dm-snap.c

/*
 * dm-snapshot.c
 *
 * Copyright (C) 2001-2002 Sistina Software (UK) Limited.
 *
 * This file is released under the GPL.
 */

#include <linux/blkdev.h>
#include <linux/device-mapper.h>
#include <linux/delay.h>
#include <linux/fs.h>
#include <linux/init.h>
#include <linux/kdev_t.h>
#include <linux/list.h>
#include <linux/mempool.h>
#include <linux/module.h>
#include <linux/slab.h>
#include <linux/vmalloc.h>
#include <linux/log2.h>
#include <linux/dm-kcopyd.h>
#include <linux/workqueue.h>

#include "dm-exception-store.h"

#define DM_MSG_PREFIX "snapshots"

static const char dm_snapshot_merge_target_name[] = "snapshot-merge";

#define dm_target_is_snapshot_merge(ti) \
	((ti)->type->name == dm_snapshot_merge_target_name)

/*
 * The percentage increment we will wake up users at
 */
#define WAKE_UP_PERCENT 5

/*
 * kcopyd priority of snapshot operations
 */
#define SNAPSHOT_COPY_PRIORITY 2

/*
 * Reserve 1MB for each snapshot initially (with minimum of 1 page).
 */
#define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1)

/*
 * The size of the mempool used to track chunks in use.
 */
#define MIN_IOS 256

#define DM_TRACKED_CHUNK_HASH_SIZE	16
#define DM_TRACKED_CHUNK_HASH(x)	((unsigned long)(x) & \
					 (DM_TRACKED_CHUNK_HASH_SIZE - 1))

struct dm_exception_table {
	uint32_t hash_mask;
	unsigned hash_shift;
	struct list_head *table;
};

struct dm_snapshot {
	struct rw_semaphore lock;

	struct dm_dev *origin;
	struct dm_dev *cow;

	struct dm_target *ti;

	/* List of snapshots per Origin */
	struct list_head list;

	/* You can't use a snapshot if this is 0 (e.g. if full) */
	int valid;

	/* Origin writes don't trigger exceptions until this is set */
	int active;

	/* Whether or not owning mapped_device is suspended */
	int suspended;

	mempool_t *pending_pool;

	atomic_t pending_exceptions_count;

	struct dm_exception_table pending;
	struct dm_exception_table complete;

	/*
	 * pe_lock protects all pending_exception operations and access
	 * as well as the snapshot_bios list.
	 */
	spinlock_t pe_lock;

	/* The on disk metadata handler */
	struct dm_exception_store *store;

	struct dm_kcopyd_client *kcopyd_client;

	/* Queue of snapshot writes for ksnapd to flush */
	struct bio_list queued_bios;
	struct work_struct queued_bios_work;

	/* Chunks with outstanding reads */
	mempool_t *tracked_chunk_pool;
	spinlock_t tracked_chunk_lock;
	struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
};

struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
{
	return s->cow;
}
EXPORT_SYMBOL(dm_snap_cow);

static struct workqueue_struct *ksnapd;
static void flush_queued_bios(struct work_struct *work);

static sector_t chunk_to_sector(struct dm_exception_store *store,
				chunk_t chunk)
{
	return chunk << store->chunk_shift;
}

static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
{
	/*
	 * There is only ever one instance of a particular block
	 * device so we can compare pointers safely.
	 */
	return lhs == rhs;
}

struct dm_snap_pending_exception {
	struct dm_exception e;

	/*
	 * Origin buffers waiting for this to complete are held
	 * in a bio list
	 */
	struct bio_list origin_bios;
	struct bio_list snapshot_bios;

	/* Pointer back to snapshot context */
	struct dm_snapshot *snap;

	/*
	 * 1 indicates the exception has already been sent to
	 * kcopyd.
	 */
	int started;
};

/*
 * Hash table mapping origin volumes to lists of snapshots and
 * a lock to protect it
 */
static struct kmem_cache *exception_cache;
static struct kmem_cache *pending_cache;

struct dm_snap_tracked_chunk {
	struct hlist_node node;
	chunk_t chunk;
};

static struct kmem_cache *tracked_chunk_cache;

static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s,
						 chunk_t chunk)
{
	struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool,
							GFP_NOIO);
	unsigned long flags;

	c->chunk = chunk;

	spin_lock_irqsave(&s->tracked_chunk_lock, flags);
	hlist_add_head(&c->node,
		       &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
	spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);

	return c;
}

static void stop_tracking_chunk(struct dm_snapshot *s,
				struct dm_snap_tracked_chunk *c)
{
	unsigned long flags;

	spin_lock_irqsave(&s->tracked_chunk_lock, flags);
	hlist_del(&c->node);
	spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);

	mempool_free(c, s->tracked_chunk_pool);
}

static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
{
	struct dm_snap_tracked_chunk *c;
	struct hlist_node *hn;
	int found = 0;

	spin_lock_irq(&s->tracked_chunk_lock);

	hlist_for_each_entry(c, hn,
	    &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
		if (c->chunk == chunk) {
			found = 1;
			break;
		}
	}

	spin_unlock_irq(&s->tracked_chunk_lock);

	return found;
}

/*
 * This conflicting I/O is extremely improbable in the caller,
 * so msleep(1) is sufficient and there is no need for a wait queue.
 */
static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
{
	while (__chunk_is_tracked(s, chunk))
		msleep(1);
}

/*
 * One of these per registered origin, held in the snapshot_origins hash
 */
struct origin {
	/* The origin device */
	struct block_device *bdev;

	struct list_head hash_list;

	/* List of snapshots for this origin */
	struct list_head snapshots;
};

/*
 * Size of the hash table for origin volumes. If we make this
 * the size of the minors list then it should be nearly perfect
 */
#define ORIGIN_HASH_SIZE 256
#define ORIGIN_MASK      0xFF
static struct list_head *_origins;
static struct rw_semaphore _origins_lock;

static int init_origin_hash(void)
{
	int i;

	_origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
			   GFP_KERNEL);
	if (!_origins) {
		DMERR("unable to allocate memory");
		return -ENOMEM;
	}

	for (i = 0; i < ORIGIN_HASH_SIZE; i++)
		INIT_LIST_HEAD(_origins + i);
	init_rwsem(&_origins_lock);

	return 0;
}

static void exit_origin_hash(void)
{
	kfree(_origins);
}

static unsigned origin_hash(struct block_device *bdev)
{
	return bdev->bd_dev & ORIGIN_MASK;
}

static struct origin *__lookup_origin(struct block_device *origin)
{
	struct list_head *ol;
	struct origin *o;

	ol = &_origins[origin_hash(origin)];
	list_for_each_entry (o, ol, hash_list)
		if (bdev_equal(o->bdev, origin))
			return o;

	return NULL;
}

static void __insert_origin(struct origin *o)
{
	struct list_head *sl = &_origins[origin_hash(o->bdev)];
	list_add_tail(&o->hash_list, sl);
}

/*
 * _origins_lock must be held when calling this function.
 * Returns number of snapshots registered using the supplied cow device, plus:
 * snap_src - a snapshot suitable for use as a source of exception handover
 * snap_dest - a snapshot capable of receiving exception handover.
 *
 * Possible return values and states:
 *   0: NULL, NULL  - first new snapshot
 *   1: snap_src, NULL - normal snapshot
 *   2: snap_src, snap_dest  - waiting for handover
 *   2: snap_src, NULL - handed over, waiting for old to be deleted
 *   1: NULL, snap_dest - source got destroyed without handover
 */
static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
					struct dm_snapshot **snap_src,
					struct dm_snapshot **snap_dest)
{
	struct dm_snapshot *s;
	struct origin *o;
	int count = 0;
	int active;

	o = __lookup_origin(snap->origin->bdev);
	if (!o)
		goto out;

	list_for_each_entry(s, &o->snapshots, list) {
		if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
			continue;

		down_read(&s->lock);
		active = s->active;
		up_read(&s->lock);

		if (active) {
			if (snap_src)
				*snap_src = s;
		} else if (snap_dest)
			*snap_dest = s;

		count++;
	}

out:
	return count;
}

/*
 * On success, returns 1 if this snapshot is a handover destination,
 * otherwise returns 0.
 */
static int __validate_exception_handover(struct dm_snapshot *snap)
{
	struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;

	/* Does snapshot need exceptions handed over to it? */
	if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest) == 2) ||
	    snap_dest) {
		snap->ti->error = "Snapshot cow pairing for exception "
				  "table handover failed";
		return -EINVAL;
	}

	/*
	 * If no snap_src was found, snap cannot become a handover
	 * destination.
	 */
	if (!snap_src)
		return 0;

	return 1;
}

static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
{
	struct dm_snapshot *l;

	/* Sort the list according to chunk size, largest-first smallest-last */
	list_for_each_entry(l, &o->snapshots, list)
		if (l->store->chunk_size < s->store->chunk_size)
			break;
	list_add_tail(&s->list, &l->list);
}

/*
 * Make a note of the snapshot and its origin so we can look it
 * up when the origin has a write on it.
 *
 * Also validate snapshot exception store handovers.
 * On success, returns 1 if this registration is a handover destination,
 * otherwise returns 0.
 */
static int register_snapshot(struct dm_snapshot *snap)
{
	struct origin *o, *new_o = NULL;
	struct block_device *bdev = snap->origin->bdev;
	int r = 0;

	new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
	if (!new_o)
		return -ENOMEM;

	down_write(&_origins_lock);

	r = __validate_exception_handover(snap);
	if (r < 0) {
		kfree(new_o);
		goto out;
	}

	o = __lookup_origin(bdev);
	if (o)
		kfree(new_o);
	else {
		/* New origin */
		o = new_o;

		/* Initialise the struct */
		INIT_LIST_HEAD(&o->snapshots);
		o->bdev = bdev;

		__insert_origin(o);
	}

	__insert_snapshot(o, snap);

out:
	up_write(&_origins_lock);

	return r;
}

/*
 * Move snapshot to correct place in list according to chunk size.
 */
static void reregister_snapshot(struct dm_snapshot *s)
{
	struct block_device *bdev = s->origin->bdev;

	down_write(&_origins_lock);

	list_del(&s->list);
	__insert_snapshot(__lookup_origin(bdev), s);

	up_write(&_origins_lock);
}

static void unregister_snapshot(struct dm_snapshot *s)
{
	struct origin *o;

	down_write(&_origins_lock);
	o = __lookup_origin(s->origin->bdev);

	list_del(&s->list);
	if (o && list_empty(&o->snapshots)) {
		list_del(&o->hash_list);
		kfree(o);
	}

	up_write(&_origins_lock);
}

/*
 * Implementation of the exception hash tables.
 * The lowest hash_shift bits of the chunk number are ignored, allowing
 * some consecutive chunks to be grouped together.
 */
static int dm_exception_table_init(struct dm_exception_table *et,
				   uint32_t size, unsigned hash_shift)
{
	unsigned int i;

	et->hash_shift = hash_shift;
	et->hash_mask = size - 1;
	et->table = dm_vcalloc(size, sizeof(struct list_head));
	if (!et->table)
		return -ENOMEM;

	for (i = 0; i < size; i++)
		INIT_LIST_HEAD(et->table + i);

	return 0;
}

static void dm_exception_table_exit(struct dm_exception_table *et,
				    struct kmem_cache *mem)
{
	struct list_head *slot;
	struct dm_exception *ex, *next;
	int i, size;

	size = et->hash_mask + 1;
	for (i = 0; i < size; i++) {
		slot = et->table + i;

		list_for_each_entry_safe (ex, next, slot, hash_list)
			kmem_cache_free(mem, ex);
	}

	vfree(et->table);
}

static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
{
	return (chunk >> et->hash_shift) & et->hash_mask;
}

static void dm_remove_exception(struct dm_exception *e)
{
	list_del(&e->hash_list);
}

/*
 * Return the exception data for a sector, or NULL if not
 * remapped.
 */
static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
						chunk_t chunk)
{
	struct list_head *slot;
	struct dm_exception *e;

	slot = &et->table[exception_hash(et, chunk)];
	list_for_each_entry (e, slot, hash_list)
		if (chunk >= e->old_chunk &&
		    chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
			return e;

	return NULL;
}

static struct dm_exception *alloc_completed_exception(void)
{
	struct dm_exception *e;

	e = kmem_cache_alloc(exception_cache, GFP_NOIO);
	if (!e)
		e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);

	return e;
}

static void free_completed_exception(struct dm_exception *e)
{
	kmem_cache_free(exception_cache, e);
}

static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
{
	struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
							     GFP_NOIO);

	atomic_inc(&s->pending_exceptions_count);
	pe->snap = s;

	return pe;
}

static void free_pending_exception(struct dm_snap_pending_exception *pe)
{
	struct dm_snapshot *s = pe->snap;

	mempool_free(pe, s->pending_pool);
	smp_mb__before_atomic_dec();
	atomic_dec(&s->pending_exceptions_count);
}

static void dm_insert_exception(struct dm_exception_table *eh,
				struct dm_exception *new_e)
{
	struct list_head *l;
	struct dm_exception *e = NULL;

	l = &eh->table[exception_hash(eh, new_e->old_chunk)];

	/* Add immediately if this table doesn't support consecutive chunks */
	if (!eh->hash_shift)
		goto out;

	/* List is ordered by old_chunk */
	list_for_each_entry_reverse(e, l, hash_list) {
		/* Insert after an existing chunk? */
		if (new_e->old_chunk == (e->old_chunk +
					 dm_consecutive_chunk_count(e) + 1) &&
		    new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
					 dm_consecutive_chunk_count(e) + 1)) {
			dm_consecutive_chunk_count_inc(e);
			free_completed_exception(new_e);
			return;
		}

		/* Insert before an existing chunk? */
		if (new_e->old_chunk == (e->old_chunk - 1) &&
		    new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
			dm_consecutive_chunk_count_inc(e);
			e->old_chunk--;
			e->new_chunk--;
			free_completed_exception(new_e);
			return;
		}

		if (new_e->old_chunk > e->old_chunk)
			break;
	}

out:
	list_add(&new_e->hash_list, e ? &e->hash_list : l);
}

/*
 * Callback used by the exception stores to load exceptions when
 * initialising.
 */
static int dm_add_exception(void *context, chunk_t old, chunk_t new)
{
	struct dm_snapshot *s = context;
	struct dm_exception *e;

	e = alloc_completed_exception();
	if (!e)
		return -ENOMEM;

	e->old_chunk = old;

	/* Consecutive_count is implicitly initialised to zero */
	e->new_chunk = new;

	dm_insert_exception(&s->complete, e);

	return 0;
}

#define min_not_zero(l, r) (((l) == 0) ? (r) : (((r) == 0) ? (l) : min(l, r)))

/*
 * Return a minimum chunk size of all snapshots that have the specified origin.
 * Return zero if the origin has no snapshots.
 */
static sector_t __minimum_chunk_size(struct origin *o)
{
	struct dm_snapshot *snap;
	unsigned chunk_size = 0;

	if (o)
		list_for_each_entry(snap, &o->snapshots, list)
			chunk_size = min_not_zero(chunk_size,
						  snap->store->chunk_size);

	return chunk_size;
}

/*
 * Hard coded magic.
 */
static int calc_max_buckets(void)
{
	/* use a fixed size of 2MB */
	unsigned long mem = 2 * 1024 * 1024;
	mem /= sizeof(struct list_head);

	return mem;
}

/*
 * Allocate room for a suitable hash table.
 */
static int init_hash_tables(struct dm_snapshot *s)
{
	sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;

	/*
	 * Calculate based on the size of the original volume or
	 * the COW volume...
	 */
	cow_dev_size = get_dev_size(s->cow->bdev);
	origin_dev_size = get_dev_size(s->origin->bdev);
	max_buckets = calc_max_buckets();

	hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift;
	hash_size = min(hash_size, max_buckets);

	if (hash_size < 64)
		hash_size = 64;
	hash_size = rounddown_pow_of_two(hash_size);
	if (dm_exception_table_init(&s->complete, hash_size,
				    DM_CHUNK_CONSECUTIVE_BITS))
		return -ENOMEM;

	/*
	 * Allocate hash table for in-flight exceptions
	 * Make this smaller than the real hash table
	 */
	hash_size >>= 3;
	if (hash_size < 64)
		hash_size = 64;

	if (dm_exception_table_init(&s->pending, hash_size, 0)) {
		dm_exception_table_exit(&s->complete, exception_cache);
		return -ENOMEM;
	}

	return 0;
}

/*
 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
 */
static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
{
	struct dm_snapshot *s;
	int i;
	int r = -EINVAL;
	char *origin_path, *cow_path;
	unsigned args_used;

	if (argc != 4) {
		ti->error = "requires exactly 4 arguments";
		r = -EINVAL;
		goto bad;
	}

	origin_path = argv[0];
	argv++;
	argc--;

	s = kmalloc(sizeof(*s), GFP_KERNEL);
	if (!s) {
		ti->error = "Cannot allocate snapshot context private "
		    "structure";
		r = -ENOMEM;
		goto bad;
	}

	cow_path = argv[0];
	argv++;
	argc--;

	r = dm_get_device(ti, cow_path, 0, 0,
			  FMODE_READ | FMODE_WRITE, &s->cow);
	if (r) {
		ti->error = "Cannot get COW device";
		goto bad_cow;
	}

	r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
	if (r) {
		ti->error = "Couldn't create exception store";
		r = -EINVAL;
		goto bad_store;
	}

	argv += args_used;
	argc -= args_used;

	r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin);
	if (r) {
		ti->error = "Cannot get origin device";
		goto bad_origin;
	}

	s->ti = ti;
	s->valid = 1;
	s->active = 0;
	s->suspended = 0;
	atomic_set(&s->pending_exceptions_count, 0);
	init_rwsem(&s->lock);
	INIT_LIST_HEAD(&s->list);
	spin_lock_init(&s->pe_lock);

	/* Allocate hash table for COW data */
	if (init_hash_tables(s)) {
		ti->error = "Unable to allocate hash table space";
		r = -ENOMEM;
		goto bad_hash_tables;
	}

	r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client);
	if (r) {
		ti->error = "Could not create kcopyd client";
		goto bad_kcopyd;
	}

	s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
	if (!s->pending_pool) {
		ti->error = "Could not allocate mempool for pending exceptions";
		goto bad_pending_pool;
	}

	s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS,
							 tracked_chunk_cache);
	if (!s->tracked_chunk_pool) {
		ti->error = "Could not allocate tracked_chunk mempool for "
			    "tracking reads";
		goto bad_tracked_chunk_pool;
	}

	for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
		INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);

	spin_lock_init(&s->tracked_chunk_lock);

	bio_list_init(&s->queued_bios);
	INIT_WORK(&s->queued_bios_work, flush_queued_bios);

	ti->private = s;
	ti->num_flush_requests = 1;

	/* Add snapshot to the list of snapshots for this origin */
	/* Exceptions aren't triggered till snapshot_resume() is called */
	r = register_snapshot(s);
	if (r == -ENOMEM) {
		ti->error = "Snapshot origin struct allocation failed";
		goto bad_load_and_register;
	} else if (r < 0) {
		/* invalid handover, register_snapshot has set ti->error */
		goto bad_load_and_register;
	}

	/*
	 * Metadata must only be loaded into one table at once, so skip this
	 * if metadata will be handed over during resume.
	 * Chunk size will be set during the handover - set it to zero to
	 * ensure it's ignored.
	 */
	if (r > 0) {
		s->store->chunk_size = 0;
		return 0;
	}

	r = s->store->type->read_metadata(s->store, dm_add_exception,
					  (void *)s);
	if (r < 0) {
		ti->error = "Failed to read snapshot metadata";
		goto bad_read_metadata;
	} else if (r > 0) {
		s->valid = 0;
		DMWARN("Snapshot is marked invalid.");
	}

	if (!s->store->chunk_size) {
		ti->error = "Chunk size not set";
		goto bad_read_metadata;
	}
	ti->split_io = s->store->chunk_size;

	return 0;

bad_read_metadata:
	unregister_snapshot(s);

bad_load_and_register:
	mempool_destroy(s->tracked_chunk_pool);

bad_tracked_chunk_pool:
	mempool_destroy(s->pending_pool);

bad_pending_pool:
	dm_kcopyd_client_destroy(s->kcopyd_client);

bad_kcopyd:
	dm_exception_table_exit(&s->pending, pending_cache);
	dm_exception_table_exit(&s->complete, exception_cache);

bad_hash_tables:
	dm_put_device(ti, s->origin);

bad_origin:
	dm_exception_store_destroy(s->store);

bad_store:
	dm_put_device(ti, s->cow);

bad_cow:
	kfree(s);

bad:
	return r;
}

static void __free_exceptions(struct dm_snapshot *s)
{
	dm_kcopyd_client_destroy(s->kcopyd_client);
	s->kcopyd_client = NULL;

	dm_exception_table_exit(&s->pending, pending_cache);
	dm_exception_table_exit(&s->complete, exception_cache);
}

static void __handover_exceptions(struct dm_snapshot *snap_src,
				  struct dm_snapshot *snap_dest)
{
	union {
		struct dm_exception_table table_swap;
		struct dm_exception_store *store_swap;
	} u;

	/*
	 * Swap all snapshot context information between the two instances.
	 */
	u.table_swap = snap_dest->complete;
	snap_dest->complete = snap_src->complete;
	snap_src->complete = u.table_swap;

	u.store_swap = snap_dest->store;
	snap_dest->store = snap_src->store;
	snap_src->store = u.store_swap;

	snap_dest->store->snap = snap_dest;
	snap_src->store->snap = snap_src;

	snap_dest->ti->split_io = snap_dest->store->chunk_size;
	snap_dest->valid = snap_src->valid;

	/*
	 * Set source invalid to ensure it receives no further I/O.
	 */
	snap_src->valid = 0;
}

static void snapshot_dtr(struct dm_target *ti)
{
#ifdef CONFIG_DM_DEBUG
	int i;
#endif
	struct dm_snapshot *s = ti->private;
	struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;

	flush_workqueue(ksnapd);

	down_read(&_origins_lock);
	/* Check whether exception handover must be cancelled */
	(void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest);
	if (snap_src && snap_dest && (s == snap_src)) {
		down_write(&snap_dest->lock);
		snap_dest->valid = 0;
		up_write(&snap_dest->lock);
		DMERR("Cancelling snapshot handover.");
	}
	up_read(&_origins_lock);

	/* Prevent further origin writes from using this snapshot. */
	/* After this returns there can be no new kcopyd jobs. */
	unregister_snapshot(s);

	while (atomic_read(&s->pending_exceptions_count))
		msleep(1);
	/*
	 * Ensure instructions in mempool_destroy aren't reordered
	 * before atomic_read.
	 */
	smp_mb();

#ifdef CONFIG_DM_DEBUG
	for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
		BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
#endif

	mempool_destroy(s->tracked_chunk_pool);

	__free_exceptions(s);

	mempool_destroy(s->pending_pool);

	dm_put_device(ti, s->origin);

	dm_exception_store_destroy(s->store);

	dm_put_device(ti, s->cow);

	kfree(s);
}

/*
 * Flush a list of buffers.
 */
static void flush_bios(struct bio *bio)
{
	struct bio *n;

	while (bio) {
		n = bio->bi_next;
		bio->bi_next = NULL;
		generic_make_request(bio);
		bio = n;
	}
}

static void flush_queued_bios(struct work_struct *work)
{
	struct dm_snapshot *s =
		container_of(work, struct dm_snapshot, queued_bios_work);
	struct bio *queued_bios;
	unsigned long flags;

	spin_lock_irqsave(&s->pe_lock, flags);
	queued_bios = bio_list_get(&s->queued_bios);
	spin_unlock_irqrestore(&s->pe_lock, flags);

	flush_bios(queued_bios);
}

static int do_origin(struct dm_dev *origin, struct bio *bio);

/*
 * Flush a list of buffers.
 */
static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
{
	struct bio *n;
	int r;

	while (bio) {
		n = bio->bi_next;
		bio->bi_next = NULL;
		r = do_origin(s->origin, bio);
		if (r == DM_MAPIO_REMAPPED)
			generic_make_request(bio);
		bio = n;
	}
}

/*
 * Error a list of buffers.
 */
static void error_bios(struct bio *bio)
{
	struct bio *n;

	while (bio) {
		n = bio->bi_next;
		bio->bi_next = NULL;
		bio_io_error(bio);
		bio = n;
	}
}

static void __invalidate_snapshot(struct dm_snapshot *s, int err)
{
	if (!s->valid)
		return;

	if (err == -EIO)
		DMERR("Invalidating snapshot: Error reading/writing.");
	else if (err == -ENOMEM)
		DMERR("Invalidating snapshot: Unable to allocate exception.");

	if (s->store->type->drop_snapshot)
		s->store->type->drop_snapshot(s->store);

	s->valid = 0;

	dm_table_event(s->ti->table);
}

static void pending_complete(struct dm_snap_pending_exception *pe, int success)
{
	struct dm_exception *e;
	struct dm_snapshot *s = pe->snap;
	struct bio *origin_bios = NULL;
	struct bio *snapshot_bios = NULL;
	int error = 0;

	if (!success) {
		/* Read/write error - snapshot is unusable */
		down_write(&s->lock);
		__invalidate_snapshot(s, -EIO);
		error = 1;
		goto out;
	}

	e = alloc_completed_exception();
	if (!e) {
		down_write(&s->lock);
		__invalidate_snapshot(s, -ENOMEM);
		error = 1;
		goto out;
	}
	*e = pe->e;

	down_write(&s->lock);
	if (!s->valid) {
		free_completed_exception(e);
		error = 1;
		goto out;
	}

	/* Check for conflicting reads */
	__check_for_conflicting_io(s, pe->e.old_chunk);

	/*
	 * Add a proper exception, and remove the
	 * in-flight exception from the list.
	 */
	dm_insert_exception(&s->complete, e);

 out:
	dm_remove_exception(&pe->e);
	snapshot_bios = bio_list_get(&pe->snapshot_bios);
	origin_bios = bio_list_get(&pe->origin_bios);
	free_pending_exception(pe);

	up_write(&s->lock);

	/* Submit any pending write bios */
	if (error)
		error_bios(snapshot_bios);
	else
		flush_bios(snapshot_bios);

	retry_origin_bios(s, origin_bios);
}

static void commit_callback(void *context, int success)
{
	struct dm_snap_pending_exception *pe = context;

	pending_complete(pe, success);
}

/*
 * Called when the copy I/O has finished.  kcopyd actually runs
 * this code so don't block.
 */
static void copy_callback(int read_err, unsigned long write_err, void *context)
{
	struct dm_snap_pending_exception *pe = context;
	struct dm_snapshot *s = pe->snap;

	if (read_err || write_err)
		pending_complete(pe, 0);

	else
		/* Update the metadata if we are persistent */
		s->store->type->commit_exception(s->store, &pe->e,
						 commit_callback, pe);
}

/*
 * Dispatches the copy operation to kcopyd.
 */
static void start_copy(struct dm_snap_pending_exception *pe)
{
	struct dm_snapshot *s = pe->snap;
	struct dm_io_region src, dest;
	struct block_device *bdev = s->origin->bdev;
	sector_t dev_size;

	dev_size = get_dev_size(bdev);

	src.bdev = bdev;
	src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
	src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);

	dest.bdev = s->cow->bdev;
	dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
	dest.count = src.count;

	/* Hand over to kcopyd */
	dm_kcopyd_copy(s->kcopyd_client,
		    &src, 1, &dest, 0, copy_callback, pe);
}

static struct dm_snap_pending_exception *
__lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
{
	struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);

	if (!e)
		return NULL;

	return container_of(e, struct dm_snap_pending_exception, e);
}

/*
 * Looks to see if this snapshot already has a pending exception
 * for this chunk, otherwise it allocates a new one and inserts
 * it into the pending table.
 *
 * NOTE: a write lock must be held on snap->lock before calling
 * this.
 */
static struct dm_snap_pending_exception *
__find_pending_exception(struct dm_snapshot *s,
			 struct dm_snap_pending_exception *pe, chunk_t chunk)
{
	struct dm_snap_pending_exception *pe2;

	pe2 = __lookup_pending_exception(s, chunk);
	if (pe2) {
		free_pending_exception(pe);
		return pe2;
	}

	pe->e.old_chunk = chunk;
	bio_list_init(&pe->origin_bios);
	bio_list_init(&pe->snapshot_bios);
	pe->started = 0;

	if (s->store->type->prepare_exception(s->store, &pe->e)) {
		free_pending_exception(pe);
		return NULL;
	}

	dm_insert_exception(&s->pending, &pe->e);

	return pe;
}

static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
			    struct bio *bio, chunk_t chunk)
{
	bio->bi_bdev = s->cow->bdev;
	bio->bi_sector = chunk_to_sector(s->store,
					 dm_chunk_number(e->new_chunk) +
					 (chunk - e->old_chunk)) +
					 (bio->bi_sector &
					  s->store->chunk_mask);
}

static int snapshot_map(struct dm_target *ti, struct bio *bio,
			union map_info *map_context)
{
	struct dm_exception *e;
	struct dm_snapshot *s = ti->private;
	int r = DM_MAPIO_REMAPPED;
	chunk_t chunk;
	struct dm_snap_pending_exception *pe = NULL;

	if (unlikely(bio_empty_barrier(bio))) {
		bio->bi_bdev = s->cow->bdev;
		return DM_MAPIO_REMAPPED;
	}

	chunk = sector_to_chunk(s->store, bio->bi_sector);

	/* Full snapshots are not usable */
	/* To get here the table must be live so s->active is always set. */
	if (!s->valid)
		return -EIO;

	/* FIXME: should only take write lock if we need
	 * to copy an exception */
	down_write(&s->lock);

	if (!s->valid) {
		r = -EIO;
		goto out_unlock;
	}

	/* If the block is already remapped - use that, else remap it */
	e = dm_lookup_exception(&s->complete, chunk);
	if (e) {
		remap_exception(s, e, bio, chunk);
		goto out_unlock;
	}

	/*
	 * Write to snapshot - higher level takes care of RW/RO
	 * flags so we should only get this if we are
	 * writeable.
	 */
	if (bio_rw(bio) == WRITE) {
		pe = __lookup_pending_exception(s, chunk);
		if (!pe) {
			up_write(&s->lock);
			pe = alloc_pending_exception(s);
			down_write(&s->lock);

			if (!s->valid) {
				free_pending_exception(pe);
				r = -EIO;
				goto out_unlock;
			}

			e = dm_lookup_exception(&s->complete, chunk);
			if (e) {
				free_pending_exception(pe);
				remap_exception(s, e, bio, chunk);
				goto out_unlock;
			}

			pe = __find_pending_exception(s, pe, chunk);
			if (!pe) {
				__invalidate_snapshot(s, -ENOMEM);
				r = -EIO;
				goto out_unlock;
			}
		}

		remap_exception(s, &pe->e, bio, chunk);
		bio_list_add(&pe->snapshot_bios, bio);

		r = DM_MAPIO_SUBMITTED;

		if (!pe->started) {
			/* this is protected by snap->lock */
			pe->started = 1;
			up_write(&s->lock);
			start_copy(pe);
			goto out;
		}
	} else {
		bio->bi_bdev = s->origin->bdev;
		map_context->ptr = track_chunk(s, chunk);
	}

 out_unlock:
	up_write(&s->lock);
 out:
	return r;
}

static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
			   int error, union map_info *map_context)
{
	struct dm_snapshot *s = ti->private;
	struct dm_snap_tracked_chunk *c = map_context->ptr;

	if (c)
		stop_tracking_chunk(s, c);

	return 0;
}

static void snapshot_postsuspend(struct dm_target *ti)
{
	struct dm_snapshot *s = ti->private;

	down_write(&s->lock);
	s->suspended = 1;
	up_write(&s->lock);
}

static int snapshot_preresume(struct dm_target *ti)
{
	int r = 0;
	struct dm_snapshot *s = ti->private;
	struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;

	down_read(&_origins_lock);
	(void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest);
	if (snap_src && snap_dest) {
		down_read(&snap_src->lock);
		if (s == snap_src) {
			DMERR("Unable to resume snapshot source until "
			      "handover completes.");
			r = -EINVAL;
		} else if (!snap_src->suspended) {
			DMERR("Unable to perform snapshot handover until "
			      "source is suspended.");
			r = -EINVAL;
		}
		up_read(&snap_src->lock);
	}
	up_read(&_origins_lock);

	return r;
}

static void snapshot_resume(struct dm_target *ti)
{
	struct dm_snapshot *s = ti->private;
	struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;

	down_read(&_origins_lock);
	(void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest);
	if (snap_src && snap_dest) {
		down_write(&snap_src->lock);
		down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
		__handover_exceptions(snap_src, snap_dest);
		up_write(&snap_dest->lock);
		up_write(&snap_src->lock);
	}
	up_read(&_origins_lock);

	/* Now we have correct chunk size, reregister */
	reregister_snapshot(s);

	down_write(&s->lock);
	s->active = 1;
	s->suspended = 0;
	up_write(&s->lock);
}

static int snapshot_status(struct dm_target *ti, status_type_t type,
			   char *result, unsigned int maxlen)
{
	unsigned sz = 0;
	struct dm_snapshot *snap = ti->private;

	switch (type) {
	case STATUSTYPE_INFO:

		down_write(&snap->lock);

		if (!snap->valid)
			DMEMIT("Invalid");
		else {
			if (snap->store->type->usage) {
				sector_t total_sectors, sectors_allocated,
					 metadata_sectors;
				snap->store->type->usage(snap->store,
							 &total_sectors,
							 &sectors_allocated,
							 &metadata_sectors);
				DMEMIT("%llu/%llu %llu",
				       (unsigned long long)sectors_allocated,
				       (unsigned long long)total_sectors,
				       (unsigned long long)metadata_sectors);
			}
			else
				DMEMIT("Unknown");
		}

		up_write(&snap->lock);

		break;

	case STATUSTYPE_TABLE:
		/*
		 * kdevname returns a static pointer so we need
		 * to make private copies if the output is to
		 * make sense.
		 */
		DMEMIT("%s %s", snap->origin->name, snap->cow->name);
		snap->store->type->status(snap->store, type, result + sz,
					  maxlen - sz);
		break;
	}

	return 0;
}

static int snapshot_iterate_devices(struct dm_target *ti,
				    iterate_devices_callout_fn fn, void *data)
{
	struct dm_snapshot *snap = ti->private;

	return fn(ti, snap->origin, 0, ti->len, data);
}


/*-----------------------------------------------------------------
 * Origin methods
 *---------------------------------------------------------------*/

/*
 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
 * supplied bio was ignored.  The caller may submit it immediately.
 * (No remapping actually occurs as the origin is always a direct linear
 * map.)
 *
 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
 * and any supplied bio is added to a list to be submitted once all
 * the necessary exceptions exist.
 */
static int __origin_write(struct list_head *snapshots, sector_t sector,
			  struct bio *bio)
{
	int r = DM_MAPIO_REMAPPED;
	struct dm_snapshot *snap;
	struct dm_exception *e;
	struct dm_snap_pending_exception *pe;
	struct dm_snap_pending_exception *pe_to_start_now = NULL;
	struct dm_snap_pending_exception *pe_to_start_last = NULL;
	chunk_t chunk;

	/* Do all the snapshots on this origin */
	list_for_each_entry (snap, snapshots, list) {
		down_write(&snap->lock);

		/* Only deal with valid and active snapshots */
		if (!snap->valid || !snap->active)
			goto next_snapshot;

		/* Nothing to do if writing beyond end of snapshot */
		if (sector >= dm_table_get_size(snap->ti->table))
			goto next_snapshot;

		/*
		 * Remember, different snapshots can have
		 * different chunk sizes.
		 */
		chunk = sector_to_chunk(snap->store, sector);

		/*
		 * Check exception table to see if block
		 * is already remapped in this snapshot
		 * and trigger an exception if not.
		 */
		e = dm_lookup_exception(&snap->complete, chunk);
		if (e)
			goto next_snapshot;

		pe = __lookup_pending_exception(snap, chunk);
		if (!pe) {
			up_write(&snap->lock);
			pe = alloc_pending_exception(snap);
			down_write(&snap->lock);

			if (!snap->valid) {
				free_pending_exception(pe);
				goto next_snapshot;
			}

			e = dm_lookup_exception(&snap->complete, chunk);
			if (e) {
				free_pending_exception(pe);
				goto next_snapshot;
			}

			pe = __find_pending_exception(snap, pe, chunk);
			if (!pe) {
				__invalidate_snapshot(snap, -ENOMEM);
				goto next_snapshot;
			}
		}

		r = DM_MAPIO_SUBMITTED;

		/*
		 * If an origin bio was supplied, queue it to wait for the
		 * completion of this exception, and start this one last,
		 * at the end of the function.
		 */
		if (bio) {
			bio_list_add(&pe->origin_bios, bio);
			bio = NULL;

			if (!pe->started) {
				pe->started = 1;
				pe_to_start_last = pe;
			}
		}

		if (!pe->started) {
			pe->started = 1;
			pe_to_start_now = pe;
		}

 next_snapshot:
		up_write(&snap->lock);

		if (pe_to_start_now) {
			start_copy(pe_to_start_now);
			pe_to_start_now = NULL;
		}
	}

	/*
	 * Submit the exception against which the bio is queued last,
	 * to give the other exceptions a head start.
	 */
	if (pe_to_start_last)
		start_copy(pe_to_start_last);

	return r;
}

/*
 * Called on a write from the origin driver.
 */
static int do_origin(struct dm_dev *origin, struct bio *bio)
{
	struct origin *o;
	int r = DM_MAPIO_REMAPPED;

	down_read(&_origins_lock);
	o = __lookup_origin(origin->bdev);
	if (o)
		r = __origin_write(&o->snapshots, bio->bi_sector, bio);
	up_read(&_origins_lock);

	return r;
}

/*
 * Origin: maps a linear range of a device, with hooks for snapshotting.
 */

/*
 * Construct an origin mapping: <dev_path>
 * The context for an origin is merely a 'struct dm_dev *'
 * pointing to the real device.
 */
static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
{
	int r;
	struct dm_dev *dev;

	if (argc != 1) {
		ti->error = "origin: incorrect number of arguments";
		return -EINVAL;
	}

	r = dm_get_device(ti, argv[0], 0, ti->len,
			  dm_table_get_mode(ti->table), &dev);
	if (r) {
		ti->error = "Cannot get target device";
		return r;
	}

	ti->private = dev;
	ti->num_flush_requests = 1;

	return 0;
}

static void origin_dtr(struct dm_target *ti)
{
	struct dm_dev *dev = ti->private;
	dm_put_device(ti, dev);
}

static int origin_map(struct dm_target *ti, struct bio *bio,
		      union map_info *map_context)
{
	struct dm_dev *dev = ti->private;
	bio->bi_bdev = dev->bdev;

	if (unlikely(bio_empty_barrier(bio)))
		return DM_MAPIO_REMAPPED;

	/* Only tell snapshots if this is a write */
	return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
}

/*
 * Set the target "split_io" field to the minimum of all the snapshots'
 * chunk sizes.
 */
static void origin_resume(struct dm_target *ti)
{
	struct dm_dev *dev = ti->private;

	down_read(&_origins_lock);

	ti->split_io = __minimum_chunk_size(__lookup_origin(dev->bdev));

	up_read(&_origins_lock);
}

static int origin_status(struct dm_target *ti, status_type_t type, char *result,
			 unsigned int maxlen)
{
	struct dm_dev *dev = ti->private;

	switch (type) {
	case STATUSTYPE_INFO:
		result[0] = '\0';
		break;

	case STATUSTYPE_TABLE:
		snprintf(result, maxlen, "%s", dev->name);
		break;
	}

	return 0;
}

static int origin_iterate_devices(struct dm_target *ti,
				  iterate_devices_callout_fn fn, void *data)
{
	struct dm_dev *dev = ti->private;

	return fn(ti, dev, 0, ti->len, data);
}

static struct target_type origin_target = {
	.name    = "snapshot-origin",
	.version = {1, 7, 0},
	.module  = THIS_MODULE,
	.ctr     = origin_ctr,
	.dtr     = origin_dtr,
	.map     = origin_map,
	.resume  = origin_resume,
	.status  = origin_status,
	.iterate_devices = origin_iterate_devices,
};

static struct target_type snapshot_target = {
	.name    = "snapshot",
	.version = {1, 9, 0},
	.module  = THIS_MODULE,
	.ctr     = snapshot_ctr,
	.dtr     = snapshot_dtr,
	.map     = snapshot_map,
	.end_io  = snapshot_end_io,
	.postsuspend = snapshot_postsuspend,
	.preresume  = snapshot_preresume,
	.resume  = snapshot_resume,
	.status  = snapshot_status,
	.iterate_devices = snapshot_iterate_devices,
};

static struct target_type merge_target = {
	.name    = dm_snapshot_merge_target_name,
	.version = {1, 0, 0},
	.module  = THIS_MODULE,
	.ctr     = snapshot_ctr,
	.dtr     = snapshot_dtr,
	.map     = snapshot_map,
	.end_io  = snapshot_end_io,
	.postsuspend = snapshot_postsuspend,
	.preresume  = snapshot_preresume,
	.resume  = snapshot_resume,
	.status  = snapshot_status,
	.iterate_devices = snapshot_iterate_devices,
};

static int __init dm_snapshot_init(void)
{
	int r;

	r = dm_exception_store_init();
	if (r) {
		DMERR("Failed to initialize exception stores");
		return r;
	}

	r = dm_register_target(&snapshot_target);
	if (r < 0) {
		DMERR("snapshot target register failed %d", r);
		goto bad_register_snapshot_target;
	}

	r = dm_register_target(&origin_target);
	if (r < 0) {
		DMERR("Origin target register failed %d", r);
		goto bad_register_origin_target;
	}

	r = dm_register_target(&merge_target);
	if (r < 0) {
		DMERR("Merge target register failed %d", r);
		goto bad_register_merge_target;
	}

	r = init_origin_hash();
	if (r) {
		DMERR("init_origin_hash failed.");
		goto bad_origin_hash;
	}

	exception_cache = KMEM_CACHE(dm_exception, 0);
	if (!exception_cache) {
		DMERR("Couldn't create exception cache.");
		r = -ENOMEM;
		goto bad_exception_cache;
	}

	pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
	if (!pending_cache) {
		DMERR("Couldn't create pending cache.");
		r = -ENOMEM;
		goto bad_pending_cache;
	}

	tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0);
	if (!tracked_chunk_cache) {
		DMERR("Couldn't create cache to track chunks in use.");
		r = -ENOMEM;
		goto bad_tracked_chunk_cache;
	}

	ksnapd = create_singlethread_workqueue("ksnapd");
	if (!ksnapd) {
		DMERR("Failed to create ksnapd workqueue.");
		r = -ENOMEM;
		goto bad_pending_pool;
	}

	return 0;

bad_pending_pool:
	kmem_cache_destroy(tracked_chunk_cache);
bad_tracked_chunk_cache:
	kmem_cache_destroy(pending_cache);
bad_pending_cache:
	kmem_cache_destroy(exception_cache);
bad_exception_cache:
	exit_origin_hash();
bad_origin_hash:
	dm_unregister_target(&merge_target);
bad_register_merge_target:
	dm_unregister_target(&origin_target);
bad_register_origin_target:
	dm_unregister_target(&snapshot_target);
bad_register_snapshot_target:
	dm_exception_store_exit();

	return r;
}

static void __exit dm_snapshot_exit(void)
{
	destroy_workqueue(ksnapd);

	dm_unregister_target(&snapshot_target);
	dm_unregister_target(&origin_target);
	dm_unregister_target(&merge_target);

	exit_origin_hash();
	kmem_cache_destroy(pending_cache);
	kmem_cache_destroy(exception_cache);
	kmem_cache_destroy(tracked_chunk_cache);

	dm_exception_store_exit();
}

/* Module hooks */
module_init(dm_snapshot_init);
module_exit(dm_snapshot_exit);

MODULE_DESCRIPTION(DM_NAME " snapshot target");
MODULE_AUTHOR("Joe Thornber");
MODULE_LICENSE("GPL");