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>
22 #include <linux/workqueue.h>
24 #include "dm-exception-store.h"
26 #define DM_MSG_PREFIX "snapshots"
28 static const char dm_snapshot_merge_target_name[] = "snapshot-merge";
30 #define dm_target_is_snapshot_merge(ti) \
31 ((ti)->type->name == dm_snapshot_merge_target_name)
34 * The percentage increment we will wake up users at
36 #define WAKE_UP_PERCENT 5
39 * kcopyd priority of snapshot operations
41 #define SNAPSHOT_COPY_PRIORITY 2
44 * Reserve 1MB for each snapshot initially (with minimum of 1 page).
46 #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1)
49 * The size of the mempool used to track chunks in use.
53 #define DM_TRACKED_CHUNK_HASH_SIZE 16
54 #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
55 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
57 struct dm_exception_table {
60 struct list_head *table;
64 struct rw_semaphore lock;
66 struct dm_dev *origin;
71 /* List of snapshots per Origin */
72 struct list_head list;
74 /* You can't use a snapshot if this is 0 (e.g. if full) */
77 /* Origin writes don't trigger exceptions until this is set */
80 /* Whether or not owning mapped_device is suspended */
83 mempool_t *pending_pool;
85 atomic_t pending_exceptions_count;
87 struct dm_exception_table pending;
88 struct dm_exception_table complete;
91 * pe_lock protects all pending_exception operations and access
92 * as well as the snapshot_bios list.
96 /* The on disk metadata handler */
97 struct dm_exception_store *store;
99 struct dm_kcopyd_client *kcopyd_client;
101 /* Queue of snapshot writes for ksnapd to flush */
102 struct bio_list queued_bios;
103 struct work_struct queued_bios_work;
105 /* Chunks with outstanding reads */
106 mempool_t *tracked_chunk_pool;
107 spinlock_t tracked_chunk_lock;
108 struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
110 /* Wait for events based on state_bits */
111 unsigned long state_bits;
113 /* Range of chunks currently being merged. */
114 chunk_t first_merging_chunk;
115 int num_merging_chunks;
118 * Incoming bios that overlap with chunks being merged must wait
119 * for them to be committed.
121 struct bio_list bios_queued_during_merge;
126 * RUNNING_MERGE - Merge operation is in progress.
127 * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped;
128 * cleared afterwards.
130 #define RUNNING_MERGE 0
131 #define SHUTDOWN_MERGE 1
133 struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
137 EXPORT_SYMBOL(dm_snap_cow);
139 static struct workqueue_struct *ksnapd;
140 static void flush_queued_bios(struct work_struct *work);
142 static sector_t chunk_to_sector(struct dm_exception_store *store,
145 return chunk << store->chunk_shift;
148 static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
151 * There is only ever one instance of a particular block
152 * device so we can compare pointers safely.
157 struct dm_snap_pending_exception {
158 struct dm_exception e;
161 * Origin buffers waiting for this to complete are held
164 struct bio_list origin_bios;
165 struct bio_list snapshot_bios;
167 /* Pointer back to snapshot context */
168 struct dm_snapshot *snap;
171 * 1 indicates the exception has already been sent to
178 * Hash table mapping origin volumes to lists of snapshots and
179 * a lock to protect it
181 static struct kmem_cache *exception_cache;
182 static struct kmem_cache *pending_cache;
184 struct dm_snap_tracked_chunk {
185 struct hlist_node node;
189 static struct kmem_cache *tracked_chunk_cache;
191 static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s,
194 struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool,
200 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
201 hlist_add_head(&c->node,
202 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
203 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
208 static void stop_tracking_chunk(struct dm_snapshot *s,
209 struct dm_snap_tracked_chunk *c)
213 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
215 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
217 mempool_free(c, s->tracked_chunk_pool);
220 static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
222 struct dm_snap_tracked_chunk *c;
223 struct hlist_node *hn;
226 spin_lock_irq(&s->tracked_chunk_lock);
228 hlist_for_each_entry(c, hn,
229 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
230 if (c->chunk == chunk) {
236 spin_unlock_irq(&s->tracked_chunk_lock);
242 * This conflicting I/O is extremely improbable in the caller,
243 * so msleep(1) is sufficient and there is no need for a wait queue.
245 static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
247 while (__chunk_is_tracked(s, chunk))
252 * One of these per registered origin, held in the snapshot_origins hash
255 /* The origin device */
256 struct block_device *bdev;
258 struct list_head hash_list;
260 /* List of snapshots for this origin */
261 struct list_head snapshots;
265 * Size of the hash table for origin volumes. If we make this
266 * the size of the minors list then it should be nearly perfect
268 #define ORIGIN_HASH_SIZE 256
269 #define ORIGIN_MASK 0xFF
270 static struct list_head *_origins;
271 static struct rw_semaphore _origins_lock;
273 static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
274 static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
275 static uint64_t _pending_exceptions_done_count;
277 static int init_origin_hash(void)
281 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
284 DMERR("unable to allocate memory");
288 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
289 INIT_LIST_HEAD(_origins + i);
290 init_rwsem(&_origins_lock);
295 static void exit_origin_hash(void)
300 static unsigned origin_hash(struct block_device *bdev)
302 return bdev->bd_dev & ORIGIN_MASK;
305 static struct origin *__lookup_origin(struct block_device *origin)
307 struct list_head *ol;
310 ol = &_origins[origin_hash(origin)];
311 list_for_each_entry (o, ol, hash_list)
312 if (bdev_equal(o->bdev, origin))
318 static void __insert_origin(struct origin *o)
320 struct list_head *sl = &_origins[origin_hash(o->bdev)];
321 list_add_tail(&o->hash_list, sl);
325 * _origins_lock must be held when calling this function.
326 * Returns number of snapshots registered using the supplied cow device, plus:
327 * snap_src - a snapshot suitable for use as a source of exception handover
328 * snap_dest - a snapshot capable of receiving exception handover.
329 * snap_merge - an existing snapshot-merge target linked to the same origin.
330 * There can be at most one snapshot-merge target. The parameter is optional.
332 * Possible return values and states of snap_src and snap_dest.
333 * 0: NULL, NULL - first new snapshot
334 * 1: snap_src, NULL - normal snapshot
335 * 2: snap_src, snap_dest - waiting for handover
336 * 2: snap_src, NULL - handed over, waiting for old to be deleted
337 * 1: NULL, snap_dest - source got destroyed without handover
339 static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
340 struct dm_snapshot **snap_src,
341 struct dm_snapshot **snap_dest,
342 struct dm_snapshot **snap_merge)
344 struct dm_snapshot *s;
349 o = __lookup_origin(snap->origin->bdev);
353 list_for_each_entry(s, &o->snapshots, list) {
354 if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
356 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
366 } else if (snap_dest)
377 * On success, returns 1 if this snapshot is a handover destination,
378 * otherwise returns 0.
380 static int __validate_exception_handover(struct dm_snapshot *snap)
382 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
383 struct dm_snapshot *snap_merge = NULL;
385 /* Does snapshot need exceptions handed over to it? */
386 if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
387 &snap_merge) == 2) ||
389 snap->ti->error = "Snapshot cow pairing for exception "
390 "table handover failed";
395 * If no snap_src was found, snap cannot become a handover
402 * Non-snapshot-merge handover?
404 if (!dm_target_is_snapshot_merge(snap->ti))
408 * Do not allow more than one merging snapshot.
411 snap->ti->error = "A snapshot is already merging.";
415 if (!snap_src->store->type->prepare_merge ||
416 !snap_src->store->type->commit_merge) {
417 snap->ti->error = "Snapshot exception store does not "
418 "support snapshot-merge.";
425 static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
427 struct dm_snapshot *l;
429 /* Sort the list according to chunk size, largest-first smallest-last */
430 list_for_each_entry(l, &o->snapshots, list)
431 if (l->store->chunk_size < s->store->chunk_size)
433 list_add_tail(&s->list, &l->list);
437 * Make a note of the snapshot and its origin so we can look it
438 * up when the origin has a write on it.
440 * Also validate snapshot exception store handovers.
441 * On success, returns 1 if this registration is a handover destination,
442 * otherwise returns 0.
444 static int register_snapshot(struct dm_snapshot *snap)
446 struct origin *o, *new_o = NULL;
447 struct block_device *bdev = snap->origin->bdev;
450 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
454 down_write(&_origins_lock);
456 r = __validate_exception_handover(snap);
462 o = __lookup_origin(bdev);
469 /* Initialise the struct */
470 INIT_LIST_HEAD(&o->snapshots);
476 __insert_snapshot(o, snap);
479 up_write(&_origins_lock);
485 * Move snapshot to correct place in list according to chunk size.
487 static void reregister_snapshot(struct dm_snapshot *s)
489 struct block_device *bdev = s->origin->bdev;
491 down_write(&_origins_lock);
494 __insert_snapshot(__lookup_origin(bdev), s);
496 up_write(&_origins_lock);
499 static void unregister_snapshot(struct dm_snapshot *s)
503 down_write(&_origins_lock);
504 o = __lookup_origin(s->origin->bdev);
507 if (o && list_empty(&o->snapshots)) {
508 list_del(&o->hash_list);
512 up_write(&_origins_lock);
516 * Implementation of the exception hash tables.
517 * The lowest hash_shift bits of the chunk number are ignored, allowing
518 * some consecutive chunks to be grouped together.
520 static int dm_exception_table_init(struct dm_exception_table *et,
521 uint32_t size, unsigned hash_shift)
525 et->hash_shift = hash_shift;
526 et->hash_mask = size - 1;
527 et->table = dm_vcalloc(size, sizeof(struct list_head));
531 for (i = 0; i < size; i++)
532 INIT_LIST_HEAD(et->table + i);
537 static void dm_exception_table_exit(struct dm_exception_table *et,
538 struct kmem_cache *mem)
540 struct list_head *slot;
541 struct dm_exception *ex, *next;
544 size = et->hash_mask + 1;
545 for (i = 0; i < size; i++) {
546 slot = et->table + i;
548 list_for_each_entry_safe (ex, next, slot, hash_list)
549 kmem_cache_free(mem, ex);
555 static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
557 return (chunk >> et->hash_shift) & et->hash_mask;
560 static void dm_remove_exception(struct dm_exception *e)
562 list_del(&e->hash_list);
566 * Return the exception data for a sector, or NULL if not
569 static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
572 struct list_head *slot;
573 struct dm_exception *e;
575 slot = &et->table[exception_hash(et, chunk)];
576 list_for_each_entry (e, slot, hash_list)
577 if (chunk >= e->old_chunk &&
578 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
584 static struct dm_exception *alloc_completed_exception(void)
586 struct dm_exception *e;
588 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
590 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
595 static void free_completed_exception(struct dm_exception *e)
597 kmem_cache_free(exception_cache, e);
600 static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
602 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
605 atomic_inc(&s->pending_exceptions_count);
611 static void free_pending_exception(struct dm_snap_pending_exception *pe)
613 struct dm_snapshot *s = pe->snap;
615 mempool_free(pe, s->pending_pool);
616 smp_mb__before_atomic_dec();
617 atomic_dec(&s->pending_exceptions_count);
620 static void dm_insert_exception(struct dm_exception_table *eh,
621 struct dm_exception *new_e)
624 struct dm_exception *e = NULL;
626 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
628 /* Add immediately if this table doesn't support consecutive chunks */
632 /* List is ordered by old_chunk */
633 list_for_each_entry_reverse(e, l, hash_list) {
634 /* Insert after an existing chunk? */
635 if (new_e->old_chunk == (e->old_chunk +
636 dm_consecutive_chunk_count(e) + 1) &&
637 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
638 dm_consecutive_chunk_count(e) + 1)) {
639 dm_consecutive_chunk_count_inc(e);
640 free_completed_exception(new_e);
644 /* Insert before an existing chunk? */
645 if (new_e->old_chunk == (e->old_chunk - 1) &&
646 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
647 dm_consecutive_chunk_count_inc(e);
650 free_completed_exception(new_e);
654 if (new_e->old_chunk > e->old_chunk)
659 list_add(&new_e->hash_list, e ? &e->hash_list : l);
663 * Callback used by the exception stores to load exceptions when
666 static int dm_add_exception(void *context, chunk_t old, chunk_t new)
668 struct dm_snapshot *s = context;
669 struct dm_exception *e;
671 e = alloc_completed_exception();
677 /* Consecutive_count is implicitly initialised to zero */
680 dm_insert_exception(&s->complete, e);
685 #define min_not_zero(l, r) (((l) == 0) ? (r) : (((r) == 0) ? (l) : min(l, r)))
688 * Return a minimum chunk size of all snapshots that have the specified origin.
689 * Return zero if the origin has no snapshots.
691 static sector_t __minimum_chunk_size(struct origin *o)
693 struct dm_snapshot *snap;
694 unsigned chunk_size = 0;
697 list_for_each_entry(snap, &o->snapshots, list)
698 chunk_size = min_not_zero(chunk_size,
699 snap->store->chunk_size);
707 static int calc_max_buckets(void)
709 /* use a fixed size of 2MB */
710 unsigned long mem = 2 * 1024 * 1024;
711 mem /= sizeof(struct list_head);
717 * Allocate room for a suitable hash table.
719 static int init_hash_tables(struct dm_snapshot *s)
721 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
724 * Calculate based on the size of the original volume or
727 cow_dev_size = get_dev_size(s->cow->bdev);
728 origin_dev_size = get_dev_size(s->origin->bdev);
729 max_buckets = calc_max_buckets();
731 hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift;
732 hash_size = min(hash_size, max_buckets);
736 hash_size = rounddown_pow_of_two(hash_size);
737 if (dm_exception_table_init(&s->complete, hash_size,
738 DM_CHUNK_CONSECUTIVE_BITS))
742 * Allocate hash table for in-flight exceptions
743 * Make this smaller than the real hash table
749 if (dm_exception_table_init(&s->pending, hash_size, 0)) {
750 dm_exception_table_exit(&s->complete, exception_cache);
757 static void merge_shutdown(struct dm_snapshot *s)
759 clear_bit_unlock(RUNNING_MERGE, &s->state_bits);
760 smp_mb__after_clear_bit();
761 wake_up_bit(&s->state_bits, RUNNING_MERGE);
764 static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s)
766 s->first_merging_chunk = 0;
767 s->num_merging_chunks = 0;
769 return bio_list_get(&s->bios_queued_during_merge);
773 * Remove one chunk from the index of completed exceptions.
775 static int __remove_single_exception_chunk(struct dm_snapshot *s,
778 struct dm_exception *e;
780 e = dm_lookup_exception(&s->complete, old_chunk);
782 DMERR("Corruption detected: exception for block %llu is "
783 "on disk but not in memory",
784 (unsigned long long)old_chunk);
789 * If this is the only chunk using this exception, remove exception.
791 if (!dm_consecutive_chunk_count(e)) {
792 dm_remove_exception(e);
793 free_completed_exception(e);
798 * The chunk may be either at the beginning or the end of a
799 * group of consecutive chunks - never in the middle. We are
800 * removing chunks in the opposite order to that in which they
801 * were added, so this should always be true.
802 * Decrement the consecutive chunk counter and adjust the
803 * starting point if necessary.
805 if (old_chunk == e->old_chunk) {
808 } else if (old_chunk != e->old_chunk +
809 dm_consecutive_chunk_count(e)) {
810 DMERR("Attempt to merge block %llu from the "
811 "middle of a chunk range [%llu - %llu]",
812 (unsigned long long)old_chunk,
813 (unsigned long long)e->old_chunk,
815 e->old_chunk + dm_consecutive_chunk_count(e));
819 dm_consecutive_chunk_count_dec(e);
824 static void flush_bios(struct bio *bio);
826 static int remove_single_exception_chunk(struct dm_snapshot *s)
828 struct bio *b = NULL;
830 chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1;
832 down_write(&s->lock);
835 * Process chunks (and associated exceptions) in reverse order
836 * so that dm_consecutive_chunk_count_dec() accounting works.
839 r = __remove_single_exception_chunk(s, old_chunk);
842 } while (old_chunk-- > s->first_merging_chunk);
844 b = __release_queued_bios_after_merge(s);
854 static int origin_write_extent(struct dm_snapshot *merging_snap,
855 sector_t sector, unsigned chunk_size);
857 static void merge_callback(int read_err, unsigned long write_err,
860 static uint64_t read_pending_exceptions_done_count(void)
862 uint64_t pending_exceptions_done;
864 spin_lock(&_pending_exceptions_done_spinlock);
865 pending_exceptions_done = _pending_exceptions_done_count;
866 spin_unlock(&_pending_exceptions_done_spinlock);
868 return pending_exceptions_done;
871 static void increment_pending_exceptions_done_count(void)
873 spin_lock(&_pending_exceptions_done_spinlock);
874 _pending_exceptions_done_count++;
875 spin_unlock(&_pending_exceptions_done_spinlock);
877 wake_up_all(&_pending_exceptions_done);
880 static void snapshot_merge_next_chunks(struct dm_snapshot *s)
882 int i, linear_chunks;
883 chunk_t old_chunk, new_chunk;
884 struct dm_io_region src, dest;
886 uint64_t previous_count;
888 BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
889 if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
893 * valid flag never changes during merge, so no lock required.
896 DMERR("Snapshot is invalid: can't merge");
900 linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
902 if (linear_chunks <= 0) {
903 if (linear_chunks < 0)
904 DMERR("Read error in exception store: "
905 "shutting down merge");
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);
991 b = __release_queued_bios_after_merge(s);
998 static void start_merge(struct dm_snapshot *s)
1000 if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits))
1001 snapshot_merge_next_chunks(s);
1004 static int wait_schedule(void *ptr)
1012 * Stop the merging process and wait until it finishes.
1014 static void stop_merge(struct dm_snapshot *s)
1016 set_bit(SHUTDOWN_MERGE, &s->state_bits);
1017 wait_on_bit(&s->state_bits, RUNNING_MERGE, wait_schedule,
1018 TASK_UNINTERRUPTIBLE);
1019 clear_bit(SHUTDOWN_MERGE, &s->state_bits);
1023 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
1025 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1027 struct dm_snapshot *s;
1030 char *origin_path, *cow_path;
1031 unsigned args_used, num_flush_requests = 1;
1032 fmode_t origin_mode = FMODE_READ;
1035 ti->error = "requires exactly 4 arguments";
1040 if (dm_target_is_snapshot_merge(ti)) {
1041 num_flush_requests = 2;
1042 origin_mode = FMODE_WRITE;
1045 origin_path = argv[0];
1049 s = kmalloc(sizeof(*s), GFP_KERNEL);
1051 ti->error = "Cannot allocate snapshot context private "
1061 r = dm_get_device(ti, cow_path, 0, 0,
1062 FMODE_READ | FMODE_WRITE, &s->cow);
1064 ti->error = "Cannot get COW device";
1068 r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
1070 ti->error = "Couldn't create exception store";
1078 r = dm_get_device(ti, origin_path, 0, ti->len, origin_mode, &s->origin);
1080 ti->error = "Cannot get origin device";
1088 atomic_set(&s->pending_exceptions_count, 0);
1089 init_rwsem(&s->lock);
1090 INIT_LIST_HEAD(&s->list);
1091 spin_lock_init(&s->pe_lock);
1093 s->first_merging_chunk = 0;
1094 s->num_merging_chunks = 0;
1095 bio_list_init(&s->bios_queued_during_merge);
1097 /* Allocate hash table for COW data */
1098 if (init_hash_tables(s)) {
1099 ti->error = "Unable to allocate hash table space";
1101 goto bad_hash_tables;
1104 r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &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);
1129 bio_list_init(&s->queued_bios);
1130 INIT_WORK(&s->queued_bios_work, flush_queued_bios);
1133 ti->num_flush_requests = num_flush_requests;
1135 /* Add snapshot to the list of snapshots for this origin */
1136 /* Exceptions aren't triggered till snapshot_resume() is called */
1137 r = register_snapshot(s);
1139 ti->error = "Snapshot origin struct allocation failed";
1140 goto bad_load_and_register;
1142 /* invalid handover, register_snapshot has set ti->error */
1143 goto bad_load_and_register;
1147 * Metadata must only be loaded into one table at once, so skip this
1148 * if metadata will be handed over during resume.
1149 * Chunk size will be set during the handover - set it to zero to
1150 * ensure it's ignored.
1153 s->store->chunk_size = 0;
1157 r = s->store->type->read_metadata(s->store, dm_add_exception,
1160 ti->error = "Failed to read snapshot metadata";
1161 goto bad_read_metadata;
1164 DMWARN("Snapshot is marked invalid.");
1167 if (!s->store->chunk_size) {
1168 ti->error = "Chunk size not set";
1169 goto bad_read_metadata;
1171 ti->split_io = s->store->chunk_size;
1176 unregister_snapshot(s);
1178 bad_load_and_register:
1179 mempool_destroy(s->tracked_chunk_pool);
1181 bad_tracked_chunk_pool:
1182 mempool_destroy(s->pending_pool);
1185 dm_kcopyd_client_destroy(s->kcopyd_client);
1188 dm_exception_table_exit(&s->pending, pending_cache);
1189 dm_exception_table_exit(&s->complete, exception_cache);
1192 dm_put_device(ti, s->origin);
1195 dm_exception_store_destroy(s->store);
1198 dm_put_device(ti, s->cow);
1207 static void __free_exceptions(struct dm_snapshot *s)
1209 dm_kcopyd_client_destroy(s->kcopyd_client);
1210 s->kcopyd_client = NULL;
1212 dm_exception_table_exit(&s->pending, pending_cache);
1213 dm_exception_table_exit(&s->complete, exception_cache);
1216 static void __handover_exceptions(struct dm_snapshot *snap_src,
1217 struct dm_snapshot *snap_dest)
1220 struct dm_exception_table table_swap;
1221 struct dm_exception_store *store_swap;
1225 * Swap all snapshot context information between the two instances.
1227 u.table_swap = snap_dest->complete;
1228 snap_dest->complete = snap_src->complete;
1229 snap_src->complete = u.table_swap;
1231 u.store_swap = snap_dest->store;
1232 snap_dest->store = snap_src->store;
1233 snap_src->store = u.store_swap;
1235 snap_dest->store->snap = snap_dest;
1236 snap_src->store->snap = snap_src;
1238 snap_dest->ti->split_io = snap_dest->store->chunk_size;
1239 snap_dest->valid = snap_src->valid;
1242 * Set source invalid to ensure it receives no further I/O.
1244 snap_src->valid = 0;
1247 static void snapshot_dtr(struct dm_target *ti)
1249 #ifdef CONFIG_DM_DEBUG
1252 struct dm_snapshot *s = ti->private;
1253 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1255 flush_workqueue(ksnapd);
1257 down_read(&_origins_lock);
1258 /* Check whether exception handover must be cancelled */
1259 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1260 if (snap_src && snap_dest && (s == snap_src)) {
1261 down_write(&snap_dest->lock);
1262 snap_dest->valid = 0;
1263 up_write(&snap_dest->lock);
1264 DMERR("Cancelling snapshot handover.");
1266 up_read(&_origins_lock);
1268 if (dm_target_is_snapshot_merge(ti))
1271 /* Prevent further origin writes from using this snapshot. */
1272 /* After this returns there can be no new kcopyd jobs. */
1273 unregister_snapshot(s);
1275 while (atomic_read(&s->pending_exceptions_count))
1278 * Ensure instructions in mempool_destroy aren't reordered
1279 * before atomic_read.
1283 #ifdef CONFIG_DM_DEBUG
1284 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1285 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
1288 mempool_destroy(s->tracked_chunk_pool);
1290 __free_exceptions(s);
1292 mempool_destroy(s->pending_pool);
1294 dm_put_device(ti, s->origin);
1296 dm_exception_store_destroy(s->store);
1298 dm_put_device(ti, s->cow);
1304 * Flush a list of buffers.
1306 static void flush_bios(struct bio *bio)
1312 bio->bi_next = NULL;
1313 generic_make_request(bio);
1318 static void flush_queued_bios(struct work_struct *work)
1320 struct dm_snapshot *s =
1321 container_of(work, struct dm_snapshot, queued_bios_work);
1322 struct bio *queued_bios;
1323 unsigned long flags;
1325 spin_lock_irqsave(&s->pe_lock, flags);
1326 queued_bios = bio_list_get(&s->queued_bios);
1327 spin_unlock_irqrestore(&s->pe_lock, flags);
1329 flush_bios(queued_bios);
1332 static int do_origin(struct dm_dev *origin, struct bio *bio);
1335 * Flush a list of buffers.
1337 static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
1344 bio->bi_next = NULL;
1345 r = do_origin(s->origin, bio);
1346 if (r == DM_MAPIO_REMAPPED)
1347 generic_make_request(bio);
1353 * Error a list of buffers.
1355 static void error_bios(struct bio *bio)
1361 bio->bi_next = NULL;
1367 static void __invalidate_snapshot(struct dm_snapshot *s, int err)
1373 DMERR("Invalidating snapshot: Error reading/writing.");
1374 else if (err == -ENOMEM)
1375 DMERR("Invalidating snapshot: Unable to allocate exception.");
1377 if (s->store->type->drop_snapshot)
1378 s->store->type->drop_snapshot(s->store);
1382 dm_table_event(s->ti->table);
1385 static void pending_complete(struct dm_snap_pending_exception *pe, int success)
1387 struct dm_exception *e;
1388 struct dm_snapshot *s = pe->snap;
1389 struct bio *origin_bios = NULL;
1390 struct bio *snapshot_bios = NULL;
1394 /* Read/write error - snapshot is unusable */
1395 down_write(&s->lock);
1396 __invalidate_snapshot(s, -EIO);
1401 e = alloc_completed_exception();
1403 down_write(&s->lock);
1404 __invalidate_snapshot(s, -ENOMEM);
1410 down_write(&s->lock);
1412 free_completed_exception(e);
1417 /* Check for conflicting reads */
1418 __check_for_conflicting_io(s, pe->e.old_chunk);
1421 * Add a proper exception, and remove the
1422 * in-flight exception from the list.
1424 dm_insert_exception(&s->complete, e);
1427 dm_remove_exception(&pe->e);
1428 snapshot_bios = bio_list_get(&pe->snapshot_bios);
1429 origin_bios = bio_list_get(&pe->origin_bios);
1430 free_pending_exception(pe);
1432 increment_pending_exceptions_done_count();
1436 /* Submit any pending write bios */
1438 error_bios(snapshot_bios);
1440 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,
1492 &src, 1, &dest, 0, copy_callback, pe);
1495 static struct dm_snap_pending_exception *
1496 __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1498 struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
1503 return container_of(e, struct dm_snap_pending_exception, e);
1507 * Looks to see if this snapshot already has a pending exception
1508 * for this chunk, otherwise it allocates a new one and inserts
1509 * it into the pending table.
1511 * NOTE: a write lock must be held on snap->lock before calling
1514 static struct dm_snap_pending_exception *
1515 __find_pending_exception(struct dm_snapshot *s,
1516 struct dm_snap_pending_exception *pe, chunk_t chunk)
1518 struct dm_snap_pending_exception *pe2;
1520 pe2 = __lookup_pending_exception(s, chunk);
1522 free_pending_exception(pe);
1526 pe->e.old_chunk = chunk;
1527 bio_list_init(&pe->origin_bios);
1528 bio_list_init(&pe->snapshot_bios);
1531 if (s->store->type->prepare_exception(s->store, &pe->e)) {
1532 free_pending_exception(pe);
1536 dm_insert_exception(&s->pending, &pe->e);
1541 static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
1542 struct bio *bio, chunk_t chunk)
1544 bio->bi_bdev = s->cow->bdev;
1545 bio->bi_sector = chunk_to_sector(s->store,
1546 dm_chunk_number(e->new_chunk) +
1547 (chunk - e->old_chunk)) +
1549 s->store->chunk_mask);
1552 static int snapshot_map(struct dm_target *ti, struct bio *bio,
1553 union map_info *map_context)
1555 struct dm_exception *e;
1556 struct dm_snapshot *s = ti->private;
1557 int r = DM_MAPIO_REMAPPED;
1559 struct dm_snap_pending_exception *pe = NULL;
1561 if (unlikely(bio_empty_barrier(bio))) {
1562 bio->bi_bdev = s->cow->bdev;
1563 return DM_MAPIO_REMAPPED;
1566 chunk = sector_to_chunk(s->store, bio->bi_sector);
1568 /* Full snapshots are not usable */
1569 /* To get here the table must be live so s->active is always set. */
1573 /* FIXME: should only take write lock if we need
1574 * to copy an exception */
1575 down_write(&s->lock);
1582 /* If the block is already remapped - use that, else remap it */
1583 e = dm_lookup_exception(&s->complete, chunk);
1585 remap_exception(s, e, bio, chunk);
1590 * Write to snapshot - higher level takes care of RW/RO
1591 * flags so we should only get this if we are
1594 if (bio_rw(bio) == WRITE) {
1595 pe = __lookup_pending_exception(s, chunk);
1598 pe = alloc_pending_exception(s);
1599 down_write(&s->lock);
1602 free_pending_exception(pe);
1607 e = dm_lookup_exception(&s->complete, chunk);
1609 free_pending_exception(pe);
1610 remap_exception(s, e, bio, chunk);
1614 pe = __find_pending_exception(s, pe, chunk);
1616 __invalidate_snapshot(s, -ENOMEM);
1622 remap_exception(s, &pe->e, bio, chunk);
1623 bio_list_add(&pe->snapshot_bios, bio);
1625 r = DM_MAPIO_SUBMITTED;
1628 /* this is protected by snap->lock */
1635 bio->bi_bdev = s->origin->bdev;
1636 map_context->ptr = track_chunk(s, chunk);
1646 * A snapshot-merge target behaves like a combination of a snapshot
1647 * target and a snapshot-origin target. It only generates new
1648 * exceptions in other snapshots and not in the one that is being
1651 * For each chunk, if there is an existing exception, it is used to
1652 * redirect I/O to the cow device. Otherwise I/O is sent to the origin,
1653 * which in turn might generate exceptions in other snapshots.
1654 * If merging is currently taking place on the chunk in question, the
1655 * I/O is deferred by adding it to s->bios_queued_during_merge.
1657 static int snapshot_merge_map(struct dm_target *ti, struct bio *bio,
1658 union map_info *map_context)
1660 struct dm_exception *e;
1661 struct dm_snapshot *s = ti->private;
1662 int r = DM_MAPIO_REMAPPED;
1665 if (unlikely(bio_empty_barrier(bio))) {
1666 if (!map_context->flush_request)
1667 bio->bi_bdev = s->origin->bdev;
1669 bio->bi_bdev = s->cow->bdev;
1670 map_context->ptr = NULL;
1671 return DM_MAPIO_REMAPPED;
1674 chunk = sector_to_chunk(s->store, bio->bi_sector);
1676 down_write(&s->lock);
1678 /* Full snapshots are not usable */
1684 /* If the block is already remapped - use that */
1685 e = dm_lookup_exception(&s->complete, chunk);
1687 /* Queue writes overlapping with chunks being merged */
1688 if (bio_rw(bio) == WRITE &&
1689 chunk >= s->first_merging_chunk &&
1690 chunk < (s->first_merging_chunk +
1691 s->num_merging_chunks)) {
1692 bio->bi_bdev = s->origin->bdev;
1693 bio_list_add(&s->bios_queued_during_merge, bio);
1694 r = DM_MAPIO_SUBMITTED;
1698 remap_exception(s, e, bio, chunk);
1700 if (bio_rw(bio) == WRITE)
1701 map_context->ptr = track_chunk(s, chunk);
1705 bio->bi_bdev = s->origin->bdev;
1707 if (bio_rw(bio) == WRITE) {
1709 return do_origin(s->origin, bio);
1718 static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1719 int error, union map_info *map_context)
1721 struct dm_snapshot *s = ti->private;
1722 struct dm_snap_tracked_chunk *c = map_context->ptr;
1725 stop_tracking_chunk(s, c);
1730 static void snapshot_merge_presuspend(struct dm_target *ti)
1732 struct dm_snapshot *s = ti->private;
1737 static void snapshot_postsuspend(struct dm_target *ti)
1739 struct dm_snapshot *s = ti->private;
1741 down_write(&s->lock);
1746 static int snapshot_preresume(struct dm_target *ti)
1749 struct dm_snapshot *s = ti->private;
1750 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1752 down_read(&_origins_lock);
1753 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1754 if (snap_src && snap_dest) {
1755 down_read(&snap_src->lock);
1756 if (s == snap_src) {
1757 DMERR("Unable to resume snapshot source until "
1758 "handover completes.");
1760 } else if (!snap_src->suspended) {
1761 DMERR("Unable to perform snapshot handover until "
1762 "source is suspended.");
1765 up_read(&snap_src->lock);
1767 up_read(&_origins_lock);
1772 static void snapshot_resume(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_write(&snap_src->lock);
1781 down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1782 __handover_exceptions(snap_src, snap_dest);
1783 up_write(&snap_dest->lock);
1784 up_write(&snap_src->lock);
1786 up_read(&_origins_lock);
1788 /* Now we have correct chunk size, reregister */
1789 reregister_snapshot(s);
1791 down_write(&s->lock);
1797 static sector_t get_origin_minimum_chunksize(struct block_device *bdev)
1799 sector_t min_chunksize;
1801 down_read(&_origins_lock);
1802 min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
1803 up_read(&_origins_lock);
1805 return min_chunksize;
1808 static void snapshot_merge_resume(struct dm_target *ti)
1810 struct dm_snapshot *s = ti->private;
1813 * Handover exceptions from existing snapshot.
1815 snapshot_resume(ti);
1818 * snapshot-merge acts as an origin, so set ti->split_io
1820 ti->split_io = get_origin_minimum_chunksize(s->origin->bdev);
1825 static int snapshot_status(struct dm_target *ti, status_type_t type,
1826 char *result, unsigned int maxlen)
1829 struct dm_snapshot *snap = ti->private;
1832 case STATUSTYPE_INFO:
1834 down_write(&snap->lock);
1839 if (snap->store->type->usage) {
1840 sector_t total_sectors, sectors_allocated,
1842 snap->store->type->usage(snap->store,
1846 DMEMIT("%llu/%llu %llu",
1847 (unsigned long long)sectors_allocated,
1848 (unsigned long long)total_sectors,
1849 (unsigned long long)metadata_sectors);
1855 up_write(&snap->lock);
1859 case STATUSTYPE_TABLE:
1861 * kdevname returns a static pointer so we need
1862 * to make private copies if the output is to
1865 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
1866 snap->store->type->status(snap->store, type, result + sz,
1874 static int snapshot_iterate_devices(struct dm_target *ti,
1875 iterate_devices_callout_fn fn, void *data)
1877 struct dm_snapshot *snap = ti->private;
1879 return fn(ti, snap->origin, 0, ti->len, data);
1883 /*-----------------------------------------------------------------
1885 *---------------------------------------------------------------*/
1888 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
1889 * supplied bio was ignored. The caller may submit it immediately.
1890 * (No remapping actually occurs as the origin is always a direct linear
1893 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
1894 * and any supplied bio is added to a list to be submitted once all
1895 * the necessary exceptions exist.
1897 static int __origin_write(struct list_head *snapshots, sector_t sector,
1900 int r = DM_MAPIO_REMAPPED;
1901 struct dm_snapshot *snap;
1902 struct dm_exception *e;
1903 struct dm_snap_pending_exception *pe;
1904 struct dm_snap_pending_exception *pe_to_start_now = NULL;
1905 struct dm_snap_pending_exception *pe_to_start_last = NULL;
1908 /* Do all the snapshots on this origin */
1909 list_for_each_entry (snap, snapshots, list) {
1911 * Don't make new exceptions in a merging snapshot
1912 * because it has effectively been deleted
1914 if (dm_target_is_snapshot_merge(snap->ti))
1917 down_write(&snap->lock);
1919 /* Only deal with valid and active snapshots */
1920 if (!snap->valid || !snap->active)
1923 /* Nothing to do if writing beyond end of snapshot */
1924 if (sector >= dm_table_get_size(snap->ti->table))
1928 * Remember, different snapshots can have
1929 * different chunk sizes.
1931 chunk = sector_to_chunk(snap->store, sector);
1934 * Check exception table to see if block
1935 * is already remapped in this snapshot
1936 * and trigger an exception if not.
1938 e = dm_lookup_exception(&snap->complete, chunk);
1942 pe = __lookup_pending_exception(snap, chunk);
1944 up_write(&snap->lock);
1945 pe = alloc_pending_exception(snap);
1946 down_write(&snap->lock);
1949 free_pending_exception(pe);
1953 e = dm_lookup_exception(&snap->complete, chunk);
1955 free_pending_exception(pe);
1959 pe = __find_pending_exception(snap, pe, chunk);
1961 __invalidate_snapshot(snap, -ENOMEM);
1966 r = DM_MAPIO_SUBMITTED;
1969 * If an origin bio was supplied, queue it to wait for the
1970 * completion of this exception, and start this one last,
1971 * at the end of the function.
1974 bio_list_add(&pe->origin_bios, bio);
1979 pe_to_start_last = pe;
1985 pe_to_start_now = pe;
1989 up_write(&snap->lock);
1991 if (pe_to_start_now) {
1992 start_copy(pe_to_start_now);
1993 pe_to_start_now = NULL;
1998 * Submit the exception against which the bio is queued last,
1999 * to give the other exceptions a head start.
2001 if (pe_to_start_last)
2002 start_copy(pe_to_start_last);
2008 * Called on a write from the origin driver.
2010 static int do_origin(struct dm_dev *origin, struct bio *bio)
2013 int r = DM_MAPIO_REMAPPED;
2015 down_read(&_origins_lock);
2016 o = __lookup_origin(origin->bdev);
2018 r = __origin_write(&o->snapshots, bio->bi_sector, bio);
2019 up_read(&_origins_lock);
2025 * Trigger exceptions in all non-merging snapshots.
2027 * The chunk size of the merging snapshot may be larger than the chunk
2028 * size of some other snapshot so we may need to reallocate multiple
2029 * chunks in other snapshots.
2031 * We scan all the overlapping exceptions in the other snapshots.
2032 * Returns 1 if anything was reallocated and must be waited for,
2033 * otherwise returns 0.
2035 * size must be a multiple of merging_snap's chunk_size.
2037 static int origin_write_extent(struct dm_snapshot *merging_snap,
2038 sector_t sector, unsigned size)
2045 * The origin's __minimum_chunk_size() got stored in split_io
2046 * by snapshot_merge_resume().
2048 down_read(&_origins_lock);
2049 o = __lookup_origin(merging_snap->origin->bdev);
2050 for (n = 0; n < size; n += merging_snap->ti->split_io)
2051 if (__origin_write(&o->snapshots, sector + n, NULL) ==
2054 up_read(&_origins_lock);
2060 * Origin: maps a linear range of a device, with hooks for snapshotting.
2064 * Construct an origin mapping: <dev_path>
2065 * The context for an origin is merely a 'struct dm_dev *'
2066 * pointing to the real device.
2068 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2074 ti->error = "origin: incorrect number of arguments";
2078 r = dm_get_device(ti, argv[0], 0, ti->len,
2079 dm_table_get_mode(ti->table), &dev);
2081 ti->error = "Cannot get target device";
2086 ti->num_flush_requests = 1;
2091 static void origin_dtr(struct dm_target *ti)
2093 struct dm_dev *dev = ti->private;
2094 dm_put_device(ti, dev);
2097 static int origin_map(struct dm_target *ti, struct bio *bio,
2098 union map_info *map_context)
2100 struct dm_dev *dev = ti->private;
2101 bio->bi_bdev = dev->bdev;
2103 if (unlikely(bio_empty_barrier(bio)))
2104 return DM_MAPIO_REMAPPED;
2106 /* Only tell snapshots if this is a write */
2107 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
2111 * Set the target "split_io" field to the minimum of all the snapshots'
2114 static void origin_resume(struct dm_target *ti)
2116 struct dm_dev *dev = ti->private;
2118 ti->split_io = get_origin_minimum_chunksize(dev->bdev);
2121 static int origin_status(struct dm_target *ti, status_type_t type, char *result,
2122 unsigned int maxlen)
2124 struct dm_dev *dev = ti->private;
2127 case STATUSTYPE_INFO:
2131 case STATUSTYPE_TABLE:
2132 snprintf(result, maxlen, "%s", dev->name);
2139 static int origin_iterate_devices(struct dm_target *ti,
2140 iterate_devices_callout_fn fn, void *data)
2142 struct dm_dev *dev = ti->private;
2144 return fn(ti, dev, 0, ti->len, data);
2147 static struct target_type origin_target = {
2148 .name = "snapshot-origin",
2149 .version = {1, 7, 0},
2150 .module = THIS_MODULE,
2154 .resume = origin_resume,
2155 .status = origin_status,
2156 .iterate_devices = origin_iterate_devices,
2159 static struct target_type snapshot_target = {
2161 .version = {1, 9, 0},
2162 .module = THIS_MODULE,
2163 .ctr = snapshot_ctr,
2164 .dtr = snapshot_dtr,
2165 .map = snapshot_map,
2166 .end_io = snapshot_end_io,
2167 .postsuspend = snapshot_postsuspend,
2168 .preresume = snapshot_preresume,
2169 .resume = snapshot_resume,
2170 .status = snapshot_status,
2171 .iterate_devices = snapshot_iterate_devices,
2174 static struct target_type merge_target = {
2175 .name = dm_snapshot_merge_target_name,
2176 .version = {1, 0, 0},
2177 .module = THIS_MODULE,
2178 .ctr = snapshot_ctr,
2179 .dtr = snapshot_dtr,
2180 .map = snapshot_merge_map,
2181 .end_io = snapshot_end_io,
2182 .presuspend = snapshot_merge_presuspend,
2183 .postsuspend = snapshot_postsuspend,
2184 .preresume = snapshot_preresume,
2185 .resume = snapshot_merge_resume,
2186 .status = snapshot_status,
2187 .iterate_devices = snapshot_iterate_devices,
2190 static int __init dm_snapshot_init(void)
2194 r = dm_exception_store_init();
2196 DMERR("Failed to initialize exception stores");
2200 r = dm_register_target(&snapshot_target);
2202 DMERR("snapshot target register failed %d", r);
2203 goto bad_register_snapshot_target;
2206 r = dm_register_target(&origin_target);
2208 DMERR("Origin target register failed %d", r);
2209 goto bad_register_origin_target;
2212 r = dm_register_target(&merge_target);
2214 DMERR("Merge target register failed %d", r);
2215 goto bad_register_merge_target;
2218 r = init_origin_hash();
2220 DMERR("init_origin_hash failed.");
2221 goto bad_origin_hash;
2224 exception_cache = KMEM_CACHE(dm_exception, 0);
2225 if (!exception_cache) {
2226 DMERR("Couldn't create exception cache.");
2228 goto bad_exception_cache;
2231 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
2232 if (!pending_cache) {
2233 DMERR("Couldn't create pending cache.");
2235 goto bad_pending_cache;
2238 tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0);
2239 if (!tracked_chunk_cache) {
2240 DMERR("Couldn't create cache to track chunks in use.");
2242 goto bad_tracked_chunk_cache;
2245 ksnapd = create_singlethread_workqueue("ksnapd");
2247 DMERR("Failed to create ksnapd workqueue.");
2249 goto bad_pending_pool;
2255 kmem_cache_destroy(tracked_chunk_cache);
2256 bad_tracked_chunk_cache:
2257 kmem_cache_destroy(pending_cache);
2259 kmem_cache_destroy(exception_cache);
2260 bad_exception_cache:
2263 dm_unregister_target(&merge_target);
2264 bad_register_merge_target:
2265 dm_unregister_target(&origin_target);
2266 bad_register_origin_target:
2267 dm_unregister_target(&snapshot_target);
2268 bad_register_snapshot_target:
2269 dm_exception_store_exit();
2274 static void __exit dm_snapshot_exit(void)
2276 destroy_workqueue(ksnapd);
2278 dm_unregister_target(&snapshot_target);
2279 dm_unregister_target(&origin_target);
2280 dm_unregister_target(&merge_target);
2283 kmem_cache_destroy(pending_cache);
2284 kmem_cache_destroy(exception_cache);
2285 kmem_cache_destroy(tracked_chunk_cache);
2287 dm_exception_store_exit();
2291 module_init(dm_snapshot_init);
2292 module_exit(dm_snapshot_exit);
2294 MODULE_DESCRIPTION(DM_NAME " snapshot target");
2295 MODULE_AUTHOR("Joe Thornber");
2296 MODULE_LICENSE("GPL");