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
176 * Hash table mapping origin volumes to lists of snapshots and
177 * a lock to protect it
179 static struct kmem_cache *exception_cache;
180 static struct kmem_cache *pending_cache;
182 struct dm_snap_tracked_chunk {
183 struct hlist_node node;
187 static struct kmem_cache *tracked_chunk_cache;
189 static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s,
192 struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool,
198 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
199 hlist_add_head(&c->node,
200 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
201 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
206 static void stop_tracking_chunk(struct dm_snapshot *s,
207 struct dm_snap_tracked_chunk *c)
211 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
213 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
215 mempool_free(c, s->tracked_chunk_pool);
218 static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
220 struct dm_snap_tracked_chunk *c;
221 struct hlist_node *hn;
224 spin_lock_irq(&s->tracked_chunk_lock);
226 hlist_for_each_entry(c, hn,
227 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
228 if (c->chunk == chunk) {
234 spin_unlock_irq(&s->tracked_chunk_lock);
240 * This conflicting I/O is extremely improbable in the caller,
241 * so msleep(1) is sufficient and there is no need for a wait queue.
243 static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
245 while (__chunk_is_tracked(s, chunk))
250 * One of these per registered origin, held in the snapshot_origins hash
253 /* The origin device */
254 struct block_device *bdev;
256 struct list_head hash_list;
258 /* List of snapshots for this origin */
259 struct list_head snapshots;
263 * Size of the hash table for origin volumes. If we make this
264 * the size of the minors list then it should be nearly perfect
266 #define ORIGIN_HASH_SIZE 256
267 #define ORIGIN_MASK 0xFF
268 static struct list_head *_origins;
269 static struct rw_semaphore _origins_lock;
271 static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
272 static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
273 static uint64_t _pending_exceptions_done_count;
275 static int init_origin_hash(void)
279 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
282 DMERR("unable to allocate memory");
286 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
287 INIT_LIST_HEAD(_origins + i);
288 init_rwsem(&_origins_lock);
293 static void exit_origin_hash(void)
298 static unsigned origin_hash(struct block_device *bdev)
300 return bdev->bd_dev & ORIGIN_MASK;
303 static struct origin *__lookup_origin(struct block_device *origin)
305 struct list_head *ol;
308 ol = &_origins[origin_hash(origin)];
309 list_for_each_entry (o, ol, hash_list)
310 if (bdev_equal(o->bdev, origin))
316 static void __insert_origin(struct origin *o)
318 struct list_head *sl = &_origins[origin_hash(o->bdev)];
319 list_add_tail(&o->hash_list, sl);
323 * _origins_lock must be held when calling this function.
324 * Returns number of snapshots registered using the supplied cow device, plus:
325 * snap_src - a snapshot suitable for use as a source of exception handover
326 * snap_dest - a snapshot capable of receiving exception handover.
327 * snap_merge - an existing snapshot-merge target linked to the same origin.
328 * There can be at most one snapshot-merge target. The parameter is optional.
330 * Possible return values and states of snap_src and snap_dest.
331 * 0: NULL, NULL - first new snapshot
332 * 1: snap_src, NULL - normal snapshot
333 * 2: snap_src, snap_dest - waiting for handover
334 * 2: snap_src, NULL - handed over, waiting for old to be deleted
335 * 1: NULL, snap_dest - source got destroyed without handover
337 static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
338 struct dm_snapshot **snap_src,
339 struct dm_snapshot **snap_dest,
340 struct dm_snapshot **snap_merge)
342 struct dm_snapshot *s;
347 o = __lookup_origin(snap->origin->bdev);
351 list_for_each_entry(s, &o->snapshots, list) {
352 if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
354 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
364 } else if (snap_dest)
375 * On success, returns 1 if this snapshot is a handover destination,
376 * otherwise returns 0.
378 static int __validate_exception_handover(struct dm_snapshot *snap)
380 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
381 struct dm_snapshot *snap_merge = NULL;
383 /* Does snapshot need exceptions handed over to it? */
384 if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
385 &snap_merge) == 2) ||
387 snap->ti->error = "Snapshot cow pairing for exception "
388 "table handover failed";
393 * If no snap_src was found, snap cannot become a handover
400 * Non-snapshot-merge handover?
402 if (!dm_target_is_snapshot_merge(snap->ti))
406 * Do not allow more than one merging snapshot.
409 snap->ti->error = "A snapshot is already merging.";
413 if (!snap_src->store->type->prepare_merge ||
414 !snap_src->store->type->commit_merge) {
415 snap->ti->error = "Snapshot exception store does not "
416 "support snapshot-merge.";
423 static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
425 struct dm_snapshot *l;
427 /* Sort the list according to chunk size, largest-first smallest-last */
428 list_for_each_entry(l, &o->snapshots, list)
429 if (l->store->chunk_size < s->store->chunk_size)
431 list_add_tail(&s->list, &l->list);
435 * Make a note of the snapshot and its origin so we can look it
436 * up when the origin has a write on it.
438 * Also validate snapshot exception store handovers.
439 * On success, returns 1 if this registration is a handover destination,
440 * otherwise returns 0.
442 static int register_snapshot(struct dm_snapshot *snap)
444 struct origin *o, *new_o = NULL;
445 struct block_device *bdev = snap->origin->bdev;
448 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
452 down_write(&_origins_lock);
454 r = __validate_exception_handover(snap);
460 o = __lookup_origin(bdev);
467 /* Initialise the struct */
468 INIT_LIST_HEAD(&o->snapshots);
474 __insert_snapshot(o, snap);
477 up_write(&_origins_lock);
483 * Move snapshot to correct place in list according to chunk size.
485 static void reregister_snapshot(struct dm_snapshot *s)
487 struct block_device *bdev = s->origin->bdev;
489 down_write(&_origins_lock);
492 __insert_snapshot(__lookup_origin(bdev), s);
494 up_write(&_origins_lock);
497 static void unregister_snapshot(struct dm_snapshot *s)
501 down_write(&_origins_lock);
502 o = __lookup_origin(s->origin->bdev);
505 if (o && list_empty(&o->snapshots)) {
506 list_del(&o->hash_list);
510 up_write(&_origins_lock);
514 * Implementation of the exception hash tables.
515 * The lowest hash_shift bits of the chunk number are ignored, allowing
516 * some consecutive chunks to be grouped together.
518 static int dm_exception_table_init(struct dm_exception_table *et,
519 uint32_t size, unsigned hash_shift)
523 et->hash_shift = hash_shift;
524 et->hash_mask = size - 1;
525 et->table = dm_vcalloc(size, sizeof(struct list_head));
529 for (i = 0; i < size; i++)
530 INIT_LIST_HEAD(et->table + i);
535 static void dm_exception_table_exit(struct dm_exception_table *et,
536 struct kmem_cache *mem)
538 struct list_head *slot;
539 struct dm_exception *ex, *next;
542 size = et->hash_mask + 1;
543 for (i = 0; i < size; i++) {
544 slot = et->table + i;
546 list_for_each_entry_safe (ex, next, slot, hash_list)
547 kmem_cache_free(mem, ex);
553 static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
555 return (chunk >> et->hash_shift) & et->hash_mask;
558 static void dm_remove_exception(struct dm_exception *e)
560 list_del(&e->hash_list);
564 * Return the exception data for a sector, or NULL if not
567 static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
570 struct list_head *slot;
571 struct dm_exception *e;
573 slot = &et->table[exception_hash(et, chunk)];
574 list_for_each_entry (e, slot, hash_list)
575 if (chunk >= e->old_chunk &&
576 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
582 static struct dm_exception *alloc_completed_exception(void)
584 struct dm_exception *e;
586 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
588 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
593 static void free_completed_exception(struct dm_exception *e)
595 kmem_cache_free(exception_cache, e);
598 static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
600 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
603 atomic_inc(&s->pending_exceptions_count);
609 static void free_pending_exception(struct dm_snap_pending_exception *pe)
611 struct dm_snapshot *s = pe->snap;
613 mempool_free(pe, s->pending_pool);
614 smp_mb__before_atomic_dec();
615 atomic_dec(&s->pending_exceptions_count);
618 static void dm_insert_exception(struct dm_exception_table *eh,
619 struct dm_exception *new_e)
622 struct dm_exception *e = NULL;
624 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
626 /* Add immediately if this table doesn't support consecutive chunks */
630 /* List is ordered by old_chunk */
631 list_for_each_entry_reverse(e, l, hash_list) {
632 /* Insert after an existing chunk? */
633 if (new_e->old_chunk == (e->old_chunk +
634 dm_consecutive_chunk_count(e) + 1) &&
635 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
636 dm_consecutive_chunk_count(e) + 1)) {
637 dm_consecutive_chunk_count_inc(e);
638 free_completed_exception(new_e);
642 /* Insert before an existing chunk? */
643 if (new_e->old_chunk == (e->old_chunk - 1) &&
644 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
645 dm_consecutive_chunk_count_inc(e);
648 free_completed_exception(new_e);
652 if (new_e->old_chunk > e->old_chunk)
657 list_add(&new_e->hash_list, e ? &e->hash_list : l);
661 * Callback used by the exception stores to load exceptions when
664 static int dm_add_exception(void *context, chunk_t old, chunk_t new)
666 struct dm_snapshot *s = context;
667 struct dm_exception *e;
669 e = alloc_completed_exception();
675 /* Consecutive_count is implicitly initialised to zero */
678 dm_insert_exception(&s->complete, e);
684 * Return a minimum chunk size of all snapshots that have the specified origin.
685 * Return zero if the origin has no snapshots.
687 static sector_t __minimum_chunk_size(struct origin *o)
689 struct dm_snapshot *snap;
690 unsigned chunk_size = 0;
693 list_for_each_entry(snap, &o->snapshots, list)
694 chunk_size = min_not_zero(chunk_size,
695 snap->store->chunk_size);
703 static int calc_max_buckets(void)
705 /* use a fixed size of 2MB */
706 unsigned long mem = 2 * 1024 * 1024;
707 mem /= sizeof(struct list_head);
713 * Allocate room for a suitable hash table.
715 static int init_hash_tables(struct dm_snapshot *s)
717 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
720 * Calculate based on the size of the original volume or
723 cow_dev_size = get_dev_size(s->cow->bdev);
724 origin_dev_size = get_dev_size(s->origin->bdev);
725 max_buckets = calc_max_buckets();
727 hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift;
728 hash_size = min(hash_size, max_buckets);
732 hash_size = rounddown_pow_of_two(hash_size);
733 if (dm_exception_table_init(&s->complete, hash_size,
734 DM_CHUNK_CONSECUTIVE_BITS))
738 * Allocate hash table for in-flight exceptions
739 * Make this smaller than the real hash table
745 if (dm_exception_table_init(&s->pending, hash_size, 0)) {
746 dm_exception_table_exit(&s->complete, exception_cache);
753 static void merge_shutdown(struct dm_snapshot *s)
755 clear_bit_unlock(RUNNING_MERGE, &s->state_bits);
756 smp_mb__after_clear_bit();
757 wake_up_bit(&s->state_bits, RUNNING_MERGE);
760 static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s)
762 s->first_merging_chunk = 0;
763 s->num_merging_chunks = 0;
765 return bio_list_get(&s->bios_queued_during_merge);
769 * Remove one chunk from the index of completed exceptions.
771 static int __remove_single_exception_chunk(struct dm_snapshot *s,
774 struct dm_exception *e;
776 e = dm_lookup_exception(&s->complete, old_chunk);
778 DMERR("Corruption detected: exception for block %llu is "
779 "on disk but not in memory",
780 (unsigned long long)old_chunk);
785 * If this is the only chunk using this exception, remove exception.
787 if (!dm_consecutive_chunk_count(e)) {
788 dm_remove_exception(e);
789 free_completed_exception(e);
794 * The chunk may be either at the beginning or the end of a
795 * group of consecutive chunks - never in the middle. We are
796 * removing chunks in the opposite order to that in which they
797 * were added, so this should always be true.
798 * Decrement the consecutive chunk counter and adjust the
799 * starting point if necessary.
801 if (old_chunk == e->old_chunk) {
804 } else if (old_chunk != e->old_chunk +
805 dm_consecutive_chunk_count(e)) {
806 DMERR("Attempt to merge block %llu from the "
807 "middle of a chunk range [%llu - %llu]",
808 (unsigned long long)old_chunk,
809 (unsigned long long)e->old_chunk,
811 e->old_chunk + dm_consecutive_chunk_count(e));
815 dm_consecutive_chunk_count_dec(e);
820 static void flush_bios(struct bio *bio);
822 static int remove_single_exception_chunk(struct dm_snapshot *s)
824 struct bio *b = NULL;
826 chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1;
828 down_write(&s->lock);
831 * Process chunks (and associated exceptions) in reverse order
832 * so that dm_consecutive_chunk_count_dec() accounting works.
835 r = __remove_single_exception_chunk(s, old_chunk);
838 } while (old_chunk-- > s->first_merging_chunk);
840 b = __release_queued_bios_after_merge(s);
850 static int origin_write_extent(struct dm_snapshot *merging_snap,
851 sector_t sector, unsigned chunk_size);
853 static void merge_callback(int read_err, unsigned long write_err,
856 static uint64_t read_pending_exceptions_done_count(void)
858 uint64_t pending_exceptions_done;
860 spin_lock(&_pending_exceptions_done_spinlock);
861 pending_exceptions_done = _pending_exceptions_done_count;
862 spin_unlock(&_pending_exceptions_done_spinlock);
864 return pending_exceptions_done;
867 static void increment_pending_exceptions_done_count(void)
869 spin_lock(&_pending_exceptions_done_spinlock);
870 _pending_exceptions_done_count++;
871 spin_unlock(&_pending_exceptions_done_spinlock);
873 wake_up_all(&_pending_exceptions_done);
876 static void snapshot_merge_next_chunks(struct dm_snapshot *s)
878 int i, linear_chunks;
879 chunk_t old_chunk, new_chunk;
880 struct dm_io_region src, dest;
882 uint64_t previous_count;
884 BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
885 if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
889 * valid flag never changes during merge, so no lock required.
892 DMERR("Snapshot is invalid: can't merge");
896 linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
898 if (linear_chunks <= 0) {
899 if (linear_chunks < 0) {
900 DMERR("Read error in exception store: "
901 "shutting down merge");
902 down_write(&s->lock);
909 /* Adjust old_chunk and new_chunk to reflect start of linear region */
910 old_chunk = old_chunk + 1 - linear_chunks;
911 new_chunk = new_chunk + 1 - linear_chunks;
914 * Use one (potentially large) I/O to copy all 'linear_chunks'
915 * from the exception store to the origin
917 io_size = linear_chunks * s->store->chunk_size;
919 dest.bdev = s->origin->bdev;
920 dest.sector = chunk_to_sector(s->store, old_chunk);
921 dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector);
923 src.bdev = s->cow->bdev;
924 src.sector = chunk_to_sector(s->store, new_chunk);
925 src.count = dest.count;
928 * Reallocate any exceptions needed in other snapshots then
929 * wait for the pending exceptions to complete.
930 * Each time any pending exception (globally on the system)
931 * completes we are woken and repeat the process to find out
932 * if we can proceed. While this may not seem a particularly
933 * efficient algorithm, it is not expected to have any
934 * significant impact on performance.
936 previous_count = read_pending_exceptions_done_count();
937 while (origin_write_extent(s, dest.sector, io_size)) {
938 wait_event(_pending_exceptions_done,
939 (read_pending_exceptions_done_count() !=
941 /* Retry after the wait, until all exceptions are done. */
942 previous_count = read_pending_exceptions_done_count();
945 down_write(&s->lock);
946 s->first_merging_chunk = old_chunk;
947 s->num_merging_chunks = linear_chunks;
950 /* Wait until writes to all 'linear_chunks' drain */
951 for (i = 0; i < linear_chunks; i++)
952 __check_for_conflicting_io(s, old_chunk + i);
954 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s);
961 static void error_bios(struct bio *bio);
963 static void merge_callback(int read_err, unsigned long write_err, void *context)
965 struct dm_snapshot *s = context;
966 struct bio *b = NULL;
968 if (read_err || write_err) {
970 DMERR("Read error: shutting down merge.");
972 DMERR("Write error: shutting down merge.");
976 if (s->store->type->commit_merge(s->store,
977 s->num_merging_chunks) < 0) {
978 DMERR("Write error in exception store: shutting down merge");
982 if (remove_single_exception_chunk(s) < 0)
985 snapshot_merge_next_chunks(s);
990 down_write(&s->lock);
992 b = __release_queued_bios_after_merge(s);
999 static void start_merge(struct dm_snapshot *s)
1001 if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits))
1002 snapshot_merge_next_chunks(s);
1005 static int wait_schedule(void *ptr)
1013 * Stop the merging process and wait until it finishes.
1015 static void stop_merge(struct dm_snapshot *s)
1017 set_bit(SHUTDOWN_MERGE, &s->state_bits);
1018 wait_on_bit(&s->state_bits, RUNNING_MERGE, wait_schedule,
1019 TASK_UNINTERRUPTIBLE);
1020 clear_bit(SHUTDOWN_MERGE, &s->state_bits);
1024 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
1026 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1028 struct dm_snapshot *s;
1031 char *origin_path, *cow_path;
1032 unsigned args_used, num_flush_requests = 1;
1033 fmode_t origin_mode = FMODE_READ;
1036 ti->error = "requires exactly 4 arguments";
1041 if (dm_target_is_snapshot_merge(ti)) {
1042 num_flush_requests = 2;
1043 origin_mode = FMODE_WRITE;
1046 s = kmalloc(sizeof(*s), GFP_KERNEL);
1048 ti->error = "Cannot allocate private snapshot structure";
1053 origin_path = argv[0];
1057 r = dm_get_device(ti, origin_path, origin_mode, &s->origin);
1059 ti->error = "Cannot get origin device";
1067 r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow);
1069 ti->error = "Cannot get COW device";
1073 r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
1075 ti->error = "Couldn't create exception store";
1086 atomic_set(&s->pending_exceptions_count, 0);
1087 init_rwsem(&s->lock);
1088 INIT_LIST_HEAD(&s->list);
1089 spin_lock_init(&s->pe_lock);
1091 s->merge_failed = 0;
1092 s->first_merging_chunk = 0;
1093 s->num_merging_chunks = 0;
1094 bio_list_init(&s->bios_queued_during_merge);
1096 /* Allocate hash table for COW data */
1097 if (init_hash_tables(s)) {
1098 ti->error = "Unable to allocate hash table space";
1100 goto bad_hash_tables;
1103 s->kcopyd_client = dm_kcopyd_client_create();
1104 if (IS_ERR(s->kcopyd_client)) {
1105 r = PTR_ERR(s->kcopyd_client);
1106 ti->error = "Could not create kcopyd client";
1110 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
1111 if (!s->pending_pool) {
1112 ti->error = "Could not allocate mempool for pending exceptions";
1113 goto bad_pending_pool;
1116 s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS,
1117 tracked_chunk_cache);
1118 if (!s->tracked_chunk_pool) {
1119 ti->error = "Could not allocate tracked_chunk mempool for "
1121 goto bad_tracked_chunk_pool;
1124 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1125 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
1127 spin_lock_init(&s->tracked_chunk_lock);
1130 ti->num_flush_requests = num_flush_requests;
1132 /* Add snapshot to the list of snapshots for this origin */
1133 /* Exceptions aren't triggered till snapshot_resume() is called */
1134 r = register_snapshot(s);
1136 ti->error = "Snapshot origin struct allocation failed";
1137 goto bad_load_and_register;
1139 /* invalid handover, register_snapshot has set ti->error */
1140 goto bad_load_and_register;
1144 * Metadata must only be loaded into one table at once, so skip this
1145 * if metadata will be handed over during resume.
1146 * Chunk size will be set during the handover - set it to zero to
1147 * ensure it's ignored.
1150 s->store->chunk_size = 0;
1154 r = s->store->type->read_metadata(s->store, dm_add_exception,
1157 ti->error = "Failed to read snapshot metadata";
1158 goto bad_read_metadata;
1161 DMWARN("Snapshot is marked invalid.");
1164 if (!s->store->chunk_size) {
1165 ti->error = "Chunk size not set";
1166 goto bad_read_metadata;
1168 ti->split_io = s->store->chunk_size;
1173 unregister_snapshot(s);
1175 bad_load_and_register:
1176 mempool_destroy(s->tracked_chunk_pool);
1178 bad_tracked_chunk_pool:
1179 mempool_destroy(s->pending_pool);
1182 dm_kcopyd_client_destroy(s->kcopyd_client);
1185 dm_exception_table_exit(&s->pending, pending_cache);
1186 dm_exception_table_exit(&s->complete, exception_cache);
1189 dm_exception_store_destroy(s->store);
1192 dm_put_device(ti, s->cow);
1195 dm_put_device(ti, s->origin);
1204 static void __free_exceptions(struct dm_snapshot *s)
1206 dm_kcopyd_client_destroy(s->kcopyd_client);
1207 s->kcopyd_client = NULL;
1209 dm_exception_table_exit(&s->pending, pending_cache);
1210 dm_exception_table_exit(&s->complete, exception_cache);
1213 static void __handover_exceptions(struct dm_snapshot *snap_src,
1214 struct dm_snapshot *snap_dest)
1217 struct dm_exception_table table_swap;
1218 struct dm_exception_store *store_swap;
1222 * Swap all snapshot context information between the two instances.
1224 u.table_swap = snap_dest->complete;
1225 snap_dest->complete = snap_src->complete;
1226 snap_src->complete = u.table_swap;
1228 u.store_swap = snap_dest->store;
1229 snap_dest->store = snap_src->store;
1230 snap_src->store = u.store_swap;
1232 snap_dest->store->snap = snap_dest;
1233 snap_src->store->snap = snap_src;
1235 snap_dest->ti->split_io = snap_dest->store->chunk_size;
1236 snap_dest->valid = snap_src->valid;
1239 * Set source invalid to ensure it receives no further I/O.
1241 snap_src->valid = 0;
1244 static void snapshot_dtr(struct dm_target *ti)
1246 #ifdef CONFIG_DM_DEBUG
1249 struct dm_snapshot *s = ti->private;
1250 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1252 down_read(&_origins_lock);
1253 /* Check whether exception handover must be cancelled */
1254 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1255 if (snap_src && snap_dest && (s == snap_src)) {
1256 down_write(&snap_dest->lock);
1257 snap_dest->valid = 0;
1258 up_write(&snap_dest->lock);
1259 DMERR("Cancelling snapshot handover.");
1261 up_read(&_origins_lock);
1263 if (dm_target_is_snapshot_merge(ti))
1266 /* Prevent further origin writes from using this snapshot. */
1267 /* After this returns there can be no new kcopyd jobs. */
1268 unregister_snapshot(s);
1270 while (atomic_read(&s->pending_exceptions_count))
1273 * Ensure instructions in mempool_destroy aren't reordered
1274 * before atomic_read.
1278 #ifdef CONFIG_DM_DEBUG
1279 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1280 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
1283 mempool_destroy(s->tracked_chunk_pool);
1285 __free_exceptions(s);
1287 mempool_destroy(s->pending_pool);
1289 dm_exception_store_destroy(s->store);
1291 dm_put_device(ti, s->cow);
1293 dm_put_device(ti, s->origin);
1299 * Flush a list of buffers.
1301 static void flush_bios(struct bio *bio)
1307 bio->bi_next = NULL;
1308 generic_make_request(bio);
1313 static int do_origin(struct dm_dev *origin, struct bio *bio);
1316 * Flush a list of buffers.
1318 static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
1325 bio->bi_next = NULL;
1326 r = do_origin(s->origin, bio);
1327 if (r == DM_MAPIO_REMAPPED)
1328 generic_make_request(bio);
1334 * Error a list of buffers.
1336 static void error_bios(struct bio *bio)
1342 bio->bi_next = NULL;
1348 static void __invalidate_snapshot(struct dm_snapshot *s, int err)
1354 DMERR("Invalidating snapshot: Error reading/writing.");
1355 else if (err == -ENOMEM)
1356 DMERR("Invalidating snapshot: Unable to allocate exception.");
1358 if (s->store->type->drop_snapshot)
1359 s->store->type->drop_snapshot(s->store);
1363 dm_table_event(s->ti->table);
1366 static void pending_complete(struct dm_snap_pending_exception *pe, int success)
1368 struct dm_exception *e;
1369 struct dm_snapshot *s = pe->snap;
1370 struct bio *origin_bios = NULL;
1371 struct bio *snapshot_bios = NULL;
1375 /* Read/write error - snapshot is unusable */
1376 down_write(&s->lock);
1377 __invalidate_snapshot(s, -EIO);
1382 e = alloc_completed_exception();
1384 down_write(&s->lock);
1385 __invalidate_snapshot(s, -ENOMEM);
1391 down_write(&s->lock);
1393 free_completed_exception(e);
1398 /* Check for conflicting reads */
1399 __check_for_conflicting_io(s, pe->e.old_chunk);
1402 * Add a proper exception, and remove the
1403 * in-flight exception from the list.
1405 dm_insert_exception(&s->complete, e);
1408 dm_remove_exception(&pe->e);
1409 snapshot_bios = bio_list_get(&pe->snapshot_bios);
1410 origin_bios = bio_list_get(&pe->origin_bios);
1411 free_pending_exception(pe);
1413 increment_pending_exceptions_done_count();
1417 /* Submit any pending write bios */
1419 error_bios(snapshot_bios);
1421 flush_bios(snapshot_bios);
1423 retry_origin_bios(s, origin_bios);
1426 static void commit_callback(void *context, int success)
1428 struct dm_snap_pending_exception *pe = context;
1430 pending_complete(pe, success);
1434 * Called when the copy I/O has finished. kcopyd actually runs
1435 * this code so don't block.
1437 static void copy_callback(int read_err, unsigned long write_err, void *context)
1439 struct dm_snap_pending_exception *pe = context;
1440 struct dm_snapshot *s = pe->snap;
1442 if (read_err || write_err)
1443 pending_complete(pe, 0);
1446 /* Update the metadata if we are persistent */
1447 s->store->type->commit_exception(s->store, &pe->e,
1448 commit_callback, pe);
1452 * Dispatches the copy operation to kcopyd.
1454 static void start_copy(struct dm_snap_pending_exception *pe)
1456 struct dm_snapshot *s = pe->snap;
1457 struct dm_io_region src, dest;
1458 struct block_device *bdev = s->origin->bdev;
1461 dev_size = get_dev_size(bdev);
1464 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
1465 src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
1467 dest.bdev = s->cow->bdev;
1468 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
1469 dest.count = src.count;
1471 /* Hand over to kcopyd */
1472 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe);
1475 static struct dm_snap_pending_exception *
1476 __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1478 struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
1483 return container_of(e, struct dm_snap_pending_exception, e);
1487 * Looks to see if this snapshot already has a pending exception
1488 * for this chunk, otherwise it allocates a new one and inserts
1489 * it into the pending table.
1491 * NOTE: a write lock must be held on snap->lock before calling
1494 static struct dm_snap_pending_exception *
1495 __find_pending_exception(struct dm_snapshot *s,
1496 struct dm_snap_pending_exception *pe, chunk_t chunk)
1498 struct dm_snap_pending_exception *pe2;
1500 pe2 = __lookup_pending_exception(s, chunk);
1502 free_pending_exception(pe);
1506 pe->e.old_chunk = chunk;
1507 bio_list_init(&pe->origin_bios);
1508 bio_list_init(&pe->snapshot_bios);
1511 if (s->store->type->prepare_exception(s->store, &pe->e)) {
1512 free_pending_exception(pe);
1516 dm_insert_exception(&s->pending, &pe->e);
1521 static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
1522 struct bio *bio, chunk_t chunk)
1524 bio->bi_bdev = s->cow->bdev;
1525 bio->bi_sector = chunk_to_sector(s->store,
1526 dm_chunk_number(e->new_chunk) +
1527 (chunk - e->old_chunk)) +
1529 s->store->chunk_mask);
1532 static int snapshot_map(struct dm_target *ti, struct bio *bio,
1533 union map_info *map_context)
1535 struct dm_exception *e;
1536 struct dm_snapshot *s = ti->private;
1537 int r = DM_MAPIO_REMAPPED;
1539 struct dm_snap_pending_exception *pe = NULL;
1541 if (bio->bi_rw & REQ_FLUSH) {
1542 bio->bi_bdev = s->cow->bdev;
1543 return DM_MAPIO_REMAPPED;
1546 chunk = sector_to_chunk(s->store, bio->bi_sector);
1548 /* Full snapshots are not usable */
1549 /* To get here the table must be live so s->active is always set. */
1553 /* FIXME: should only take write lock if we need
1554 * to copy an exception */
1555 down_write(&s->lock);
1562 /* If the block is already remapped - use that, else remap it */
1563 e = dm_lookup_exception(&s->complete, chunk);
1565 remap_exception(s, e, bio, chunk);
1570 * Write to snapshot - higher level takes care of RW/RO
1571 * flags so we should only get this if we are
1574 if (bio_rw(bio) == WRITE) {
1575 pe = __lookup_pending_exception(s, chunk);
1578 pe = alloc_pending_exception(s);
1579 down_write(&s->lock);
1582 free_pending_exception(pe);
1587 e = dm_lookup_exception(&s->complete, chunk);
1589 free_pending_exception(pe);
1590 remap_exception(s, e, bio, chunk);
1594 pe = __find_pending_exception(s, pe, chunk);
1596 __invalidate_snapshot(s, -ENOMEM);
1602 remap_exception(s, &pe->e, bio, chunk);
1603 bio_list_add(&pe->snapshot_bios, bio);
1605 r = DM_MAPIO_SUBMITTED;
1608 /* this is protected by snap->lock */
1615 bio->bi_bdev = s->origin->bdev;
1616 map_context->ptr = track_chunk(s, chunk);
1626 * A snapshot-merge target behaves like a combination of a snapshot
1627 * target and a snapshot-origin target. It only generates new
1628 * exceptions in other snapshots and not in the one that is being
1631 * For each chunk, if there is an existing exception, it is used to
1632 * redirect I/O to the cow device. Otherwise I/O is sent to the origin,
1633 * which in turn might generate exceptions in other snapshots.
1634 * If merging is currently taking place on the chunk in question, the
1635 * I/O is deferred by adding it to s->bios_queued_during_merge.
1637 static int snapshot_merge_map(struct dm_target *ti, struct bio *bio,
1638 union map_info *map_context)
1640 struct dm_exception *e;
1641 struct dm_snapshot *s = ti->private;
1642 int r = DM_MAPIO_REMAPPED;
1645 if (bio->bi_rw & REQ_FLUSH) {
1646 if (!map_context->target_request_nr)
1647 bio->bi_bdev = s->origin->bdev;
1649 bio->bi_bdev = s->cow->bdev;
1650 map_context->ptr = NULL;
1651 return DM_MAPIO_REMAPPED;
1654 chunk = sector_to_chunk(s->store, bio->bi_sector);
1656 down_write(&s->lock);
1658 /* Full merging snapshots are redirected to the origin */
1660 goto redirect_to_origin;
1662 /* If the block is already remapped - use that */
1663 e = dm_lookup_exception(&s->complete, chunk);
1665 /* Queue writes overlapping with chunks being merged */
1666 if (bio_rw(bio) == WRITE &&
1667 chunk >= s->first_merging_chunk &&
1668 chunk < (s->first_merging_chunk +
1669 s->num_merging_chunks)) {
1670 bio->bi_bdev = s->origin->bdev;
1671 bio_list_add(&s->bios_queued_during_merge, bio);
1672 r = DM_MAPIO_SUBMITTED;
1676 remap_exception(s, e, bio, chunk);
1678 if (bio_rw(bio) == WRITE)
1679 map_context->ptr = track_chunk(s, chunk);
1684 bio->bi_bdev = s->origin->bdev;
1686 if (bio_rw(bio) == WRITE) {
1688 return do_origin(s->origin, bio);
1697 static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1698 int error, union map_info *map_context)
1700 struct dm_snapshot *s = ti->private;
1701 struct dm_snap_tracked_chunk *c = map_context->ptr;
1704 stop_tracking_chunk(s, c);
1709 static void snapshot_merge_presuspend(struct dm_target *ti)
1711 struct dm_snapshot *s = ti->private;
1716 static int snapshot_preresume(struct dm_target *ti)
1719 struct dm_snapshot *s = ti->private;
1720 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1722 down_read(&_origins_lock);
1723 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1724 if (snap_src && snap_dest) {
1725 down_read(&snap_src->lock);
1726 if (s == snap_src) {
1727 DMERR("Unable to resume snapshot source until "
1728 "handover completes.");
1730 } else if (!dm_suspended(snap_src->ti)) {
1731 DMERR("Unable to perform snapshot handover until "
1732 "source is suspended.");
1735 up_read(&snap_src->lock);
1737 up_read(&_origins_lock);
1742 static void snapshot_resume(struct dm_target *ti)
1744 struct dm_snapshot *s = ti->private;
1745 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1747 down_read(&_origins_lock);
1748 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1749 if (snap_src && snap_dest) {
1750 down_write(&snap_src->lock);
1751 down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1752 __handover_exceptions(snap_src, snap_dest);
1753 up_write(&snap_dest->lock);
1754 up_write(&snap_src->lock);
1756 up_read(&_origins_lock);
1758 /* Now we have correct chunk size, reregister */
1759 reregister_snapshot(s);
1761 down_write(&s->lock);
1766 static sector_t get_origin_minimum_chunksize(struct block_device *bdev)
1768 sector_t min_chunksize;
1770 down_read(&_origins_lock);
1771 min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
1772 up_read(&_origins_lock);
1774 return min_chunksize;
1777 static void snapshot_merge_resume(struct dm_target *ti)
1779 struct dm_snapshot *s = ti->private;
1782 * Handover exceptions from existing snapshot.
1784 snapshot_resume(ti);
1787 * snapshot-merge acts as an origin, so set ti->split_io
1789 ti->split_io = get_origin_minimum_chunksize(s->origin->bdev);
1794 static int snapshot_status(struct dm_target *ti, status_type_t type,
1795 char *result, unsigned int maxlen)
1798 struct dm_snapshot *snap = ti->private;
1801 case STATUSTYPE_INFO:
1803 down_write(&snap->lock);
1807 else if (snap->merge_failed)
1808 DMEMIT("Merge failed");
1810 if (snap->store->type->usage) {
1811 sector_t total_sectors, sectors_allocated,
1813 snap->store->type->usage(snap->store,
1817 DMEMIT("%llu/%llu %llu",
1818 (unsigned long long)sectors_allocated,
1819 (unsigned long long)total_sectors,
1820 (unsigned long long)metadata_sectors);
1826 up_write(&snap->lock);
1830 case STATUSTYPE_TABLE:
1832 * kdevname returns a static pointer so we need
1833 * to make private copies if the output is to
1836 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
1837 snap->store->type->status(snap->store, type, result + sz,
1845 static int snapshot_iterate_devices(struct dm_target *ti,
1846 iterate_devices_callout_fn fn, void *data)
1848 struct dm_snapshot *snap = ti->private;
1851 r = fn(ti, snap->origin, 0, ti->len, data);
1854 r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data);
1860 /*-----------------------------------------------------------------
1862 *---------------------------------------------------------------*/
1865 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
1866 * supplied bio was ignored. The caller may submit it immediately.
1867 * (No remapping actually occurs as the origin is always a direct linear
1870 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
1871 * and any supplied bio is added to a list to be submitted once all
1872 * the necessary exceptions exist.
1874 static int __origin_write(struct list_head *snapshots, sector_t sector,
1877 int r = DM_MAPIO_REMAPPED;
1878 struct dm_snapshot *snap;
1879 struct dm_exception *e;
1880 struct dm_snap_pending_exception *pe;
1881 struct dm_snap_pending_exception *pe_to_start_now = NULL;
1882 struct dm_snap_pending_exception *pe_to_start_last = NULL;
1885 /* Do all the snapshots on this origin */
1886 list_for_each_entry (snap, snapshots, list) {
1888 * Don't make new exceptions in a merging snapshot
1889 * because it has effectively been deleted
1891 if (dm_target_is_snapshot_merge(snap->ti))
1894 down_write(&snap->lock);
1896 /* Only deal with valid and active snapshots */
1897 if (!snap->valid || !snap->active)
1900 /* Nothing to do if writing beyond end of snapshot */
1901 if (sector >= dm_table_get_size(snap->ti->table))
1905 * Remember, different snapshots can have
1906 * different chunk sizes.
1908 chunk = sector_to_chunk(snap->store, sector);
1911 * Check exception table to see if block
1912 * is already remapped in this snapshot
1913 * and trigger an exception if not.
1915 e = dm_lookup_exception(&snap->complete, chunk);
1919 pe = __lookup_pending_exception(snap, chunk);
1921 up_write(&snap->lock);
1922 pe = alloc_pending_exception(snap);
1923 down_write(&snap->lock);
1926 free_pending_exception(pe);
1930 e = dm_lookup_exception(&snap->complete, chunk);
1932 free_pending_exception(pe);
1936 pe = __find_pending_exception(snap, pe, chunk);
1938 __invalidate_snapshot(snap, -ENOMEM);
1943 r = DM_MAPIO_SUBMITTED;
1946 * If an origin bio was supplied, queue it to wait for the
1947 * completion of this exception, and start this one last,
1948 * at the end of the function.
1951 bio_list_add(&pe->origin_bios, bio);
1956 pe_to_start_last = pe;
1962 pe_to_start_now = pe;
1966 up_write(&snap->lock);
1968 if (pe_to_start_now) {
1969 start_copy(pe_to_start_now);
1970 pe_to_start_now = NULL;
1975 * Submit the exception against which the bio is queued last,
1976 * to give the other exceptions a head start.
1978 if (pe_to_start_last)
1979 start_copy(pe_to_start_last);
1985 * Called on a write from the origin driver.
1987 static int do_origin(struct dm_dev *origin, struct bio *bio)
1990 int r = DM_MAPIO_REMAPPED;
1992 down_read(&_origins_lock);
1993 o = __lookup_origin(origin->bdev);
1995 r = __origin_write(&o->snapshots, bio->bi_sector, bio);
1996 up_read(&_origins_lock);
2002 * Trigger exceptions in all non-merging snapshots.
2004 * The chunk size of the merging snapshot may be larger than the chunk
2005 * size of some other snapshot so we may need to reallocate multiple
2006 * chunks in other snapshots.
2008 * We scan all the overlapping exceptions in the other snapshots.
2009 * Returns 1 if anything was reallocated and must be waited for,
2010 * otherwise returns 0.
2012 * size must be a multiple of merging_snap's chunk_size.
2014 static int origin_write_extent(struct dm_snapshot *merging_snap,
2015 sector_t sector, unsigned size)
2022 * The origin's __minimum_chunk_size() got stored in split_io
2023 * by snapshot_merge_resume().
2025 down_read(&_origins_lock);
2026 o = __lookup_origin(merging_snap->origin->bdev);
2027 for (n = 0; n < size; n += merging_snap->ti->split_io)
2028 if (__origin_write(&o->snapshots, sector + n, NULL) ==
2031 up_read(&_origins_lock);
2037 * Origin: maps a linear range of a device, with hooks for snapshotting.
2041 * Construct an origin mapping: <dev_path>
2042 * The context for an origin is merely a 'struct dm_dev *'
2043 * pointing to the real device.
2045 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2051 ti->error = "origin: incorrect number of arguments";
2055 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev);
2057 ti->error = "Cannot get target device";
2062 ti->num_flush_requests = 1;
2067 static void origin_dtr(struct dm_target *ti)
2069 struct dm_dev *dev = ti->private;
2070 dm_put_device(ti, dev);
2073 static int origin_map(struct dm_target *ti, struct bio *bio,
2074 union map_info *map_context)
2076 struct dm_dev *dev = ti->private;
2077 bio->bi_bdev = dev->bdev;
2079 if (bio->bi_rw & REQ_FLUSH)
2080 return DM_MAPIO_REMAPPED;
2082 /* Only tell snapshots if this is a write */
2083 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
2087 * Set the target "split_io" field to the minimum of all the snapshots'
2090 static void origin_resume(struct dm_target *ti)
2092 struct dm_dev *dev = ti->private;
2094 ti->split_io = get_origin_minimum_chunksize(dev->bdev);
2097 static int origin_status(struct dm_target *ti, status_type_t type, char *result,
2098 unsigned int maxlen)
2100 struct dm_dev *dev = ti->private;
2103 case STATUSTYPE_INFO:
2107 case STATUSTYPE_TABLE:
2108 snprintf(result, maxlen, "%s", dev->name);
2115 static int origin_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2116 struct bio_vec *biovec, int max_size)
2118 struct dm_dev *dev = ti->private;
2119 struct request_queue *q = bdev_get_queue(dev->bdev);
2121 if (!q->merge_bvec_fn)
2124 bvm->bi_bdev = dev->bdev;
2125 bvm->bi_sector = bvm->bi_sector;
2127 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2130 static int origin_iterate_devices(struct dm_target *ti,
2131 iterate_devices_callout_fn fn, void *data)
2133 struct dm_dev *dev = ti->private;
2135 return fn(ti, dev, 0, ti->len, data);
2138 static struct target_type origin_target = {
2139 .name = "snapshot-origin",
2140 .version = {1, 7, 1},
2141 .module = THIS_MODULE,
2145 .resume = origin_resume,
2146 .status = origin_status,
2147 .merge = origin_merge,
2148 .iterate_devices = origin_iterate_devices,
2151 static struct target_type snapshot_target = {
2153 .version = {1, 10, 0},
2154 .module = THIS_MODULE,
2155 .ctr = snapshot_ctr,
2156 .dtr = snapshot_dtr,
2157 .map = snapshot_map,
2158 .end_io = snapshot_end_io,
2159 .preresume = snapshot_preresume,
2160 .resume = snapshot_resume,
2161 .status = snapshot_status,
2162 .iterate_devices = snapshot_iterate_devices,
2165 static struct target_type merge_target = {
2166 .name = dm_snapshot_merge_target_name,
2167 .version = {1, 1, 0},
2168 .module = THIS_MODULE,
2169 .ctr = snapshot_ctr,
2170 .dtr = snapshot_dtr,
2171 .map = snapshot_merge_map,
2172 .end_io = snapshot_end_io,
2173 .presuspend = snapshot_merge_presuspend,
2174 .preresume = snapshot_preresume,
2175 .resume = snapshot_merge_resume,
2176 .status = snapshot_status,
2177 .iterate_devices = snapshot_iterate_devices,
2180 static int __init dm_snapshot_init(void)
2184 r = dm_exception_store_init();
2186 DMERR("Failed to initialize exception stores");
2190 r = dm_register_target(&snapshot_target);
2192 DMERR("snapshot target register failed %d", r);
2193 goto bad_register_snapshot_target;
2196 r = dm_register_target(&origin_target);
2198 DMERR("Origin target register failed %d", r);
2199 goto bad_register_origin_target;
2202 r = dm_register_target(&merge_target);
2204 DMERR("Merge target register failed %d", r);
2205 goto bad_register_merge_target;
2208 r = init_origin_hash();
2210 DMERR("init_origin_hash failed.");
2211 goto bad_origin_hash;
2214 exception_cache = KMEM_CACHE(dm_exception, 0);
2215 if (!exception_cache) {
2216 DMERR("Couldn't create exception cache.");
2218 goto bad_exception_cache;
2221 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
2222 if (!pending_cache) {
2223 DMERR("Couldn't create pending cache.");
2225 goto bad_pending_cache;
2228 tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0);
2229 if (!tracked_chunk_cache) {
2230 DMERR("Couldn't create cache to track chunks in use.");
2232 goto bad_tracked_chunk_cache;
2237 bad_tracked_chunk_cache:
2238 kmem_cache_destroy(pending_cache);
2240 kmem_cache_destroy(exception_cache);
2241 bad_exception_cache:
2244 dm_unregister_target(&merge_target);
2245 bad_register_merge_target:
2246 dm_unregister_target(&origin_target);
2247 bad_register_origin_target:
2248 dm_unregister_target(&snapshot_target);
2249 bad_register_snapshot_target:
2250 dm_exception_store_exit();
2255 static void __exit dm_snapshot_exit(void)
2257 dm_unregister_target(&snapshot_target);
2258 dm_unregister_target(&origin_target);
2259 dm_unregister_target(&merge_target);
2262 kmem_cache_destroy(pending_cache);
2263 kmem_cache_destroy(exception_cache);
2264 kmem_cache_destroy(tracked_chunk_cache);
2266 dm_exception_store_exit();
2270 module_init(dm_snapshot_init);
2271 module_exit(dm_snapshot_exit);
2273 MODULE_DESCRIPTION(DM_NAME " snapshot target");
2274 MODULE_AUTHOR("Joe Thornber");
2275 MODULE_LICENSE("GPL");