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>
23 #include "dm-exception-store.h"
25 #define DM_MSG_PREFIX "snapshots"
27 static const char dm_snapshot_merge_target_name[] = "snapshot-merge";
29 #define dm_target_is_snapshot_merge(ti) \
30 ((ti)->type->name == dm_snapshot_merge_target_name)
33 * The size of the mempool used to track chunks in use.
37 #define DM_TRACKED_CHUNK_HASH_SIZE 16
38 #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
39 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
41 struct dm_exception_table {
44 struct list_head *table;
48 struct rw_semaphore lock;
50 struct dm_dev *origin;
55 /* List of snapshots per Origin */
56 struct list_head list;
59 * You can't use a snapshot if this is 0 (e.g. if full).
60 * A snapshot-merge target never clears this.
64 /* Origin writes don't trigger exceptions until this is set */
67 atomic_t pending_exceptions_count;
69 mempool_t *pending_pool;
71 struct dm_exception_table pending;
72 struct dm_exception_table complete;
75 * pe_lock protects all pending_exception operations and access
76 * as well as the snapshot_bios list.
80 /* Chunks with outstanding reads */
81 spinlock_t tracked_chunk_lock;
82 mempool_t *tracked_chunk_pool;
83 struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
85 /* The on disk metadata handler */
86 struct dm_exception_store *store;
88 struct dm_kcopyd_client *kcopyd_client;
90 /* Wait for events based on state_bits */
91 unsigned long state_bits;
93 /* Range of chunks currently being merged. */
94 chunk_t first_merging_chunk;
95 int num_merging_chunks;
98 * The merge operation failed if this flag is set.
99 * Failure modes are handled as follows:
100 * - I/O error reading the header
101 * => don't load the target; abort.
102 * - Header does not have "valid" flag set
103 * => use the origin; forget about the snapshot.
104 * - I/O error when reading exceptions
105 * => don't load the target; abort.
106 * (We can't use the intermediate origin state.)
107 * - I/O error while merging
108 * => stop merging; set merge_failed; process I/O normally.
113 * Incoming bios that overlap with chunks being merged must wait
114 * for them to be committed.
116 struct bio_list bios_queued_during_merge;
121 * RUNNING_MERGE - Merge operation is in progress.
122 * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped;
123 * cleared afterwards.
125 #define RUNNING_MERGE 0
126 #define SHUTDOWN_MERGE 1
128 struct dm_dev *dm_snap_origin(struct dm_snapshot *s)
132 EXPORT_SYMBOL(dm_snap_origin);
134 struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
138 EXPORT_SYMBOL(dm_snap_cow);
140 static sector_t chunk_to_sector(struct dm_exception_store *store,
143 return chunk << store->chunk_shift;
146 static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
149 * There is only ever one instance of a particular block
150 * device so we can compare pointers safely.
155 struct dm_snap_pending_exception {
156 struct dm_exception e;
159 * Origin buffers waiting for this to complete are held
162 struct bio_list origin_bios;
163 struct bio_list snapshot_bios;
165 /* Pointer back to snapshot context */
166 struct dm_snapshot *snap;
169 * 1 indicates the exception has already been sent to
175 * For writing a complete chunk, bypassing the copy.
177 struct bio *full_bio;
178 bio_end_io_t *full_bio_end_io;
179 void *full_bio_private;
183 * Hash table mapping origin volumes to lists of snapshots and
184 * a lock to protect it
186 static struct kmem_cache *exception_cache;
187 static struct kmem_cache *pending_cache;
189 struct dm_snap_tracked_chunk {
190 struct hlist_node node;
194 static struct kmem_cache *tracked_chunk_cache;
196 static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s,
199 struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool,
205 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
206 hlist_add_head(&c->node,
207 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
208 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
213 static void stop_tracking_chunk(struct dm_snapshot *s,
214 struct dm_snap_tracked_chunk *c)
218 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
220 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
222 mempool_free(c, s->tracked_chunk_pool);
225 static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
227 struct dm_snap_tracked_chunk *c;
228 struct hlist_node *hn;
231 spin_lock_irq(&s->tracked_chunk_lock);
233 hlist_for_each_entry(c, hn,
234 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
235 if (c->chunk == chunk) {
241 spin_unlock_irq(&s->tracked_chunk_lock);
247 * This conflicting I/O is extremely improbable in the caller,
248 * so msleep(1) is sufficient and there is no need for a wait queue.
250 static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
252 while (__chunk_is_tracked(s, chunk))
257 * One of these per registered origin, held in the snapshot_origins hash
260 /* The origin device */
261 struct block_device *bdev;
263 struct list_head hash_list;
265 /* List of snapshots for this origin */
266 struct list_head snapshots;
270 * Size of the hash table for origin volumes. If we make this
271 * the size of the minors list then it should be nearly perfect
273 #define ORIGIN_HASH_SIZE 256
274 #define ORIGIN_MASK 0xFF
275 static struct list_head *_origins;
276 static struct rw_semaphore _origins_lock;
278 static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
279 static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
280 static uint64_t _pending_exceptions_done_count;
282 static int init_origin_hash(void)
286 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
289 DMERR("unable to allocate memory");
293 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
294 INIT_LIST_HEAD(_origins + i);
295 init_rwsem(&_origins_lock);
300 static void exit_origin_hash(void)
305 static unsigned origin_hash(struct block_device *bdev)
307 return bdev->bd_dev & ORIGIN_MASK;
310 static struct origin *__lookup_origin(struct block_device *origin)
312 struct list_head *ol;
315 ol = &_origins[origin_hash(origin)];
316 list_for_each_entry (o, ol, hash_list)
317 if (bdev_equal(o->bdev, origin))
323 static void __insert_origin(struct origin *o)
325 struct list_head *sl = &_origins[origin_hash(o->bdev)];
326 list_add_tail(&o->hash_list, sl);
330 * _origins_lock must be held when calling this function.
331 * Returns number of snapshots registered using the supplied cow device, plus:
332 * snap_src - a snapshot suitable for use as a source of exception handover
333 * snap_dest - a snapshot capable of receiving exception handover.
334 * snap_merge - an existing snapshot-merge target linked to the same origin.
335 * There can be at most one snapshot-merge target. The parameter is optional.
337 * Possible return values and states of snap_src and snap_dest.
338 * 0: NULL, NULL - first new snapshot
339 * 1: snap_src, NULL - normal snapshot
340 * 2: snap_src, snap_dest - waiting for handover
341 * 2: snap_src, NULL - handed over, waiting for old to be deleted
342 * 1: NULL, snap_dest - source got destroyed without handover
344 static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
345 struct dm_snapshot **snap_src,
346 struct dm_snapshot **snap_dest,
347 struct dm_snapshot **snap_merge)
349 struct dm_snapshot *s;
354 o = __lookup_origin(snap->origin->bdev);
358 list_for_each_entry(s, &o->snapshots, list) {
359 if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
361 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
371 } else if (snap_dest)
382 * On success, returns 1 if this snapshot is a handover destination,
383 * otherwise returns 0.
385 static int __validate_exception_handover(struct dm_snapshot *snap)
387 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
388 struct dm_snapshot *snap_merge = NULL;
390 /* Does snapshot need exceptions handed over to it? */
391 if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
392 &snap_merge) == 2) ||
394 snap->ti->error = "Snapshot cow pairing for exception "
395 "table handover failed";
400 * If no snap_src was found, snap cannot become a handover
407 * Non-snapshot-merge handover?
409 if (!dm_target_is_snapshot_merge(snap->ti))
413 * Do not allow more than one merging snapshot.
416 snap->ti->error = "A snapshot is already merging.";
420 if (!snap_src->store->type->prepare_merge ||
421 !snap_src->store->type->commit_merge) {
422 snap->ti->error = "Snapshot exception store does not "
423 "support snapshot-merge.";
430 static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
432 struct dm_snapshot *l;
434 /* Sort the list according to chunk size, largest-first smallest-last */
435 list_for_each_entry(l, &o->snapshots, list)
436 if (l->store->chunk_size < s->store->chunk_size)
438 list_add_tail(&s->list, &l->list);
442 * Make a note of the snapshot and its origin so we can look it
443 * up when the origin has a write on it.
445 * Also validate snapshot exception store handovers.
446 * On success, returns 1 if this registration is a handover destination,
447 * otherwise returns 0.
449 static int register_snapshot(struct dm_snapshot *snap)
451 struct origin *o, *new_o = NULL;
452 struct block_device *bdev = snap->origin->bdev;
455 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
459 down_write(&_origins_lock);
461 r = __validate_exception_handover(snap);
467 o = __lookup_origin(bdev);
474 /* Initialise the struct */
475 INIT_LIST_HEAD(&o->snapshots);
481 __insert_snapshot(o, snap);
484 up_write(&_origins_lock);
490 * Move snapshot to correct place in list according to chunk size.
492 static void reregister_snapshot(struct dm_snapshot *s)
494 struct block_device *bdev = s->origin->bdev;
496 down_write(&_origins_lock);
499 __insert_snapshot(__lookup_origin(bdev), s);
501 up_write(&_origins_lock);
504 static void unregister_snapshot(struct dm_snapshot *s)
508 down_write(&_origins_lock);
509 o = __lookup_origin(s->origin->bdev);
512 if (o && list_empty(&o->snapshots)) {
513 list_del(&o->hash_list);
517 up_write(&_origins_lock);
521 * Implementation of the exception hash tables.
522 * The lowest hash_shift bits of the chunk number are ignored, allowing
523 * some consecutive chunks to be grouped together.
525 static int dm_exception_table_init(struct dm_exception_table *et,
526 uint32_t size, unsigned hash_shift)
530 et->hash_shift = hash_shift;
531 et->hash_mask = size - 1;
532 et->table = dm_vcalloc(size, sizeof(struct list_head));
536 for (i = 0; i < size; i++)
537 INIT_LIST_HEAD(et->table + i);
542 static void dm_exception_table_exit(struct dm_exception_table *et,
543 struct kmem_cache *mem)
545 struct list_head *slot;
546 struct dm_exception *ex, *next;
549 size = et->hash_mask + 1;
550 for (i = 0; i < size; i++) {
551 slot = et->table + i;
553 list_for_each_entry_safe (ex, next, slot, hash_list)
554 kmem_cache_free(mem, ex);
560 static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
562 return (chunk >> et->hash_shift) & et->hash_mask;
565 static void dm_remove_exception(struct dm_exception *e)
567 list_del(&e->hash_list);
571 * Return the exception data for a sector, or NULL if not
574 static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
577 struct list_head *slot;
578 struct dm_exception *e;
580 slot = &et->table[exception_hash(et, chunk)];
581 list_for_each_entry (e, slot, hash_list)
582 if (chunk >= e->old_chunk &&
583 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
589 static struct dm_exception *alloc_completed_exception(void)
591 struct dm_exception *e;
593 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
595 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
600 static void free_completed_exception(struct dm_exception *e)
602 kmem_cache_free(exception_cache, e);
605 static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
607 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
610 atomic_inc(&s->pending_exceptions_count);
616 static void free_pending_exception(struct dm_snap_pending_exception *pe)
618 struct dm_snapshot *s = pe->snap;
620 mempool_free(pe, s->pending_pool);
621 smp_mb__before_atomic_dec();
622 atomic_dec(&s->pending_exceptions_count);
625 static void dm_insert_exception(struct dm_exception_table *eh,
626 struct dm_exception *new_e)
629 struct dm_exception *e = NULL;
631 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
633 /* Add immediately if this table doesn't support consecutive chunks */
637 /* List is ordered by old_chunk */
638 list_for_each_entry_reverse(e, l, hash_list) {
639 /* Insert after an existing chunk? */
640 if (new_e->old_chunk == (e->old_chunk +
641 dm_consecutive_chunk_count(e) + 1) &&
642 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
643 dm_consecutive_chunk_count(e) + 1)) {
644 dm_consecutive_chunk_count_inc(e);
645 free_completed_exception(new_e);
649 /* Insert before an existing chunk? */
650 if (new_e->old_chunk == (e->old_chunk - 1) &&
651 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
652 dm_consecutive_chunk_count_inc(e);
655 free_completed_exception(new_e);
659 if (new_e->old_chunk > e->old_chunk)
664 list_add(&new_e->hash_list, e ? &e->hash_list : l);
668 * Callback used by the exception stores to load exceptions when
671 static int dm_add_exception(void *context, chunk_t old, chunk_t new)
673 struct dm_snapshot *s = context;
674 struct dm_exception *e;
676 e = alloc_completed_exception();
682 /* Consecutive_count is implicitly initialised to zero */
685 dm_insert_exception(&s->complete, e);
691 * Return a minimum chunk size of all snapshots that have the specified origin.
692 * Return zero if the origin has no snapshots.
694 static sector_t __minimum_chunk_size(struct origin *o)
696 struct dm_snapshot *snap;
697 unsigned chunk_size = 0;
700 list_for_each_entry(snap, &o->snapshots, list)
701 chunk_size = min_not_zero(chunk_size,
702 snap->store->chunk_size);
710 static int calc_max_buckets(void)
712 /* use a fixed size of 2MB */
713 unsigned long mem = 2 * 1024 * 1024;
714 mem /= sizeof(struct list_head);
720 * Allocate room for a suitable hash table.
722 static int init_hash_tables(struct dm_snapshot *s)
724 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
727 * Calculate based on the size of the original volume or
730 cow_dev_size = get_dev_size(s->cow->bdev);
731 origin_dev_size = get_dev_size(s->origin->bdev);
732 max_buckets = calc_max_buckets();
734 hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift;
735 hash_size = min(hash_size, max_buckets);
739 hash_size = rounddown_pow_of_two(hash_size);
740 if (dm_exception_table_init(&s->complete, hash_size,
741 DM_CHUNK_CONSECUTIVE_BITS))
745 * Allocate hash table for in-flight exceptions
746 * Make this smaller than the real hash table
752 if (dm_exception_table_init(&s->pending, hash_size, 0)) {
753 dm_exception_table_exit(&s->complete, exception_cache);
760 static void merge_shutdown(struct dm_snapshot *s)
762 clear_bit_unlock(RUNNING_MERGE, &s->state_bits);
763 smp_mb__after_clear_bit();
764 wake_up_bit(&s->state_bits, RUNNING_MERGE);
767 static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s)
769 s->first_merging_chunk = 0;
770 s->num_merging_chunks = 0;
772 return bio_list_get(&s->bios_queued_during_merge);
776 * Remove one chunk from the index of completed exceptions.
778 static int __remove_single_exception_chunk(struct dm_snapshot *s,
781 struct dm_exception *e;
783 e = dm_lookup_exception(&s->complete, old_chunk);
785 DMERR("Corruption detected: exception for block %llu is "
786 "on disk but not in memory",
787 (unsigned long long)old_chunk);
792 * If this is the only chunk using this exception, remove exception.
794 if (!dm_consecutive_chunk_count(e)) {
795 dm_remove_exception(e);
796 free_completed_exception(e);
801 * The chunk may be either at the beginning or the end of a
802 * group of consecutive chunks - never in the middle. We are
803 * removing chunks in the opposite order to that in which they
804 * were added, so this should always be true.
805 * Decrement the consecutive chunk counter and adjust the
806 * starting point if necessary.
808 if (old_chunk == e->old_chunk) {
811 } else if (old_chunk != e->old_chunk +
812 dm_consecutive_chunk_count(e)) {
813 DMERR("Attempt to merge block %llu from the "
814 "middle of a chunk range [%llu - %llu]",
815 (unsigned long long)old_chunk,
816 (unsigned long long)e->old_chunk,
818 e->old_chunk + dm_consecutive_chunk_count(e));
822 dm_consecutive_chunk_count_dec(e);
827 static void flush_bios(struct bio *bio);
829 static int remove_single_exception_chunk(struct dm_snapshot *s)
831 struct bio *b = NULL;
833 chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1;
835 down_write(&s->lock);
838 * Process chunks (and associated exceptions) in reverse order
839 * so that dm_consecutive_chunk_count_dec() accounting works.
842 r = __remove_single_exception_chunk(s, old_chunk);
845 } while (old_chunk-- > s->first_merging_chunk);
847 b = __release_queued_bios_after_merge(s);
857 static int origin_write_extent(struct dm_snapshot *merging_snap,
858 sector_t sector, unsigned chunk_size);
860 static void merge_callback(int read_err, unsigned long write_err,
863 static uint64_t read_pending_exceptions_done_count(void)
865 uint64_t pending_exceptions_done;
867 spin_lock(&_pending_exceptions_done_spinlock);
868 pending_exceptions_done = _pending_exceptions_done_count;
869 spin_unlock(&_pending_exceptions_done_spinlock);
871 return pending_exceptions_done;
874 static void increment_pending_exceptions_done_count(void)
876 spin_lock(&_pending_exceptions_done_spinlock);
877 _pending_exceptions_done_count++;
878 spin_unlock(&_pending_exceptions_done_spinlock);
880 wake_up_all(&_pending_exceptions_done);
883 static void snapshot_merge_next_chunks(struct dm_snapshot *s)
885 int i, linear_chunks;
886 chunk_t old_chunk, new_chunk;
887 struct dm_io_region src, dest;
889 uint64_t previous_count;
891 BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
892 if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
896 * valid flag never changes during merge, so no lock required.
899 DMERR("Snapshot is invalid: can't merge");
903 linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
905 if (linear_chunks <= 0) {
906 if (linear_chunks < 0) {
907 DMERR("Read error in exception store: "
908 "shutting down merge");
909 down_write(&s->lock);
916 /* Adjust old_chunk and new_chunk to reflect start of linear region */
917 old_chunk = old_chunk + 1 - linear_chunks;
918 new_chunk = new_chunk + 1 - linear_chunks;
921 * Use one (potentially large) I/O to copy all 'linear_chunks'
922 * from the exception store to the origin
924 io_size = linear_chunks * s->store->chunk_size;
926 dest.bdev = s->origin->bdev;
927 dest.sector = chunk_to_sector(s->store, old_chunk);
928 dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector);
930 src.bdev = s->cow->bdev;
931 src.sector = chunk_to_sector(s->store, new_chunk);
932 src.count = dest.count;
935 * Reallocate any exceptions needed in other snapshots then
936 * wait for the pending exceptions to complete.
937 * Each time any pending exception (globally on the system)
938 * completes we are woken and repeat the process to find out
939 * if we can proceed. While this may not seem a particularly
940 * efficient algorithm, it is not expected to have any
941 * significant impact on performance.
943 previous_count = read_pending_exceptions_done_count();
944 while (origin_write_extent(s, dest.sector, io_size)) {
945 wait_event(_pending_exceptions_done,
946 (read_pending_exceptions_done_count() !=
948 /* Retry after the wait, until all exceptions are done. */
949 previous_count = read_pending_exceptions_done_count();
952 down_write(&s->lock);
953 s->first_merging_chunk = old_chunk;
954 s->num_merging_chunks = linear_chunks;
957 /* Wait until writes to all 'linear_chunks' drain */
958 for (i = 0; i < linear_chunks; i++)
959 __check_for_conflicting_io(s, old_chunk + i);
961 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s);
968 static void error_bios(struct bio *bio);
970 static void merge_callback(int read_err, unsigned long write_err, void *context)
972 struct dm_snapshot *s = context;
973 struct bio *b = NULL;
975 if (read_err || write_err) {
977 DMERR("Read error: shutting down merge.");
979 DMERR("Write error: shutting down merge.");
983 if (s->store->type->commit_merge(s->store,
984 s->num_merging_chunks) < 0) {
985 DMERR("Write error in exception store: shutting down merge");
989 if (remove_single_exception_chunk(s) < 0)
992 snapshot_merge_next_chunks(s);
997 down_write(&s->lock);
999 b = __release_queued_bios_after_merge(s);
1006 static void start_merge(struct dm_snapshot *s)
1008 if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits))
1009 snapshot_merge_next_chunks(s);
1012 static int wait_schedule(void *ptr)
1020 * Stop the merging process and wait until it finishes.
1022 static void stop_merge(struct dm_snapshot *s)
1024 set_bit(SHUTDOWN_MERGE, &s->state_bits);
1025 wait_on_bit(&s->state_bits, RUNNING_MERGE, wait_schedule,
1026 TASK_UNINTERRUPTIBLE);
1027 clear_bit(SHUTDOWN_MERGE, &s->state_bits);
1031 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
1033 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1035 struct dm_snapshot *s;
1038 char *origin_path, *cow_path;
1039 unsigned args_used, num_flush_requests = 1;
1040 fmode_t origin_mode = FMODE_READ;
1043 ti->error = "requires exactly 4 arguments";
1048 if (dm_target_is_snapshot_merge(ti)) {
1049 num_flush_requests = 2;
1050 origin_mode = FMODE_WRITE;
1053 s = kmalloc(sizeof(*s), GFP_KERNEL);
1055 ti->error = "Cannot allocate private snapshot structure";
1060 origin_path = argv[0];
1064 r = dm_get_device(ti, origin_path, origin_mode, &s->origin);
1066 ti->error = "Cannot get origin device";
1074 r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow);
1076 ti->error = "Cannot get COW device";
1080 r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
1082 ti->error = "Couldn't create exception store";
1093 atomic_set(&s->pending_exceptions_count, 0);
1094 init_rwsem(&s->lock);
1095 INIT_LIST_HEAD(&s->list);
1096 spin_lock_init(&s->pe_lock);
1098 s->merge_failed = 0;
1099 s->first_merging_chunk = 0;
1100 s->num_merging_chunks = 0;
1101 bio_list_init(&s->bios_queued_during_merge);
1103 /* Allocate hash table for COW data */
1104 if (init_hash_tables(s)) {
1105 ti->error = "Unable to allocate hash table space";
1107 goto bad_hash_tables;
1110 s->kcopyd_client = dm_kcopyd_client_create();
1111 if (IS_ERR(s->kcopyd_client)) {
1112 r = PTR_ERR(s->kcopyd_client);
1113 ti->error = "Could not create kcopyd client";
1117 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
1118 if (!s->pending_pool) {
1119 ti->error = "Could not allocate mempool for pending exceptions";
1121 goto bad_pending_pool;
1124 s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS,
1125 tracked_chunk_cache);
1126 if (!s->tracked_chunk_pool) {
1127 ti->error = "Could not allocate tracked_chunk mempool for "
1129 goto bad_tracked_chunk_pool;
1132 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1133 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
1135 spin_lock_init(&s->tracked_chunk_lock);
1138 ti->num_flush_requests = num_flush_requests;
1140 /* Add snapshot to the list of snapshots for this origin */
1141 /* Exceptions aren't triggered till snapshot_resume() is called */
1142 r = register_snapshot(s);
1144 ti->error = "Snapshot origin struct allocation failed";
1145 goto bad_load_and_register;
1147 /* invalid handover, register_snapshot has set ti->error */
1148 goto bad_load_and_register;
1152 * Metadata must only be loaded into one table at once, so skip this
1153 * if metadata will be handed over during resume.
1154 * Chunk size will be set during the handover - set it to zero to
1155 * ensure it's ignored.
1158 s->store->chunk_size = 0;
1162 r = s->store->type->read_metadata(s->store, dm_add_exception,
1165 ti->error = "Failed to read snapshot metadata";
1166 goto bad_read_metadata;
1169 DMWARN("Snapshot is marked invalid.");
1172 if (!s->store->chunk_size) {
1173 ti->error = "Chunk size not set";
1174 goto bad_read_metadata;
1176 ti->split_io = s->store->chunk_size;
1181 unregister_snapshot(s);
1183 bad_load_and_register:
1184 mempool_destroy(s->tracked_chunk_pool);
1186 bad_tracked_chunk_pool:
1187 mempool_destroy(s->pending_pool);
1190 dm_kcopyd_client_destroy(s->kcopyd_client);
1193 dm_exception_table_exit(&s->pending, pending_cache);
1194 dm_exception_table_exit(&s->complete, exception_cache);
1197 dm_exception_store_destroy(s->store);
1200 dm_put_device(ti, s->cow);
1203 dm_put_device(ti, s->origin);
1212 static void __free_exceptions(struct dm_snapshot *s)
1214 dm_kcopyd_client_destroy(s->kcopyd_client);
1215 s->kcopyd_client = NULL;
1217 dm_exception_table_exit(&s->pending, pending_cache);
1218 dm_exception_table_exit(&s->complete, exception_cache);
1221 static void __handover_exceptions(struct dm_snapshot *snap_src,
1222 struct dm_snapshot *snap_dest)
1225 struct dm_exception_table table_swap;
1226 struct dm_exception_store *store_swap;
1230 * Swap all snapshot context information between the two instances.
1232 u.table_swap = snap_dest->complete;
1233 snap_dest->complete = snap_src->complete;
1234 snap_src->complete = u.table_swap;
1236 u.store_swap = snap_dest->store;
1237 snap_dest->store = snap_src->store;
1238 snap_src->store = u.store_swap;
1240 snap_dest->store->snap = snap_dest;
1241 snap_src->store->snap = snap_src;
1243 snap_dest->ti->split_io = snap_dest->store->chunk_size;
1244 snap_dest->valid = snap_src->valid;
1247 * Set source invalid to ensure it receives no further I/O.
1249 snap_src->valid = 0;
1252 static void snapshot_dtr(struct dm_target *ti)
1254 #ifdef CONFIG_DM_DEBUG
1257 struct dm_snapshot *s = ti->private;
1258 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1260 down_read(&_origins_lock);
1261 /* Check whether exception handover must be cancelled */
1262 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1263 if (snap_src && snap_dest && (s == snap_src)) {
1264 down_write(&snap_dest->lock);
1265 snap_dest->valid = 0;
1266 up_write(&snap_dest->lock);
1267 DMERR("Cancelling snapshot handover.");
1269 up_read(&_origins_lock);
1271 if (dm_target_is_snapshot_merge(ti))
1274 /* Prevent further origin writes from using this snapshot. */
1275 /* After this returns there can be no new kcopyd jobs. */
1276 unregister_snapshot(s);
1278 while (atomic_read(&s->pending_exceptions_count))
1281 * Ensure instructions in mempool_destroy aren't reordered
1282 * before atomic_read.
1286 #ifdef CONFIG_DM_DEBUG
1287 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1288 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
1291 mempool_destroy(s->tracked_chunk_pool);
1293 __free_exceptions(s);
1295 mempool_destroy(s->pending_pool);
1297 dm_exception_store_destroy(s->store);
1299 dm_put_device(ti, s->cow);
1301 dm_put_device(ti, s->origin);
1307 * Flush a list of buffers.
1309 static void flush_bios(struct bio *bio)
1315 bio->bi_next = NULL;
1316 generic_make_request(bio);
1321 static int do_origin(struct dm_dev *origin, struct bio *bio);
1324 * Flush a list of buffers.
1326 static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
1333 bio->bi_next = NULL;
1334 r = do_origin(s->origin, bio);
1335 if (r == DM_MAPIO_REMAPPED)
1336 generic_make_request(bio);
1342 * Error a list of buffers.
1344 static void error_bios(struct bio *bio)
1350 bio->bi_next = NULL;
1356 static void __invalidate_snapshot(struct dm_snapshot *s, int err)
1362 DMERR("Invalidating snapshot: Error reading/writing.");
1363 else if (err == -ENOMEM)
1364 DMERR("Invalidating snapshot: Unable to allocate exception.");
1366 if (s->store->type->drop_snapshot)
1367 s->store->type->drop_snapshot(s->store);
1371 dm_table_event(s->ti->table);
1374 static void pending_complete(struct dm_snap_pending_exception *pe, int success)
1376 struct dm_exception *e;
1377 struct dm_snapshot *s = pe->snap;
1378 struct bio *origin_bios = NULL;
1379 struct bio *snapshot_bios = NULL;
1380 struct bio *full_bio = NULL;
1384 /* Read/write error - snapshot is unusable */
1385 down_write(&s->lock);
1386 __invalidate_snapshot(s, -EIO);
1391 e = alloc_completed_exception();
1393 down_write(&s->lock);
1394 __invalidate_snapshot(s, -ENOMEM);
1400 down_write(&s->lock);
1402 free_completed_exception(e);
1407 /* Check for conflicting reads */
1408 __check_for_conflicting_io(s, pe->e.old_chunk);
1411 * Add a proper exception, and remove the
1412 * in-flight exception from the list.
1414 dm_insert_exception(&s->complete, e);
1417 dm_remove_exception(&pe->e);
1418 snapshot_bios = bio_list_get(&pe->snapshot_bios);
1419 origin_bios = bio_list_get(&pe->origin_bios);
1420 full_bio = pe->full_bio;
1422 full_bio->bi_end_io = pe->full_bio_end_io;
1423 full_bio->bi_private = pe->full_bio_private;
1425 free_pending_exception(pe);
1427 increment_pending_exceptions_done_count();
1431 /* Submit any pending write bios */
1434 bio_io_error(full_bio);
1435 error_bios(snapshot_bios);
1438 bio_endio(full_bio, 0);
1439 flush_bios(snapshot_bios);
1442 retry_origin_bios(s, origin_bios);
1445 static void commit_callback(void *context, int success)
1447 struct dm_snap_pending_exception *pe = context;
1449 pending_complete(pe, success);
1453 * Called when the copy I/O has finished. kcopyd actually runs
1454 * this code so don't block.
1456 static void copy_callback(int read_err, unsigned long write_err, void *context)
1458 struct dm_snap_pending_exception *pe = context;
1459 struct dm_snapshot *s = pe->snap;
1461 if (read_err || write_err)
1462 pending_complete(pe, 0);
1465 /* Update the metadata if we are persistent */
1466 s->store->type->commit_exception(s->store, &pe->e,
1467 commit_callback, pe);
1471 * Dispatches the copy operation to kcopyd.
1473 static void start_copy(struct dm_snap_pending_exception *pe)
1475 struct dm_snapshot *s = pe->snap;
1476 struct dm_io_region src, dest;
1477 struct block_device *bdev = s->origin->bdev;
1480 dev_size = get_dev_size(bdev);
1483 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
1484 src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
1486 dest.bdev = s->cow->bdev;
1487 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
1488 dest.count = src.count;
1490 /* Hand over to kcopyd */
1491 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe);
1494 static void full_bio_end_io(struct bio *bio, int error)
1496 void *callback_data = bio->bi_private;
1498 dm_kcopyd_do_callback(callback_data, 0, error ? 1 : 0);
1501 static void start_full_bio(struct dm_snap_pending_exception *pe,
1504 struct dm_snapshot *s = pe->snap;
1505 void *callback_data;
1508 pe->full_bio_end_io = bio->bi_end_io;
1509 pe->full_bio_private = bio->bi_private;
1511 callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client,
1514 bio->bi_end_io = full_bio_end_io;
1515 bio->bi_private = callback_data;
1517 generic_make_request(bio);
1520 static struct dm_snap_pending_exception *
1521 __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1523 struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
1528 return container_of(e, struct dm_snap_pending_exception, e);
1532 * Looks to see if this snapshot already has a pending exception
1533 * for this chunk, otherwise it allocates a new one and inserts
1534 * it into the pending table.
1536 * NOTE: a write lock must be held on snap->lock before calling
1539 static struct dm_snap_pending_exception *
1540 __find_pending_exception(struct dm_snapshot *s,
1541 struct dm_snap_pending_exception *pe, chunk_t chunk)
1543 struct dm_snap_pending_exception *pe2;
1545 pe2 = __lookup_pending_exception(s, chunk);
1547 free_pending_exception(pe);
1551 pe->e.old_chunk = chunk;
1552 bio_list_init(&pe->origin_bios);
1553 bio_list_init(&pe->snapshot_bios);
1555 pe->full_bio = NULL;
1557 if (s->store->type->prepare_exception(s->store, &pe->e)) {
1558 free_pending_exception(pe);
1562 dm_insert_exception(&s->pending, &pe->e);
1567 static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
1568 struct bio *bio, chunk_t chunk)
1570 bio->bi_bdev = s->cow->bdev;
1571 bio->bi_sector = chunk_to_sector(s->store,
1572 dm_chunk_number(e->new_chunk) +
1573 (chunk - e->old_chunk)) +
1575 s->store->chunk_mask);
1578 static int snapshot_map(struct dm_target *ti, struct bio *bio,
1579 union map_info *map_context)
1581 struct dm_exception *e;
1582 struct dm_snapshot *s = ti->private;
1583 int r = DM_MAPIO_REMAPPED;
1585 struct dm_snap_pending_exception *pe = NULL;
1587 if (bio->bi_rw & REQ_FLUSH) {
1588 bio->bi_bdev = s->cow->bdev;
1589 return DM_MAPIO_REMAPPED;
1592 chunk = sector_to_chunk(s->store, bio->bi_sector);
1594 /* Full snapshots are not usable */
1595 /* To get here the table must be live so s->active is always set. */
1599 /* FIXME: should only take write lock if we need
1600 * to copy an exception */
1601 down_write(&s->lock);
1608 /* If the block is already remapped - use that, else remap it */
1609 e = dm_lookup_exception(&s->complete, chunk);
1611 remap_exception(s, e, bio, chunk);
1616 * Write to snapshot - higher level takes care of RW/RO
1617 * flags so we should only get this if we are
1620 if (bio_rw(bio) == WRITE) {
1621 pe = __lookup_pending_exception(s, chunk);
1624 pe = alloc_pending_exception(s);
1625 down_write(&s->lock);
1628 free_pending_exception(pe);
1633 e = dm_lookup_exception(&s->complete, chunk);
1635 free_pending_exception(pe);
1636 remap_exception(s, e, bio, chunk);
1640 pe = __find_pending_exception(s, pe, chunk);
1642 __invalidate_snapshot(s, -ENOMEM);
1648 remap_exception(s, &pe->e, bio, chunk);
1650 r = DM_MAPIO_SUBMITTED;
1653 bio->bi_size == (s->store->chunk_size << SECTOR_SHIFT)) {
1656 start_full_bio(pe, bio);
1660 bio_list_add(&pe->snapshot_bios, bio);
1663 /* this is protected by snap->lock */
1670 bio->bi_bdev = s->origin->bdev;
1671 map_context->ptr = track_chunk(s, chunk);
1681 * A snapshot-merge target behaves like a combination of a snapshot
1682 * target and a snapshot-origin target. It only generates new
1683 * exceptions in other snapshots and not in the one that is being
1686 * For each chunk, if there is an existing exception, it is used to
1687 * redirect I/O to the cow device. Otherwise I/O is sent to the origin,
1688 * which in turn might generate exceptions in other snapshots.
1689 * If merging is currently taking place on the chunk in question, the
1690 * I/O is deferred by adding it to s->bios_queued_during_merge.
1692 static int snapshot_merge_map(struct dm_target *ti, struct bio *bio,
1693 union map_info *map_context)
1695 struct dm_exception *e;
1696 struct dm_snapshot *s = ti->private;
1697 int r = DM_MAPIO_REMAPPED;
1700 if (bio->bi_rw & REQ_FLUSH) {
1701 if (!map_context->target_request_nr)
1702 bio->bi_bdev = s->origin->bdev;
1704 bio->bi_bdev = s->cow->bdev;
1705 map_context->ptr = NULL;
1706 return DM_MAPIO_REMAPPED;
1709 chunk = sector_to_chunk(s->store, bio->bi_sector);
1711 down_write(&s->lock);
1713 /* Full merging snapshots are redirected to the origin */
1715 goto redirect_to_origin;
1717 /* If the block is already remapped - use that */
1718 e = dm_lookup_exception(&s->complete, chunk);
1720 /* Queue writes overlapping with chunks being merged */
1721 if (bio_rw(bio) == WRITE &&
1722 chunk >= s->first_merging_chunk &&
1723 chunk < (s->first_merging_chunk +
1724 s->num_merging_chunks)) {
1725 bio->bi_bdev = s->origin->bdev;
1726 bio_list_add(&s->bios_queued_during_merge, bio);
1727 r = DM_MAPIO_SUBMITTED;
1731 remap_exception(s, e, bio, chunk);
1733 if (bio_rw(bio) == WRITE)
1734 map_context->ptr = track_chunk(s, chunk);
1739 bio->bi_bdev = s->origin->bdev;
1741 if (bio_rw(bio) == WRITE) {
1743 return do_origin(s->origin, bio);
1752 static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1753 int error, union map_info *map_context)
1755 struct dm_snapshot *s = ti->private;
1756 struct dm_snap_tracked_chunk *c = map_context->ptr;
1759 stop_tracking_chunk(s, c);
1764 static void snapshot_merge_presuspend(struct dm_target *ti)
1766 struct dm_snapshot *s = ti->private;
1771 static int snapshot_preresume(struct dm_target *ti)
1774 struct dm_snapshot *s = ti->private;
1775 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1777 down_read(&_origins_lock);
1778 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1779 if (snap_src && snap_dest) {
1780 down_read(&snap_src->lock);
1781 if (s == snap_src) {
1782 DMERR("Unable to resume snapshot source until "
1783 "handover completes.");
1785 } else if (!dm_suspended(snap_src->ti)) {
1786 DMERR("Unable to perform snapshot handover until "
1787 "source is suspended.");
1790 up_read(&snap_src->lock);
1792 up_read(&_origins_lock);
1797 static void snapshot_resume(struct dm_target *ti)
1799 struct dm_snapshot *s = ti->private;
1800 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1802 down_read(&_origins_lock);
1803 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1804 if (snap_src && snap_dest) {
1805 down_write(&snap_src->lock);
1806 down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1807 __handover_exceptions(snap_src, snap_dest);
1808 up_write(&snap_dest->lock);
1809 up_write(&snap_src->lock);
1811 up_read(&_origins_lock);
1813 /* Now we have correct chunk size, reregister */
1814 reregister_snapshot(s);
1816 down_write(&s->lock);
1821 static sector_t get_origin_minimum_chunksize(struct block_device *bdev)
1823 sector_t min_chunksize;
1825 down_read(&_origins_lock);
1826 min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
1827 up_read(&_origins_lock);
1829 return min_chunksize;
1832 static void snapshot_merge_resume(struct dm_target *ti)
1834 struct dm_snapshot *s = ti->private;
1837 * Handover exceptions from existing snapshot.
1839 snapshot_resume(ti);
1842 * snapshot-merge acts as an origin, so set ti->split_io
1844 ti->split_io = get_origin_minimum_chunksize(s->origin->bdev);
1849 static void snapshot_status(struct dm_target *ti, status_type_t type,
1850 char *result, unsigned maxlen)
1853 struct dm_snapshot *snap = ti->private;
1856 case STATUSTYPE_INFO:
1858 down_write(&snap->lock);
1862 else if (snap->merge_failed)
1863 DMEMIT("Merge failed");
1865 if (snap->store->type->usage) {
1866 sector_t total_sectors, sectors_allocated,
1868 snap->store->type->usage(snap->store,
1872 DMEMIT("%llu/%llu %llu",
1873 (unsigned long long)sectors_allocated,
1874 (unsigned long long)total_sectors,
1875 (unsigned long long)metadata_sectors);
1881 up_write(&snap->lock);
1885 case STATUSTYPE_TABLE:
1887 * kdevname returns a static pointer so we need
1888 * to make private copies if the output is to
1891 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
1892 snap->store->type->status(snap->store, type, result + sz,
1898 static int snapshot_iterate_devices(struct dm_target *ti,
1899 iterate_devices_callout_fn fn, void *data)
1901 struct dm_snapshot *snap = ti->private;
1904 r = fn(ti, snap->origin, 0, ti->len, data);
1907 r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data);
1913 /*-----------------------------------------------------------------
1915 *---------------------------------------------------------------*/
1918 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
1919 * supplied bio was ignored. The caller may submit it immediately.
1920 * (No remapping actually occurs as the origin is always a direct linear
1923 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
1924 * and any supplied bio is added to a list to be submitted once all
1925 * the necessary exceptions exist.
1927 static int __origin_write(struct list_head *snapshots, sector_t sector,
1930 int r = DM_MAPIO_REMAPPED;
1931 struct dm_snapshot *snap;
1932 struct dm_exception *e;
1933 struct dm_snap_pending_exception *pe;
1934 struct dm_snap_pending_exception *pe_to_start_now = NULL;
1935 struct dm_snap_pending_exception *pe_to_start_last = NULL;
1938 /* Do all the snapshots on this origin */
1939 list_for_each_entry (snap, snapshots, list) {
1941 * Don't make new exceptions in a merging snapshot
1942 * because it has effectively been deleted
1944 if (dm_target_is_snapshot_merge(snap->ti))
1947 down_write(&snap->lock);
1949 /* Only deal with valid and active snapshots */
1950 if (!snap->valid || !snap->active)
1953 /* Nothing to do if writing beyond end of snapshot */
1954 if (sector >= dm_table_get_size(snap->ti->table))
1958 * Remember, different snapshots can have
1959 * different chunk sizes.
1961 chunk = sector_to_chunk(snap->store, sector);
1964 * Check exception table to see if block
1965 * is already remapped in this snapshot
1966 * and trigger an exception if not.
1968 e = dm_lookup_exception(&snap->complete, chunk);
1972 pe = __lookup_pending_exception(snap, chunk);
1974 up_write(&snap->lock);
1975 pe = alloc_pending_exception(snap);
1976 down_write(&snap->lock);
1979 free_pending_exception(pe);
1983 e = dm_lookup_exception(&snap->complete, chunk);
1985 free_pending_exception(pe);
1989 pe = __find_pending_exception(snap, pe, chunk);
1991 __invalidate_snapshot(snap, -ENOMEM);
1996 r = DM_MAPIO_SUBMITTED;
1999 * If an origin bio was supplied, queue it to wait for the
2000 * completion of this exception, and start this one last,
2001 * at the end of the function.
2004 bio_list_add(&pe->origin_bios, bio);
2009 pe_to_start_last = pe;
2015 pe_to_start_now = pe;
2019 up_write(&snap->lock);
2021 if (pe_to_start_now) {
2022 start_copy(pe_to_start_now);
2023 pe_to_start_now = NULL;
2028 * Submit the exception against which the bio is queued last,
2029 * to give the other exceptions a head start.
2031 if (pe_to_start_last)
2032 start_copy(pe_to_start_last);
2038 * Called on a write from the origin driver.
2040 static int do_origin(struct dm_dev *origin, struct bio *bio)
2043 int r = DM_MAPIO_REMAPPED;
2045 down_read(&_origins_lock);
2046 o = __lookup_origin(origin->bdev);
2048 r = __origin_write(&o->snapshots, bio->bi_sector, bio);
2049 up_read(&_origins_lock);
2055 * Trigger exceptions in all non-merging snapshots.
2057 * The chunk size of the merging snapshot may be larger than the chunk
2058 * size of some other snapshot so we may need to reallocate multiple
2059 * chunks in other snapshots.
2061 * We scan all the overlapping exceptions in the other snapshots.
2062 * Returns 1 if anything was reallocated and must be waited for,
2063 * otherwise returns 0.
2065 * size must be a multiple of merging_snap's chunk_size.
2067 static int origin_write_extent(struct dm_snapshot *merging_snap,
2068 sector_t sector, unsigned size)
2075 * The origin's __minimum_chunk_size() got stored in split_io
2076 * by snapshot_merge_resume().
2078 down_read(&_origins_lock);
2079 o = __lookup_origin(merging_snap->origin->bdev);
2080 for (n = 0; n < size; n += merging_snap->ti->split_io)
2081 if (__origin_write(&o->snapshots, sector + n, NULL) ==
2084 up_read(&_origins_lock);
2090 * Origin: maps a linear range of a device, with hooks for snapshotting.
2094 * Construct an origin mapping: <dev_path>
2095 * The context for an origin is merely a 'struct dm_dev *'
2096 * pointing to the real device.
2098 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2104 ti->error = "origin: incorrect number of arguments";
2108 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev);
2110 ti->error = "Cannot get target device";
2115 ti->num_flush_requests = 1;
2120 static void origin_dtr(struct dm_target *ti)
2122 struct dm_dev *dev = ti->private;
2123 dm_put_device(ti, dev);
2126 static int origin_map(struct dm_target *ti, struct bio *bio,
2127 union map_info *map_context)
2129 struct dm_dev *dev = ti->private;
2130 bio->bi_bdev = dev->bdev;
2132 if (bio->bi_rw & REQ_FLUSH)
2133 return DM_MAPIO_REMAPPED;
2135 /* Only tell snapshots if this is a write */
2136 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
2140 * Set the target "split_io" field to the minimum of all the snapshots'
2143 static void origin_resume(struct dm_target *ti)
2145 struct dm_dev *dev = ti->private;
2147 ti->split_io = get_origin_minimum_chunksize(dev->bdev);
2150 static void origin_status(struct dm_target *ti, status_type_t type,
2151 char *result, unsigned maxlen)
2153 struct dm_dev *dev = ti->private;
2156 case STATUSTYPE_INFO:
2160 case STATUSTYPE_TABLE:
2161 snprintf(result, maxlen, "%s", dev->name);
2166 static int origin_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2167 struct bio_vec *biovec, int max_size)
2169 struct dm_dev *dev = ti->private;
2170 struct request_queue *q = bdev_get_queue(dev->bdev);
2172 if (!q->merge_bvec_fn)
2175 bvm->bi_bdev = dev->bdev;
2176 bvm->bi_sector = bvm->bi_sector;
2178 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2181 static int origin_iterate_devices(struct dm_target *ti,
2182 iterate_devices_callout_fn fn, void *data)
2184 struct dm_dev *dev = ti->private;
2186 return fn(ti, dev, 0, ti->len, data);
2189 static struct target_type origin_target = {
2190 .name = "snapshot-origin",
2191 .version = {1, 7, 2},
2192 .module = THIS_MODULE,
2196 .resume = origin_resume,
2197 .status = origin_status,
2198 .merge = origin_merge,
2199 .iterate_devices = origin_iterate_devices,
2202 static struct target_type snapshot_target = {
2204 .version = {1, 10, 1},
2205 .module = THIS_MODULE,
2206 .ctr = snapshot_ctr,
2207 .dtr = snapshot_dtr,
2208 .map = snapshot_map,
2209 .end_io = snapshot_end_io,
2210 .preresume = snapshot_preresume,
2211 .resume = snapshot_resume,
2212 .status = snapshot_status,
2213 .iterate_devices = snapshot_iterate_devices,
2216 static struct target_type merge_target = {
2217 .name = dm_snapshot_merge_target_name,
2218 .version = {1, 1, 0},
2219 .module = THIS_MODULE,
2220 .ctr = snapshot_ctr,
2221 .dtr = snapshot_dtr,
2222 .map = snapshot_merge_map,
2223 .end_io = snapshot_end_io,
2224 .presuspend = snapshot_merge_presuspend,
2225 .preresume = snapshot_preresume,
2226 .resume = snapshot_merge_resume,
2227 .status = snapshot_status,
2228 .iterate_devices = snapshot_iterate_devices,
2231 static int __init dm_snapshot_init(void)
2235 r = dm_exception_store_init();
2237 DMERR("Failed to initialize exception stores");
2241 r = dm_register_target(&snapshot_target);
2243 DMERR("snapshot target register failed %d", r);
2244 goto bad_register_snapshot_target;
2247 r = dm_register_target(&origin_target);
2249 DMERR("Origin target register failed %d", r);
2250 goto bad_register_origin_target;
2253 r = dm_register_target(&merge_target);
2255 DMERR("Merge target register failed %d", r);
2256 goto bad_register_merge_target;
2259 r = init_origin_hash();
2261 DMERR("init_origin_hash failed.");
2262 goto bad_origin_hash;
2265 exception_cache = KMEM_CACHE(dm_exception, 0);
2266 if (!exception_cache) {
2267 DMERR("Couldn't create exception cache.");
2269 goto bad_exception_cache;
2272 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
2273 if (!pending_cache) {
2274 DMERR("Couldn't create pending cache.");
2276 goto bad_pending_cache;
2279 tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0);
2280 if (!tracked_chunk_cache) {
2281 DMERR("Couldn't create cache to track chunks in use.");
2283 goto bad_tracked_chunk_cache;
2288 bad_tracked_chunk_cache:
2289 kmem_cache_destroy(pending_cache);
2291 kmem_cache_destroy(exception_cache);
2292 bad_exception_cache:
2295 dm_unregister_target(&merge_target);
2296 bad_register_merge_target:
2297 dm_unregister_target(&origin_target);
2298 bad_register_origin_target:
2299 dm_unregister_target(&snapshot_target);
2300 bad_register_snapshot_target:
2301 dm_exception_store_exit();
2306 static void __exit dm_snapshot_exit(void)
2308 dm_unregister_target(&snapshot_target);
2309 dm_unregister_target(&origin_target);
2310 dm_unregister_target(&merge_target);
2313 kmem_cache_destroy(pending_cache);
2314 kmem_cache_destroy(exception_cache);
2315 kmem_cache_destroy(tracked_chunk_cache);
2317 dm_exception_store_exit();
2321 module_init(dm_snapshot_init);
2322 module_exit(dm_snapshot_exit);
2324 MODULE_DESCRIPTION(DM_NAME " snapshot target");
2325 MODULE_AUTHOR("Joe Thornber");
2326 MODULE_LICENSE("GPL");
2327 MODULE_ALIAS("dm-snapshot-origin");
2328 MODULE_ALIAS("dm-snapshot-merge");