
+	bio_list_add(&cell->bios, inmate);
+	spin_unlock_irqrestore(&prison->lock, flags);
+
+	if (cell2)
+		mempool_free(cell2, prison->cell_pool);
+
+	*ref = cell;
+
+	return r;
+}
+
+/*
+ * @inmates must have been initialised prior to this call
+ */
+static void __cell_release(struct cell *cell, struct bio_list *inmates)
+{
+	struct bio_prison *prison = cell->prison;
+
+	hlist_del(&cell->list);
+
+	if (inmates)
+		bio_list_merge(inmates, &cell->bios);
+
+	mempool_free(cell, prison->cell_pool);
+}
+
+static void cell_release(struct cell *cell, struct bio_list *bios)
+{
+	unsigned long flags;
+	struct bio_prison *prison = cell->prison;
+
+	spin_lock_irqsave(&prison->lock, flags);
+	__cell_release(cell, bios);
+	spin_unlock_irqrestore(&prison->lock, flags);
+}
+
+/*
+ * There are a couple of places where we put a bio into a cell briefly
+ * before taking it out again.  In these situations we know that no other
+ * bio may be in the cell.  This function releases the cell, and also does
+ * a sanity check.
+ */
+static void cell_release_singleton(struct cell *cell, struct bio *bio)
+{
+	struct bio_prison *prison = cell->prison;
+	struct bio_list bios;
+	struct bio *b;
+	unsigned long flags;
+
+	bio_list_init(&bios);
+
+	spin_lock_irqsave(&prison->lock, flags);
+	__cell_release(cell, &bios);
+	spin_unlock_irqrestore(&prison->lock, flags);
+
+	b = bio_list_pop(&bios);
+	BUG_ON(b != bio);
+	BUG_ON(!bio_list_empty(&bios));
+}
+
+static void cell_error(struct cell *cell)
+{
+	struct bio_prison *prison = cell->prison;
+	struct bio_list bios;
+	struct bio *bio;
+	unsigned long flags;
+
+	bio_list_init(&bios);
+
+	spin_lock_irqsave(&prison->lock, flags);
+	__cell_release(cell, &bios);
+	spin_unlock_irqrestore(&prison->lock, flags);
+
+	while ((bio = bio_list_pop(&bios)))
+		bio_io_error(bio);
+}
+
+/*----------------------------------------------------------------*/
+
+/*
+ * We use the deferred set to keep track of pending reads to shared blocks.
+ * We do this to ensure the new mapping caused by a write isn't performed
+ * until these prior reads have completed.  Otherwise the insertion of the
+ * new mapping could free the old block that the read bios are mapped to.
+ */
+
+struct deferred_set;
+struct deferred_entry {
+	struct deferred_set *ds;
+	unsigned count;
+	struct list_head work_items;
+};
+
+struct deferred_set {
+	spinlock_t lock;
+	unsigned current_entry;
+	unsigned sweeper;
+	struct deferred_entry entries[DEFERRED_SET_SIZE];
+};
+
+static void ds_init(struct deferred_set *ds)
+{
+	int i;
+
+	spin_lock_init(&ds->lock);
+	ds->current_entry = 0;
+	ds->sweeper = 0;
+	for (i = 0; i < DEFERRED_SET_SIZE; i++) {
+		ds->entries[i].ds = ds;
+		ds->entries[i].count = 0;
+		INIT_LIST_HEAD(&ds->entries[i].work_items);
+	}
+}
+
+static struct deferred_entry *ds_inc(struct deferred_set *ds)
+{
+	unsigned long flags;
+	struct deferred_entry *entry;
+
+	spin_lock_irqsave(&ds->lock, flags);
+	entry = ds->entries + ds->current_entry;
+	entry->count++;
+	spin_unlock_irqrestore(&ds->lock, flags);
+
+	return entry;
+}
+
+static unsigned ds_next(unsigned index)
+{
+	return (index + 1) % DEFERRED_SET_SIZE;
+}
+
+static void __sweep(struct deferred_set *ds, struct list_head *head)
+{
+	while ((ds->sweeper != ds->current_entry) &&
+	       !ds->entries[ds->sweeper].count) {
+		list_splice_init(&ds->entries[ds->sweeper].work_items, head);
+		ds->sweeper = ds_next(ds->sweeper);
+	}
+
+	if ((ds->sweeper == ds->current_entry) && !ds->entries[ds->sweeper].count)
+		list_splice_init(&ds->entries[ds->sweeper].work_items, head);
+}
+
+static void ds_dec(struct deferred_entry *entry, struct list_head *head)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&entry->ds->lock, flags);
+	BUG_ON(!entry->count);
+	--entry->count;
+	__sweep(entry->ds, head);
+	spin_unlock_irqrestore(&entry->ds->lock, flags);
+}
+
+/*
+ * Returns 1 if deferred or 0 if no pending items to delay job.
+ */
+static int ds_add_work(struct deferred_set *ds, struct list_head *work)
+{
+	int r = 1;
+	unsigned long flags;
+	unsigned next_entry;
+
+	spin_lock_irqsave(&ds->lock, flags);
+	if ((ds->sweeper == ds->current_entry) &&
+	    !ds->entries[ds->current_entry].count)
+		r = 0;
+	else {
+		list_add(work, &ds->entries[ds->current_entry].work_items);
+		next_entry = ds_next(ds->current_entry);
+		if (!ds->entries[next_entry].count)
+			ds->current_entry = next_entry;
+	}
+	spin_unlock_irqrestore(&ds->lock, flags);
+
+	return r;
+}
+
+/*----------------------------------------------------------------*/
+
+/*
+ * Key building.
+ */
+static void build_data_key(struct dm_thin_device *td,
+			   dm_block_t b, struct cell_key *key)
+{
+	key->virtual = 0;
+	key->dev = dm_thin_dev_id(td);
+	key->block = b;
+}
+
+static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
+			      struct cell_key *key)
+{
+	key->virtual = 1;
+	key->dev = dm_thin_dev_id(td);
+	key->block = b;
+}
+
+/*----------------------------------------------------------------*/
+
+/*
+ * A pool device ties together a metadata device and a data device.  It
+ * also provides the interface for creating and destroying internal
+ * devices.
+ */
+struct new_mapping;
+struct pool {
+	struct list_head list;
+	struct dm_target *ti;	/* Only set if a pool target is bound */
+
+	struct mapped_device *pool_md;
+	struct block_device *md_dev;
+	struct dm_pool_metadata *pmd;
+
+	uint32_t sectors_per_block;
+	unsigned block_shift;
+	dm_block_t offset_mask;
+	dm_block_t low_water_blocks;
+
+	unsigned zero_new_blocks:1;
+	unsigned low_water_triggered:1;	/* A dm event has been sent */
+	unsigned no_free_space:1;	/* A -ENOSPC warning has been issued */
+
+	struct bio_prison *prison;
+	struct dm_kcopyd_client *copier;
+
+	struct workqueue_struct *wq;
+	struct work_struct worker;
+
+	unsigned ref_count;
+
+	spinlock_t lock;
+	struct bio_list deferred_bios;
+	struct bio_list deferred_flush_bios;
+	struct list_head prepared_mappings;
+
+	struct bio_list retry_on_resume_list;
+
+	struct deferred_set ds;	/* FIXME: move to thin_c */
+
+	struct new_mapping *next_mapping;
+	mempool_t *mapping_pool;
+	mempool_t *endio_hook_pool;
+};
+
+/*
+ * Target context for a pool.
+ */
+struct pool_c {
+	struct dm_target *ti;
+	struct pool *pool;
+	struct dm_dev *data_dev;
+	struct dm_dev *metadata_dev;
+	struct dm_target_callbacks callbacks;
+
+	dm_block_t low_water_blocks;
+	unsigned zero_new_blocks:1;
+};
+
+/*
+ * Target context for a thin.
+ */
+struct thin_c {
+	struct dm_dev *pool_dev;
+	dm_thin_id dev_id;
+
+	struct pool *pool;
+	struct dm_thin_device *td;
+};
+
+/*----------------------------------------------------------------*/
+
+/*
+ * A global list of pools that uses a struct mapped_device as a key.
+ */
+static struct dm_thin_pool_table {
+	struct mutex mutex;
+	struct list_head pools;
+} dm_thin_pool_table;
+
+static void pool_table_init(void)
+{
+	mutex_init(&dm_thin_pool_table.mutex);
+	INIT_LIST_HEAD(&dm_thin_pool_table.pools);
+}
+
+static void __pool_table_insert(struct pool *pool)
+{
+	BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
+	list_add(&pool->list, &dm_thin_pool_table.pools);
+}
+
+static void __pool_table_remove(struct pool *pool)
+{
+	BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
+	list_del(&pool->list);
+}
+
+static struct pool *__pool_table_lookup(struct mapped_device *md)
+{
+	struct pool *pool = NULL, *tmp;
+
+	BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
+
+	list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
+		if (tmp->pool_md == md) {
+			pool = tmp;
+			break;
+		}
+	}
+
+	return pool;
+}
+
+static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
+{
+	struct pool *pool = NULL, *tmp;
+
+	BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
+
+	list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
+		if (tmp->md_dev == md_dev) {
+			pool = tmp;
+			break;
+		}
+	}
+
+	return pool;
+}
+
+/*----------------------------------------------------------------*/
+
+static void __requeue_bio_list(struct thin_c *tc, struct bio_list *master)
+{
+	struct bio *bio;
+	struct bio_list bios;
+
+	bio_list_init(&bios);
+	bio_list_merge(&bios, master);
+	bio_list_init(master);
+
+	while ((bio = bio_list_pop(&bios))) {
+		if (dm_get_mapinfo(bio)->ptr == tc)
+			bio_endio(bio, DM_ENDIO_REQUEUE);
+		else
+			bio_list_add(master, bio);
+	}
+}
+
+static void requeue_io(struct thin_c *tc)
+{
+	struct pool *pool = tc->pool;
+	unsigned long flags;
+
+	spin_lock_irqsave(&pool->lock, flags);
+	__requeue_bio_list(tc, &pool->deferred_bios);
+	__requeue_bio_list(tc, &pool->retry_on_resume_list);
+	spin_unlock_irqrestore(&pool->lock, flags);
+}
+
+/*
+ * This section of code contains the logic for processing a thin device's IO.
+ * Much of the code depends on pool object resources (lists, workqueues, etc)
+ * but most is exclusively called from the thin target rather than the thin-pool
+ * target.
+ */
+
+static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
+{
+	return bio->bi_sector >> tc->pool->block_shift;
+}
+
+static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
+{
+	struct pool *pool = tc->pool;
+
+	bio->bi_bdev = tc->pool_dev->bdev;
+	bio->bi_sector = (block << pool->block_shift) +
+		(bio->bi_sector & pool->offset_mask);
+}
+
+static void remap_and_issue(struct thin_c *tc, struct bio *bio,
+			    dm_block_t block)
+{
+	struct pool *pool = tc->pool;
+	unsigned long flags;
+
+	remap(tc, bio, block);
+
+	/*
+	 * Batch together any FUA/FLUSH bios we find and then issue
+	 * a single commit for them in process_deferred_bios().
+	 */
+	if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
+		spin_lock_irqsave(&pool->lock, flags);
+		bio_list_add(&pool->deferred_flush_bios, bio);
+		spin_unlock_irqrestore(&pool->lock, flags);
+	} else
+		generic_make_request(bio);
+}
+
+/*
+ * wake_worker() is used when new work is queued and when pool_resume is
+ * ready to continue deferred IO processing.
+ */
+static void wake_worker(struct pool *pool)
+{
+	queue_work(pool->wq, &pool->worker);
+}
+
+/*----------------------------------------------------------------*/
+
+/*
+ * Bio endio functions.
+ */
+struct endio_hook {
+	struct thin_c *tc;
+	bio_end_io_t *saved_bi_end_io;
+	struct deferred_entry *entry;
+};
+
+struct new_mapping {
+	struct list_head list;
+
+	int prepared;
+
+	struct thin_c *tc;
+	dm_block_t virt_block;
+	dm_block_t data_block;
+	struct cell *cell;
+	int err;
+
+	/*
+	 * If the bio covers the whole area of a block then we can avoid
+	 * zeroing or copying.  Instead this bio is hooked.  The bio will
+	 * still be in the cell, so care has to be taken to avoid issuing
+	 * the bio twice.
+	 */
+	struct bio *bio;
+	bio_end_io_t *saved_bi_end_io;
+};
+
+static void __maybe_add_mapping(struct new_mapping *m)
+{
+	struct pool *pool = m->tc->pool;
+
+	if (list_empty(&m->list) && m->prepared) {
+		list_add(&m->list, &pool->prepared_mappings);
+		wake_worker(pool);
+	}
+}
+
+static void copy_complete(int read_err, unsigned long write_err, void *context)
+{
+	unsigned long flags;
+	struct new_mapping *m = context;
+	struct pool *pool = m->tc->pool;
+
+	m->err = read_err || write_err ? -EIO : 0;
+
+	spin_lock_irqsave(&pool->lock, flags);
+	m->prepared = 1;
+	__maybe_add_mapping(m);
+	spin_unlock_irqrestore(&pool->lock, flags);
+}
+
+static void overwrite_endio(struct bio *bio, int err)
+{
+	unsigned long flags;
+	struct new_mapping *m = dm_get_mapinfo(bio)->ptr;
+	struct pool *pool = m->tc->pool;
+
+	m->err = err;
+
+	spin_lock_irqsave(&pool->lock, flags);
+	m->prepared = 1;
+	__maybe_add_mapping(m);
+	spin_unlock_irqrestore(&pool->lock, flags);
+}
+
+static void shared_read_endio(struct bio *bio, int err)
+{
+	struct list_head mappings;
+	struct new_mapping *m, *tmp;
+	struct endio_hook *h = dm_get_mapinfo(bio)->ptr;
+	unsigned long flags;
+	struct pool *pool = h->tc->pool;
+
+	bio->bi_end_io = h->saved_bi_end_io;
+	bio_endio(bio, err);
+
+	INIT_LIST_HEAD(&mappings);
+	ds_dec(h->entry, &mappings);
+
+	spin_lock_irqsave(&pool->lock, flags);
+	list_for_each_entry_safe(m, tmp, &mappings, list) {
+		list_del(&m->list);
+		INIT_LIST_HEAD(&m->list);
+		__maybe_add_mapping(m);
+	}
+	spin_unlock_irqrestore(&pool->lock, flags);
+
+	mempool_free(h, pool->endio_hook_pool);
+}
+
+/*----------------------------------------------------------------*/
+
+/*
+ * Workqueue.
+ */
+
+/*
+ * Prepared mapping jobs.
+ */
+
+/*
+ * This sends the bios in the cell back to the deferred_bios list.
+ */
+static void cell_defer(struct thin_c *tc, struct cell *cell,
+		       dm_block_t data_block)
+{
+	struct pool *pool = tc->pool;
+	unsigned long flags;
+
+	spin_lock_irqsave(&pool->lock, flags);
+	cell_release(cell, &pool->deferred_bios);
+	spin_unlock_irqrestore(&tc->pool->lock, flags);
+
+	wake_worker(pool);
+}
+
+/*
+ * Same as cell_defer above, except it omits one particular detainee,
+ * a write bio that covers the block and has already been processed.
+ */
+static void cell_defer_except(struct thin_c *tc, struct cell *cell,
+			      struct bio *exception)
+{
+	struct bio_list bios;
+	struct bio *bio;
+	struct pool *pool = tc->pool;
+	unsigned long flags;
+
+	bio_list_init(&bios);
+	cell_release(cell, &bios);
+
+	spin_lock_irqsave(&pool->lock, flags);
+	while ((bio = bio_list_pop(&bios)))
+		if (bio != exception)
+			bio_list_add(&pool->deferred_bios, bio);
+	spin_unlock_irqrestore(&pool->lock, flags);
+
+	wake_worker(pool);
+}
+
+static void process_prepared_mapping(struct new_mapping *m)
+{
+	struct thin_c *tc = m->tc;
+	struct bio *bio;
+	int r;
+
+	bio = m->bio;
+	if (bio)
+		bio->bi_end_io = m->saved_bi_end_io;
+
+	if (m->err) {
+		cell_error(m->cell);
+		return;
+	}
+
+	/*
+	 * Commit the prepared block into the mapping btree.
+	 * Any I/O for this block arriving after this point will get
+	 * remapped to it directly.
+	 */
+	r = dm_thin_insert_block(tc->td, m->virt_block, m->data_block);
+	if (r) {
+		DMERR("dm_thin_insert_block() failed");
+		cell_error(m->cell);
+		return;
+	}
+
+	/*
+	 * Release any bios held while the block was being provisioned.
+	 * If we are processing a write bio that completely covers the block,
+	 * we already processed it so can ignore it now when processing
+	 * the bios in the cell.
+	 */
+	if (bio) {
+		cell_defer_except(tc, m->cell, bio);
+		bio_endio(bio, 0);
+	} else
+		cell_defer(tc, m->cell, m->data_block);
+
+	list_del(&m->list);
+	mempool_free(m, tc->pool->mapping_pool);
+}
+
+static void process_prepared_mappings(struct pool *pool)
+{
+	unsigned long flags;
+	struct list_head maps;
+	struct new_mapping *m, *tmp;
+
+	INIT_LIST_HEAD(&maps);
+	spin_lock_irqsave(&pool->lock, flags);
+	list_splice_init(&pool->prepared_mappings, &maps);
+	spin_unlock_irqrestore(&pool->lock, flags);
+
+	list_for_each_entry_safe(m, tmp, &maps, list)
+		process_prepared_mapping(m);
+}
+
+/*
+ * Deferred bio jobs.
+ */
+static int io_overwrites_block(struct pool *pool, struct bio *bio)
+{
+	return ((bio_data_dir(bio) == WRITE) &&
+		!(bio->bi_sector & pool->offset_mask)) &&
+		(bio->bi_size == (pool->sectors_per_block << SECTOR_SHIFT));
+}
+
+static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
+			       bio_end_io_t *fn)
+{
+	*save = bio->bi_end_io;
+	bio->bi_end_io = fn;
+}
+
+static int ensure_next_mapping(struct pool *pool)
+{
+	if (pool->next_mapping)
+		return 0;
+
+	pool->next_mapping = mempool_alloc(pool->mapping_pool, GFP_ATOMIC);
+
+	return pool->next_mapping ? 0 : -ENOMEM;
+}
+
+static struct new_mapping *get_next_mapping(struct pool *pool)
+{
+	struct new_mapping *r = pool->next_mapping;
+
+	BUG_ON(!pool->next_mapping);
+
+	pool->next_mapping = NULL;
+
+	return r;
+}
+
+static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
+			  dm_block_t data_origin, dm_block_t data_dest,
+			  struct cell *cell, struct bio *bio)
+{
+	int r;
+	struct pool *pool = tc->pool;
+	struct new_mapping *m = get_next_mapping(pool);
+
+	INIT_LIST_HEAD(&m->list);
+	m->prepared = 0;
+	m->tc = tc;
+	m->virt_block = virt_block;
+	m->data_block = data_dest;
+	m->cell = cell;
+	m->err = 0;
+	m->bio = NULL;
+
+	ds_add_work(&pool->ds, &m->list);
+
+	/*
+	 * IO to pool_dev remaps to the pool target's data_dev.
+	 *
+	 * If the whole block of data is being overwritten, we can issue the
+	 * bio immediately. Otherwise we use kcopyd to clone the data first.
+	 */
+	if (io_overwrites_block(pool, bio)) {
+		m->bio = bio;
+		save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
+		dm_get_mapinfo(bio)->ptr = m;
+		remap_and_issue(tc, bio, data_dest);
+	} else {
+		struct dm_io_region from, to;
+
+		from.bdev = tc->pool_dev->bdev;
+		from.sector = data_origin * pool->sectors_per_block;
+		from.count = pool->sectors_per_block;
+
+		to.bdev = tc->pool_dev->bdev;
+		to.sector = data_dest * pool->sectors_per_block;
+		to.count = pool->sectors_per_block;
+
+		r = dm_kcopyd_copy(pool->copier, &from, 1, &to,
+				   0, copy_complete, m);
+		if (r < 0) {
+			mempool_free(m, pool->mapping_pool);
+			DMERR("dm_kcopyd_copy() failed");
+			cell_error(cell);
+		}
+	}
+}
+
+static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
+			  dm_block_t data_block, struct cell *cell,
+			  struct bio *bio)
+{
+	struct pool *pool = tc->pool;
+	struct new_mapping *m = get_next_mapping(pool);
+
+	INIT_LIST_HEAD(&m->list);
+	m->prepared = 0;
+	m->tc = tc;
+	m->virt_block = virt_block;
+	m->data_block = data_block;
+	m->cell = cell;
+	m->err = 0;
+	m->bio = NULL;
+
+	/*
+	 * If the whole block of data is being overwritten or we are not
+	 * zeroing pre-existing data, we can issue the bio immediately.
+	 * Otherwise we use kcopyd to zero the data first.
+	 */
+	if (!pool->zero_new_blocks)
+		process_prepared_mapping(m);
+
+	else if (io_overwrites_block(pool, bio)) {
+		m->bio = bio;
+		save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
+		dm_get_mapinfo(bio)->ptr = m;
+		remap_and_issue(tc, bio, data_block);
+
+	} else {
+		int r;
+		struct dm_io_region to;
+
+		to.bdev = tc->pool_dev->bdev;
+		to.sector = data_block * pool->sectors_per_block;
+		to.count = pool->sectors_per_block;
+
+		r = dm_kcopyd_zero(pool->copier, 1, &to, 0, copy_complete, m);
+		if (r < 0) {
+			mempool_free(m, pool->mapping_pool);
+			DMERR("dm_kcopyd_zero() failed");
+			cell_error(cell);
+		}
+	}
+}
+
+static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
+{
+	int r;
+	dm_block_t free_blocks;
+	unsigned long flags;
+	struct pool *pool = tc->pool;
+
+	r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
+	if (r)
+		return r;
+
+	if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
+		DMWARN("%s: reached low water mark, sending event.",
+		       dm_device_name(pool->pool_md));
+		spin_lock_irqsave(&pool->lock, flags);
+		pool->low_water_triggered = 1;
+		spin_unlock_irqrestore(&pool->lock, flags);
+		dm_table_event(pool->ti->table);
+	}
+
+	if (!free_blocks) {
+		if (pool->no_free_space)
+			return -ENOSPC;
+		else {
+			/*
+			 * Try to commit to see if that will free up some
+			 * more space.
+			 */
+			r = dm_pool_commit_metadata(pool->pmd);
+			if (r) {
+				DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
+				      __func__, r);
+				return r;
+			}
+
+			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;
+			}
+		}
+	}
+
+	r = dm_pool_alloc_data_block(pool->pmd, result);
+	if (r)
+		return r;
+
+	return 0;
+}
+
+/*
+ * If we have run out of space, queue bios until the device is
+ * resumed, presumably after having been reloaded with more space.
+ */
+static void retry_on_resume(struct bio *bio)
+{
+	struct thin_c *tc = dm_get_mapinfo(bio)->ptr;
+	struct pool *pool = tc->pool;
+	unsigned long flags;
+
+	spin_lock_irqsave(&pool->lock, flags);
+	bio_list_add(&pool->retry_on_resume_list, bio);
+	spin_unlock_irqrestore(&pool->lock, flags);
+}
+
+static void no_space(struct cell *cell)
+{
+	struct bio *bio;
+	struct bio_list bios;
+
+	bio_list_init(&bios);
+	cell_release(cell, &bios);
+
+	while ((bio = bio_list_pop(&bios)))
+		retry_on_resume(bio);
+}
+
+static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
+			  struct cell_key *key,
+			  struct dm_thin_lookup_result *lookup_result,
+			  struct cell *cell)
+{
+	int r;
+	dm_block_t data_block;
+
+	r = alloc_data_block(tc, &data_block);
+	switch (r) {
+	case 0:
+		schedule_copy(tc, block, lookup_result->block,
+			      data_block, cell, bio);
+		break;
+
+	case -ENOSPC:
+		no_space(cell);
+		break;
+
+	default:
+		DMERR("%s: alloc_data_block() failed, error = %d", __func__, r);
+		cell_error(cell);
+		break;
+	}
+}
+
+static void process_shared_bio(struct thin_c *tc, struct bio *bio,
+			       dm_block_t block,
+			       struct dm_thin_lookup_result *lookup_result)
+{
+	struct cell *cell;
+	struct pool *pool = tc->pool;
+	struct cell_key key;
+
+	/*
+	 * If cell is already occupied, then sharing is already in the process
+	 * of being broken so we have nothing further to do here.
+	 */
+	build_data_key(tc->td, lookup_result->block, &key);
+	if (bio_detain(pool->prison, &key, bio, &cell))
+		return;
+
+	if (bio_data_dir(bio) == WRITE)
+		break_sharing(tc, bio, block, &key, lookup_result, cell);
+	else {
+		struct endio_hook *h;
+		h = mempool_alloc(pool->endio_hook_pool, GFP_NOIO);
+
+		h->tc = tc;
+		h->entry = ds_inc(&pool->ds);
+		save_and_set_endio(bio, &h->saved_bi_end_io, shared_read_endio);
+		dm_get_mapinfo(bio)->ptr = h;
+
+		cell_release_singleton(cell, bio);
+		remap_and_issue(tc, bio, lookup_result->block);
+	}
+}
+
+static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
+			    struct cell *cell)
+{
+	int r;
+	dm_block_t data_block;
+
+	/*
+	 * Remap empty bios (flushes) immediately, without provisioning.
+	 */
+	if (!bio->bi_size) {
+		cell_release_singleton(cell, bio);
+		remap_and_issue(tc, bio, 0);
+		return;
+	}
+
+	/*
+	 * Fill read bios with zeroes and complete them immediately.
+	 */
+	if (bio_data_dir(bio) == READ) {
+		zero_fill_bio(bio);
+		cell_release_singleton(cell, bio);
+		bio_endio(bio, 0);
+		return;
+	}
+
+	r = alloc_data_block(tc, &data_block);
+	switch (r) {
+	case 0:
+		schedule_zero(tc, block, data_block, cell, bio);
+		break;
+
+	case -ENOSPC:
+		no_space(cell);
+		break;
+
+	default:
+		DMERR("%s: alloc_data_block() failed, error = %d", __func__, r);
+		cell_error(cell);
+		break;
+	}
+}
+
+static void process_bio(struct thin_c *tc, struct bio *bio)
+{
+	int r;
+	dm_block_t block = get_bio_block(tc, bio);
+	struct cell *cell;
+	struct cell_key key;
+	struct dm_thin_lookup_result lookup_result;
+
+	/*
+	 * If cell is already occupied, then the block is already
+	 * being provisioned so we have nothing further to do here.
+	 */
+	build_virtual_key(tc->td, block, &key);
+	if (bio_detain(tc->pool->prison, &key, bio, &cell))
+		return;
+
+	r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
+	switch (r) {
+	case 0:
+		/*
+		 * We can release this cell now.  This thread is the only
+		 * one that puts bios into a cell, and we know there were
+		 * no preceding bios.
+		 */
+		/*
+		 * TODO: this will probably have to change when discard goes
+		 * back in.
+		 */
+		cell_release_singleton(cell, bio);
+
+		if (lookup_result.shared)
+			process_shared_bio(tc, bio, block, &lookup_result);
+		else
+			remap_and_issue(tc, bio, lookup_result.block);
+		break;
+
+	case -ENODATA:
+		provision_block(tc, bio, block, cell);
+		break;
+
+	default:
+		DMERR("dm_thin_find_block() failed, error = %d", r);
+		bio_io_error(bio);
+		break;
+	}
+}
+
+static void process_deferred_bios(struct pool *pool)
+{
+	unsigned long flags;
+	struct bio *bio;
+	struct bio_list bios;
+	int r;
+
+	bio_list_init(&bios);
+
+	spin_lock_irqsave(&pool->lock, flags);
+	bio_list_merge(&bios, &pool->deferred_bios);
+	bio_list_init(&pool->deferred_bios);
+	spin_unlock_irqrestore(&pool->lock, flags);
+
+	while ((bio = bio_list_pop(&bios))) {
+		struct thin_c *tc = dm_get_mapinfo(bio)->ptr;
+		/*
+		 * If we've got no free new_mapping structs, and processing
+		 * this bio might require one, we pause until there are some
+		 * prepared mappings to process.
+		 */
+		if (ensure_next_mapping(pool)) {
+			spin_lock_irqsave(&pool->lock, flags);
+			bio_list_merge(&pool->deferred_bios, &bios);
+			spin_unlock_irqrestore(&pool->lock, flags);
+
+			break;
+		}
+		process_bio(tc, bio);
+	}
+
+	/*
+	 * If there are any deferred flush bios, we must commit
+	 * the metadata before issuing them.
+	 */
+	bio_list_init(&bios);
+	spin_lock_irqsave(&pool->lock, flags);
+	bio_list_merge(&bios, &pool->deferred_flush_bios);
+	bio_list_init(&pool->deferred_flush_bios);
+	spin_unlock_irqrestore(&pool->lock, flags);
+
+	if (bio_list_empty(&bios))
+		return;
+
+	r = dm_pool_commit_metadata(pool->pmd);
+	if (r) {
+		DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
+		      __func__, r);
+		while ((bio = bio_list_pop(&bios)))
+			bio_io_error(bio);
+		return;
+	}
+
+	while ((bio = bio_list_pop(&bios)))
+		generic_make_request(bio);
+}
+
+static void do_worker(struct work_struct *ws)
+{
+	struct pool *pool = container_of(ws, struct pool, worker);
+
+	process_prepared_mappings(pool);
+	process_deferred_bios(pool);
+}
+
+/*----------------------------------------------------------------*/
+
+/*
+ * Mapping functions.
+ */
+
+/*
+ * Called only while mapping a thin bio to hand it over to the workqueue.
+ */
+static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
+{
+	unsigned long flags;
+	struct pool *pool = tc->pool;
+
+	spin_lock_irqsave(&pool->lock, flags);
+	bio_list_add(&pool->deferred_bios, bio);
+	spin_unlock_irqrestore(&pool->lock, flags);
+
+	wake_worker(pool);
+}
+
+/*
+ * Non-blocking function called from the thin target's map function.
+ */
+static int thin_bio_map(struct dm_target *ti, struct bio *bio,
+			union map_info *map_context)
+{
+	int r;
+	struct thin_c *tc = ti->private;
+	dm_block_t block = get_bio_block(tc, bio);
+	struct dm_thin_device *td = tc->td;
+	struct dm_thin_lookup_result result;
+
+	/*
+	 * Save the thin context for easy access from the deferred bio later.
+	 */
+	map_context->ptr = tc;
+
+	if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
+		thin_defer_bio(tc, bio);
+		return DM_MAPIO_SUBMITTED;
+	}
+
+	r = dm_thin_find_block(td, block, 0, &result);
+
+	/*
+	 * Note that we defer readahead too.
+	 */
+	switch (r) {
+	case 0:
+		if (unlikely(result.shared)) {
+			/*
+			 * We have a race condition here between the
+			 * result.shared value returned by the lookup and
+			 * snapshot creation, which may cause new
+			 * sharing.
+			 *
+			 * To avoid this always quiesce the origin before
+			 * taking the snap.  You want to do this anyway to
+			 * ensure a consistent application view
+			 * (i.e. lockfs).
+			 *
+			 * More distant ancestors are irrelevant. The
+			 * shared flag will be set in their case.
+			 */
+			thin_defer_bio(tc, bio);
+			r = DM_MAPIO_SUBMITTED;
+		} else {
+			remap(tc, bio, result.block);
+			r = DM_MAPIO_REMAPPED;
+		}
+		break;
+
+	case -ENODATA:
+		/*
+		 * In future, the failed dm_thin_find_block above could
+		 * provide the hint to load the metadata into cache.
+		 */
+	case -EWOULDBLOCK:
+		thin_defer_bio(tc, bio);
+		r = DM_MAPIO_SUBMITTED;
+		break;
+	}
+
+	return r;
+}
+
+static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
+{
+	int r;
+	unsigned long flags;
+	struct pool_c *pt = container_of(cb, struct pool_c, callbacks);
+
+	spin_lock_irqsave(&pt->pool->lock, flags);
+	r = !bio_list_empty(&pt->pool->retry_on_resume_list);
+	spin_unlock_irqrestore(&pt->pool->lock, flags);
+
+	if (!r) {
+		struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
+		r = bdi_congested(&q->backing_dev_info, bdi_bits);
+	}
+
+	return r;
+}
+
+static void __requeue_bios(struct pool *pool)
+{
+	bio_list_merge(&pool->deferred_bios, &pool->retry_on_resume_list);
+	bio_list_init(&pool->retry_on_resume_list);
+}
+
+/*----------------------------------------------------------------
+ * Binding of control targets to a pool object
+ *--------------------------------------------------------------*/
+static int bind_control_target(struct pool *pool, struct dm_target *ti)
+{
+	struct pool_c *pt = ti->private;
+
+	pool->ti = ti;
+	pool->low_water_blocks = pt->low_water_blocks;
+	pool->zero_new_blocks = pt->zero_new_blocks;
+
+	return 0;
+}
+
+static void unbind_control_target(struct pool *pool, struct dm_target *ti)
+{
+	if (pool->ti == ti)
+		pool->ti = NULL;
+}
+
+/*----------------------------------------------------------------
+ * Pool creation
+ *--------------------------------------------------------------*/
+static void __pool_destroy(struct pool *pool)
+{
+	__pool_table_remove(pool);
+
+	if (dm_pool_metadata_close(pool->pmd) < 0)
+		DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
+
+	prison_destroy(pool->prison);
+	dm_kcopyd_client_destroy(pool->copier);
+
+	if (pool->wq)
+		destroy_workqueue(pool->wq);
+
+	if (pool->next_mapping)
+		mempool_free(pool->next_mapping, pool->mapping_pool);
+	mempool_destroy(pool->mapping_pool);
+	mempool_destroy(pool->endio_hook_pool);
+	kfree(pool);
+}
+
+static struct pool *pool_create(struct mapped_device *pool_md,
+				struct block_device *metadata_dev,
+				unsigned long block_size, char **error)
+{
+	int r;
+	void *err_p;
+	struct pool *pool;
+	struct dm_pool_metadata *pmd;
+
+	pmd = dm_pool_metadata_open(metadata_dev, block_size);
+	if (IS_ERR(pmd)) {
+		*error = "Error creating metadata object";
+		return (struct pool *)pmd;
+	}
+
+	pool = kmalloc(sizeof(*pool), GFP_KERNEL);
+	if (!pool) {
+		*error = "Error allocating memory for pool";
+		err_p = ERR_PTR(-ENOMEM);
+		goto bad_pool;
+	}
+
+	pool->pmd = pmd;
+	pool->sectors_per_block = block_size;
+	pool->block_shift = ffs(block_size) - 1;
+	pool->offset_mask = block_size - 1;
+	pool->low_water_blocks = 0;
+	pool->zero_new_blocks = 1;
+	pool->prison = prison_create(PRISON_CELLS);
+	if (!pool->prison) {
+		*error = "Error creating pool's bio prison";
+		err_p = ERR_PTR(-ENOMEM);
+		goto bad_prison;
+	}
+
+	pool->copier = dm_kcopyd_client_create();
+	if (IS_ERR(pool->copier)) {
+		r = PTR_ERR(pool->copier);
+		*error = "Error creating pool's kcopyd client";
+		err_p = ERR_PTR(r);
+		goto bad_kcopyd_client;
+	}
+
+	/*
+	 * Create singlethreaded workqueue that will service all devices
+	 * that use this metadata.
+	 */
+	pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
+	if (!pool->wq) {
+		*error = "Error creating pool's workqueue";
+		err_p = ERR_PTR(-ENOMEM);
+		goto bad_wq;
+	}
+
+	INIT_WORK(&pool->worker, do_worker);
+	spin_lock_init(&pool->lock);
+	bio_list_init(&pool->deferred_bios);
+	bio_list_init(&pool->deferred_flush_bios);
+	INIT_LIST_HEAD(&pool->prepared_mappings);
+	pool->low_water_triggered = 0;
+	pool->no_free_space = 0;
+	bio_list_init(&pool->retry_on_resume_list);
+	ds_init(&pool->ds);
+
+	pool->next_mapping = NULL;
+	pool->mapping_pool =
+		mempool_create_kmalloc_pool(MAPPING_POOL_SIZE, sizeof(struct new_mapping));
+	if (!pool->mapping_pool) {
+		*error = "Error creating pool's mapping mempool";
+		err_p = ERR_PTR(-ENOMEM);
+		goto bad_mapping_pool;
+	}
+
+	pool->endio_hook_pool =
+		mempool_create_kmalloc_pool(ENDIO_HOOK_POOL_SIZE, sizeof(struct endio_hook));
+	if (!pool->endio_hook_pool) {
+		*error = "Error creating pool's endio_hook mempool";
+		err_p = ERR_PTR(-ENOMEM);
+		goto bad_endio_hook_pool;
+	}
+	pool->ref_count = 1;
+	pool->pool_md = pool_md;
+	pool->md_dev = metadata_dev;
+	__pool_table_insert(pool);
+
+	return pool;
+
+bad_endio_hook_pool:
+	mempool_destroy(pool->mapping_pool);
+bad_mapping_pool:
+	destroy_workqueue(pool->wq);
+bad_wq:
+	dm_kcopyd_client_destroy(pool->copier);
+bad_kcopyd_client:
+	prison_destroy(pool->prison);
+bad_prison:
+	kfree(pool);
+bad_pool:
+	if (dm_pool_metadata_close(pmd))
+		DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
+
+	return err_p;
+}
+
+static void __pool_inc(struct pool *pool)
+{
+	BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
+	pool->ref_count++;
+}
+
+static void __pool_dec(struct pool *pool)
+{
+	BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
+	BUG_ON(!pool->ref_count);
+	if (!--pool->ref_count)
+		__pool_destroy(pool);
+}
+
+static struct pool *__pool_find(struct mapped_device *pool_md,
+				struct block_device *metadata_dev,
+				unsigned long block_size, char **error)
+{
+	struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
+
+	if (pool) {
+		if (pool->pool_md != pool_md)
+			return ERR_PTR(-EBUSY);
+		__pool_inc(pool);
+
+	} else {
+		pool = __pool_table_lookup(pool_md);
+		if (pool) {
+			if (pool->md_dev != metadata_dev)
+				return ERR_PTR(-EINVAL);
+			__pool_inc(pool);
+
+		} else
+			pool = pool_create(pool_md, metadata_dev, block_size, error);
+	}
+
+	return pool;
+}
+
+/*----------------------------------------------------------------
+ * Pool target methods
+ *--------------------------------------------------------------*/
+static void pool_dtr(struct dm_target *ti)
+{
+	struct pool_c *pt = ti->private;
+
+	mutex_lock(&dm_thin_pool_table.mutex);
+
+	unbind_control_target(pt->pool, ti);
+	__pool_dec(pt->pool);
+	dm_put_device(ti, pt->metadata_dev);
+	dm_put_device(ti, pt->data_dev);
+	kfree(pt);
+
+	mutex_unlock(&dm_thin_pool_table.mutex);
+}
+
+struct pool_features {
+	unsigned zero_new_blocks:1;
+};
+
+static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
+			       struct dm_target *ti)
+{
+	int r;
+	unsigned argc;
+	const char *arg_name;
+
+	static struct dm_arg _args[] = {
+		{0, 1, "Invalid number of pool feature arguments"},
+	};
+
+	/*
+	 * No feature arguments supplied.
+	 */
+	if (!as->argc)
+		return 0;
+
+	r = dm_read_arg_group(_args, as, &argc, &ti->error);
+	if (r)
+		return -EINVAL;
+
+	while (argc && !r) {
+		arg_name = dm_shift_arg(as);
+		argc--;
+
+		if (!strcasecmp(arg_name, "skip_block_zeroing")) {
+			pf->zero_new_blocks = 0;
+			continue;
+		}
+
+		ti->error = "Unrecognised pool feature requested";
+		r = -EINVAL;
+	}
+
+	return r;
+}
+
+/*
+ * thin-pool <metadata dev> <data dev>
+ *	     <data block size (sectors)>
+ *	     <low water mark (blocks)>
+ *	     [<#feature args> [<arg>]*]
+ *
+ * Optional feature arguments are:
+ *	     skip_block_zeroing: skips the zeroing of newly-provisioned blocks.
+ */
+static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
+{
+	int r;
+	struct pool_c *pt;
+	struct pool *pool;
+	struct pool_features pf;
+	struct dm_arg_set as;
+	struct dm_dev *data_dev;
+	unsigned long block_size;
+	dm_block_t low_water_blocks;
+	struct dm_dev *metadata_dev;
+	sector_t metadata_dev_size;
+
+	/*
+	 * FIXME Remove validation from scope of lock.
+	 */
+	mutex_lock(&dm_thin_pool_table.mutex);
+
+	if (argc < 4) {
+		ti->error = "Invalid argument count";
+		r = -EINVAL;
+		goto out_unlock;
+	}
+	as.argc = argc;
+	as.argv = argv;
+
+	r = dm_get_device(ti, argv[0], FMODE_READ | FMODE_WRITE, &metadata_dev);
+	if (r) {
+		ti->error = "Error opening metadata block device";
+		goto out_unlock;
+	}
+
+	metadata_dev_size = i_size_read(metadata_dev->bdev->bd_inode) >> SECTOR_SHIFT;
+	if (metadata_dev_size > METADATA_DEV_MAX_SECTORS) {
+		ti->error = "Metadata device is too large";
+		r = -EINVAL;
+		goto out_metadata;
+	}
+
+	r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
+	if (r) {
+		ti->error = "Error getting data device";
+		goto out_metadata;
+	}
+
+	if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
+	    block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
+	    block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
+	    !is_power_of_2(block_size)) {
+		ti->error = "Invalid block size";
+		r = -EINVAL;
+		goto out;
+	}
+
+	if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
+		ti->error = "Invalid low water mark";
+		r = -EINVAL;
+		goto out;
+	}
+
+	/*
+	 * Set default pool features.
+	 */
+	memset(&pf, 0, sizeof(pf));
+	pf.zero_new_blocks = 1;
+
+	dm_consume_args(&as, 4);
+	r = parse_pool_features(&as, &pf, ti);
+	if (r)
+		goto out;
+
+	pt = kzalloc(sizeof(*pt), GFP_KERNEL);
+	if (!pt) {
+		r = -ENOMEM;
+		goto out;
+	}
+
+	pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
+			   block_size, &ti->error);
+	if (IS_ERR(pool)) {
+		r = PTR_ERR(pool);
+		goto out_free_pt;
+	}
+
+	pt->pool = pool;
+	pt->ti = ti;
+	pt->metadata_dev = metadata_dev;
+	pt->data_dev = data_dev;
+	pt->low_water_blocks = low_water_blocks;
+	pt->zero_new_blocks = pf.zero_new_blocks;
+	ti->num_flush_requests = 1;
+	ti->num_discard_requests = 0;
+	ti->private = pt;
+
+	pt->callbacks.congested_fn = pool_is_congested;
+	dm_table_add_target_callbacks(ti->table, &pt->callbacks);
+
+	mutex_unlock(&dm_thin_pool_table.mutex);
+
+	return 0;
+
+out_free_pt:
+	kfree(pt);
+out:
+	dm_put_device(ti, data_dev);
+out_metadata:
+	dm_put_device(ti, metadata_dev);
+out_unlock:
+	mutex_unlock(&dm_thin_pool_table.mutex);
+
+	return r;
+}
+
+static int pool_map(struct dm_target *ti, struct bio *bio,
+		    union map_info *map_context)
+{
+	int r;
+	struct pool_c *pt = ti->private;
+	struct pool *pool = pt->pool;
+	unsigned long flags;
+
+	/*
+	 * As this is a singleton target, ti->begin is always zero.
+	 */
+	spin_lock_irqsave(&pool->lock, flags);
+	bio->bi_bdev = pt->data_dev->bdev;
+	r = DM_MAPIO_REMAPPED;
+	spin_unlock_irqrestore(&pool->lock, flags);
+
+	return r;
+}
+
+/*
+ * Retrieves the number of blocks of the data device from
+ * the superblock and compares it to the actual device size,
+ * thus resizing the data device in case it has grown.
+ *
+ * This both copes with opening preallocated data devices in the ctr
+ * being followed by a resume
+ * -and-
+ * calling the resume method individually after userspace has
+ * grown the data device in reaction to a table event.
+ */
+static int pool_preresume(struct dm_target *ti)
+{
+	int r;
+	struct pool_c *pt = ti->private;
+	struct pool *pool = pt->pool;
+	dm_block_t data_size, sb_data_size;
+
+	/*
+	 * Take control of the pool object.
+	 */
+	r = bind_control_target(pool, ti);
+	if (r)
+		return r;
+
+	data_size = ti->len >> pool->block_shift;
+	r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
+	if (r) {
+		DMERR("failed to retrieve data device size");
+		return r;
+	}
+
+	if (data_size < sb_data_size) {
+		DMERR("pool target too small, is %llu blocks (expected %llu)",
+		      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");
+			return r;
+		}
+
+		r = dm_pool_commit_metadata(pool->pmd);
+		if (r) {
+			DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
+			      __func__, r);
+			return r;
+		}
+	}
+
+	return 0;
+}
+
+static void pool_resume(struct dm_target *ti)
+{
+	struct pool_c *pt = ti->private;
+	struct pool *pool = pt->pool;
+	unsigned long flags;
+
+	spin_lock_irqsave(&pool->lock, flags);
+	pool->low_water_triggered = 0;
+	pool->no_free_space = 0;
+	__requeue_bios(pool);
+	spin_unlock_irqrestore(&pool->lock, flags);
+
+	wake_worker(pool);
+}
+
+static void pool_postsuspend(struct dm_target *ti)
+{
+	int r;
+	struct pool_c *pt = ti->private;
+	struct pool *pool = pt->pool;
+
+	flush_workqueue(pool->wq);
+
+	r = dm_pool_commit_metadata(pool->pmd);
+	if (r < 0) {
+		DMERR("%s: dm_pool_commit_metadata() failed, error = %d",
+		      __func__, r);
+		/* FIXME: invalidate device? error the next FUA or FLUSH bio ?*/
+	}
+}
+
+static int check_arg_count(unsigned argc, unsigned args_required)
+{
+	if (argc != args_required) {
+		DMWARN("Message received with %u arguments instead of %u.",
+		       argc, args_required);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
+{
+	if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
+	    *dev_id <= MAX_DEV_ID)
+		return 0;
+
+	if (warning)
+		DMWARN("Message received with invalid device id: %s", arg);
+
+	return -EINVAL;
+}
+
+static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
+{
+	dm_thin_id dev_id;
+	int r;
+
+	r = check_arg_count(argc, 2);
+	if (r)
+		return r;
+
+	r = read_dev_id(argv[1], &dev_id, 1);
+	if (r)
+		return r;
+
+	r = dm_pool_create_thin(pool->pmd, dev_id);
+	if (r) {
+		DMWARN("Creation of new thinly-provisioned device with id %s failed.",
+		       argv[1]);
+		return r;
+	}
+
+	return 0;
+}
+
+static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
+{
+	dm_thin_id dev_id;
+	dm_thin_id origin_dev_id;
+	int r;
+
+	r = check_arg_count(argc, 3);
+	if (r)
+		return r;
+
+	r = read_dev_id(argv[1], &dev_id, 1);
+	if (r)
+		return r;
+
+	r = read_dev_id(argv[2], &origin_dev_id, 1);
+	if (r)
+		return r;
+
+	r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
+	if (r) {
+		DMWARN("Creation of new snapshot %s of device %s failed.",
+		       argv[1], argv[2]);
+		return r;
+	}
+
+	return 0;
+}
+
+static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
+{
+	dm_thin_id dev_id;
+	int r;
+
+	r = check_arg_count(argc, 2);
+	if (r)
+		return r;
+
+	r = read_dev_id(argv[1], &dev_id, 1);
+	if (r)
+		return r;
+
+	r = dm_pool_delete_thin_device(pool->pmd, dev_id);
+	if (r)
+		DMWARN("Deletion of thin device %s failed.", argv[1]);
+
+	return r;
+}
+
+static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
+{
+	dm_thin_id old_id, new_id;
+	int r;
+
+	r = check_arg_count(argc, 3);
+	if (r)
+		return r;
+
+	if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
+		DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
+		return -EINVAL;
+	}
+
+	if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
+		DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
+		return -EINVAL;
+	}
+
+	r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
+	if (r) {
+		DMWARN("Failed to change transaction id from %s to %s.",
+		       argv[1], argv[2]);
+		return r;
+	}
+
+	return 0;
+}
+
+/*
+ * Messages supported:
+ *   create_thin	<dev_id>
+ *   create_snap	<dev_id> <origin_id>
+ *   delete		<dev_id>
+ *   trim		<dev_id> <new_size_in_sectors>
+ *   set_transaction_id <current_trans_id> <new_trans_id>
+ */
+static int pool_message(struct dm_target *ti, unsigned argc, char **argv)
+{
+	int r = -EINVAL;
+	struct pool_c *pt = ti->private;
+	struct pool *pool = pt->pool;
+
+	if (!strcasecmp(argv[0], "create_thin"))
+		r = process_create_thin_mesg(argc, argv, pool);
+
+	else if (!strcasecmp(argv[0], "create_snap"))
+		r = process_create_snap_mesg(argc, argv, pool);
+
+	else if (!strcasecmp(argv[0], "delete"))
+		r = process_delete_mesg(argc, argv, pool);
+
+	else if (!strcasecmp(argv[0], "set_transaction_id"))
+		r = process_set_transaction_id_mesg(argc, argv, pool);
+
+	else
+		DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
+
+	if (!r) {
+		r = dm_pool_commit_metadata(pool->pmd);
+		if (r)
+			DMERR("%s message: dm_pool_commit_metadata() failed, error = %d",
+			      argv[0], r);
+	}
+
+	return r;
+}
+
+/*
+ * Status line is:
+ *    <transaction id> <used metadata sectors>/<total metadata sectors>
+ *    <used data sectors>/<total data sectors> <held metadata root>
+ */
+static int pool_status(struct dm_target *ti, status_type_t type,
+		       char *result, unsigned maxlen)
+{
+	int r;
+	unsigned sz = 0;
+	uint64_t transaction_id;
+	dm_block_t nr_free_blocks_data;
+	dm_block_t nr_free_blocks_metadata;
+	dm_block_t nr_blocks_data;
+	dm_block_t nr_blocks_metadata;
+	dm_block_t held_root;
+	char buf[BDEVNAME_SIZE];
+	char buf2[BDEVNAME_SIZE];
+	struct pool_c *pt = ti->private;
+	struct pool *pool = pt->pool;
+
+	switch (type) {
+	case STATUSTYPE_INFO:
+		r = dm_pool_get_metadata_transaction_id(pool->pmd,
+							&transaction_id);
+		if (r)
+			return r;
+
+		r = dm_pool_get_free_metadata_block_count(pool->pmd,
+							  &nr_free_blocks_metadata);
+		if (r)
+			return r;
+
+		r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
+		if (r)
+			return r;
+
+		r = dm_pool_get_free_block_count(pool->pmd,
+						 &nr_free_blocks_data);
+		if (r)
+			return r;
+
+		r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
+		if (r)
+			return r;
+
+		r = dm_pool_get_held_metadata_root(pool->pmd, &held_root);
+		if (r)
+			return r;
+
+		DMEMIT("%llu %llu/%llu %llu/%llu ",
+		       (unsigned long long)transaction_id,
+		       (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
+		       (unsigned long long)nr_blocks_metadata,
+		       (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
+		       (unsigned long long)nr_blocks_data);
+
+		if (held_root)
+			DMEMIT("%llu", held_root);
+		else
+			DMEMIT("-");
+
+		break;
+
+	case STATUSTYPE_TABLE:
+		DMEMIT("%s %s %lu %llu ",
+		       format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
+		       format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
+		       (unsigned long)pool->sectors_per_block,
+		       (unsigned long long)pt->low_water_blocks);
+
+		DMEMIT("%u ", !pool->zero_new_blocks);
+
+		if (!pool->zero_new_blocks)
+			DMEMIT("skip_block_zeroing ");
+		break;
+	}
+
+	return 0;
+}
+
+static int pool_iterate_devices(struct dm_target *ti,
+				iterate_devices_callout_fn fn, void *data)
+{
+	struct pool_c *pt = ti->private;
+
+	return fn(ti, pt->data_dev, 0, ti->len, data);
+}
+
+static int pool_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
+		      struct bio_vec *biovec, int max_size)
+{
+	struct pool_c *pt = ti->private;
+	struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
+
+	if (!q->merge_bvec_fn)
+		return max_size;
+
+	bvm->bi_bdev = pt->data_dev->bdev;
+
+	return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
+}
+
+static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
+{
+	struct pool_c *pt = ti->private;
+	struct pool *pool = pt->pool;
+
+	blk_limits_io_min(limits, 0);
+	blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
+}
+
+static struct target_type pool_target = {
+	.name = "thin-pool",
+	.features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
+		    DM_TARGET_IMMUTABLE,
+	.version = {1, 0, 0},
+	.module = THIS_MODULE,
+	.ctr = pool_ctr,
+	.dtr = pool_dtr,
+	.map = pool_map,
+	.postsuspend = pool_postsuspend,
+	.preresume = pool_preresume,
+	.resume = pool_resume,
+	.message = pool_message,
+	.status = pool_status,
+	.merge = pool_merge,
+	.iterate_devices = pool_iterate_devices,
+	.io_hints = pool_io_hints,
+};
+
+/*----------------------------------------------------------------
+ * Thin target methods
+ *--------------------------------------------------------------*/
+static void thin_dtr(struct dm_target *ti)
+{
+	struct thin_c *tc = ti->private;
+
+	mutex_lock(&dm_thin_pool_table.mutex);
+
+	__pool_dec(tc->pool);
+	dm_pool_close_thin_device(tc->td);
+	dm_put_device(ti, tc->pool_dev);
+	kfree(tc);
+
+	mutex_unlock(&dm_thin_pool_table.mutex);
+}
+
+/*
+ * Thin target parameters:
+ *
+ * <pool_dev> <dev_id>
+ *
+ * pool_dev: the path to the pool (eg, /dev/mapper/my_pool)
+ * dev_id: the internal device identifier
+ */
+static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
+{
+	int r;
+	struct thin_c *tc;
+	struct dm_dev *pool_dev;
+	struct mapped_device *pool_md;
+
+	mutex_lock(&dm_thin_pool_table.mutex);
+
+	if (argc != 2) {
+		ti->error = "Invalid argument count";
+		r = -EINVAL;
+		goto out_unlock;
+	}
+
+	tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
+	if (!tc) {
+		ti->error = "Out of memory";
+		r = -ENOMEM;
+		goto out_unlock;
+	}
+
+	r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
+	if (r) {
+		ti->error = "Error opening pool device";
+		goto bad_pool_dev;
+	}
+	tc->pool_dev = pool_dev;
+
+	if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
+		ti->error = "Invalid device id";
+		r = -EINVAL;
+		goto bad_common;
+	}
+
+	pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
+	if (!pool_md) {
+		ti->error = "Couldn't get pool mapped device";
+		r = -EINVAL;
+		goto bad_common;
+	}
+
+	tc->pool = __pool_table_lookup(pool_md);
+	if (!tc->pool) {
+		ti->error = "Couldn't find pool object";
+		r = -EINVAL;
+		goto bad_pool_lookup;
+	}
+	__pool_inc(tc->pool);
+
+	r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
+	if (r) {
+		ti->error = "Couldn't open thin internal device";
+		goto bad_thin_open;
+	}
+
+	ti->split_io = tc->pool->sectors_per_block;
+	ti->num_flush_requests = 1;
+	ti->num_discard_requests = 0;
+	ti->discards_supported = 0;
+
+	dm_put(pool_md);
+
+	mutex_unlock(&dm_thin_pool_table.mutex);
+
+	return 0;
+
+bad_thin_open:
+	__pool_dec(tc->pool);
+bad_pool_lookup:
+	dm_put(pool_md);
+bad_common:
+	dm_put_device(ti, tc->pool_dev);
+bad_pool_dev:
+	kfree(tc);
+out_unlock:
+	mutex_unlock(&dm_thin_pool_table.mutex);
+
+	return r;
+}
+
+static int thin_map(struct dm_target *ti, struct bio *bio,
+		    union map_info *map_context)
+{
+	bio->bi_sector -= ti->begin;
+
+	return thin_bio_map(ti, bio, map_context);
+}
+
+static void thin_postsuspend(struct dm_target *ti)
+{
+	if (dm_noflush_suspending(ti))
+		requeue_io((struct thin_c *)ti->private);
+}
+
+/*
+ * <nr mapped sectors> <highest mapped sector>
+ */
+static int thin_status(struct dm_target *ti, status_type_t type,
+		       char *result, unsigned maxlen)
+{
+	int r;
+	ssize_t sz = 0;
+	dm_block_t mapped, highest;
+	char buf[BDEVNAME_SIZE];
+	struct thin_c *tc = ti->private;
+
+	if (!tc->td)
+		DMEMIT("-");
+	else {
+		switch (type) {
+		case STATUSTYPE_INFO:
+			r = dm_thin_get_mapped_count(tc->td, &mapped);
+			if (r)
+				return r;
+
+			r = dm_thin_get_highest_mapped_block(tc->td, &highest);
+			if (r < 0)
+				return r;
+
+			DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
+			if (r)
+				DMEMIT("%llu", ((highest + 1) *
+						tc->pool->sectors_per_block) - 1);
+			else
+				DMEMIT("-");
+			break;
+
+		case STATUSTYPE_TABLE:
+			DMEMIT("%s %lu",
+			       format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
+			       (unsigned long) tc->dev_id);
+			break;
+		}
+	}
+
+	return 0;
+}
+
+static int thin_iterate_devices(struct dm_target *ti,
+				iterate_devices_callout_fn fn, void *data)
+{
+	dm_block_t blocks;
+	struct thin_c *tc = ti->private;
+
+	/*
+	 * We can't call dm_pool_get_data_dev_size() since that blocks.  So
+	 * we follow a more convoluted path through to the pool's target.
+	 */
+	if (!tc->pool->ti)
+		return 0;	/* nothing is bound */
+
+	blocks = tc->pool->ti->len >> tc->pool->block_shift;
+	if (blocks)
+		return fn(ti, tc->pool_dev, 0, tc->pool->sectors_per_block * blocks, data);
+
+	return 0;
+}
+
+static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits)
+{
+	struct thin_c *tc = ti->private;
+
+	blk_limits_io_min(limits, 0);
+	blk_limits_io_opt(limits, tc->pool->sectors_per_block << SECTOR_SHIFT);
+}
+
+static struct target_type thin_target = {
+	.name = "thin",
+	.version = {1, 0, 0},
+	.module	= THIS_MODULE,
+	.ctr = thin_ctr,
+	.dtr = thin_dtr,
+	.map = thin_map,
+	.postsuspend = thin_postsuspend,
+	.status = thin_status,
+	.iterate_devices = thin_iterate_devices,
+	.io_hints = thin_io_hints,
+};
+
+/*----------------------------------------------------------------*/
+
+static int __init dm_thin_init(void)
+{
+	int r;
+
+	pool_table_init();
+
+	r = dm_register_target(&thin_target);
+	if (r)
+		return r;
+
+	r = dm_register_target(&pool_target);
+	if (r)
+		dm_unregister_target(&thin_target);
+
+	return r;
+}
+
+static void dm_thin_exit(void)
+{
+	dm_unregister_target(&thin_target);
+	dm_unregister_target(&pool_target);
+}
+
+module_init(dm_thin_init);
+module_exit(dm_thin_exit);
+
+MODULE_DESCRIPTION(DM_NAME "device-mapper thin provisioning target");
+MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
+MODULE_LICENSE("GPL");