4 * Copyright (C) 2001-2002 Sistina Software (UK) Limited.
6 * This file is released under the GPL.
9 #include <linux/blkdev.h>
10 #include <linux/device-mapper.h>
11 #include <linux/delay.h>
13 #include <linux/init.h>
14 #include <linux/kdev_t.h>
15 #include <linux/list.h>
16 #include <linux/mempool.h>
17 #include <linux/module.h>
18 #include <linux/slab.h>
19 #include <linux/vmalloc.h>
20 #include <linux/log2.h>
21 #include <linux/dm-kcopyd.h>
25 #include "dm-exception-store.h"
27 #define DM_MSG_PREFIX "snapshots"
29 static const char dm_snapshot_merge_target_name[] = "snapshot-merge";
31 #define dm_target_is_snapshot_merge(ti) \
32 ((ti)->type->name == dm_snapshot_merge_target_name)
35 * The size of the mempool used to track chunks in use.
39 #define DM_TRACKED_CHUNK_HASH_SIZE 16
40 #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
41 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
43 struct dm_exception_table {
46 struct list_head *table;
52 struct dm_dev *origin;
57 /* List of snapshots per Origin */
58 struct list_head list;
61 * You can't use a snapshot if this is 0 (e.g. if full).
62 * A snapshot-merge target never clears this.
67 * The snapshot overflowed because of a write to the snapshot device.
68 * We don't have to invalidate the snapshot in this case, but we need
69 * to prevent further writes.
71 int snapshot_overflowed;
73 /* Origin writes don't trigger exceptions until this is set */
76 atomic_t pending_exceptions_count;
78 /* Protected by "lock" */
79 sector_t exception_start_sequence;
81 /* Protected by kcopyd single-threaded callback */
82 sector_t exception_complete_sequence;
85 * A list of pending exceptions that completed out of order.
86 * Protected by kcopyd single-threaded callback.
88 struct list_head out_of_order_list;
90 mempool_t *pending_pool;
92 struct dm_exception_table pending;
93 struct dm_exception_table complete;
96 * pe_lock protects all pending_exception operations and access
97 * as well as the snapshot_bios list.
101 /* Chunks with outstanding reads */
102 spinlock_t tracked_chunk_lock;
103 struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
105 /* The on disk metadata handler */
106 struct dm_exception_store *store;
108 unsigned in_progress;
109 wait_queue_head_t in_progress_wait;
111 struct dm_kcopyd_client *kcopyd_client;
113 /* Wait for events based on state_bits */
114 unsigned long state_bits;
116 /* Range of chunks currently being merged. */
117 chunk_t first_merging_chunk;
118 int num_merging_chunks;
121 * The merge operation failed if this flag is set.
122 * Failure modes are handled as follows:
123 * - I/O error reading the header
124 * => don't load the target; abort.
125 * - Header does not have "valid" flag set
126 * => use the origin; forget about the snapshot.
127 * - I/O error when reading exceptions
128 * => don't load the target; abort.
129 * (We can't use the intermediate origin state.)
130 * - I/O error while merging
131 * => stop merging; set merge_failed; process I/O normally.
136 * Incoming bios that overlap with chunks being merged must wait
137 * for them to be committed.
139 struct bio_list bios_queued_during_merge;
144 * RUNNING_MERGE - Merge operation is in progress.
145 * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped;
146 * cleared afterwards.
148 #define RUNNING_MERGE 0
149 #define SHUTDOWN_MERGE 1
152 * Maximum number of chunks being copied on write.
154 * The value was decided experimentally as a trade-off between memory
155 * consumption, stalling the kernel's workqueues and maintaining a high enough
158 #define DEFAULT_COW_THRESHOLD 2048
160 static unsigned cow_threshold = DEFAULT_COW_THRESHOLD;
161 module_param_named(snapshot_cow_threshold, cow_threshold, uint, 0644);
162 MODULE_PARM_DESC(snapshot_cow_threshold, "Maximum number of chunks being copied on write");
164 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
165 "A percentage of time allocated for copy on write");
167 struct dm_dev *dm_snap_origin(struct dm_snapshot *s)
171 EXPORT_SYMBOL(dm_snap_origin);
173 struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
177 EXPORT_SYMBOL(dm_snap_cow);
179 static sector_t chunk_to_sector(struct dm_exception_store *store,
182 return chunk << store->chunk_shift;
185 static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
188 * There is only ever one instance of a particular block
189 * device so we can compare pointers safely.
194 struct dm_snap_pending_exception {
195 struct dm_exception e;
198 * Origin buffers waiting for this to complete are held
201 struct bio_list origin_bios;
202 struct bio_list snapshot_bios;
204 /* Pointer back to snapshot context */
205 struct dm_snapshot *snap;
208 * 1 indicates the exception has already been sent to
213 /* There was copying error. */
216 /* A sequence number, it is used for in-order completion. */
217 sector_t exception_sequence;
219 struct list_head out_of_order_entry;
222 * For writing a complete chunk, bypassing the copy.
224 struct bio *full_bio;
225 bio_end_io_t *full_bio_end_io;
226 void *full_bio_private;
230 * Hash table mapping origin volumes to lists of snapshots and
231 * a lock to protect it
233 static struct kmem_cache *exception_cache;
234 static struct kmem_cache *pending_cache;
236 struct dm_snap_tracked_chunk {
237 struct hlist_node node;
241 static void init_tracked_chunk(struct bio *bio)
243 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
244 INIT_HLIST_NODE(&c->node);
247 static bool is_bio_tracked(struct bio *bio)
249 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
250 return !hlist_unhashed(&c->node);
253 static void track_chunk(struct dm_snapshot *s, struct bio *bio, chunk_t chunk)
255 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
259 spin_lock_irq(&s->tracked_chunk_lock);
260 hlist_add_head(&c->node,
261 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
262 spin_unlock_irq(&s->tracked_chunk_lock);
265 static void stop_tracking_chunk(struct dm_snapshot *s, struct bio *bio)
267 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
270 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
272 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
275 static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
277 struct dm_snap_tracked_chunk *c;
280 spin_lock_irq(&s->tracked_chunk_lock);
282 hlist_for_each_entry(c,
283 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
284 if (c->chunk == chunk) {
290 spin_unlock_irq(&s->tracked_chunk_lock);
296 * This conflicting I/O is extremely improbable in the caller,
297 * so msleep(1) is sufficient and there is no need for a wait queue.
299 static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
301 while (__chunk_is_tracked(s, chunk))
306 * One of these per registered origin, held in the snapshot_origins hash
309 /* The origin device */
310 struct block_device *bdev;
312 struct list_head hash_list;
314 /* List of snapshots for this origin */
315 struct list_head snapshots;
319 * This structure is allocated for each origin target
323 struct dm_target *ti;
324 unsigned split_boundary;
325 struct list_head hash_list;
329 * Size of the hash table for origin volumes. If we make this
330 * the size of the minors list then it should be nearly perfect
332 #define ORIGIN_HASH_SIZE 256
333 #define ORIGIN_MASK 0xFF
334 static struct list_head *_origins;
335 static struct list_head *_dm_origins;
336 static struct rw_semaphore _origins_lock;
338 static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
339 static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
340 static uint64_t _pending_exceptions_done_count;
342 static int init_origin_hash(void)
346 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
349 DMERR("unable to allocate memory for _origins");
352 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
353 INIT_LIST_HEAD(_origins + i);
355 _dm_origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
358 DMERR("unable to allocate memory for _dm_origins");
362 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
363 INIT_LIST_HEAD(_dm_origins + i);
365 init_rwsem(&_origins_lock);
370 static void exit_origin_hash(void)
376 static unsigned origin_hash(struct block_device *bdev)
378 return bdev->bd_dev & ORIGIN_MASK;
381 static struct origin *__lookup_origin(struct block_device *origin)
383 struct list_head *ol;
386 ol = &_origins[origin_hash(origin)];
387 list_for_each_entry (o, ol, hash_list)
388 if (bdev_equal(o->bdev, origin))
394 static void __insert_origin(struct origin *o)
396 struct list_head *sl = &_origins[origin_hash(o->bdev)];
397 list_add_tail(&o->hash_list, sl);
400 static struct dm_origin *__lookup_dm_origin(struct block_device *origin)
402 struct list_head *ol;
405 ol = &_dm_origins[origin_hash(origin)];
406 list_for_each_entry (o, ol, hash_list)
407 if (bdev_equal(o->dev->bdev, origin))
413 static void __insert_dm_origin(struct dm_origin *o)
415 struct list_head *sl = &_dm_origins[origin_hash(o->dev->bdev)];
416 list_add_tail(&o->hash_list, sl);
419 static void __remove_dm_origin(struct dm_origin *o)
421 list_del(&o->hash_list);
425 * _origins_lock must be held when calling this function.
426 * Returns number of snapshots registered using the supplied cow device, plus:
427 * snap_src - a snapshot suitable for use as a source of exception handover
428 * snap_dest - a snapshot capable of receiving exception handover.
429 * snap_merge - an existing snapshot-merge target linked to the same origin.
430 * There can be at most one snapshot-merge target. The parameter is optional.
432 * Possible return values and states of snap_src and snap_dest.
433 * 0: NULL, NULL - first new snapshot
434 * 1: snap_src, NULL - normal snapshot
435 * 2: snap_src, snap_dest - waiting for handover
436 * 2: snap_src, NULL - handed over, waiting for old to be deleted
437 * 1: NULL, snap_dest - source got destroyed without handover
439 static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
440 struct dm_snapshot **snap_src,
441 struct dm_snapshot **snap_dest,
442 struct dm_snapshot **snap_merge)
444 struct dm_snapshot *s;
449 o = __lookup_origin(snap->origin->bdev);
453 list_for_each_entry(s, &o->snapshots, list) {
454 if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
456 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
459 mutex_lock(&s->lock);
461 mutex_unlock(&s->lock);
466 } else if (snap_dest)
477 * On success, returns 1 if this snapshot is a handover destination,
478 * otherwise returns 0.
480 static int __validate_exception_handover(struct dm_snapshot *snap)
482 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
483 struct dm_snapshot *snap_merge = NULL;
485 /* Does snapshot need exceptions handed over to it? */
486 if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
487 &snap_merge) == 2) ||
489 snap->ti->error = "Snapshot cow pairing for exception "
490 "table handover failed";
495 * If no snap_src was found, snap cannot become a handover
502 * Non-snapshot-merge handover?
504 if (!dm_target_is_snapshot_merge(snap->ti))
508 * Do not allow more than one merging snapshot.
511 snap->ti->error = "A snapshot is already merging.";
515 if (!snap_src->store->type->prepare_merge ||
516 !snap_src->store->type->commit_merge) {
517 snap->ti->error = "Snapshot exception store does not "
518 "support snapshot-merge.";
525 static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
527 struct dm_snapshot *l;
529 /* Sort the list according to chunk size, largest-first smallest-last */
530 list_for_each_entry(l, &o->snapshots, list)
531 if (l->store->chunk_size < s->store->chunk_size)
533 list_add_tail(&s->list, &l->list);
537 * Make a note of the snapshot and its origin so we can look it
538 * up when the origin has a write on it.
540 * Also validate snapshot exception store handovers.
541 * On success, returns 1 if this registration is a handover destination,
542 * otherwise returns 0.
544 static int register_snapshot(struct dm_snapshot *snap)
546 struct origin *o, *new_o = NULL;
547 struct block_device *bdev = snap->origin->bdev;
550 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
554 down_write(&_origins_lock);
556 r = __validate_exception_handover(snap);
562 o = __lookup_origin(bdev);
569 /* Initialise the struct */
570 INIT_LIST_HEAD(&o->snapshots);
576 __insert_snapshot(o, snap);
579 up_write(&_origins_lock);
585 * Move snapshot to correct place in list according to chunk size.
587 static void reregister_snapshot(struct dm_snapshot *s)
589 struct block_device *bdev = s->origin->bdev;
591 down_write(&_origins_lock);
594 __insert_snapshot(__lookup_origin(bdev), s);
596 up_write(&_origins_lock);
599 static void unregister_snapshot(struct dm_snapshot *s)
603 down_write(&_origins_lock);
604 o = __lookup_origin(s->origin->bdev);
607 if (o && list_empty(&o->snapshots)) {
608 list_del(&o->hash_list);
612 up_write(&_origins_lock);
616 * Implementation of the exception hash tables.
617 * The lowest hash_shift bits of the chunk number are ignored, allowing
618 * some consecutive chunks to be grouped together.
620 static int dm_exception_table_init(struct dm_exception_table *et,
621 uint32_t size, unsigned hash_shift)
625 et->hash_shift = hash_shift;
626 et->hash_mask = size - 1;
627 et->table = dm_vcalloc(size, sizeof(struct list_head));
631 for (i = 0; i < size; i++)
632 INIT_LIST_HEAD(et->table + i);
637 static void dm_exception_table_exit(struct dm_exception_table *et,
638 struct kmem_cache *mem)
640 struct list_head *slot;
641 struct dm_exception *ex, *next;
644 size = et->hash_mask + 1;
645 for (i = 0; i < size; i++) {
646 slot = et->table + i;
648 list_for_each_entry_safe (ex, next, slot, hash_list)
649 kmem_cache_free(mem, ex);
655 static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
657 return (chunk >> et->hash_shift) & et->hash_mask;
660 static void dm_remove_exception(struct dm_exception *e)
662 list_del(&e->hash_list);
666 * Return the exception data for a sector, or NULL if not
669 static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
672 struct list_head *slot;
673 struct dm_exception *e;
675 slot = &et->table[exception_hash(et, chunk)];
676 list_for_each_entry (e, slot, hash_list)
677 if (chunk >= e->old_chunk &&
678 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
684 static struct dm_exception *alloc_completed_exception(gfp_t gfp)
686 struct dm_exception *e;
688 e = kmem_cache_alloc(exception_cache, gfp);
689 if (!e && gfp == GFP_NOIO)
690 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
695 static void free_completed_exception(struct dm_exception *e)
697 kmem_cache_free(exception_cache, e);
700 static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
702 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
705 atomic_inc(&s->pending_exceptions_count);
711 static void free_pending_exception(struct dm_snap_pending_exception *pe)
713 struct dm_snapshot *s = pe->snap;
715 mempool_free(pe, s->pending_pool);
716 smp_mb__before_atomic();
717 atomic_dec(&s->pending_exceptions_count);
720 static void dm_insert_exception(struct dm_exception_table *eh,
721 struct dm_exception *new_e)
724 struct dm_exception *e = NULL;
726 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
728 /* Add immediately if this table doesn't support consecutive chunks */
732 /* List is ordered by old_chunk */
733 list_for_each_entry_reverse(e, l, hash_list) {
734 /* Insert after an existing chunk? */
735 if (new_e->old_chunk == (e->old_chunk +
736 dm_consecutive_chunk_count(e) + 1) &&
737 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
738 dm_consecutive_chunk_count(e) + 1)) {
739 dm_consecutive_chunk_count_inc(e);
740 free_completed_exception(new_e);
744 /* Insert before an existing chunk? */
745 if (new_e->old_chunk == (e->old_chunk - 1) &&
746 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
747 dm_consecutive_chunk_count_inc(e);
750 free_completed_exception(new_e);
754 if (new_e->old_chunk > e->old_chunk)
759 list_add(&new_e->hash_list, e ? &e->hash_list : l);
763 * Callback used by the exception stores to load exceptions when
766 static int dm_add_exception(void *context, chunk_t old, chunk_t new)
768 struct dm_snapshot *s = context;
769 struct dm_exception *e;
771 e = alloc_completed_exception(GFP_KERNEL);
777 /* Consecutive_count is implicitly initialised to zero */
780 dm_insert_exception(&s->complete, e);
786 * Return a minimum chunk size of all snapshots that have the specified origin.
787 * Return zero if the origin has no snapshots.
789 static uint32_t __minimum_chunk_size(struct origin *o)
791 struct dm_snapshot *snap;
792 unsigned chunk_size = rounddown_pow_of_two(UINT_MAX);
795 list_for_each_entry(snap, &o->snapshots, list)
796 chunk_size = min_not_zero(chunk_size,
797 snap->store->chunk_size);
799 return (uint32_t) chunk_size;
805 static int calc_max_buckets(void)
807 /* use a fixed size of 2MB */
808 unsigned long mem = 2 * 1024 * 1024;
809 mem /= sizeof(struct list_head);
815 * Allocate room for a suitable hash table.
817 static int init_hash_tables(struct dm_snapshot *s)
819 sector_t hash_size, cow_dev_size, max_buckets;
822 * Calculate based on the size of the original volume or
825 cow_dev_size = get_dev_size(s->cow->bdev);
826 max_buckets = calc_max_buckets();
828 hash_size = cow_dev_size >> s->store->chunk_shift;
829 hash_size = min(hash_size, max_buckets);
833 hash_size = rounddown_pow_of_two(hash_size);
834 if (dm_exception_table_init(&s->complete, hash_size,
835 DM_CHUNK_CONSECUTIVE_BITS))
839 * Allocate hash table for in-flight exceptions
840 * Make this smaller than the real hash table
846 if (dm_exception_table_init(&s->pending, hash_size, 0)) {
847 dm_exception_table_exit(&s->complete, exception_cache);
854 static void merge_shutdown(struct dm_snapshot *s)
856 clear_bit_unlock(RUNNING_MERGE, &s->state_bits);
857 smp_mb__after_atomic();
858 wake_up_bit(&s->state_bits, RUNNING_MERGE);
861 static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s)
863 s->first_merging_chunk = 0;
864 s->num_merging_chunks = 0;
866 return bio_list_get(&s->bios_queued_during_merge);
870 * Remove one chunk from the index of completed exceptions.
872 static int __remove_single_exception_chunk(struct dm_snapshot *s,
875 struct dm_exception *e;
877 e = dm_lookup_exception(&s->complete, old_chunk);
879 DMERR("Corruption detected: exception for block %llu is "
880 "on disk but not in memory",
881 (unsigned long long)old_chunk);
886 * If this is the only chunk using this exception, remove exception.
888 if (!dm_consecutive_chunk_count(e)) {
889 dm_remove_exception(e);
890 free_completed_exception(e);
895 * The chunk may be either at the beginning or the end of a
896 * group of consecutive chunks - never in the middle. We are
897 * removing chunks in the opposite order to that in which they
898 * were added, so this should always be true.
899 * Decrement the consecutive chunk counter and adjust the
900 * starting point if necessary.
902 if (old_chunk == e->old_chunk) {
905 } else if (old_chunk != e->old_chunk +
906 dm_consecutive_chunk_count(e)) {
907 DMERR("Attempt to merge block %llu from the "
908 "middle of a chunk range [%llu - %llu]",
909 (unsigned long long)old_chunk,
910 (unsigned long long)e->old_chunk,
912 e->old_chunk + dm_consecutive_chunk_count(e));
916 dm_consecutive_chunk_count_dec(e);
921 static void flush_bios(struct bio *bio);
923 static int remove_single_exception_chunk(struct dm_snapshot *s)
925 struct bio *b = NULL;
927 chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1;
929 mutex_lock(&s->lock);
932 * Process chunks (and associated exceptions) in reverse order
933 * so that dm_consecutive_chunk_count_dec() accounting works.
936 r = __remove_single_exception_chunk(s, old_chunk);
939 } while (old_chunk-- > s->first_merging_chunk);
941 b = __release_queued_bios_after_merge(s);
944 mutex_unlock(&s->lock);
951 static int origin_write_extent(struct dm_snapshot *merging_snap,
952 sector_t sector, unsigned chunk_size);
954 static void merge_callback(int read_err, unsigned long write_err,
957 static uint64_t read_pending_exceptions_done_count(void)
959 uint64_t pending_exceptions_done;
961 spin_lock(&_pending_exceptions_done_spinlock);
962 pending_exceptions_done = _pending_exceptions_done_count;
963 spin_unlock(&_pending_exceptions_done_spinlock);
965 return pending_exceptions_done;
968 static void increment_pending_exceptions_done_count(void)
970 spin_lock(&_pending_exceptions_done_spinlock);
971 _pending_exceptions_done_count++;
972 spin_unlock(&_pending_exceptions_done_spinlock);
974 wake_up_all(&_pending_exceptions_done);
977 static void snapshot_merge_next_chunks(struct dm_snapshot *s)
979 int i, linear_chunks;
980 chunk_t old_chunk, new_chunk;
981 struct dm_io_region src, dest;
983 uint64_t previous_count;
985 BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
986 if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
990 * valid flag never changes during merge, so no lock required.
993 DMERR("Snapshot is invalid: can't merge");
997 linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
999 if (linear_chunks <= 0) {
1000 if (linear_chunks < 0) {
1001 DMERR("Read error in exception store: "
1002 "shutting down merge");
1003 mutex_lock(&s->lock);
1004 s->merge_failed = 1;
1005 mutex_unlock(&s->lock);
1010 /* Adjust old_chunk and new_chunk to reflect start of linear region */
1011 old_chunk = old_chunk + 1 - linear_chunks;
1012 new_chunk = new_chunk + 1 - linear_chunks;
1015 * Use one (potentially large) I/O to copy all 'linear_chunks'
1016 * from the exception store to the origin
1018 io_size = linear_chunks * s->store->chunk_size;
1020 dest.bdev = s->origin->bdev;
1021 dest.sector = chunk_to_sector(s->store, old_chunk);
1022 dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector);
1024 src.bdev = s->cow->bdev;
1025 src.sector = chunk_to_sector(s->store, new_chunk);
1026 src.count = dest.count;
1029 * Reallocate any exceptions needed in other snapshots then
1030 * wait for the pending exceptions to complete.
1031 * Each time any pending exception (globally on the system)
1032 * completes we are woken and repeat the process to find out
1033 * if we can proceed. While this may not seem a particularly
1034 * efficient algorithm, it is not expected to have any
1035 * significant impact on performance.
1037 previous_count = read_pending_exceptions_done_count();
1038 while (origin_write_extent(s, dest.sector, io_size)) {
1039 wait_event(_pending_exceptions_done,
1040 (read_pending_exceptions_done_count() !=
1042 /* Retry after the wait, until all exceptions are done. */
1043 previous_count = read_pending_exceptions_done_count();
1046 mutex_lock(&s->lock);
1047 s->first_merging_chunk = old_chunk;
1048 s->num_merging_chunks = linear_chunks;
1049 mutex_unlock(&s->lock);
1051 /* Wait until writes to all 'linear_chunks' drain */
1052 for (i = 0; i < linear_chunks; i++)
1053 __check_for_conflicting_io(s, old_chunk + i);
1055 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s);
1062 static void error_bios(struct bio *bio);
1064 static void merge_callback(int read_err, unsigned long write_err, void *context)
1066 struct dm_snapshot *s = context;
1067 struct bio *b = NULL;
1069 if (read_err || write_err) {
1071 DMERR("Read error: shutting down merge.");
1073 DMERR("Write error: shutting down merge.");
1077 if (s->store->type->commit_merge(s->store,
1078 s->num_merging_chunks) < 0) {
1079 DMERR("Write error in exception store: shutting down merge");
1083 if (remove_single_exception_chunk(s) < 0)
1086 snapshot_merge_next_chunks(s);
1091 mutex_lock(&s->lock);
1092 s->merge_failed = 1;
1093 b = __release_queued_bios_after_merge(s);
1094 mutex_unlock(&s->lock);
1100 static void start_merge(struct dm_snapshot *s)
1102 if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits))
1103 snapshot_merge_next_chunks(s);
1107 * Stop the merging process and wait until it finishes.
1109 static void stop_merge(struct dm_snapshot *s)
1111 set_bit(SHUTDOWN_MERGE, &s->state_bits);
1112 wait_on_bit(&s->state_bits, RUNNING_MERGE, TASK_UNINTERRUPTIBLE);
1113 clear_bit(SHUTDOWN_MERGE, &s->state_bits);
1117 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p|po|n> <chunk-size>
1119 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1121 struct dm_snapshot *s;
1124 char *origin_path, *cow_path;
1125 dev_t origin_dev, cow_dev;
1126 unsigned args_used, num_flush_bios = 1;
1127 fmode_t origin_mode = FMODE_READ;
1130 ti->error = "requires exactly 4 arguments";
1135 if (dm_target_is_snapshot_merge(ti)) {
1137 origin_mode = FMODE_WRITE;
1140 s = kzalloc(sizeof(*s), GFP_KERNEL);
1142 ti->error = "Cannot allocate private snapshot structure";
1147 origin_path = argv[0];
1151 r = dm_get_device(ti, origin_path, origin_mode, &s->origin);
1153 ti->error = "Cannot get origin device";
1156 origin_dev = s->origin->bdev->bd_dev;
1162 cow_dev = dm_get_dev_t(cow_path);
1163 if (cow_dev && cow_dev == origin_dev) {
1164 ti->error = "COW device cannot be the same as origin device";
1169 r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow);
1171 ti->error = "Cannot get COW device";
1175 r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
1177 ti->error = "Couldn't create exception store";
1187 s->snapshot_overflowed = 0;
1189 atomic_set(&s->pending_exceptions_count, 0);
1190 s->exception_start_sequence = 0;
1191 s->exception_complete_sequence = 0;
1192 INIT_LIST_HEAD(&s->out_of_order_list);
1193 mutex_init(&s->lock);
1194 INIT_LIST_HEAD(&s->list);
1195 spin_lock_init(&s->pe_lock);
1197 s->merge_failed = 0;
1198 s->first_merging_chunk = 0;
1199 s->num_merging_chunks = 0;
1200 bio_list_init(&s->bios_queued_during_merge);
1202 /* Allocate hash table for COW data */
1203 if (init_hash_tables(s)) {
1204 ti->error = "Unable to allocate hash table space";
1206 goto bad_hash_tables;
1209 init_waitqueue_head(&s->in_progress_wait);
1211 s->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle);
1212 if (IS_ERR(s->kcopyd_client)) {
1213 r = PTR_ERR(s->kcopyd_client);
1214 ti->error = "Could not create kcopyd client";
1218 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
1219 if (!s->pending_pool) {
1220 ti->error = "Could not allocate mempool for pending exceptions";
1222 goto bad_pending_pool;
1225 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1226 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
1228 spin_lock_init(&s->tracked_chunk_lock);
1231 ti->num_flush_bios = num_flush_bios;
1232 ti->per_bio_data_size = sizeof(struct dm_snap_tracked_chunk);
1234 /* Add snapshot to the list of snapshots for this origin */
1235 /* Exceptions aren't triggered till snapshot_resume() is called */
1236 r = register_snapshot(s);
1238 ti->error = "Snapshot origin struct allocation failed";
1239 goto bad_load_and_register;
1241 /* invalid handover, register_snapshot has set ti->error */
1242 goto bad_load_and_register;
1246 * Metadata must only be loaded into one table at once, so skip this
1247 * if metadata will be handed over during resume.
1248 * Chunk size will be set during the handover - set it to zero to
1249 * ensure it's ignored.
1252 s->store->chunk_size = 0;
1256 r = s->store->type->read_metadata(s->store, dm_add_exception,
1259 ti->error = "Failed to read snapshot metadata";
1260 goto bad_read_metadata;
1263 DMWARN("Snapshot is marked invalid.");
1266 if (!s->store->chunk_size) {
1267 ti->error = "Chunk size not set";
1269 goto bad_read_metadata;
1272 r = dm_set_target_max_io_len(ti, s->store->chunk_size);
1274 goto bad_read_metadata;
1279 unregister_snapshot(s);
1281 bad_load_and_register:
1282 mempool_destroy(s->pending_pool);
1285 dm_kcopyd_client_destroy(s->kcopyd_client);
1288 dm_exception_table_exit(&s->pending, pending_cache);
1289 dm_exception_table_exit(&s->complete, exception_cache);
1292 dm_exception_store_destroy(s->store);
1295 dm_put_device(ti, s->cow);
1298 dm_put_device(ti, s->origin);
1307 static void __free_exceptions(struct dm_snapshot *s)
1309 dm_kcopyd_client_destroy(s->kcopyd_client);
1310 s->kcopyd_client = NULL;
1312 dm_exception_table_exit(&s->pending, pending_cache);
1313 dm_exception_table_exit(&s->complete, exception_cache);
1316 static void __handover_exceptions(struct dm_snapshot *snap_src,
1317 struct dm_snapshot *snap_dest)
1320 struct dm_exception_table table_swap;
1321 struct dm_exception_store *store_swap;
1325 * Swap all snapshot context information between the two instances.
1327 u.table_swap = snap_dest->complete;
1328 snap_dest->complete = snap_src->complete;
1329 snap_src->complete = u.table_swap;
1331 u.store_swap = snap_dest->store;
1332 snap_dest->store = snap_src->store;
1333 snap_dest->store->userspace_supports_overflow = u.store_swap->userspace_supports_overflow;
1334 snap_src->store = u.store_swap;
1336 snap_dest->store->snap = snap_dest;
1337 snap_src->store->snap = snap_src;
1339 snap_dest->ti->max_io_len = snap_dest->store->chunk_size;
1340 snap_dest->valid = snap_src->valid;
1341 snap_dest->snapshot_overflowed = snap_src->snapshot_overflowed;
1344 * Set source invalid to ensure it receives no further I/O.
1346 snap_src->valid = 0;
1349 static void snapshot_dtr(struct dm_target *ti)
1351 #ifdef CONFIG_DM_DEBUG
1354 struct dm_snapshot *s = ti->private;
1355 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1357 down_read(&_origins_lock);
1358 /* Check whether exception handover must be cancelled */
1359 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1360 if (snap_src && snap_dest && (s == snap_src)) {
1361 mutex_lock(&snap_dest->lock);
1362 snap_dest->valid = 0;
1363 mutex_unlock(&snap_dest->lock);
1364 DMERR("Cancelling snapshot handover.");
1366 up_read(&_origins_lock);
1368 if (dm_target_is_snapshot_merge(ti))
1371 /* Prevent further origin writes from using this snapshot. */
1372 /* After this returns there can be no new kcopyd jobs. */
1373 unregister_snapshot(s);
1375 while (atomic_read(&s->pending_exceptions_count))
1378 * Ensure instructions in mempool_destroy aren't reordered
1379 * before atomic_read.
1383 #ifdef CONFIG_DM_DEBUG
1384 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1385 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
1388 __free_exceptions(s);
1390 mempool_destroy(s->pending_pool);
1392 dm_exception_store_destroy(s->store);
1394 mutex_destroy(&s->lock);
1396 dm_put_device(ti, s->cow);
1398 dm_put_device(ti, s->origin);
1400 WARN_ON(s->in_progress);
1405 static void account_start_copy(struct dm_snapshot *s)
1407 spin_lock(&s->in_progress_wait.lock);
1409 spin_unlock(&s->in_progress_wait.lock);
1412 static void account_end_copy(struct dm_snapshot *s)
1414 spin_lock(&s->in_progress_wait.lock);
1415 BUG_ON(!s->in_progress);
1417 if (likely(s->in_progress <= cow_threshold) &&
1418 unlikely(waitqueue_active(&s->in_progress_wait)))
1419 wake_up_locked(&s->in_progress_wait);
1420 spin_unlock(&s->in_progress_wait.lock);
1423 static bool wait_for_in_progress(struct dm_snapshot *s, bool unlock_origins)
1425 if (unlikely(s->in_progress > cow_threshold)) {
1426 spin_lock(&s->in_progress_wait.lock);
1427 if (likely(s->in_progress > cow_threshold)) {
1429 * NOTE: this throttle doesn't account for whether
1430 * the caller is servicing an IO that will trigger a COW
1431 * so excess throttling may result for chunks not required
1432 * to be COW'd. But if cow_threshold was reached, extra
1433 * throttling is unlikely to negatively impact performance.
1435 DECLARE_WAITQUEUE(wait, current);
1436 __add_wait_queue(&s->in_progress_wait, &wait);
1437 __set_current_state(TASK_UNINTERRUPTIBLE);
1438 spin_unlock(&s->in_progress_wait.lock);
1440 up_read(&_origins_lock);
1442 remove_wait_queue(&s->in_progress_wait, &wait);
1445 spin_unlock(&s->in_progress_wait.lock);
1451 * Flush a list of buffers.
1453 static void flush_bios(struct bio *bio)
1459 bio->bi_next = NULL;
1460 generic_make_request(bio);
1465 static int do_origin(struct dm_dev *origin, struct bio *bio, bool limit);
1468 * Flush a list of buffers.
1470 static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
1477 bio->bi_next = NULL;
1478 r = do_origin(s->origin, bio, false);
1479 if (r == DM_MAPIO_REMAPPED)
1480 generic_make_request(bio);
1486 * Error a list of buffers.
1488 static void error_bios(struct bio *bio)
1494 bio->bi_next = NULL;
1500 static void __invalidate_snapshot(struct dm_snapshot *s, int err)
1506 DMERR("Invalidating snapshot: Error reading/writing.");
1507 else if (err == -ENOMEM)
1508 DMERR("Invalidating snapshot: Unable to allocate exception.");
1510 if (s->store->type->drop_snapshot)
1511 s->store->type->drop_snapshot(s->store);
1515 dm_table_event(s->ti->table);
1518 static void pending_complete(void *context, int success)
1520 struct dm_snap_pending_exception *pe = context;
1521 struct dm_exception *e;
1522 struct dm_snapshot *s = pe->snap;
1523 struct bio *origin_bios = NULL;
1524 struct bio *snapshot_bios = NULL;
1525 struct bio *full_bio = NULL;
1529 /* Read/write error - snapshot is unusable */
1530 mutex_lock(&s->lock);
1531 __invalidate_snapshot(s, -EIO);
1536 e = alloc_completed_exception(GFP_NOIO);
1538 mutex_lock(&s->lock);
1539 __invalidate_snapshot(s, -ENOMEM);
1545 mutex_lock(&s->lock);
1547 free_completed_exception(e);
1552 /* Check for conflicting reads */
1553 __check_for_conflicting_io(s, pe->e.old_chunk);
1556 * Add a proper exception, and remove the
1557 * in-flight exception from the list.
1559 dm_insert_exception(&s->complete, e);
1562 dm_remove_exception(&pe->e);
1563 snapshot_bios = bio_list_get(&pe->snapshot_bios);
1564 origin_bios = bio_list_get(&pe->origin_bios);
1565 full_bio = pe->full_bio;
1567 full_bio->bi_end_io = pe->full_bio_end_io;
1568 full_bio->bi_private = pe->full_bio_private;
1570 increment_pending_exceptions_done_count();
1572 mutex_unlock(&s->lock);
1574 /* Submit any pending write bios */
1577 bio_io_error(full_bio);
1578 error_bios(snapshot_bios);
1581 bio_endio(full_bio);
1582 flush_bios(snapshot_bios);
1585 retry_origin_bios(s, origin_bios);
1587 free_pending_exception(pe);
1590 static void complete_exception(struct dm_snap_pending_exception *pe)
1592 struct dm_snapshot *s = pe->snap;
1594 /* Update the metadata if we are persistent */
1595 s->store->type->commit_exception(s->store, &pe->e, !pe->copy_error,
1596 pending_complete, pe);
1600 * Called when the copy I/O has finished. kcopyd actually runs
1601 * this code so don't block.
1603 static void copy_callback(int read_err, unsigned long write_err, void *context)
1605 struct dm_snap_pending_exception *pe = context;
1606 struct dm_snapshot *s = pe->snap;
1608 pe->copy_error = read_err || write_err;
1610 if (pe->exception_sequence == s->exception_complete_sequence) {
1611 s->exception_complete_sequence++;
1612 complete_exception(pe);
1614 while (!list_empty(&s->out_of_order_list)) {
1615 pe = list_entry(s->out_of_order_list.next,
1616 struct dm_snap_pending_exception, out_of_order_entry);
1617 if (pe->exception_sequence != s->exception_complete_sequence)
1619 s->exception_complete_sequence++;
1620 list_del(&pe->out_of_order_entry);
1621 complete_exception(pe);
1624 struct list_head *lh;
1625 struct dm_snap_pending_exception *pe2;
1627 list_for_each_prev(lh, &s->out_of_order_list) {
1628 pe2 = list_entry(lh, struct dm_snap_pending_exception, out_of_order_entry);
1629 if (pe2->exception_sequence < pe->exception_sequence)
1632 list_add(&pe->out_of_order_entry, lh);
1634 account_end_copy(s);
1638 * Dispatches the copy operation to kcopyd.
1640 static void start_copy(struct dm_snap_pending_exception *pe)
1642 struct dm_snapshot *s = pe->snap;
1643 struct dm_io_region src, dest;
1644 struct block_device *bdev = s->origin->bdev;
1647 dev_size = get_dev_size(bdev);
1650 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
1651 src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
1653 dest.bdev = s->cow->bdev;
1654 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
1655 dest.count = src.count;
1657 /* Hand over to kcopyd */
1658 account_start_copy(s);
1659 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe);
1662 static void full_bio_end_io(struct bio *bio)
1664 void *callback_data = bio->bi_private;
1666 dm_kcopyd_do_callback(callback_data, 0, bio->bi_error ? 1 : 0);
1669 static void start_full_bio(struct dm_snap_pending_exception *pe,
1672 struct dm_snapshot *s = pe->snap;
1673 void *callback_data;
1676 pe->full_bio_end_io = bio->bi_end_io;
1677 pe->full_bio_private = bio->bi_private;
1679 account_start_copy(s);
1680 callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client,
1683 bio->bi_end_io = full_bio_end_io;
1684 bio->bi_private = callback_data;
1686 generic_make_request(bio);
1689 static struct dm_snap_pending_exception *
1690 __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1692 struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
1697 return container_of(e, struct dm_snap_pending_exception, e);
1701 * Looks to see if this snapshot already has a pending exception
1702 * for this chunk, otherwise it allocates a new one and inserts
1703 * it into the pending table.
1705 * NOTE: a write lock must be held on snap->lock before calling
1708 static struct dm_snap_pending_exception *
1709 __find_pending_exception(struct dm_snapshot *s,
1710 struct dm_snap_pending_exception *pe, chunk_t chunk)
1712 struct dm_snap_pending_exception *pe2;
1714 pe2 = __lookup_pending_exception(s, chunk);
1716 free_pending_exception(pe);
1720 pe->e.old_chunk = chunk;
1721 bio_list_init(&pe->origin_bios);
1722 bio_list_init(&pe->snapshot_bios);
1724 pe->full_bio = NULL;
1726 if (s->store->type->prepare_exception(s->store, &pe->e)) {
1727 free_pending_exception(pe);
1731 pe->exception_sequence = s->exception_start_sequence++;
1733 dm_insert_exception(&s->pending, &pe->e);
1738 static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
1739 struct bio *bio, chunk_t chunk)
1741 bio->bi_bdev = s->cow->bdev;
1742 bio->bi_iter.bi_sector =
1743 chunk_to_sector(s->store, dm_chunk_number(e->new_chunk) +
1744 (chunk - e->old_chunk)) +
1745 (bio->bi_iter.bi_sector & s->store->chunk_mask);
1748 static int snapshot_map(struct dm_target *ti, struct bio *bio)
1750 struct dm_exception *e;
1751 struct dm_snapshot *s = ti->private;
1752 int r = DM_MAPIO_REMAPPED;
1754 struct dm_snap_pending_exception *pe = NULL;
1756 init_tracked_chunk(bio);
1758 if (bio->bi_rw & REQ_FLUSH) {
1759 bio->bi_bdev = s->cow->bdev;
1760 return DM_MAPIO_REMAPPED;
1763 chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector);
1765 /* Full snapshots are not usable */
1766 /* To get here the table must be live so s->active is always set. */
1770 if (bio_data_dir(bio) == WRITE) {
1771 while (unlikely(!wait_for_in_progress(s, false)))
1772 ; /* wait_for_in_progress() has slept */
1775 mutex_lock(&s->lock);
1777 if (!s->valid || (unlikely(s->snapshot_overflowed) && bio_rw(bio) == WRITE)) {
1782 /* If the block is already remapped - use that, else remap it */
1783 e = dm_lookup_exception(&s->complete, chunk);
1785 remap_exception(s, e, bio, chunk);
1790 * Write to snapshot - higher level takes care of RW/RO
1791 * flags so we should only get this if we are
1794 if (bio_rw(bio) == WRITE) {
1795 pe = __lookup_pending_exception(s, chunk);
1797 mutex_unlock(&s->lock);
1798 pe = alloc_pending_exception(s);
1799 mutex_lock(&s->lock);
1801 if (!s->valid || s->snapshot_overflowed) {
1802 free_pending_exception(pe);
1807 e = dm_lookup_exception(&s->complete, chunk);
1809 free_pending_exception(pe);
1810 remap_exception(s, e, bio, chunk);
1814 pe = __find_pending_exception(s, pe, chunk);
1816 if (s->store->userspace_supports_overflow) {
1817 s->snapshot_overflowed = 1;
1818 DMERR("Snapshot overflowed: Unable to allocate exception.");
1820 __invalidate_snapshot(s, -ENOMEM);
1826 remap_exception(s, &pe->e, bio, chunk);
1828 r = DM_MAPIO_SUBMITTED;
1831 bio->bi_iter.bi_size ==
1832 (s->store->chunk_size << SECTOR_SHIFT)) {
1834 mutex_unlock(&s->lock);
1835 start_full_bio(pe, bio);
1839 bio_list_add(&pe->snapshot_bios, bio);
1842 /* this is protected by snap->lock */
1844 mutex_unlock(&s->lock);
1849 bio->bi_bdev = s->origin->bdev;
1850 track_chunk(s, bio, chunk);
1854 mutex_unlock(&s->lock);
1860 * A snapshot-merge target behaves like a combination of a snapshot
1861 * target and a snapshot-origin target. It only generates new
1862 * exceptions in other snapshots and not in the one that is being
1865 * For each chunk, if there is an existing exception, it is used to
1866 * redirect I/O to the cow device. Otherwise I/O is sent to the origin,
1867 * which in turn might generate exceptions in other snapshots.
1868 * If merging is currently taking place on the chunk in question, the
1869 * I/O is deferred by adding it to s->bios_queued_during_merge.
1871 static int snapshot_merge_map(struct dm_target *ti, struct bio *bio)
1873 struct dm_exception *e;
1874 struct dm_snapshot *s = ti->private;
1875 int r = DM_MAPIO_REMAPPED;
1878 init_tracked_chunk(bio);
1880 if (bio->bi_rw & REQ_FLUSH) {
1881 if (!dm_bio_get_target_bio_nr(bio))
1882 bio->bi_bdev = s->origin->bdev;
1884 bio->bi_bdev = s->cow->bdev;
1885 return DM_MAPIO_REMAPPED;
1888 chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector);
1890 mutex_lock(&s->lock);
1892 /* Full merging snapshots are redirected to the origin */
1894 goto redirect_to_origin;
1896 /* If the block is already remapped - use that */
1897 e = dm_lookup_exception(&s->complete, chunk);
1899 /* Queue writes overlapping with chunks being merged */
1900 if (bio_rw(bio) == WRITE &&
1901 chunk >= s->first_merging_chunk &&
1902 chunk < (s->first_merging_chunk +
1903 s->num_merging_chunks)) {
1904 bio->bi_bdev = s->origin->bdev;
1905 bio_list_add(&s->bios_queued_during_merge, bio);
1906 r = DM_MAPIO_SUBMITTED;
1910 remap_exception(s, e, bio, chunk);
1912 if (bio_rw(bio) == WRITE)
1913 track_chunk(s, bio, chunk);
1918 bio->bi_bdev = s->origin->bdev;
1920 if (bio_rw(bio) == WRITE) {
1921 mutex_unlock(&s->lock);
1922 return do_origin(s->origin, bio, false);
1926 mutex_unlock(&s->lock);
1931 static int snapshot_end_io(struct dm_target *ti, struct bio *bio, int error)
1933 struct dm_snapshot *s = ti->private;
1935 if (is_bio_tracked(bio))
1936 stop_tracking_chunk(s, bio);
1941 static void snapshot_merge_presuspend(struct dm_target *ti)
1943 struct dm_snapshot *s = ti->private;
1948 static int snapshot_preresume(struct dm_target *ti)
1951 struct dm_snapshot *s = ti->private;
1952 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1954 down_read(&_origins_lock);
1955 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1956 if (snap_src && snap_dest) {
1957 mutex_lock(&snap_src->lock);
1958 if (s == snap_src) {
1959 DMERR("Unable to resume snapshot source until "
1960 "handover completes.");
1962 } else if (!dm_suspended(snap_src->ti)) {
1963 DMERR("Unable to perform snapshot handover until "
1964 "source is suspended.");
1967 mutex_unlock(&snap_src->lock);
1969 up_read(&_origins_lock);
1974 static void snapshot_resume(struct dm_target *ti)
1976 struct dm_snapshot *s = ti->private;
1977 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL, *snap_merging = NULL;
1978 struct dm_origin *o;
1979 struct mapped_device *origin_md = NULL;
1980 bool must_restart_merging = false;
1982 down_read(&_origins_lock);
1984 o = __lookup_dm_origin(s->origin->bdev);
1986 origin_md = dm_table_get_md(o->ti->table);
1988 (void) __find_snapshots_sharing_cow(s, NULL, NULL, &snap_merging);
1990 origin_md = dm_table_get_md(snap_merging->ti->table);
1992 if (origin_md == dm_table_get_md(ti->table))
1995 if (dm_hold(origin_md))
1999 up_read(&_origins_lock);
2002 dm_internal_suspend_fast(origin_md);
2003 if (snap_merging && test_bit(RUNNING_MERGE, &snap_merging->state_bits)) {
2004 must_restart_merging = true;
2005 stop_merge(snap_merging);
2009 down_read(&_origins_lock);
2011 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
2012 if (snap_src && snap_dest) {
2013 mutex_lock(&snap_src->lock);
2014 mutex_lock_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
2015 __handover_exceptions(snap_src, snap_dest);
2016 mutex_unlock(&snap_dest->lock);
2017 mutex_unlock(&snap_src->lock);
2020 up_read(&_origins_lock);
2023 if (must_restart_merging)
2024 start_merge(snap_merging);
2025 dm_internal_resume_fast(origin_md);
2029 /* Now we have correct chunk size, reregister */
2030 reregister_snapshot(s);
2032 mutex_lock(&s->lock);
2034 mutex_unlock(&s->lock);
2037 static uint32_t get_origin_minimum_chunksize(struct block_device *bdev)
2039 uint32_t min_chunksize;
2041 down_read(&_origins_lock);
2042 min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
2043 up_read(&_origins_lock);
2045 return min_chunksize;
2048 static void snapshot_merge_resume(struct dm_target *ti)
2050 struct dm_snapshot *s = ti->private;
2053 * Handover exceptions from existing snapshot.
2055 snapshot_resume(ti);
2058 * snapshot-merge acts as an origin, so set ti->max_io_len
2060 ti->max_io_len = get_origin_minimum_chunksize(s->origin->bdev);
2065 static void snapshot_status(struct dm_target *ti, status_type_t type,
2066 unsigned status_flags, char *result, unsigned maxlen)
2069 struct dm_snapshot *snap = ti->private;
2072 case STATUSTYPE_INFO:
2074 mutex_lock(&snap->lock);
2078 else if (snap->merge_failed)
2079 DMEMIT("Merge failed");
2080 else if (snap->snapshot_overflowed)
2083 if (snap->store->type->usage) {
2084 sector_t total_sectors, sectors_allocated,
2086 snap->store->type->usage(snap->store,
2090 DMEMIT("%llu/%llu %llu",
2091 (unsigned long long)sectors_allocated,
2092 (unsigned long long)total_sectors,
2093 (unsigned long long)metadata_sectors);
2099 mutex_unlock(&snap->lock);
2103 case STATUSTYPE_TABLE:
2105 * kdevname returns a static pointer so we need
2106 * to make private copies if the output is to
2109 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
2110 snap->store->type->status(snap->store, type, result + sz,
2116 static int snapshot_iterate_devices(struct dm_target *ti,
2117 iterate_devices_callout_fn fn, void *data)
2119 struct dm_snapshot *snap = ti->private;
2122 r = fn(ti, snap->origin, 0, ti->len, data);
2125 r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data);
2131 /*-----------------------------------------------------------------
2133 *---------------------------------------------------------------*/
2136 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
2137 * supplied bio was ignored. The caller may submit it immediately.
2138 * (No remapping actually occurs as the origin is always a direct linear
2141 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
2142 * and any supplied bio is added to a list to be submitted once all
2143 * the necessary exceptions exist.
2145 static int __origin_write(struct list_head *snapshots, sector_t sector,
2148 int r = DM_MAPIO_REMAPPED;
2149 struct dm_snapshot *snap;
2150 struct dm_exception *e;
2151 struct dm_snap_pending_exception *pe;
2152 struct dm_snap_pending_exception *pe_to_start_now = NULL;
2153 struct dm_snap_pending_exception *pe_to_start_last = NULL;
2156 /* Do all the snapshots on this origin */
2157 list_for_each_entry (snap, snapshots, list) {
2159 * Don't make new exceptions in a merging snapshot
2160 * because it has effectively been deleted
2162 if (dm_target_is_snapshot_merge(snap->ti))
2165 mutex_lock(&snap->lock);
2167 /* Only deal with valid and active snapshots */
2168 if (!snap->valid || !snap->active)
2171 /* Nothing to do if writing beyond end of snapshot */
2172 if (sector >= dm_table_get_size(snap->ti->table))
2176 * Remember, different snapshots can have
2177 * different chunk sizes.
2179 chunk = sector_to_chunk(snap->store, sector);
2182 * Check exception table to see if block
2183 * is already remapped in this snapshot
2184 * and trigger an exception if not.
2186 e = dm_lookup_exception(&snap->complete, chunk);
2190 pe = __lookup_pending_exception(snap, chunk);
2192 mutex_unlock(&snap->lock);
2193 pe = alloc_pending_exception(snap);
2194 mutex_lock(&snap->lock);
2197 free_pending_exception(pe);
2201 e = dm_lookup_exception(&snap->complete, chunk);
2203 free_pending_exception(pe);
2207 pe = __find_pending_exception(snap, pe, chunk);
2209 __invalidate_snapshot(snap, -ENOMEM);
2214 r = DM_MAPIO_SUBMITTED;
2217 * If an origin bio was supplied, queue it to wait for the
2218 * completion of this exception, and start this one last,
2219 * at the end of the function.
2222 bio_list_add(&pe->origin_bios, bio);
2227 pe_to_start_last = pe;
2233 pe_to_start_now = pe;
2237 mutex_unlock(&snap->lock);
2239 if (pe_to_start_now) {
2240 start_copy(pe_to_start_now);
2241 pe_to_start_now = NULL;
2246 * Submit the exception against which the bio is queued last,
2247 * to give the other exceptions a head start.
2249 if (pe_to_start_last)
2250 start_copy(pe_to_start_last);
2256 * Called on a write from the origin driver.
2258 static int do_origin(struct dm_dev *origin, struct bio *bio, bool limit)
2261 int r = DM_MAPIO_REMAPPED;
2264 down_read(&_origins_lock);
2265 o = __lookup_origin(origin->bdev);
2268 struct dm_snapshot *s;
2269 list_for_each_entry(s, &o->snapshots, list)
2270 if (unlikely(!wait_for_in_progress(s, true)))
2274 r = __origin_write(&o->snapshots, bio->bi_iter.bi_sector, bio);
2276 up_read(&_origins_lock);
2282 * Trigger exceptions in all non-merging snapshots.
2284 * The chunk size of the merging snapshot may be larger than the chunk
2285 * size of some other snapshot so we may need to reallocate multiple
2286 * chunks in other snapshots.
2288 * We scan all the overlapping exceptions in the other snapshots.
2289 * Returns 1 if anything was reallocated and must be waited for,
2290 * otherwise returns 0.
2292 * size must be a multiple of merging_snap's chunk_size.
2294 static int origin_write_extent(struct dm_snapshot *merging_snap,
2295 sector_t sector, unsigned size)
2302 * The origin's __minimum_chunk_size() got stored in max_io_len
2303 * by snapshot_merge_resume().
2305 down_read(&_origins_lock);
2306 o = __lookup_origin(merging_snap->origin->bdev);
2307 for (n = 0; n < size; n += merging_snap->ti->max_io_len)
2308 if (__origin_write(&o->snapshots, sector + n, NULL) ==
2311 up_read(&_origins_lock);
2317 * Origin: maps a linear range of a device, with hooks for snapshotting.
2321 * Construct an origin mapping: <dev_path>
2322 * The context for an origin is merely a 'struct dm_dev *'
2323 * pointing to the real device.
2325 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2328 struct dm_origin *o;
2331 ti->error = "origin: incorrect number of arguments";
2335 o = kmalloc(sizeof(struct dm_origin), GFP_KERNEL);
2337 ti->error = "Cannot allocate private origin structure";
2342 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &o->dev);
2344 ti->error = "Cannot get target device";
2350 ti->num_flush_bios = 1;
2360 static void origin_dtr(struct dm_target *ti)
2362 struct dm_origin *o = ti->private;
2364 dm_put_device(ti, o->dev);
2368 static int origin_map(struct dm_target *ti, struct bio *bio)
2370 struct dm_origin *o = ti->private;
2371 unsigned available_sectors;
2373 bio->bi_bdev = o->dev->bdev;
2375 if (unlikely(bio->bi_rw & REQ_FLUSH))
2376 return DM_MAPIO_REMAPPED;
2378 if (bio_rw(bio) != WRITE)
2379 return DM_MAPIO_REMAPPED;
2381 available_sectors = o->split_boundary -
2382 ((unsigned)bio->bi_iter.bi_sector & (o->split_boundary - 1));
2384 if (bio_sectors(bio) > available_sectors)
2385 dm_accept_partial_bio(bio, available_sectors);
2387 /* Only tell snapshots if this is a write */
2388 return do_origin(o->dev, bio, true);
2392 * Set the target "max_io_len" field to the minimum of all the snapshots'
2395 static void origin_resume(struct dm_target *ti)
2397 struct dm_origin *o = ti->private;
2399 o->split_boundary = get_origin_minimum_chunksize(o->dev->bdev);
2401 down_write(&_origins_lock);
2402 __insert_dm_origin(o);
2403 up_write(&_origins_lock);
2406 static void origin_postsuspend(struct dm_target *ti)
2408 struct dm_origin *o = ti->private;
2410 down_write(&_origins_lock);
2411 __remove_dm_origin(o);
2412 up_write(&_origins_lock);
2415 static void origin_status(struct dm_target *ti, status_type_t type,
2416 unsigned status_flags, char *result, unsigned maxlen)
2418 struct dm_origin *o = ti->private;
2421 case STATUSTYPE_INFO:
2425 case STATUSTYPE_TABLE:
2426 snprintf(result, maxlen, "%s", o->dev->name);
2431 static int origin_iterate_devices(struct dm_target *ti,
2432 iterate_devices_callout_fn fn, void *data)
2434 struct dm_origin *o = ti->private;
2436 return fn(ti, o->dev, 0, ti->len, data);
2439 static struct target_type origin_target = {
2440 .name = "snapshot-origin",
2441 .version = {1, 9, 0},
2442 .module = THIS_MODULE,
2446 .resume = origin_resume,
2447 .postsuspend = origin_postsuspend,
2448 .status = origin_status,
2449 .iterate_devices = origin_iterate_devices,
2452 static struct target_type snapshot_target = {
2454 .version = {1, 15, 0},
2455 .module = THIS_MODULE,
2456 .ctr = snapshot_ctr,
2457 .dtr = snapshot_dtr,
2458 .map = snapshot_map,
2459 .end_io = snapshot_end_io,
2460 .preresume = snapshot_preresume,
2461 .resume = snapshot_resume,
2462 .status = snapshot_status,
2463 .iterate_devices = snapshot_iterate_devices,
2466 static struct target_type merge_target = {
2467 .name = dm_snapshot_merge_target_name,
2468 .version = {1, 4, 0},
2469 .module = THIS_MODULE,
2470 .ctr = snapshot_ctr,
2471 .dtr = snapshot_dtr,
2472 .map = snapshot_merge_map,
2473 .end_io = snapshot_end_io,
2474 .presuspend = snapshot_merge_presuspend,
2475 .preresume = snapshot_preresume,
2476 .resume = snapshot_merge_resume,
2477 .status = snapshot_status,
2478 .iterate_devices = snapshot_iterate_devices,
2481 static int __init dm_snapshot_init(void)
2485 r = dm_exception_store_init();
2487 DMERR("Failed to initialize exception stores");
2491 r = dm_register_target(&snapshot_target);
2493 DMERR("snapshot target register failed %d", r);
2494 goto bad_register_snapshot_target;
2497 r = dm_register_target(&origin_target);
2499 DMERR("Origin target register failed %d", r);
2500 goto bad_register_origin_target;
2503 r = dm_register_target(&merge_target);
2505 DMERR("Merge target register failed %d", r);
2506 goto bad_register_merge_target;
2509 r = init_origin_hash();
2511 DMERR("init_origin_hash failed.");
2512 goto bad_origin_hash;
2515 exception_cache = KMEM_CACHE(dm_exception, 0);
2516 if (!exception_cache) {
2517 DMERR("Couldn't create exception cache.");
2519 goto bad_exception_cache;
2522 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
2523 if (!pending_cache) {
2524 DMERR("Couldn't create pending cache.");
2526 goto bad_pending_cache;
2532 kmem_cache_destroy(exception_cache);
2533 bad_exception_cache:
2536 dm_unregister_target(&merge_target);
2537 bad_register_merge_target:
2538 dm_unregister_target(&origin_target);
2539 bad_register_origin_target:
2540 dm_unregister_target(&snapshot_target);
2541 bad_register_snapshot_target:
2542 dm_exception_store_exit();
2547 static void __exit dm_snapshot_exit(void)
2549 dm_unregister_target(&snapshot_target);
2550 dm_unregister_target(&origin_target);
2551 dm_unregister_target(&merge_target);
2554 kmem_cache_destroy(pending_cache);
2555 kmem_cache_destroy(exception_cache);
2557 dm_exception_store_exit();
2561 module_init(dm_snapshot_init);
2562 module_exit(dm_snapshot_exit);
2564 MODULE_DESCRIPTION(DM_NAME " snapshot target");
2565 MODULE_AUTHOR("Joe Thornber");
2566 MODULE_LICENSE("GPL");
2567 MODULE_ALIAS("dm-snapshot-origin");
2568 MODULE_ALIAS("dm-snapshot-merge");