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>

1 files changed, 1391 insertions, 0 deletions

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;
+}