#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
*/
struct bio_list origin_bios;
struct bio_list snapshot_bios;
- /*
- * Short-term queue of pending exceptions prior to submission.
- */
- struct list_head list;
-
- /*
- * The primary pending_exception is the one that holds
- * the ref_count and the list of origin_bios for a
- * group of pending_exceptions. It is always last to get freed.
- * These fields get set up when writing to the origin.
- */
- struct dm_snap_pending_exception *primary_pe;
-
- /*
- * Number of pending_exceptions processing this chunk.
- * When this drops to zero we must complete the origin bios.
- * If incrementing or decrementing this, hold pe->snap->lock for
- * the sibling concerned and not pe->primary_pe->snap->lock unless
- * they are the same.
- */
- atomic_t ref_count;
-
/* Pointer back to snapshot context */
struct dm_snapshot *snap;
}
/*
+ * 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 {
}
/*
+ * _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 dm_snapshot *l;
- struct origin *o, *new_o;
+ 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);
- o = __lookup_origin(bdev);
+ r = __validate_exception_handover(snap);
+ if (r < 0) {
+ kfree(new_o);
+ goto out;
+ }
+
+ o = __lookup_origin(bdev);
if (o)
kfree(new_o);
else {
__insert_origin(o);
}
- /* Sort the list according to chunk size, largest-first smallest-last */
- list_for_each_entry(l, &o->snapshots, list)
- if (l->store->chunk_size < snap->store->chunk_size)
- break;
- list_add_tail(&snap->list, &l->list);
+ __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);
- return 0;
}
static void unregister_snapshot(struct dm_snapshot *s)
o = __lookup_origin(s->origin->bdev);
list_del(&s->list);
- if (list_empty(&o->snapshots)) {
+ if (o && list_empty(&o->snapshots)) {
list_del(&o->hash_list);
kfree(o);
}
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 */
spin_lock_init(&s->tracked_chunk_lock);
- /* Metadata must only be loaded into one table at once */
+ 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_load_and_register;
+ goto bad_read_metadata;
} else if (r > 0) {
s->valid = 0;
DMWARN("Snapshot is marked invalid.");
}
- bio_list_init(&s->queued_bios);
- INIT_WORK(&s->queued_bios_work, flush_queued_bios);
-
if (!s->store->chunk_size) {
ti->error = "Chunk size not set";
- goto bad_load_and_register;
- }
-
- /* Add snapshot to the list of snapshots for this origin */
- /* Exceptions aren't triggered till snapshot_resume() is called */
- if (register_snapshot(s)) {
- r = -EINVAL;
- ti->error = "Cannot register snapshot origin";
- goto bad_load_and_register;
+ goto bad_read_metadata;
}
-
- ti->private = s;
ti->split_io = s->store->chunk_size;
- ti->num_flush_requests = 1;
return 0;
+bad_read_metadata:
+ unregister_snapshot(s);
+
bad_load_and_register:
mempool_destroy(s->tracked_chunk_pool);
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);
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.
*/
dm_table_event(s->ti->table);
}
-static void get_pending_exception(struct dm_snap_pending_exception *pe)
-{
- atomic_inc(&pe->ref_count);
-}
-
-static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe)
-{
- struct dm_snap_pending_exception *primary_pe;
- struct bio *origin_bios = NULL;
-
- primary_pe = pe->primary_pe;
-
- /*
- * If this pe is involved in a write to the origin and
- * it is the last sibling to complete then release
- * the bios for the original write to the origin.
- */
- if (primary_pe &&
- atomic_dec_and_test(&primary_pe->ref_count)) {
- origin_bios = bio_list_get(&primary_pe->origin_bios);
- free_pending_exception(primary_pe);
- }
-
- /*
- * Free the pe if it's not linked to an origin write or if
- * it's not itself a primary pe.
- */
- if (!primary_pe || primary_pe != pe)
- free_pending_exception(pe);
-
- return origin_bios;
-}
-
static void pending_complete(struct dm_snap_pending_exception *pe, int success)
{
struct dm_exception *e;
goto out;
}
- /*
- * Check for conflicting reads. This is extremely improbable,
- * so msleep(1) is sufficient and there is no need for a wait queue.
- */
- while (__chunk_is_tracked(s, pe->e.old_chunk))
- msleep(1);
+ /* Check for conflicting reads */
+ __check_for_conflicting_io(s, pe->e.old_chunk);
/*
* Add a proper exception, and remove the
out:
dm_remove_exception(&pe->e);
snapshot_bios = bio_list_get(&pe->snapshot_bios);
- origin_bios = put_pending_exception(pe);
+ origin_bios = bio_list_get(&pe->origin_bios);
+ free_pending_exception(pe);
up_write(&s->lock);
else
flush_bios(snapshot_bios);
- flush_bios(origin_bios);
+ retry_origin_bios(s, origin_bios);
}
static void commit_callback(void *context, int success)
pe->e.old_chunk = chunk;
bio_list_init(&pe->origin_bios);
bio_list_init(&pe->snapshot_bios);
- pe->primary_pe = NULL;
- atomic_set(&pe->ref_count, 0);
pe->started = 0;
if (s->store->type->prepare_exception(s->store, &pe->e)) {
return NULL;
}
- get_pending_exception(pe);
dm_insert_exception(&s->pending, &pe->e);
return pe;
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;
/*-----------------------------------------------------------------
* Origin methods
*---------------------------------------------------------------*/
-static int __origin_write(struct list_head *snapshots, struct bio *bio)
+
+/*
+ * 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, first = 0;
+ int r = DM_MAPIO_REMAPPED;
struct dm_snapshot *snap;
struct dm_exception *e;
- struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL;
+ 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;
- LIST_HEAD(pe_queue);
/* 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 */
goto next_snapshot;
/* Nothing to do if writing beyond end of snapshot */
- if (bio->bi_sector >= dm_table_get_size(snap->ti->table))
+ 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, bio->bi_sector);
+ 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.
- *
- * ref_count is initialised to 1 so pending_complete()
- * won't destroy the primary_pe while we're inside this loop.
*/
e = dm_lookup_exception(&snap->complete, chunk);
if (e)
}
}
- if (!primary_pe) {
- /*
- * Either every pe here has same
- * primary_pe or none has one yet.
- */
- if (pe->primary_pe)
- primary_pe = pe->primary_pe;
- else {
- primary_pe = pe;
- first = 1;
- }
-
- bio_list_add(&primary_pe->origin_bios, bio);
+ r = DM_MAPIO_SUBMITTED;
- 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->primary_pe) {
- pe->primary_pe = primary_pe;
- get_pending_exception(primary_pe);
+ if (!pe->started) {
+ pe->started = 1;
+ pe_to_start_last = pe;
+ }
}
if (!pe->started) {
pe->started = 1;
- list_add_tail(&pe->list, &pe_queue);
+ pe_to_start_now = pe;
}
next_snapshot:
up_write(&snap->lock);
- }
-
- if (!primary_pe)
- return r;
-
- /*
- * If this is the first time we're processing this chunk and
- * ref_count is now 1 it means all the pending exceptions
- * got completed while we were in the loop above, so it falls to
- * us here to remove the primary_pe and submit any origin_bios.
- */
- if (first && atomic_dec_and_test(&primary_pe->ref_count)) {
- flush_bios(bio_list_get(&primary_pe->origin_bios));
- free_pending_exception(primary_pe);
- /* If we got here, pe_queue is necessarily empty. */
- return r;
+ if (pe_to_start_now) {
+ start_copy(pe_to_start_now);
+ pe_to_start_now = NULL;
+ }
}
/*
- * Now that we have a complete pe list we can start the copying.
+ * Submit the exception against which the bio is queued last,
+ * to give the other exceptions a head start.
*/
- list_for_each_entry_safe(pe, next_pe, &pe_queue, list)
- start_copy(pe);
+ if (pe_to_start_last)
+ start_copy(pe_to_start_last);
return r;
}
down_read(&_origins_lock);
o = __lookup_origin(origin->bdev);
if (o)
- r = __origin_write(&o->snapshots, bio);
+ r = __origin_write(&o->snapshots, bio->bi_sector, bio);
up_read(&_origins_lock);
return r;
.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,
}
r = dm_register_target(&snapshot_target);
- if (r) {
+ 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 bad1;
+ 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 bad2;
+ goto bad_origin_hash;
}
exception_cache = KMEM_CACHE(dm_exception, 0);
if (!exception_cache) {
DMERR("Couldn't create exception cache.");
r = -ENOMEM;
- goto bad3;
+ 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 bad4;
+ 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 bad5;
+ goto bad_tracked_chunk_cache;
}
ksnapd = create_singlethread_workqueue("ksnapd");
bad_pending_pool:
kmem_cache_destroy(tracked_chunk_cache);
-bad5:
+bad_tracked_chunk_cache:
kmem_cache_destroy(pending_cache);
-bad4:
+bad_pending_cache:
kmem_cache_destroy(exception_cache);
-bad3:
+bad_exception_cache:
exit_origin_hash();
-bad2:
+bad_origin_hash:
+ dm_unregister_target(&merge_target);
+bad_register_merge_target:
dm_unregister_target(&origin_target);
-bad1:
+bad_register_origin_target:
dm_unregister_target(&snapshot_target);
-
bad_register_snapshot_target:
dm_exception_store_exit();
+
return r;
}
dm_unregister_target(&snapshot_target);
dm_unregister_target(&origin_target);
+ dm_unregister_target(&merge_target);
exit_origin_hash();
kmem_cache_destroy(pending_cache);