/* List of snapshots per Origin */
struct list_head list;
- /* You can't use a snapshot if this is 0 (e.g. if full) */
+ /*
+ * You can't use a snapshot if this is 0 (e.g. if full).
+ * A snapshot-merge target never clears this.
+ */
int valid;
/* Origin writes don't trigger exceptions until this is set */
/* Whether or not owning mapped_device is suspended */
int suspended;
- mempool_t *pending_pool;
-
atomic_t pending_exceptions_count;
+ mempool_t *pending_pool;
+
struct dm_exception_table pending;
struct dm_exception_table complete;
*/
spinlock_t pe_lock;
+ /* Chunks with outstanding reads */
+ spinlock_t tracked_chunk_lock;
+ mempool_t *tracked_chunk_pool;
+ struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
+
/* The on disk metadata handler */
struct dm_exception_store *store;
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];
+ /* Wait for events based on state_bits */
+ unsigned long state_bits;
+
+ /* Range of chunks currently being merged. */
+ chunk_t first_merging_chunk;
+ int num_merging_chunks;
+
+ /*
+ * The merge operation failed if this flag is set.
+ * Failure modes are handled as follows:
+ * - I/O error reading the header
+ * => don't load the target; abort.
+ * - Header does not have "valid" flag set
+ * => use the origin; forget about the snapshot.
+ * - I/O error when reading exceptions
+ * => don't load the target; abort.
+ * (We can't use the intermediate origin state.)
+ * - I/O error while merging
+ * => stop merging; set merge_failed; process I/O normally.
+ */
+ int merge_failed;
+
+ /*
+ * Incoming bios that overlap with chunks being merged must wait
+ * for them to be committed.
+ */
+ struct bio_list bios_queued_during_merge;
};
+/*
+ * state_bits:
+ * RUNNING_MERGE - Merge operation is in progress.
+ * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped;
+ * cleared afterwards.
+ */
+#define RUNNING_MERGE 0
+#define SHUTDOWN_MERGE 1
+
struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
{
return s->cow;
static struct list_head *_origins;
static struct rw_semaphore _origins_lock;
+static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
+static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
+static uint64_t _pending_exceptions_done_count;
+
static int init_origin_hash(void)
{
int i;
* 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.
+ * snap_merge - an existing snapshot-merge target linked to the same origin.
+ * There can be at most one snapshot-merge target. The parameter is optional.
*
- * Possible return values and states:
+ * Possible return values and states of snap_src and snap_dest.
* 0: NULL, NULL - first new snapshot
* 1: snap_src, NULL - normal snapshot
* 2: snap_src, snap_dest - waiting for handover
*/
static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
struct dm_snapshot **snap_src,
- struct dm_snapshot **snap_dest)
+ struct dm_snapshot **snap_dest,
+ struct dm_snapshot **snap_merge)
{
struct dm_snapshot *s;
struct origin *o;
goto out;
list_for_each_entry(s, &o->snapshots, list) {
+ if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
+ *snap_merge = s;
if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
continue;
static int __validate_exception_handover(struct dm_snapshot *snap)
{
struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
+ struct dm_snapshot *snap_merge = NULL;
/* Does snapshot need exceptions handed over to it? */
- if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest) == 2) ||
+ if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
+ &snap_merge) == 2) ||
snap_dest) {
snap->ti->error = "Snapshot cow pairing for exception "
"table handover failed";
if (!snap_src)
return 0;
+ /*
+ * Non-snapshot-merge handover?
+ */
+ if (!dm_target_is_snapshot_merge(snap->ti))
+ return 1;
+
+ /*
+ * Do not allow more than one merging snapshot.
+ */
+ if (snap_merge) {
+ snap->ti->error = "A snapshot is already merging.";
+ return -EINVAL;
+ }
+
+ if (!snap_src->store->type->prepare_merge ||
+ !snap_src->store->type->commit_merge) {
+ snap->ti->error = "Snapshot exception store does not "
+ "support snapshot-merge.";
+ return -EINVAL;
+ }
+
return 1;
}
return 0;
}
+static void merge_shutdown(struct dm_snapshot *s)
+{
+ clear_bit_unlock(RUNNING_MERGE, &s->state_bits);
+ smp_mb__after_clear_bit();
+ wake_up_bit(&s->state_bits, RUNNING_MERGE);
+}
+
+static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s)
+{
+ s->first_merging_chunk = 0;
+ s->num_merging_chunks = 0;
+
+ return bio_list_get(&s->bios_queued_during_merge);
+}
+
+/*
+ * Remove one chunk from the index of completed exceptions.
+ */
+static int __remove_single_exception_chunk(struct dm_snapshot *s,
+ chunk_t old_chunk)
+{
+ struct dm_exception *e;
+
+ e = dm_lookup_exception(&s->complete, old_chunk);
+ if (!e) {
+ DMERR("Corruption detected: exception for block %llu is "
+ "on disk but not in memory",
+ (unsigned long long)old_chunk);
+ return -EINVAL;
+ }
+
+ /*
+ * If this is the only chunk using this exception, remove exception.
+ */
+ if (!dm_consecutive_chunk_count(e)) {
+ dm_remove_exception(e);
+ free_completed_exception(e);
+ return 0;
+ }
+
+ /*
+ * The chunk may be either at the beginning or the end of a
+ * group of consecutive chunks - never in the middle. We are
+ * removing chunks in the opposite order to that in which they
+ * were added, so this should always be true.
+ * Decrement the consecutive chunk counter and adjust the
+ * starting point if necessary.
+ */
+ if (old_chunk == e->old_chunk) {
+ e->old_chunk++;
+ e->new_chunk++;
+ } else if (old_chunk != e->old_chunk +
+ dm_consecutive_chunk_count(e)) {
+ DMERR("Attempt to merge block %llu from the "
+ "middle of a chunk range [%llu - %llu]",
+ (unsigned long long)old_chunk,
+ (unsigned long long)e->old_chunk,
+ (unsigned long long)
+ e->old_chunk + dm_consecutive_chunk_count(e));
+ return -EINVAL;
+ }
+
+ dm_consecutive_chunk_count_dec(e);
+
+ return 0;
+}
+
+static void flush_bios(struct bio *bio);
+
+static int remove_single_exception_chunk(struct dm_snapshot *s)
+{
+ struct bio *b = NULL;
+ int r;
+ chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1;
+
+ down_write(&s->lock);
+
+ /*
+ * Process chunks (and associated exceptions) in reverse order
+ * so that dm_consecutive_chunk_count_dec() accounting works.
+ */
+ do {
+ r = __remove_single_exception_chunk(s, old_chunk);
+ if (r)
+ goto out;
+ } while (old_chunk-- > s->first_merging_chunk);
+
+ b = __release_queued_bios_after_merge(s);
+
+out:
+ up_write(&s->lock);
+ if (b)
+ flush_bios(b);
+
+ return r;
+}
+
+static int origin_write_extent(struct dm_snapshot *merging_snap,
+ sector_t sector, unsigned chunk_size);
+
+static void merge_callback(int read_err, unsigned long write_err,
+ void *context);
+
+static uint64_t read_pending_exceptions_done_count(void)
+{
+ uint64_t pending_exceptions_done;
+
+ spin_lock(&_pending_exceptions_done_spinlock);
+ pending_exceptions_done = _pending_exceptions_done_count;
+ spin_unlock(&_pending_exceptions_done_spinlock);
+
+ return pending_exceptions_done;
+}
+
+static void increment_pending_exceptions_done_count(void)
+{
+ spin_lock(&_pending_exceptions_done_spinlock);
+ _pending_exceptions_done_count++;
+ spin_unlock(&_pending_exceptions_done_spinlock);
+
+ wake_up_all(&_pending_exceptions_done);
+}
+
+static void snapshot_merge_next_chunks(struct dm_snapshot *s)
+{
+ int i, linear_chunks;
+ chunk_t old_chunk, new_chunk;
+ struct dm_io_region src, dest;
+ sector_t io_size;
+ uint64_t previous_count;
+
+ BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
+ if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
+ goto shut;
+
+ /*
+ * valid flag never changes during merge, so no lock required.
+ */
+ if (!s->valid) {
+ DMERR("Snapshot is invalid: can't merge");
+ goto shut;
+ }
+
+ linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
+ &new_chunk);
+ if (linear_chunks <= 0) {
+ if (linear_chunks < 0) {
+ DMERR("Read error in exception store: "
+ "shutting down merge");
+ down_write(&s->lock);
+ s->merge_failed = 1;
+ up_write(&s->lock);
+ }
+ goto shut;
+ }
+
+ /* Adjust old_chunk and new_chunk to reflect start of linear region */
+ old_chunk = old_chunk + 1 - linear_chunks;
+ new_chunk = new_chunk + 1 - linear_chunks;
+
+ /*
+ * Use one (potentially large) I/O to copy all 'linear_chunks'
+ * from the exception store to the origin
+ */
+ io_size = linear_chunks * s->store->chunk_size;
+
+ dest.bdev = s->origin->bdev;
+ dest.sector = chunk_to_sector(s->store, old_chunk);
+ dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector);
+
+ src.bdev = s->cow->bdev;
+ src.sector = chunk_to_sector(s->store, new_chunk);
+ src.count = dest.count;
+
+ /*
+ * Reallocate any exceptions needed in other snapshots then
+ * wait for the pending exceptions to complete.
+ * Each time any pending exception (globally on the system)
+ * completes we are woken and repeat the process to find out
+ * if we can proceed. While this may not seem a particularly
+ * efficient algorithm, it is not expected to have any
+ * significant impact on performance.
+ */
+ previous_count = read_pending_exceptions_done_count();
+ while (origin_write_extent(s, dest.sector, io_size)) {
+ wait_event(_pending_exceptions_done,
+ (read_pending_exceptions_done_count() !=
+ previous_count));
+ /* Retry after the wait, until all exceptions are done. */
+ previous_count = read_pending_exceptions_done_count();
+ }
+
+ down_write(&s->lock);
+ s->first_merging_chunk = old_chunk;
+ s->num_merging_chunks = linear_chunks;
+ up_write(&s->lock);
+
+ /* Wait until writes to all 'linear_chunks' drain */
+ for (i = 0; i < linear_chunks; i++)
+ __check_for_conflicting_io(s, old_chunk + i);
+
+ dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s);
+ return;
+
+shut:
+ merge_shutdown(s);
+}
+
+static void error_bios(struct bio *bio);
+
+static void merge_callback(int read_err, unsigned long write_err, void *context)
+{
+ struct dm_snapshot *s = context;
+ struct bio *b = NULL;
+
+ if (read_err || write_err) {
+ if (read_err)
+ DMERR("Read error: shutting down merge.");
+ else
+ DMERR("Write error: shutting down merge.");
+ goto shut;
+ }
+
+ if (s->store->type->commit_merge(s->store,
+ s->num_merging_chunks) < 0) {
+ DMERR("Write error in exception store: shutting down merge");
+ goto shut;
+ }
+
+ if (remove_single_exception_chunk(s) < 0)
+ goto shut;
+
+ snapshot_merge_next_chunks(s);
+
+ return;
+
+shut:
+ down_write(&s->lock);
+ s->merge_failed = 1;
+ b = __release_queued_bios_after_merge(s);
+ up_write(&s->lock);
+ error_bios(b);
+
+ merge_shutdown(s);
+}
+
+static void start_merge(struct dm_snapshot *s)
+{
+ if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits))
+ snapshot_merge_next_chunks(s);
+}
+
+static int wait_schedule(void *ptr)
+{
+ schedule();
+
+ return 0;
+}
+
+/*
+ * Stop the merging process and wait until it finishes.
+ */
+static void stop_merge(struct dm_snapshot *s)
+{
+ set_bit(SHUTDOWN_MERGE, &s->state_bits);
+ wait_on_bit(&s->state_bits, RUNNING_MERGE, wait_schedule,
+ TASK_UNINTERRUPTIBLE);
+ clear_bit(SHUTDOWN_MERGE, &s->state_bits);
+}
+
/*
* Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
*/
int i;
int r = -EINVAL;
char *origin_path, *cow_path;
- unsigned args_used;
+ unsigned args_used, num_flush_requests = 1;
+ fmode_t origin_mode = FMODE_READ;
if (argc != 4) {
ti->error = "requires exactly 4 arguments";
goto bad;
}
+ if (dm_target_is_snapshot_merge(ti)) {
+ num_flush_requests = 2;
+ origin_mode = FMODE_WRITE;
+ }
+
origin_path = argv[0];
argv++;
argc--;
argv++;
argc--;
- r = dm_get_device(ti, cow_path, 0, 0,
- FMODE_READ | FMODE_WRITE, &s->cow);
+ r = dm_get_device(ti, cow_path, FMODE_READ | FMODE_WRITE, &s->cow);
if (r) {
ti->error = "Cannot get COW device";
goto bad_cow;
argv += args_used;
argc -= args_used;
- r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin);
+ r = dm_get_device(ti, origin_path, origin_mode, &s->origin);
if (r) {
ti->error = "Cannot get origin device";
goto bad_origin;
init_rwsem(&s->lock);
INIT_LIST_HEAD(&s->list);
spin_lock_init(&s->pe_lock);
+ s->state_bits = 0;
+ s->merge_failed = 0;
+ s->first_merging_chunk = 0;
+ s->num_merging_chunks = 0;
+ bio_list_init(&s->bios_queued_during_merge);
/* Allocate hash table for COW data */
if (init_hash_tables(s)) {
INIT_WORK(&s->queued_bios_work, flush_queued_bios);
ti->private = s;
- ti->num_flush_requests = 1;
+ ti->num_flush_requests = num_flush_requests;
/* Add snapshot to the list of snapshots for this origin */
/* Exceptions aren't triggered till snapshot_resume() is called */
down_read(&_origins_lock);
/* Check whether exception handover must be cancelled */
- (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest);
+ (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
if (snap_src && snap_dest && (s == snap_src)) {
down_write(&snap_dest->lock);
snap_dest->valid = 0;
}
up_read(&_origins_lock);
+ if (dm_target_is_snapshot_merge(ti))
+ stop_merge(s);
+
/* Prevent further origin writes from using this snapshot. */
/* After this returns there can be no new kcopyd jobs. */
unregister_snapshot(s);
origin_bios = bio_list_get(&pe->origin_bios);
free_pending_exception(pe);
+ increment_pending_exceptions_done_count();
+
up_write(&s->lock);
/* Submit any pending write bios */
* For each chunk, if there is an existing exception, it is used to
* redirect I/O to the cow device. Otherwise I/O is sent to the origin,
* which in turn might generate exceptions in other snapshots.
+ * If merging is currently taking place on the chunk in question, the
+ * I/O is deferred by adding it to s->bios_queued_during_merge.
*/
static int snapshot_merge_map(struct dm_target *ti, struct bio *bio,
union map_info *map_context)
int r = DM_MAPIO_REMAPPED;
chunk_t chunk;
+ if (unlikely(bio_empty_barrier(bio))) {
+ if (!map_context->flush_request)
+ bio->bi_bdev = s->origin->bdev;
+ else
+ bio->bi_bdev = s->cow->bdev;
+ map_context->ptr = NULL;
+ return DM_MAPIO_REMAPPED;
+ }
+
chunk = sector_to_chunk(s->store, bio->bi_sector);
- down_read(&s->lock);
+ down_write(&s->lock);
- /* Full snapshots are not usable */
- if (!s->valid) {
- r = -EIO;
- goto out_unlock;
- }
+ /* Full merging snapshots are redirected to the origin */
+ if (!s->valid)
+ goto redirect_to_origin;
/* If the block is already remapped - use that */
e = dm_lookup_exception(&s->complete, chunk);
if (e) {
+ /* Queue writes overlapping with chunks being merged */
+ if (bio_rw(bio) == WRITE &&
+ chunk >= s->first_merging_chunk &&
+ chunk < (s->first_merging_chunk +
+ s->num_merging_chunks)) {
+ bio->bi_bdev = s->origin->bdev;
+ bio_list_add(&s->bios_queued_during_merge, bio);
+ r = DM_MAPIO_SUBMITTED;
+ goto out_unlock;
+ }
+
remap_exception(s, e, bio, chunk);
+
+ if (bio_rw(bio) == WRITE)
+ map_context->ptr = track_chunk(s, chunk);
goto out_unlock;
}
+redirect_to_origin:
bio->bi_bdev = s->origin->bdev;
if (bio_rw(bio) == WRITE) {
- up_read(&s->lock);
+ up_write(&s->lock);
return do_origin(s->origin, bio);
}
out_unlock:
- up_read(&s->lock);
+ up_write(&s->lock);
return r;
}
return 0;
}
+static void snapshot_merge_presuspend(struct dm_target *ti)
+{
+ struct dm_snapshot *s = ti->private;
+
+ stop_merge(s);
+}
+
static void snapshot_postsuspend(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);
+ (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
if (snap_src && snap_dest) {
down_read(&snap_src->lock);
if (s == snap_src) {
struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
down_read(&_origins_lock);
- (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest);
+ (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
if (snap_src && snap_dest) {
down_write(&snap_src->lock);
down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
up_write(&s->lock);
}
+static sector_t get_origin_minimum_chunksize(struct block_device *bdev)
+{
+ sector_t min_chunksize;
+
+ down_read(&_origins_lock);
+ min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
+ up_read(&_origins_lock);
+
+ return min_chunksize;
+}
+
+static void snapshot_merge_resume(struct dm_target *ti)
+{
+ struct dm_snapshot *s = ti->private;
+
+ /*
+ * Handover exceptions from existing snapshot.
+ */
+ snapshot_resume(ti);
+
+ /*
+ * snapshot-merge acts as an origin, so set ti->split_io
+ */
+ ti->split_io = get_origin_minimum_chunksize(s->origin->bdev);
+
+ start_merge(s);
+}
+
static int snapshot_status(struct dm_target *ti, status_type_t type,
char *result, unsigned int maxlen)
{
if (!snap->valid)
DMEMIT("Invalid");
+ else if (snap->merge_failed)
+ DMEMIT("Merge failed");
else {
if (snap->store->type->usage) {
sector_t total_sectors, sectors_allocated,
}
/*
+ * Trigger exceptions in all non-merging snapshots.
+ *
+ * The chunk size of the merging snapshot may be larger than the chunk
+ * size of some other snapshot so we may need to reallocate multiple
+ * chunks in other snapshots.
+ *
+ * We scan all the overlapping exceptions in the other snapshots.
+ * Returns 1 if anything was reallocated and must be waited for,
+ * otherwise returns 0.
+ *
+ * size must be a multiple of merging_snap's chunk_size.
+ */
+static int origin_write_extent(struct dm_snapshot *merging_snap,
+ sector_t sector, unsigned size)
+{
+ int must_wait = 0;
+ sector_t n;
+ struct origin *o;
+
+ /*
+ * The origin's __minimum_chunk_size() got stored in split_io
+ * by snapshot_merge_resume().
+ */
+ down_read(&_origins_lock);
+ o = __lookup_origin(merging_snap->origin->bdev);
+ for (n = 0; n < size; n += merging_snap->ti->split_io)
+ if (__origin_write(&o->snapshots, sector + n, NULL) ==
+ DM_MAPIO_SUBMITTED)
+ must_wait = 1;
+ up_read(&_origins_lock);
+
+ return must_wait;
+}
+
+/*
* Origin: maps a linear range of a device, with hooks for snapshotting.
*/
return -EINVAL;
}
- r = dm_get_device(ti, argv[0], 0, ti->len,
- dm_table_get_mode(ti->table), &dev);
+ r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev);
if (r) {
ti->error = "Cannot get target device";
return r;
{
struct dm_dev *dev = ti->private;
- down_read(&_origins_lock);
-
- ti->split_io = __minimum_chunk_size(__lookup_origin(dev->bdev));
-
- up_read(&_origins_lock);
+ ti->split_io = get_origin_minimum_chunksize(dev->bdev);
}
static int origin_status(struct dm_target *ti, status_type_t type, char *result,
.dtr = snapshot_dtr,
.map = snapshot_merge_map,
.end_io = snapshot_end_io,
+ .presuspend = snapshot_merge_presuspend,
.postsuspend = snapshot_postsuspend,
.preresume = snapshot_preresume,
- .resume = snapshot_resume,
+ .resume = snapshot_merge_resume,
.status = snapshot_status,
.iterate_devices = snapshot_iterate_devices,
};