dm: add era target

dm-era is a target that behaves similar to the linear target.  In
addition it keeps track of which blocks were written within a user
defined period of time called an 'era'.  Each era target instance
maintains the current era as a monotonically increasing 32-bit
counter.

Use cases include tracking changed blocks for backup software, and
partially invalidating the contents of a cache to restore cache
coherency after rolling back a vendor snapshot.

dm-era is primarily expected to be paired with the dm-cache target.

Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

3 files changed, 1743 insertions, 0 deletions

diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig
index 95ad936e604..5bdedf6df15 100644
--- a/drivers/md/Kconfig
+++ b/drivers/md/Kconfig
@@ -285,6 +285,17 @@ config DM_CACHE_CLEANER
          A simple cache policy that writes back all data to the
          origin.  Used when decommissioning a dm-cache.
 
+config DM_ERA
+       tristate "Era target (EXPERIMENTAL)"
+       depends on BLK_DEV_DM
+       default n
+       select DM_PERSISTENT_DATA
+       select DM_BIO_PRISON
+       ---help---
+         dm-era tracks which parts of a block device are written to
+         over time.  Useful for maintaining cache coherency when using
+         vendor snapshots.
+
 config DM_MIRROR
        tristate "Mirror target"
        depends on BLK_DEV_DM
diff --git a/drivers/md/Makefile b/drivers/md/Makefile
index f26d8329257..a2da532b1c2 100644
--- a/drivers/md/Makefile
+++ b/drivers/md/Makefile
@@ -14,6 +14,7 @@ dm-thin-pool-y	+= dm-thin.o dm-thin-metadata.o
 dm-cache-y	+= dm-cache-target.o dm-cache-metadata.o dm-cache-policy.o
 dm-cache-mq-y   += dm-cache-policy-mq.o
 dm-cache-cleaner-y += dm-cache-policy-cleaner.o
+dm-era-y	+= dm-era-target.o
 md-mod-y	+= md.o bitmap.o
 raid456-y	+= raid5.o
 
@@ -53,6 +54,7 @@ obj-$(CONFIG_DM_VERITY)		+= dm-verity.o
 obj-$(CONFIG_DM_CACHE)		+= dm-cache.o
 obj-$(CONFIG_DM_CACHE_MQ)	+= dm-cache-mq.o
 obj-$(CONFIG_DM_CACHE_CLEANER)	+= dm-cache-cleaner.o
+obj-$(CONFIG_DM_ERA)		+= dm-era.o
 
 ifeq ($(CONFIG_DM_UEVENT),y)
 dm-mod-objs			+= dm-uevent.o
diff --git a/drivers/md/dm-era-target.c b/drivers/md/dm-era-target.c
new file mode 100644
index 00000000000..03d9560bfd9
--- /dev/null
+++ b/drivers/md/dm-era-target.c
@@ -0,0 +1,1730 @@
+#include "dm.h"
+#include "persistent-data/dm-transaction-manager.h"
+#include "persistent-data/dm-bitset.h"
+#include "persistent-data/dm-space-map.h"
+
+#include <linux/dm-io.h>
+#include <linux/dm-kcopyd.h>
+#include <linux/init.h>
+#include <linux/mempool.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+
+#define DM_MSG_PREFIX "era"
+
+#define SUPERBLOCK_LOCATION 0
+#define SUPERBLOCK_MAGIC 2126579579
+#define SUPERBLOCK_CSUM_XOR 146538381
+#define MIN_ERA_VERSION 1
+#define MAX_ERA_VERSION 1
+#define INVALID_WRITESET_ROOT SUPERBLOCK_LOCATION
+#define MIN_BLOCK_SIZE 8
+
+/*----------------------------------------------------------------
+ * Writeset
+ *--------------------------------------------------------------*/
+struct writeset_metadata {
+	uint32_t nr_bits;
+	dm_block_t root;
+};
+
+struct writeset {
+	struct writeset_metadata md;
+
+	/*
+	 * An in core copy of the bits to save constantly doing look ups on
+	 * disk.
+	 */
+	unsigned long *bits;
+};
+
+/*
+ * This does not free off the on disk bitset as this will normally be done
+ * after digesting into the era array.
+ */
+static void writeset_free(struct writeset *ws)
+{
+	vfree(ws->bits);
+}
+
+static int setup_on_disk_bitset(struct dm_disk_bitset *info,
+				unsigned nr_bits, dm_block_t *root)
+{
+	int r;
+
+	r = dm_bitset_empty(info, root);
+	if (r)
+		return r;
+
+	return dm_bitset_resize(info, *root, 0, nr_bits, false, root);
+}
+
+static size_t bitset_size(unsigned nr_bits)
+{
+	return sizeof(unsigned long) * dm_div_up(nr_bits, BITS_PER_LONG);
+}
+
+/*
+ * Allocates memory for the in core bitset.
+ */
+static int writeset_alloc(struct writeset *ws, dm_block_t nr_blocks)
+{
+	ws->md.nr_bits = nr_blocks;
+	ws->md.root = INVALID_WRITESET_ROOT;
+	ws->bits = vzalloc(bitset_size(nr_blocks));
+	if (!ws->bits) {
+		DMERR("%s: couldn't allocate in memory bitset", __func__);
+		return -ENOMEM;
+	}
+
+	return 0;
+}
+
+/*
+ * Wipes the in-core bitset, and creates a new on disk bitset.
+ */
+static int writeset_init(struct dm_disk_bitset *info, struct writeset *ws)
+{
+	int r;
+
+	memset(ws->bits, 0, bitset_size(ws->md.nr_bits));
+
+	r = setup_on_disk_bitset(info, ws->md.nr_bits, &ws->md.root);
+	if (r) {
+		DMERR("%s: setup_on_disk_bitset failed", __func__);
+		return r;
+	}
+
+	return 0;
+}
+
+static bool writeset_marked(struct writeset *ws, dm_block_t block)
+{
+	return test_bit(block, ws->bits);
+}
+
+static int writeset_marked_on_disk(struct dm_disk_bitset *info,
+				   struct writeset_metadata *m, dm_block_t block,
+				   bool *result)
+{
+	dm_block_t old = m->root;
+
+	/*
+	 * The bitset was flushed when it was archived, so we know there'll
+	 * be no change to the root.
+	 */
+	int r = dm_bitset_test_bit(info, m->root, block, &m->root, result);
+	if (r) {
+		DMERR("%s: dm_bitset_test_bit failed", __func__);
+		return r;
+	}
+
+	BUG_ON(m->root != old);
+
+	return r;
+}
+
+/*
+ * Returns < 0 on error, 0 if the bit wasn't previously set, 1 if it was.
+ */
+static int writeset_test_and_set(struct dm_disk_bitset *info,
+				 struct writeset *ws, uint32_t block)
+{
+	int r;
+
+	if (!test_and_set_bit(block, ws->bits)) {
+		r = dm_bitset_set_bit(info, ws->md.root, block, &ws->md.root);
+		if (r) {
+			/* FIXME: fail mode */
+			return r;
+		}
+
+		return 0;
+	}
+
+	return 1;
+}
+
+/*----------------------------------------------------------------
+ * On disk metadata layout
+ *--------------------------------------------------------------*/
+#define SPACE_MAP_ROOT_SIZE 128
+#define UUID_LEN 16
+
+struct writeset_disk {
+	__le32 nr_bits;
+	__le64 root;
+} __packed;
+
+struct superblock_disk {
+	__le32 csum;
+	__le32 flags;
+	__le64 blocknr;
+
+	__u8 uuid[UUID_LEN];
+	__le64 magic;
+	__le32 version;
+
+	__u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
+
+	__le32 data_block_size;
+	__le32 metadata_block_size;
+	__le32 nr_blocks;
+
+	__le32 current_era;
+	struct writeset_disk current_writeset;
+
+	/*
+	 * Only these two fields are valid within the metadata snapshot.
+	 */
+	__le64 writeset_tree_root;
+	__le64 era_array_root;
+
+	__le64 metadata_snap;
+} __packed;
+
+/*----------------------------------------------------------------
+ * Superblock validation
+ *--------------------------------------------------------------*/
+static void sb_prepare_for_write(struct dm_block_validator *v,
+				 struct dm_block *b,
+				 size_t sb_block_size)
+{
+	struct superblock_disk *disk = dm_block_data(b);
+
+	disk->blocknr = cpu_to_le64(dm_block_location(b));
+	disk->csum = cpu_to_le32(dm_bm_checksum(&disk->flags,
+						sb_block_size - sizeof(__le32),
+						SUPERBLOCK_CSUM_XOR));
+}
+
+static int check_metadata_version(struct superblock_disk *disk)
+{
+	uint32_t metadata_version = le32_to_cpu(disk->version);
+	if (metadata_version < MIN_ERA_VERSION || metadata_version > MAX_ERA_VERSION) {
+		DMERR("Era metadata version %u found, but only versions between %u and %u supported.",
+		      metadata_version, MIN_ERA_VERSION, MAX_ERA_VERSION);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int sb_check(struct dm_block_validator *v,
+		    struct dm_block *b,
+		    size_t sb_block_size)
+{
+	struct superblock_disk *disk = dm_block_data(b);
+	__le32 csum_le;
+
+	if (dm_block_location(b) != le64_to_cpu(disk->blocknr)) {
+		DMERR("sb_check failed: blocknr %llu: wanted %llu",
+		      le64_to_cpu(disk->blocknr),
+		      (unsigned long long)dm_block_location(b));
+		return -ENOTBLK;
+	}
+
+	if (le64_to_cpu(disk->magic) != SUPERBLOCK_MAGIC) {
+		DMERR("sb_check failed: magic %llu: wanted %llu",
+		      le64_to_cpu(disk->magic),
+		      (unsigned long long) SUPERBLOCK_MAGIC);
+		return -EILSEQ;
+	}
+
+	csum_le = cpu_to_le32(dm_bm_checksum(&disk->flags,
+					     sb_block_size - sizeof(__le32),
+					     SUPERBLOCK_CSUM_XOR));
+	if (csum_le != disk->csum) {
+		DMERR("sb_check failed: csum %u: wanted %u",
+		      le32_to_cpu(csum_le), le32_to_cpu(disk->csum));
+		return -EILSEQ;
+	}
+
+	return check_metadata_version(disk);
+}
+
+static struct dm_block_validator sb_validator = {
+	.name = "superblock",
+	.prepare_for_write = sb_prepare_for_write,
+	.check = sb_check
+};
+
+/*----------------------------------------------------------------
+ * Low level metadata handling
+ *--------------------------------------------------------------*/
+#define DM_ERA_METADATA_BLOCK_SIZE 4096
+#define DM_ERA_METADATA_CACHE_SIZE 64
+#define ERA_MAX_CONCURRENT_LOCKS 5
+
+struct era_metadata {
+	struct block_device *bdev;
+	struct dm_block_manager *bm;
+	struct dm_space_map *sm;
+	struct dm_transaction_manager *tm;
+
+	dm_block_t block_size;
+	uint32_t nr_blocks;
+
+	uint32_t current_era;
+
+	/*
+	 * We preallocate 2 writesets.  When an era rolls over we
+	 * switch between them. This means the allocation is done at
+	 * preresume time, rather than on the io path.
+	 */
+	struct writeset writesets[2];
+	struct writeset *current_writeset;
+
+	dm_block_t writeset_tree_root;
+	dm_block_t era_array_root;
+
+	struct dm_disk_bitset bitset_info;
+	struct dm_btree_info writeset_tree_info;
+	struct dm_array_info era_array_info;
+
+	dm_block_t metadata_snap;
+
+	/*
+	 * A flag that is set whenever a writeset has been archived.
+	 */
+	bool archived_writesets;
+};
+
+static int superblock_read_lock(struct era_metadata *md,
+				struct dm_block **sblock)
+{
+	return dm_bm_read_lock(md->bm, SUPERBLOCK_LOCATION,
+			       &sb_validator, sblock);
+}
+
+static int superblock_lock_zero(struct era_metadata *md,
+				struct dm_block **sblock)
+{
+	return dm_bm_write_lock_zero(md->bm, SUPERBLOCK_LOCATION,
+				     &sb_validator, sblock);
+}
+
+static int superblock_lock(struct era_metadata *md,
+			   struct dm_block **sblock)
+{
+	return dm_bm_write_lock(md->bm, SUPERBLOCK_LOCATION,
+				&sb_validator, sblock);
+}
+
+/* FIXME: duplication with cache and thin */
+static int superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
+{
+	int r;
+	unsigned i;
+	struct dm_block *b;
+	__le64 *data_le, zero = cpu_to_le64(0);
+	unsigned sb_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, SUPERBLOCK_LOCATION, NULL, &b);
+	if (r)
+		return r;
+
+	data_le = dm_block_data(b);
+	*result = true;
+	for (i = 0; i < sb_block_size; i++) {
+		if (data_le[i] != zero) {
+			*result = false;
+			break;
+		}
+	}
+
+	return dm_bm_unlock(b);
+}
+
+/*----------------------------------------------------------------*/
+
+static void ws_pack(const struct writeset_metadata *core, struct writeset_disk *disk)
+{
+	disk->nr_bits = cpu_to_le32(core->nr_bits);
+	disk->root = cpu_to_le64(core->root);
+}
+
+static void ws_unpack(const struct writeset_disk *disk, struct writeset_metadata *core)
+{
+	core->nr_bits = le32_to_cpu(disk->nr_bits);
+	core->root = le64_to_cpu(disk->root);
+}
+
+static void ws_inc(void *context, const void *value)
+{
+	struct era_metadata *md = context;
+	struct writeset_disk ws_d;
+	dm_block_t b;
+
+	memcpy(&ws_d, value, sizeof(ws_d));
+	b = le64_to_cpu(ws_d.root);
+
+	dm_tm_inc(md->tm, b);
+}
+
+static void ws_dec(void *context, const void *value)
+{
+	struct era_metadata *md = context;
+	struct writeset_disk ws_d;
+	dm_block_t b;
+
+	memcpy(&ws_d, value, sizeof(ws_d));
+	b = le64_to_cpu(ws_d.root);
+
+	dm_bitset_del(&md->bitset_info, b);
+}
+
+static int ws_eq(void *context, const void *value1, const void *value2)
+{
+	return !memcmp(value1, value2, sizeof(struct writeset_metadata));
+}
+
+/*----------------------------------------------------------------*/
+
+static void setup_writeset_tree_info(struct era_metadata *md)
+{
+	struct dm_btree_value_type *vt = &md->writeset_tree_info.value_type;
+	md->writeset_tree_info.tm = md->tm;
+	md->writeset_tree_info.levels = 1;
+	vt->context = md;
+	vt->size = sizeof(struct writeset_disk);
+	vt->inc = ws_inc;
+	vt->dec = ws_dec;
+	vt->equal = ws_eq;
+}
+
+static void setup_era_array_info(struct era_metadata *md)
+
+{
+	struct dm_btree_value_type vt;
+	vt.context = NULL;
+	vt.size = sizeof(__le32);
+	vt.inc = NULL;
+	vt.dec = NULL;
+	vt.equal = NULL;
+
+	dm_array_info_init(&md->era_array_info, md->tm, &vt);
+}
+
+static void setup_infos(struct era_metadata *md)
+{
+	dm_disk_bitset_init(md->tm, &md->bitset_info);
+	setup_writeset_tree_info(md);
+	setup_era_array_info(md);
+}
+
+/*----------------------------------------------------------------*/
+
+static int create_fresh_metadata(struct era_metadata *md)
+{
+	int r;
+
+	r = dm_tm_create_with_sm(md->bm, SUPERBLOCK_LOCATION,
+				 &md->tm, &md->sm);
+	if (r < 0) {
+		DMERR("dm_tm_create_with_sm failed");
+		return r;
+	}
+
+	setup_infos(md);
+
+	r = dm_btree_empty(&md->writeset_tree_info, &md->writeset_tree_root);
+	if (r) {
+		DMERR("couldn't create new writeset tree");
+		goto bad;
+	}
+
+	r = dm_array_empty(&md->era_array_info, &md->era_array_root);
+	if (r) {
+		DMERR("couldn't create era array");
+		goto bad;
+	}
+
+	return 0;
+
+bad:
+	dm_sm_destroy(md->sm);
+	dm_tm_destroy(md->tm);
+
+	return r;
+}
+
+/*
+ * Writes a superblock, including the static fields that don't get updated
+ * with every commit (possible optimisation here).  'md' should be fully
+ * constructed when this is called.
+ */
+static int prepare_superblock(struct era_metadata *md, struct superblock_disk *disk)
+{
+	int r;
+	size_t metadata_len;
+
+	disk->magic = cpu_to_le64(SUPERBLOCK_MAGIC);
+	disk->flags = cpu_to_le32(0ul);
+
+	/* FIXME: can't keep blanking the uuid (uuid is currently unused though) */
+	memset(disk->uuid, 0, sizeof(disk->uuid));
+	disk->version = cpu_to_le32(MAX_ERA_VERSION);
+
+	r = dm_sm_root_size(md->sm, &metadata_len);
+	if (r < 0)
+		return r;
+
+	r = dm_sm_copy_root(md->sm, &disk->metadata_space_map_root,
+			    metadata_len);
+	if (r < 0)
+		return r;
+
+	disk->data_block_size = cpu_to_le32(md->block_size);
+	disk->metadata_block_size = cpu_to_le32(DM_ERA_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
+	disk->nr_blocks = cpu_to_le32(md->nr_blocks);
+	disk->current_era = cpu_to_le32(md->current_era);
+
+	ws_pack(&md->current_writeset->md, &disk->current_writeset);
+	disk->writeset_tree_root = cpu_to_le64(md->writeset_tree_root);
+	disk->era_array_root = cpu_to_le64(md->era_array_root);
+	disk->metadata_snap = cpu_to_le64(md->metadata_snap);
+
+	return 0;
+}
+
+static int write_superblock(struct era_metadata *md)
+{
+	int r;
+	struct dm_block *sblock;
+	struct superblock_disk *disk;
+
+	r = superblock_lock_zero(md, &sblock);
+	if (r)
+		return r;
+
+	disk = dm_block_data(sblock);
+	r = prepare_superblock(md, disk);
+	if (r) {
+		DMERR("%s: prepare_superblock failed", __func__);
+		dm_bm_unlock(sblock); /* FIXME: does this commit? */
+		return r;
+	}
+
+	return dm_tm_commit(md->tm, sblock);
+}
+
+/*
+ * Assumes block_size and the infos are set.
+ */
+static int format_metadata(struct era_metadata *md)
+{
+	int r;
+
+	r = create_fresh_metadata(md);
+	if (r)
+		return r;
+
+	r = write_superblock(md);
+	if (r) {
+		dm_sm_destroy(md->sm);
+		dm_tm_destroy(md->tm);
+		return r;
+	}
+
+	return 0;
+}
+
+static int open_metadata(struct era_metadata *md)
+{
+	int r;
+	struct dm_block *sblock;
+	struct superblock_disk *disk;
+
+	r = superblock_read_lock(md, &sblock);
+	if (r) {
+		DMERR("couldn't read_lock superblock");
+		return r;
+	}
+
+	disk = dm_block_data(sblock);
+	r = dm_tm_open_with_sm(md->bm, SUPERBLOCK_LOCATION,
+			       disk->metadata_space_map_root,
+			       sizeof(disk->metadata_space_map_root),
+			       &md->tm, &md->sm);
+	if (r) {
+		DMERR("dm_tm_open_with_sm failed");
+		goto bad;
+	}
+
+	setup_infos(md);
+
+	md->block_size = le32_to_cpu(disk->data_block_size);
+	md->nr_blocks = le32_to_cpu(disk->nr_blocks);
+	md->current_era = le32_to_cpu(disk->current_era);
+
+	md->writeset_tree_root = le64_to_cpu(disk->writeset_tree_root);
+	md->era_array_root = le64_to_cpu(disk->era_array_root);
+	md->metadata_snap = le64_to_cpu(disk->metadata_snap);
+	md->archived_writesets = true;
+
+	return dm_bm_unlock(sblock);
+
+bad:
+	dm_bm_unlock(sblock);
+	return r;
+}
+
+static int open_or_format_metadata(struct era_metadata *md,
+				   bool may_format)
+{
+	int r;
+	bool unformatted = false;
+
+	r = superblock_all_zeroes(md->bm, &unformatted);
+	if (r)
+		return r;
+
+	if (unformatted)
+		return may_format ? format_metadata(md) : -EPERM;
+
+	return open_metadata(md);
+}
+
+static int create_persistent_data_objects(struct era_metadata *md,
+					  bool may_format)
+{
+	int r;
+
+	md->bm = dm_block_manager_create(md->bdev, DM_ERA_METADATA_BLOCK_SIZE,
+					 DM_ERA_METADATA_CACHE_SIZE,
+					 ERA_MAX_CONCURRENT_LOCKS);
+	if (IS_ERR(md->bm)) {
+		DMERR("could not create block manager");
+		return PTR_ERR(md->bm);
+	}
+
+	r = open_or_format_metadata(md, may_format);
+	if (r)
+		dm_block_manager_destroy(md->bm);
+
+	return r;
+}
+
+static void destroy_persistent_data_objects(struct era_metadata *md)
+{
+	dm_sm_destroy(md->sm);
+	dm_tm_destroy(md->tm);
+	dm_block_manager_destroy(md->bm);
+}
+
+/*
+ * This waits until all era_map threads have picked up the new filter.
+ */
+static void swap_writeset(struct era_metadata *md, struct writeset *new_writeset)
+{
+	rcu_assign_pointer(md->current_writeset, new_writeset);
+	synchronize_rcu();
+}
+
+/*----------------------------------------------------------------
+ * Writesets get 'digested' into the main era array.
+ *
+ * We're using a coroutine here so the worker thread can do the digestion,
+ * thus avoiding synchronisation of the metadata.  Digesting a whole
+ * writeset in one go would cause too much latency.
+ *--------------------------------------------------------------*/
+struct digest {
+	uint32_t era;
+	unsigned nr_bits, current_bit;
+	struct writeset_metadata writeset;
+	__le32 value;
+	struct dm_disk_bitset info;
+
+	int (*step)(struct era_metadata *, struct digest *);
+};
+
+static int metadata_digest_lookup_writeset(struct era_metadata *md,
+					   struct digest *d);
+
+static int metadata_digest_remove_writeset(struct era_metadata *md,
+					   struct digest *d)
+{
+	int r;
+	uint64_t key = d->era;
+
+	r = dm_btree_remove(&md->writeset_tree_info, md->writeset_tree_root,
+			    &key, &md->writeset_tree_root);
+	if (r) {
+		DMERR("%s: dm_btree_remove failed", __func__);
+		return r;
+	}
+
+	d->step = metadata_digest_lookup_writeset;
+	return 0;
+}
+
+#define INSERTS_PER_STEP 100
+
+static int metadata_digest_transcribe_writeset(struct era_metadata *md,
+					       struct digest *d)
+{
+	int r;
+	bool marked;
+	unsigned b, e = min(d->current_bit + INSERTS_PER_STEP, d->nr_bits);
+
+	for (b = d->current_bit; b < e; b++) {
+		r = writeset_marked_on_disk(&d->info, &d->writeset, b, &marked);
+		if (r) {
+			DMERR("%s: writeset_marked_on_disk failed", __func__);
+			return r;
+		}
+
+		if (!marked)
+			continue;
+
+		__dm_bless_for_disk(&d->value);
+		r = dm_array_set_value(&md->era_array_info, md->era_array_root,
+				       b, &d->value, &md->era_array_root);
+		if (r) {
+			DMERR("%s: dm_array_set_value failed", __func__);
+			return r;
+		}
+	}
+
+	if (b == d->nr_bits)
+		d->step = metadata_digest_remove_writeset;
+	else
+		d->current_bit = b;
+
+	return 0;
+}
+
+static int metadata_digest_lookup_writeset(struct era_metadata *md,
+					   struct digest *d)
+{
+	int r;
+	uint64_t key;
+	struct writeset_disk disk;
+
+	r = dm_btree_find_lowest_key(&md->writeset_tree_info,
+				     md->writeset_tree_root, &key);
+	if (r < 0)
+		return r;
+
+	d->era = key;
+
+	r = dm_btree_lookup(&md->writeset_tree_info,
+			    md->writeset_tree_root, &key, &disk);
+	if (r) {
+		if (r == -ENODATA) {
+			d->step = NULL;
+			return 0;
+		}
+
+		DMERR("%s: dm_btree_lookup failed", __func__);
+		return r;
+	}
+
+	ws_unpack(&disk, &d->writeset);
+	d->value = cpu_to_le32(key);
+
+	d->nr_bits = min(d->writeset.nr_bits, md->nr_blocks);
+	d->current_bit = 0;
+	d->step = metadata_digest_transcribe_writeset;
+
+	return 0;
+}
+
+static int metadata_digest_start(struct era_metadata *md, struct digest *d)
+{
+	if (d->step)
+		return 0;
+
+	memset(d, 0, sizeof(*d));
+
+	/*
+	 * We initialise another bitset info to avoid any caching side
+	 * effects with the previous one.
+	 */
+	dm_disk_bitset_init(md->tm, &d->info);
+	d->step = metadata_digest_lookup_writeset;
+
+	return 0;
+}
+
+/*----------------------------------------------------------------
+ * High level metadata interface.  Target methods should use these, and not
+ * the lower level ones.
+ *--------------------------------------------------------------*/
+static struct era_metadata *metadata_open(struct block_device *bdev,
+					  sector_t block_size,
+					  bool may_format)
+{
+	int r;
+	struct era_metadata *md = kzalloc(sizeof(*md), GFP_KERNEL);
+
+	if (!md)
+		return NULL;
+
+	md->bdev = bdev;
+	md->block_size = block_size;
+
+	md->writesets[0].md.root = INVALID_WRITESET_ROOT;
+	md->writesets[1].md.root = INVALID_WRITESET_ROOT;
+	md->current_writeset = &md->writesets[0];
+
+	r = create_persistent_data_objects(md, may_format);
+	if (r) {
+		kfree(md);
+		return ERR_PTR(r);
+	}
+
+	return md;
+}
+
+static void metadata_close(struct era_metadata *md)
+{
+	destroy_persistent_data_objects(md);
+	kfree(md);
+}
+
+static bool valid_nr_blocks(dm_block_t n)
+{
+	/*
+	 * dm_bitset restricts us to 2^32.  test_bit & co. restrict us
+	 * further to 2^31 - 1
+	 */
+	return n < (1ull << 31);
+}
+
+static int metadata_resize(struct era_metadata *md, void *arg)
+{
+	int r;
+	dm_block_t *new_size = arg;
+	__le32 value;
+
+	if (!valid_nr_blocks(*new_size)) {
+		DMERR("Invalid number of origin blocks %llu",
+		      (unsigned long long) *new_size);
+		return -EINVAL;
+	}
+
+	writeset_free(&md->writesets[0]);
+	writeset_free(&md->writesets[1]);
+
+	r = writeset_alloc(&md->writesets[0], *new_size);
+	if (r) {
+		DMERR("%s: writeset_alloc failed for writeset 0", __func__);
+		return r;
+	}
+
+	r = writeset_alloc(&md->writesets[1], *new_size);
+	if (r) {
+		DMERR("%s: writeset_alloc failed for writeset 1", __func__);
+		return r;
+	}
+
+	value = cpu_to_le32(0u);
+	__dm_bless_for_disk(&value);
+	r = dm_array_resize(&md->era_array_info, md->era_array_root,
+			    md->nr_blocks, *new_size,
+			    &value, &md->era_array_root);
+	if (r) {
+		DMERR("%s: dm_array_resize failed", __func__);
+		return r;
+	}
+
+	md->nr_blocks = *new_size;
+	return 0;
+}
+
+static int metadata_era_archive(struct era_metadata *md)
+{
+	int r;
+	uint64_t keys[1];
+	struct writeset_disk value;
+
+	r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root,
+			    &md->current_writeset->md.root);
+	if (r) {
+		DMERR("%s: dm_bitset_flush failed", __func__);
+		return r;
+	}
+
+	ws_pack(&md->current_writeset->md, &value);
+	md->current_writeset->md.root = INVALID_WRITESET_ROOT;
+
+	keys[0] = md->current_era;
+	__dm_bless_for_disk(&value);
+	r = dm_btree_insert(&md->writeset_tree_info, md->writeset_tree_root,
+			    keys, &value, &md->writeset_tree_root);
+	if (r) {
+		DMERR("%s: couldn't insert writeset into btree", __func__);
+		/* FIXME: fail mode */
+		return r;
+	}
+
+	md->archived_writesets = true;
+
+	return 0;
+}
+
+static struct writeset *next_writeset(struct era_metadata *md)
+{
+	return (md->current_writeset == &md->writesets[0]) ?
+		&md->writesets[1] : &md->writesets[0];
+}
+
+static int metadata_new_era(struct era_metadata *md)
+{
+	int r;
+	struct writeset *new_writeset = next_writeset(md);
+
+	r = writeset_init(&md->bitset_info, new_writeset);
+	if (r) {
+		DMERR("%s: writeset_init failed", __func__);
+		return r;
+	}
+
+	swap_writeset(md, new_writeset);
+	md->current_era++;
+
+	return 0;
+}
+
+static int metadata_era_rollover(struct era_metadata *md)
+{
+	int r;
+
+	if (md->current_writeset->md.root != INVALID_WRITESET_ROOT) {
+		r = metadata_era_archive(md);
+		if (r) {
+			DMERR("%s: metadata_archive_era failed", __func__);
+			/* FIXME: fail mode? */
+			return r;
+		}
+	}
+
+	r = metadata_new_era(md);
+	if (r) {
+		DMERR("%s: new era failed", __func__);
+		/* FIXME: fail mode */
+		return r;
+	}
+
+	return 0;
+}
+
+static bool metadata_current_marked(struct era_metadata *md, dm_block_t block)
+{
+	bool r;
+	struct writeset *ws;
+
+	rcu_read_lock();
+	ws = rcu_dereference(md->current_writeset);
+	r = writeset_marked(ws, block);
+	rcu_read_unlock();
+
+	return r;
+}
+
+static int metadata_commit(struct era_metadata *md)
+{
+	int r;
+	struct dm_block *sblock;
+
+	if (md->current_writeset->md.root != SUPERBLOCK_LOCATION) {
+		r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root,
+				    &md->current_writeset->md.root);
+		if (r) {
+			DMERR("%s: bitset flush failed", __func__);
+			return r;
+		}
+	}
+
+	r = dm_tm_pre_commit(md->tm);
+	if (r) {
+		DMERR("%s: pre commit failed", __func__);
+		return r;
+	}
+
+	r = superblock_lock(md, &sblock);
+	if (r) {
+		DMERR("%s: superblock lock failed", __func__);
+		return r;
+	}
+
+	r = prepare_superblock(md, dm_block_data(sblock));
+	if (r) {
+		DMERR("%s: prepare_superblock failed", __func__);
+		dm_bm_unlock(sblock); /* FIXME: does this commit? */
+		return r;
+	}
+
+	return dm_tm_commit(md->tm, sblock);
+}
+
+static int metadata_checkpoint(struct era_metadata *md)
+{
+	/*
+	 * For now we just rollover, but later I want to put a check in to
+	 * avoid this if the filter is still pretty fresh.
+	 */
+	return metadata_era_rollover(md);
+}
+
+/*
+ * Metadata snapshots allow userland to access era data.
+ */
+static int metadata_take_snap(struct era_metadata *md)
+{
+	int r, inc;
+	struct dm_block *clone;
+
+	if (md->metadata_snap != SUPERBLOCK_LOCATION) {
+		DMERR("%s: metadata snapshot already exists", __func__);
+		return -EINVAL;
+	}
+
+	r = metadata_era_rollover(md);
+	if (r) {
+		DMERR("%s: era rollover failed", __func__);
+		return r;
+	}
+
+	r = metadata_commit(md);
+	if (r) {
+		DMERR("%s: pre commit failed", __func__);
+		return r;
+	}
+
+	r = dm_sm_inc_block(md->sm, SUPERBLOCK_LOCATION);
+	if (r) {
+		DMERR("%s: couldn't increment superblock", __func__);
+		return r;
+	}
+
+	r = dm_tm_shadow_block(md->tm, SUPERBLOCK_LOCATION,
+			       &sb_validator, &clone, &inc);
+	if (r) {
+		DMERR("%s: couldn't shadow superblock", __func__);
+		dm_sm_dec_block(md->sm, SUPERBLOCK_LOCATION);
+		return r;
+	}
+	BUG_ON(!inc);
+
+	r = dm_sm_inc_block(md->sm, md->writeset_tree_root);
+	if (r) {
+		DMERR("%s: couldn't inc writeset tree root", __func__);
+		dm_tm_unlock(md->tm, clone);
+		return r;
+	}
+
+	r = dm_sm_inc_block(md->sm, md->era_array_root);
+	if (r) {
+		DMERR("%s: couldn't inc era tree root", __func__);
+		dm_sm_dec_block(md->sm, md->writeset_tree_root);
+		dm_tm_unlock(md->tm, clone);
+		return r;
+	}
+
+	md->metadata_snap = dm_block_location(clone);
+
+	r = dm_tm_unlock(md->tm, clone);
+	if (r) {
+		DMERR("%s: couldn't unlock clone", __func__);
+		md->metadata_snap = SUPERBLOCK_LOCATION;
+		return r;
+	}
+
+	return 0;
+}
+
+static int metadata_drop_snap(struct era_metadata *md)
+{
+	int r;
+	dm_block_t location;
+	struct dm_block *clone;
+	struct superblock_disk *disk;
+
+	if (md->metadata_snap == SUPERBLOCK_LOCATION) {
+		DMERR("%s: no snap to drop", __func__);
+		return -EINVAL;
+	}
+
+	r = dm_tm_read_lock(md->tm, md->metadata_snap, &sb_validator, &clone);
+	if (r) {
+		DMERR("%s: couldn't read lock superblock clone", __func__);
+		return r;
+	}
+
+	/*
+	 * Whatever happens now we'll commit with no record of the metadata
+	 * snap.
+	 */
+	md->metadata_snap = SUPERBLOCK_LOCATION;
+
+	disk = dm_block_data(clone);
+	r = dm_btree_del(&md->writeset_tree_info,
+			 le64_to_cpu(disk->writeset_tree_root));
+	if (r) {
+		DMERR("%s: error deleting writeset tree clone", __func__);
+		dm_tm_unlock(md->tm, clone);
+		return r;
+	}
+
+	r = dm_array_del(&md->era_array_info, le64_to_cpu(disk->era_array_root));
+	if (r) {
+		DMERR("%s: error deleting era array clone", __func__);
+		dm_tm_unlock(md->tm, clone);
+		return r;
+	}
+
+	location = dm_block_location(clone);
+	dm_tm_unlock(md->tm, clone);
+
+	return dm_sm_dec_block(md->sm, location);
+}
+
+struct metadata_stats {
+	dm_block_t used;
+	dm_block_t total;
+	dm_block_t snap;
+	uint32_t era;
+};
+
+static int metadata_get_stats(struct era_metadata *md, void *ptr)
+{
+	int r;
+	struct metadata_stats *s = ptr;
+	dm_block_t nr_free, nr_total;
+
+	r = dm_sm_get_nr_free(md->sm, &nr_free);
+	if (r) {
+		DMERR("dm_sm_get_nr_free returned %d", r);
+		return r;
+	}
+
+	r = dm_sm_get_nr_blocks(md->sm, &nr_total);
+	if (r) {
+		DMERR("dm_pool_get_metadata_dev_size returned %d", r);
+		return r;
+	}
+
+	s->used = nr_total - nr_free;
+	s->total = nr_total;
+	s->snap = md->metadata_snap;
+	s->era = md->current_era;
+
+	return 0;
+}
+
+/*----------------------------------------------------------------*/
+
+struct era {
+	struct dm_target *ti;
+	struct dm_target_callbacks callbacks;
+
+	struct dm_dev *metadata_dev;
+	struct dm_dev *origin_dev;
+
+	dm_block_t nr_blocks;
+	uint32_t sectors_per_block;
+	int sectors_per_block_shift;
+	struct era_metadata *md;
+
+	struct workqueue_struct *wq;
+	struct work_struct worker;
+
+	spinlock_t deferred_lock;
+	struct bio_list deferred_bios;
+
+	spinlock_t rpc_lock;
+	struct list_head rpc_calls;
+
+	struct digest digest;
+	atomic_t suspended;
+};
+
+struct rpc {
+	struct list_head list;
+
+	int (*fn0)(struct era_metadata *);
+	int (*fn1)(struct era_metadata *, void *);
+	void *arg;
+	int result;
+
+	struct completion complete;
+};
+
+/*----------------------------------------------------------------
+ * Remapping.
+ *---------------------------------------------------------------*/
+static bool block_size_is_power_of_two(struct era *era)
+{
+	return era->sectors_per_block_shift >= 0;
+}
+
+static dm_block_t get_block(struct era *era, struct bio *bio)
+{
+	sector_t block_nr = bio->bi_iter.bi_sector;
+
+	if (!block_size_is_power_of_two(era))
+		(void) sector_div(block_nr, era-&