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 percentage increment we will wake up users at
35 #define WAKE_UP_PERCENT 5
38 * kcopyd priority of snapshot operations
40 #define SNAPSHOT_COPY_PRIORITY 2
43 * Reserve 1MB for each snapshot initially (with minimum of 1 page).
45 #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1)
48 * The size of the mempool used to track chunks in use.
52 #define DM_TRACKED_CHUNK_HASH_SIZE 16
53 #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
54 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
56 struct dm_exception_table {
59 struct list_head *table;
63 struct rw_semaphore lock;
65 struct dm_dev *origin;
70 /* List of snapshots per Origin */
71 struct list_head list;
74 * You can't use a snapshot if this is 0 (e.g. if full).
75 * A snapshot-merge target never clears this.
79 /* Origin writes don't trigger exceptions until this is set */
82 /* Whether or not owning mapped_device is suspended */
85 atomic_t pending_exceptions_count;
87 mempool_t *pending_pool;
89 struct dm_exception_table pending;
90 struct dm_exception_table complete;
93 * pe_lock protects all pending_exception operations and access
94 * as well as the snapshot_bios list.
98 /* Chunks with outstanding reads */
99 spinlock_t tracked_chunk_lock;
100 mempool_t *tracked_chunk_pool;
101 struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
103 /* The on disk metadata handler */
104 struct dm_exception_store *store;
106 struct dm_kcopyd_client *kcopyd_client;
108 /* Wait for events based on state_bits */
109 unsigned long state_bits;
111 /* Range of chunks currently being merged. */
112 chunk_t first_merging_chunk;
113 int num_merging_chunks;
116 * The merge operation failed if this flag is set.
117 * Failure modes are handled as follows:
118 * - I/O error reading the header
119 * => don't load the target; abort.
120 * - Header does not have "valid" flag set
121 * => use the origin; forget about the snapshot.
122 * - I/O error when reading exceptions
123 * => don't load the target; abort.
124 * (We can't use the intermediate origin state.)
125 * - I/O error while merging
126 * => stop merging; set merge_failed; process I/O normally.
131 * Incoming bios that overlap with chunks being merged must wait
132 * for them to be committed.
134 struct bio_list bios_queued_during_merge;
139 * RUNNING_MERGE - Merge operation is in progress.
140 * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped;
141 * cleared afterwards.
143 #define RUNNING_MERGE 0
144 #define SHUTDOWN_MERGE 1
146 struct dm_dev *dm_snap_origin(struct dm_snapshot *s)
150 EXPORT_SYMBOL(dm_snap_origin);
152 struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
156 EXPORT_SYMBOL(dm_snap_cow);
158 static sector_t chunk_to_sector(struct dm_exception_store *store,
161 return chunk << store->chunk_shift;
164 static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
167 * There is only ever one instance of a particular block
168 * device so we can compare pointers safely.
173 struct dm_snap_pending_exception {
174 struct dm_exception e;
177 * Origin buffers waiting for this to complete are held
180 struct bio_list origin_bios;
181 struct bio_list snapshot_bios;
183 /* Pointer back to snapshot context */
184 struct dm_snapshot *snap;
187 * 1 indicates the exception has already been sent to
194 * Hash table mapping origin volumes to lists of snapshots and
195 * a lock to protect it
197 static struct kmem_cache *exception_cache;
198 static struct kmem_cache *pending_cache;
200 struct dm_snap_tracked_chunk {
201 struct hlist_node node;
205 static struct kmem_cache *tracked_chunk_cache;
207 static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s,
210 struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool,
216 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
217 hlist_add_head(&c->node,
218 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
219 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
224 static void stop_tracking_chunk(struct dm_snapshot *s,
225 struct dm_snap_tracked_chunk *c)
229 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
231 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
233 mempool_free(c, s->tracked_chunk_pool);
236 static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
238 struct dm_snap_tracked_chunk *c;
239 struct hlist_node *hn;
242 spin_lock_irq(&s->tracked_chunk_lock);
244 hlist_for_each_entry(c, hn,
245 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
246 if (c->chunk == chunk) {
252 spin_unlock_irq(&s->tracked_chunk_lock);
258 * This conflicting I/O is extremely improbable in the caller,
259 * so msleep(1) is sufficient and there is no need for a wait queue.
261 static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
263 while (__chunk_is_tracked(s, chunk))
268 * One of these per registered origin, held in the snapshot_origins hash
271 /* The origin device */
272 struct block_device *bdev;
274 struct list_head hash_list;
276 /* List of snapshots for this origin */
277 struct list_head snapshots;
281 * Size of the hash table for origin volumes. If we make this
282 * the size of the minors list then it should be nearly perfect
284 #define ORIGIN_HASH_SIZE 256
285 #define ORIGIN_MASK 0xFF
286 static struct list_head *_origins;
287 static struct rw_semaphore _origins_lock;
289 static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
290 static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
291 static uint64_t _pending_exceptions_done_count;
293 static int init_origin_hash(void)
297 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
300 DMERR("unable to allocate memory");
304 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
305 INIT_LIST_HEAD(_origins + i);
306 init_rwsem(&_origins_lock);
311 static void exit_origin_hash(void)
316 static unsigned origin_hash(struct block_device *bdev)
318 return bdev->bd_dev & ORIGIN_MASK;
321 static struct origin *__lookup_origin(struct block_device *origin)
323 struct list_head *ol;
326 ol = &_origins[origin_hash(origin)];
327 list_for_each_entry (o, ol, hash_list)
328 if (bdev_equal(o->bdev, origin))
334 static void __insert_origin(struct origin *o)
336 struct list_head *sl = &_origins[origin_hash(o->bdev)];
337 list_add_tail(&o->hash_list, sl);
341 * _origins_lock must be held when calling this function.
342 * Returns number of snapshots registered using the supplied cow device, plus:
343 * snap_src - a snapshot suitable for use as a source of exception handover
344 * snap_dest - a snapshot capable of receiving exception handover.
345 * snap_merge - an existing snapshot-merge target linked to the same origin.
346 * There can be at most one snapshot-merge target. The parameter is optional.
348 * Possible return values and states of snap_src and snap_dest.
349 * 0: NULL, NULL - first new snapshot
350 * 1: snap_src, NULL - normal snapshot
351 * 2: snap_src, snap_dest - waiting for handover
352 * 2: snap_src, NULL - handed over, waiting for old to be deleted
353 * 1: NULL, snap_dest - source got destroyed without handover
355 static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
356 struct dm_snapshot **snap_src,
357 struct dm_snapshot **snap_dest,
358 struct dm_snapshot **snap_merge)
360 struct dm_snapshot *s;
365 o = __lookup_origin(snap->origin->bdev);
369 list_for_each_entry(s, &o->snapshots, list) {
370 if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
372 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
382 } else if (snap_dest)
393 * On success, returns 1 if this snapshot is a handover destination,
394 * otherwise returns 0.
396 static int __validate_exception_handover(struct dm_snapshot *snap)
398 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
399 struct dm_snapshot *snap_merge = NULL;
401 /* Does snapshot need exceptions handed over to it? */
402 if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
403 &snap_merge) == 2) ||
405 snap->ti->error = "Snapshot cow pairing for exception "
406 "table handover failed";
411 * If no snap_src was found, snap cannot become a handover
418 * Non-snapshot-merge handover?
420 if (!dm_target_is_snapshot_merge(snap->ti))
424 * Do not allow more than one merging snapshot.
427 snap->ti->error = "A snapshot is already merging.";
431 if (!snap_src->store->type->prepare_merge ||
432 !snap_src->store->type->commit_merge) {
433 snap->ti->error = "Snapshot exception store does not "
434 "support snapshot-merge.";
441 static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
443 struct dm_snapshot *l;
445 /* Sort the list according to chunk size, largest-first smallest-last */
446 list_for_each_entry(l, &o->snapshots, list)
447 if (l->store->chunk_size < s->store->chunk_size)
449 list_add_tail(&s->list, &l->list);
453 * Make a note of the snapshot and its origin so we can look it
454 * up when the origin has a write on it.
456 * Also validate snapshot exception store handovers.
457 * On success, returns 1 if this registration is a handover destination,
458 * otherwise returns 0.
460 static int register_snapshot(struct dm_snapshot *snap)
462 struct origin *o, *new_o = NULL;
463 struct block_device *bdev = snap->origin->bdev;
466 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
470 down_write(&_origins_lock);
472 r = __validate_exception_handover(snap);
478 o = __lookup_origin(bdev);
485 /* Initialise the struct */
486 INIT_LIST_HEAD(&o->snapshots);
492 __insert_snapshot(o, snap);
495 up_write(&_origins_lock);
501 * Move snapshot to correct place in list according to chunk size.
503 static void reregister_snapshot(struct dm_snapshot *s)
505 struct block_device *bdev = s->origin->bdev;
507 down_write(&_origins_lock);
510 __insert_snapshot(__lookup_origin(bdev), s);
512 up_write(&_origins_lock);
515 static void unregister_snapshot(struct dm_snapshot *s)
519 down_write(&_origins_lock);
520 o = __lookup_origin(s->origin->bdev);
523 if (o && list_empty(&o->snapshots)) {
524 list_del(&o->hash_list);
528 up_write(&_origins_lock);
532 * Implementation of the exception hash tables.
533 * The lowest hash_shift bits of the chunk number are ignored, allowing
534 * some consecutive chunks to be grouped together.
536 static int dm_exception_table_init(struct dm_exception_table *et,
537 uint32_t size, unsigned hash_shift)
541 et->hash_shift = hash_shift;
542 et->hash_mask = size - 1;
543 et->table = dm_vcalloc(size, sizeof(struct list_head));
547 for (i = 0; i < size; i++)
548 INIT_LIST_HEAD(et->table + i);
553 static void dm_exception_table_exit(struct dm_exception_table *et,
554 struct kmem_cache *mem)
556 struct list_head *slot;
557 struct dm_exception *ex, *next;
560 size = et->hash_mask + 1;
561 for (i = 0; i < size; i++) {
562 slot = et->table + i;
564 list_for_each_entry_safe (ex, next, slot, hash_list)
565 kmem_cache_free(mem, ex);
571 static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
573 return (chunk >> et->hash_shift) & et->hash_mask;
576 static void dm_remove_exception(struct dm_exception *e)
578 list_del(&e->hash_list);
582 * Return the exception data for a sector, or NULL if not
585 static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
588 struct list_head *slot;
589 struct dm_exception *e;
591 slot = &et->table[exception_hash(et, chunk)];
592 list_for_each_entry (e, slot, hash_list)
593 if (chunk >= e->old_chunk &&
594 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
600 static struct dm_exception *alloc_completed_exception(void)
602 struct dm_exception *e;
604 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
606 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
611 static void free_completed_exception(struct dm_exception *e)
613 kmem_cache_free(exception_cache, e);
616 static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
618 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
621 atomic_inc(&s->pending_exceptions_count);
627 static void free_pending_exception(struct dm_snap_pending_exception *pe)
629 struct dm_snapshot *s = pe->snap;
631 mempool_free(pe, s->pending_pool);
632 smp_mb__before_atomic_dec();
633 atomic_dec(&s->pending_exceptions_count);
636 static void dm_insert_exception(struct dm_exception_table *eh,
637 struct dm_exception *new_e)
640 struct dm_exception *e = NULL;
642 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
644 /* Add immediately if this table doesn't support consecutive chunks */
648 /* List is ordered by old_chunk */
649 list_for_each_entry_reverse(e, l, hash_list) {
650 /* Insert after an existing chunk? */
651 if (new_e->old_chunk == (e->old_chunk +
652 dm_consecutive_chunk_count(e) + 1) &&
653 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
654 dm_consecutive_chunk_count(e) + 1)) {
655 dm_consecutive_chunk_count_inc(e);
656 free_completed_exception(new_e);
660 /* Insert before an existing chunk? */
661 if (new_e->old_chunk == (e->old_chunk - 1) &&
662 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
663 dm_consecutive_chunk_count_inc(e);
666 free_completed_exception(new_e);
670 if (new_e->old_chunk > e->old_chunk)
675 list_add(&new_e->hash_list, e ? &e->hash_list : l);
679 * Callback used by the exception stores to load exceptions when
682 static int dm_add_exception(void *context, chunk_t old, chunk_t new)
684 struct dm_snapshot *s = context;
685 struct dm_exception *e;
687 e = alloc_completed_exception();
693 /* Consecutive_count is implicitly initialised to zero */
696 dm_insert_exception(&s->complete, e);
702 * Return a minimum chunk size of all snapshots that have the specified origin.
703 * Return zero if the origin has no snapshots.
705 static sector_t __minimum_chunk_size(struct origin *o)
707 struct dm_snapshot *snap;
708 unsigned chunk_size = 0;
711 list_for_each_entry(snap, &o->snapshots, list)
712 chunk_size = min_not_zero(chunk_size,
713 snap->store->chunk_size);
721 static int calc_max_buckets(void)
723 /* use a fixed size of 2MB */
724 unsigned long mem = 2 * 1024 * 1024;
725 mem /= sizeof(struct list_head);
731 * Allocate room for a suitable hash table.
733 static int init_hash_tables(struct dm_snapshot *s)
735 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
738 * Calculate based on the size of the original volume or
741 cow_dev_size = get_dev_size(s->cow->bdev);
742 origin_dev_size = get_dev_size(s->origin->bdev);
743 max_buckets = calc_max_buckets();
745 hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift;
746 hash_size = min(hash_size, max_buckets);
750 hash_size = rounddown_pow_of_two(hash_size);
751 if (dm_exception_table_init(&s->complete, hash_size,
752 DM_CHUNK_CONSECUTIVE_BITS))
756 * Allocate hash table for in-flight exceptions
757 * Make this smaller than the real hash table
763 if (dm_exception_table_init(&s->pending, hash_size, 0)) {
764 dm_exception_table_exit(&s->complete, exception_cache);
771 static void merge_shutdown(struct dm_snapshot *s)
773 clear_bit_unlock(RUNNING_MERGE, &s->state_bits);
774 smp_mb__after_clear_bit();
775 wake_up_bit(&s->state_bits, RUNNING_MERGE);
778 static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s)
780 s->first_merging_chunk = 0;
781 s->num_merging_chunks = 0;
783 return bio_list_get(&s->bios_queued_during_merge);
787 * Remove one chunk from the index of completed exceptions.
789 static int __remove_single_exception_chunk(struct dm_snapshot *s,
792 struct dm_exception *e;
794 e = dm_lookup_exception(&s->complete, old_chunk);
796 DMERR("Corruption detected: exception for block %llu is "
797 "on disk but not in memory",
798 (unsigned long long)old_chunk);
803 * If this is the only chunk using this exception, remove exception.
805 if (!dm_consecutive_chunk_count(e)) {
806 dm_remove_exception(e);
807 free_completed_exception(e);
812 * The chunk may be either at the beginning or the end of a
813 * group of consecutive chunks - never in the middle. We are
814 * removing chunks in the opposite order to that in which they
815 * were added, so this should always be true.
816 * Decrement the consecutive chunk counter and adjust the
817 * starting point if necessary.
819 if (old_chunk == e->old_chunk) {
822 } else if (old_chunk != e->old_chunk +
823 dm_consecutive_chunk_count(e)) {
824 DMERR("Attempt to merge block %llu from the "
825 "middle of a chunk range [%llu - %llu]",
826 (unsigned long long)old_chunk,
827 (unsigned long long)e->old_chunk,
829 e->old_chunk + dm_consecutive_chunk_count(e));
833 dm_consecutive_chunk_count_dec(e);
838 static void flush_bios(struct bio *bio);
840 static int remove_single_exception_chunk(struct dm_snapshot *s)
842 struct bio *b = NULL;
844 chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1;
846 down_write(&s->lock);
849 * Process chunks (and associated exceptions) in reverse order
850 * so that dm_consecutive_chunk_count_dec() accounting works.
853 r = __remove_single_exception_chunk(s, old_chunk);
856 } while (old_chunk-- > s->first_merging_chunk);
858 b = __release_queued_bios_after_merge(s);
868 static int origin_write_extent(struct dm_snapshot *merging_snap,
869 sector_t sector, unsigned chunk_size);
871 static void merge_callback(int read_err, unsigned long write_err,
874 static uint64_t read_pending_exceptions_done_count(void)
876 uint64_t pending_exceptions_done;
878 spin_lock(&_pending_exceptions_done_spinlock);
879 pending_exceptions_done = _pending_exceptions_done_count;
880 spin_unlock(&_pending_exceptions_done_spinlock);
882 return pending_exceptions_done;
885 static void increment_pending_exceptions_done_count(void)
887 spin_lock(&_pending_exceptions_done_spinlock);
888 _pending_exceptions_done_count++;
889 spin_unlock(&_pending_exceptions_done_spinlock);
891 wake_up_all(&_pending_exceptions_done);
894 static void snapshot_merge_next_chunks(struct dm_snapshot *s)
896 int i, linear_chunks;
897 chunk_t old_chunk, new_chunk;
898 struct dm_io_region src, dest;
900 uint64_t previous_count;
902 BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
903 if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
907 * valid flag never changes during merge, so no lock required.
910 DMERR("Snapshot is invalid: can't merge");
914 linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
916 if (linear_chunks <= 0) {
917 if (linear_chunks < 0) {
918 DMERR("Read error in exception store: "
919 "shutting down merge");
920 down_write(&s->lock);
927 /* Adjust old_chunk and new_chunk to reflect start of linear region */
928 old_chunk = old_chunk + 1 - linear_chunks;
929 new_chunk = new_chunk + 1 - linear_chunks;
932 * Use one (potentially large) I/O to copy all 'linear_chunks'
933 * from the exception store to the origin
935 io_size = linear_chunks * s->store->chunk_size;
937 dest.bdev = s->origin->bdev;
938 dest.sector = chunk_to_sector(s->store, old_chunk);
939 dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector);
941 src.bdev = s->cow->bdev;
942 src.sector = chunk_to_sector(s->store, new_chunk);
943 src.count = dest.count;
946 * Reallocate any exceptions needed in other snapshots then
947 * wait for the pending exceptions to complete.
948 * Each time any pending exception (globally on the system)
949 * completes we are woken and repeat the process to find out
950 * if we can proceed. While this may not seem a particularly
951 * efficient algorithm, it is not expected to have any
952 * significant impact on performance.
954 previous_count = read_pending_exceptions_done_count();
955 while (origin_write_extent(s, dest.sector, io_size)) {
956 wait_event(_pending_exceptions_done,
957 (read_pending_exceptions_done_count() !=
959 /* Retry after the wait, until all exceptions are done. */
960 previous_count = read_pending_exceptions_done_count();
963 down_write(&s->lock);
964 s->first_merging_chunk = old_chunk;
965 s->num_merging_chunks = linear_chunks;
968 /* Wait until writes to all 'linear_chunks' drain */
969 for (i = 0; i < linear_chunks; i++)
970 __check_for_conflicting_io(s, old_chunk + i);
972 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s);
979 static void error_bios(struct bio *bio);
981 static void merge_callback(int read_err, unsigned long write_err, void *context)
983 struct dm_snapshot *s = context;
984 struct bio *b = NULL;
986 if (read_err || write_err) {
988 DMERR("Read error: shutting down merge.");
990 DMERR("Write error: shutting down merge.");
994 if (s->store->type->commit_merge(s->store,
995 s->num_merging_chunks) < 0) {
996 DMERR("Write error in exception store: shutting down merge");
1000 if (remove_single_exception_chunk(s) < 0)
1003 snapshot_merge_next_chunks(s);
1008 down_write(&s->lock);
1009 s->merge_failed = 1;
1010 b = __release_queued_bios_after_merge(s);
1017 static void start_merge(struct dm_snapshot *s)
1019 if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits))
1020 snapshot_merge_next_chunks(s);
1023 static int wait_schedule(void *ptr)
1031 * Stop the merging process and wait until it finishes.
1033 static void stop_merge(struct dm_snapshot *s)
1035 set_bit(SHUTDOWN_MERGE, &s->state_bits);
1036 wait_on_bit(&s->state_bits, RUNNING_MERGE, wait_schedule,
1037 TASK_UNINTERRUPTIBLE);
1038 clear_bit(SHUTDOWN_MERGE, &s->state_bits);
1042 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
1044 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1046 struct dm_snapshot *s;
1049 char *origin_path, *cow_path;
1050 unsigned args_used, num_flush_requests = 1;
1051 fmode_t origin_mode = FMODE_READ;
1054 ti->error = "requires exactly 4 arguments";
1059 if (dm_target_is_snapshot_merge(ti)) {
1060 num_flush_requests = 2;
1061 origin_mode = FMODE_WRITE;
1064 s = kmalloc(sizeof(*s), GFP_KERNEL);
1066 ti->error = "Cannot allocate snapshot context private "
1072 origin_path = argv[0];
1076 r = dm_get_device(ti, origin_path, origin_mode, &s->origin);
1078 ti->error = "Cannot get origin device";
1086 r = dm_get_device(ti, cow_path, FMODE_READ | FMODE_WRITE, &s->cow);
1088 ti->error = "Cannot get COW device";
1092 r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
1094 ti->error = "Couldn't create exception store";
1106 atomic_set(&s->pending_exceptions_count, 0);
1107 init_rwsem(&s->lock);
1108 INIT_LIST_HEAD(&s->list);
1109 spin_lock_init(&s->pe_lock);
1111 s->merge_failed = 0;
1112 s->first_merging_chunk = 0;
1113 s->num_merging_chunks = 0;
1114 bio_list_init(&s->bios_queued_during_merge);
1116 /* Allocate hash table for COW data */
1117 if (init_hash_tables(s)) {
1118 ti->error = "Unable to allocate hash table space";
1120 goto bad_hash_tables;
1123 r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client);
1125 ti->error = "Could not create kcopyd client";
1129 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
1130 if (!s->pending_pool) {
1131 ti->error = "Could not allocate mempool for pending exceptions";
1132 goto bad_pending_pool;
1135 s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS,
1136 tracked_chunk_cache);
1137 if (!s->tracked_chunk_pool) {
1138 ti->error = "Could not allocate tracked_chunk mempool for "
1140 goto bad_tracked_chunk_pool;
1143 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1144 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
1146 spin_lock_init(&s->tracked_chunk_lock);
1149 ti->num_flush_requests = num_flush_requests;
1151 /* Add snapshot to the list of snapshots for this origin */
1152 /* Exceptions aren't triggered till snapshot_resume() is called */
1153 r = register_snapshot(s);
1155 ti->error = "Snapshot origin struct allocation failed";
1156 goto bad_load_and_register;
1158 /* invalid handover, register_snapshot has set ti->error */
1159 goto bad_load_and_register;
1163 * Metadata must only be loaded into one table at once, so skip this
1164 * if metadata will be handed over during resume.
1165 * Chunk size will be set during the handover - set it to zero to
1166 * ensure it's ignored.
1169 s->store->chunk_size = 0;
1173 r = s->store->type->read_metadata(s->store, dm_add_exception,
1176 ti->error = "Failed to read snapshot metadata";
1177 goto bad_read_metadata;
1180 DMWARN("Snapshot is marked invalid.");
1183 if (!s->store->chunk_size) {
1184 ti->error = "Chunk size not set";
1185 goto bad_read_metadata;
1187 ti->split_io = s->store->chunk_size;
1192 unregister_snapshot(s);
1194 bad_load_and_register:
1195 mempool_destroy(s->tracked_chunk_pool);
1197 bad_tracked_chunk_pool:
1198 mempool_destroy(s->pending_pool);
1201 dm_kcopyd_client_destroy(s->kcopyd_client);
1204 dm_exception_table_exit(&s->pending, pending_cache);
1205 dm_exception_table_exit(&s->complete, exception_cache);
1208 dm_exception_store_destroy(s->store);
1211 dm_put_device(ti, s->cow);
1214 dm_put_device(ti, s->origin);
1223 static void __free_exceptions(struct dm_snapshot *s)
1225 dm_kcopyd_client_destroy(s->kcopyd_client);
1226 s->kcopyd_client = NULL;
1228 dm_exception_table_exit(&s->pending, pending_cache);
1229 dm_exception_table_exit(&s->complete, exception_cache);
1232 static void __handover_exceptions(struct dm_snapshot *snap_src,
1233 struct dm_snapshot *snap_dest)
1236 struct dm_exception_table table_swap;
1237 struct dm_exception_store *store_swap;
1241 * Swap all snapshot context information between the two instances.
1243 u.table_swap = snap_dest->complete;
1244 snap_dest->complete = snap_src->complete;
1245 snap_src->complete = u.table_swap;
1247 u.store_swap = snap_dest->store;
1248 snap_dest->store = snap_src->store;
1249 snap_src->store = u.store_swap;
1251 snap_dest->store->snap = snap_dest;
1252 snap_src->store->snap = snap_src;
1254 snap_dest->ti->split_io = snap_dest->store->chunk_size;
1255 snap_dest->valid = snap_src->valid;
1258 * Set source invalid to ensure it receives no further I/O.
1260 snap_src->valid = 0;
1263 static void snapshot_dtr(struct dm_target *ti)
1265 #ifdef CONFIG_DM_DEBUG
1268 struct dm_snapshot *s = ti->private;
1269 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1271 down_read(&_origins_lock);
1272 /* Check whether exception handover must be cancelled */
1273 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1274 if (snap_src && snap_dest && (s == snap_src)) {
1275 down_write(&snap_dest->lock);
1276 snap_dest->valid = 0;
1277 up_write(&snap_dest->lock);
1278 DMERR("Cancelling snapshot handover.");
1280 up_read(&_origins_lock);
1282 if (dm_target_is_snapshot_merge(ti))
1285 /* Prevent further origin writes from using this snapshot. */
1286 /* After this returns there can be no new kcopyd jobs. */
1287 unregister_snapshot(s);
1289 while (atomic_read(&s->pending_exceptions_count))
1292 * Ensure instructions in mempool_destroy aren't reordered
1293 * before atomic_read.
1297 #ifdef CONFIG_DM_DEBUG
1298 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1299 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
1302 mempool_destroy(s->tracked_chunk_pool);
1304 __free_exceptions(s);
1306 mempool_destroy(s->pending_pool);
1308 dm_exception_store_destroy(s->store);
1310 dm_put_device(ti, s->cow);
1312 dm_put_device(ti, s->origin);
1318 * Flush a list of buffers.
1320 static void flush_bios(struct bio *bio)
1326 bio->bi_next = NULL;
1327 generic_make_request(bio);
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 (bio->bi_rw & REQ_FLUSH) {
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 (bio->bi_rw & REQ_FLUSH) {
1666 if (!map_context->target_request_nr)
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 merging snapshots are redirected to the origin */
1680 goto redirect_to_origin;
1682 /* If the block is already remapped - use that */
1683 e = dm_lookup_exception(&s->complete, chunk);
1685 /* Queue writes overlapping with chunks being merged */
1686 if (bio_rw(bio) == WRITE &&
1687 chunk >= s->first_merging_chunk &&
1688 chunk < (s->first_merging_chunk +
1689 s->num_merging_chunks)) {
1690 bio->bi_bdev = s->origin->bdev;
1691 bio_list_add(&s->bios_queued_during_merge, bio);
1692 r = DM_MAPIO_SUBMITTED;
1696 remap_exception(s, e, bio, chunk);
1698 if (bio_rw(bio) == WRITE)
1699 map_context->ptr = track_chunk(s, chunk);
1704 bio->bi_bdev = s->origin->bdev;
1706 if (bio_rw(bio) == WRITE) {
1708 return do_origin(s->origin, bio);
1717 static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1718 int error, union map_info *map_context)
1720 struct dm_snapshot *s = ti->private;
1721 struct dm_snap_tracked_chunk *c = map_context->ptr;
1724 stop_tracking_chunk(s, c);
1729 static void snapshot_merge_presuspend(struct dm_target *ti)
1731 struct dm_snapshot *s = ti->private;
1736 static void snapshot_postsuspend(struct dm_target *ti)
1738 struct dm_snapshot *s = ti->private;
1740 down_write(&s->lock);
1745 static int snapshot_preresume(struct dm_target *ti)
1748 struct dm_snapshot *s = ti->private;
1749 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1751 down_read(&_origins_lock);
1752 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1753 if (snap_src && snap_dest) {
1754 down_read(&snap_src->lock);
1755 if (s == snap_src) {
1756 DMERR("Unable to resume snapshot source until "
1757 "handover completes.");
1759 } else if (!snap_src->suspended) {
1760 DMERR("Unable to perform snapshot handover until "
1761 "source is suspended.");
1764 up_read(&snap_src->lock);
1766 up_read(&_origins_lock);
1771 static void snapshot_resume(struct dm_target *ti)
1773 struct dm_snapshot *s = ti->private;
1774 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1776 down_read(&_origins_lock);
1777 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1778 if (snap_src && snap_dest) {
1779 down_write(&snap_src->lock);
1780 down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1781 __handover_exceptions(snap_src, snap_dest);
1782 up_write(&snap_dest->lock);
1783 up_write(&snap_src->lock);
1785 up_read(&_origins_lock);
1787 /* Now we have correct chunk size, reregister */
1788 reregister_snapshot(s);
1790 down_write(&s->lock);
1796 static sector_t get_origin_minimum_chunksize(struct block_device *bdev)
1798 sector_t min_chunksize;
1800 down_read(&_origins_lock);
1801 min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
1802 up_read(&_origins_lock);
1804 return min_chunksize;
1807 static void snapshot_merge_resume(struct dm_target *ti)
1809 struct dm_snapshot *s = ti->private;
1812 * Handover exceptions from existing snapshot.
1814 snapshot_resume(ti);
1817 * snapshot-merge acts as an origin, so set ti->split_io
1819 ti->split_io = get_origin_minimum_chunksize(s->origin->bdev);
1824 static int snapshot_status(struct dm_target *ti, status_type_t type,
1825 char *result, unsigned int maxlen)
1828 struct dm_snapshot *snap = ti->private;
1831 case STATUSTYPE_INFO:
1833 down_write(&snap->lock);
1837 else if (snap->merge_failed)
1838 DMEMIT("Merge failed");
1840 if (snap->store->type->usage) {
1841 sector_t total_sectors, sectors_allocated,
1843 snap->store->type->usage(snap->store,
1847 DMEMIT("%llu/%llu %llu",
1848 (unsigned long long)sectors_allocated,
1849 (unsigned long long)total_sectors,
1850 (unsigned long long)metadata_sectors);
1856 up_write(&snap->lock);
1860 case STATUSTYPE_TABLE:
1862 * kdevname returns a static pointer so we need
1863 * to make private copies if the output is to
1866 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
1867 snap->store->type->status(snap->store, type, result + sz,
1875 static int snapshot_iterate_devices(struct dm_target *ti,
1876 iterate_devices_callout_fn fn, void *data)
1878 struct dm_snapshot *snap = ti->private;
1881 r = fn(ti, snap->origin, 0, ti->len, data);
1884 r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data);
1890 /*-----------------------------------------------------------------
1892 *---------------------------------------------------------------*/
1895 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
1896 * supplied bio was ignored. The caller may submit it immediately.
1897 * (No remapping actually occurs as the origin is always a direct linear
1900 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
1901 * and any supplied bio is added to a list to be submitted once all
1902 * the necessary exceptions exist.
1904 static int __origin_write(struct list_head *snapshots, sector_t sector,
1907 int r = DM_MAPIO_REMAPPED;
1908 struct dm_snapshot *snap;
1909 struct dm_exception *e;
1910 struct dm_snap_pending_exception *pe;
1911 struct dm_snap_pending_exception *pe_to_start_now = NULL;
1912 struct dm_snap_pending_exception *pe_to_start_last = NULL;
1915 /* Do all the snapshots on this origin */
1916 list_for_each_entry (snap, snapshots, list) {
1918 * Don't make new exceptions in a merging snapshot
1919 * because it has effectively been deleted
1921 if (dm_target_is_snapshot_merge(snap->ti))
1924 down_write(&snap->lock);
1926 /* Only deal with valid and active snapshots */
1927 if (!snap->valid || !snap->active)
1930 /* Nothing to do if writing beyond end of snapshot */
1931 if (sector >= dm_table_get_size(snap->ti->table))
1935 * Remember, different snapshots can have
1936 * different chunk sizes.
1938 chunk = sector_to_chunk(snap->store, sector);
1941 * Check exception table to see if block
1942 * is already remapped in this snapshot
1943 * and trigger an exception if not.
1945 e = dm_lookup_exception(&snap->complete, chunk);
1949 pe = __lookup_pending_exception(snap, chunk);
1951 up_write(&snap->lock);
1952 pe = alloc_pending_exception(snap);
1953 down_write(&snap->lock);
1956 free_pending_exception(pe);
1960 e = dm_lookup_exception(&snap->complete, chunk);
1962 free_pending_exception(pe);
1966 pe = __find_pending_exception(snap, pe, chunk);
1968 __invalidate_snapshot(snap, -ENOMEM);
1973 r = DM_MAPIO_SUBMITTED;
1976 * If an origin bio was supplied, queue it to wait for the
1977 * completion of this exception, and start this one last,
1978 * at the end of the function.
1981 bio_list_add(&pe->origin_bios, bio);
1986 pe_to_start_last = pe;
1992 pe_to_start_now = pe;
1996 up_write(&snap->lock);
1998 if (pe_to_start_now) {
1999 start_copy(pe_to_start_now);
2000 pe_to_start_now = NULL;
2005 * Submit the exception against which the bio is queued last,
2006 * to give the other exceptions a head start.
2008 if (pe_to_start_last)
2009 start_copy(pe_to_start_last);
2015 * Called on a write from the origin driver.
2017 static int do_origin(struct dm_dev *origin, struct bio *bio)
2020 int r = DM_MAPIO_REMAPPED;
2022 down_read(&_origins_lock);
2023 o = __lookup_origin(origin->bdev);
2025 r = __origin_write(&o->snapshots, bio->bi_sector, bio);
2026 up_read(&_origins_lock);
2032 * Trigger exceptions in all non-merging snapshots.
2034 * The chunk size of the merging snapshot may be larger than the chunk
2035 * size of some other snapshot so we may need to reallocate multiple
2036 * chunks in other snapshots.
2038 * We scan all the overlapping exceptions in the other snapshots.
2039 * Returns 1 if anything was reallocated and must be waited for,
2040 * otherwise returns 0.
2042 * size must be a multiple of merging_snap's chunk_size.
2044 static int origin_write_extent(struct dm_snapshot *merging_snap,
2045 sector_t sector, unsigned size)
2052 * The origin's __minimum_chunk_size() got stored in split_io
2053 * by snapshot_merge_resume().
2055 down_read(&_origins_lock);
2056 o = __lookup_origin(merging_snap->origin->bdev);
2057 for (n = 0; n < size; n += merging_snap->ti->split_io)
2058 if (__origin_write(&o->snapshots, sector + n, NULL) ==
2061 up_read(&_origins_lock);
2067 * Origin: maps a linear range of a device, with hooks for snapshotting.
2071 * Construct an origin mapping: <dev_path>
2072 * The context for an origin is merely a 'struct dm_dev *'
2073 * pointing to the real device.
2075 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2081 ti->error = "origin: incorrect number of arguments";
2085 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev);
2087 ti->error = "Cannot get target device";
2092 ti->num_flush_requests = 1;
2097 static void origin_dtr(struct dm_target *ti)
2099 struct dm_dev *dev = ti->private;
2100 dm_put_device(ti, dev);
2103 static int origin_map(struct dm_target *ti, struct bio *bio,
2104 union map_info *map_context)
2106 struct dm_dev *dev = ti->private;
2107 bio->bi_bdev = dev->bdev;
2109 if (bio->bi_rw & REQ_FLUSH)
2110 return DM_MAPIO_REMAPPED;
2112 /* Only tell snapshots if this is a write */
2113 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
2117 * Set the target "split_io" field to the minimum of all the snapshots'
2120 static void origin_resume(struct dm_target *ti)
2122 struct dm_dev *dev = ti->private;
2124 ti->split_io = get_origin_minimum_chunksize(dev->bdev);
2127 static int origin_status(struct dm_target *ti, status_type_t type, char *result,
2128 unsigned int maxlen)
2130 struct dm_dev *dev = ti->private;
2133 case STATUSTYPE_INFO:
2137 case STATUSTYPE_TABLE:
2138 snprintf(result, maxlen, "%s", dev->name);
2145 static int origin_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2146 struct bio_vec *biovec, int max_size)
2148 struct dm_dev *dev = ti->private;
2149 struct request_queue *q = bdev_get_queue(dev->bdev);
2151 if (!q->merge_bvec_fn)
2154 bvm->bi_bdev = dev->bdev;
2155 bvm->bi_sector = bvm->bi_sector;
2157 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2160 static int origin_iterate_devices(struct dm_target *ti,
2161 iterate_devices_callout_fn fn, void *data)
2163 struct dm_dev *dev = ti->private;
2165 return fn(ti, dev, 0, ti->len, data);
2168 static struct target_type origin_target = {
2169 .name = "snapshot-origin",
2170 .version = {1, 7, 0},
2171 .module = THIS_MODULE,
2175 .resume = origin_resume,
2176 .status = origin_status,
2177 .merge = origin_merge,
2178 .iterate_devices = origin_iterate_devices,
2181 static struct target_type snapshot_target = {
2183 .version = {1, 9, 0},
2184 .module = THIS_MODULE,
2185 .ctr = snapshot_ctr,
2186 .dtr = snapshot_dtr,
2187 .map = snapshot_map,
2188 .end_io = snapshot_end_io,
2189 .postsuspend = snapshot_postsuspend,
2190 .preresume = snapshot_preresume,
2191 .resume = snapshot_resume,
2192 .status = snapshot_status,
2193 .iterate_devices = snapshot_iterate_devices,
2196 static struct target_type merge_target = {
2197 .name = dm_snapshot_merge_target_name,
2198 .version = {1, 0, 0},
2199 .module = THIS_MODULE,
2200 .ctr = snapshot_ctr,
2201 .dtr = snapshot_dtr,
2202 .map = snapshot_merge_map,
2203 .end_io = snapshot_end_io,
2204 .presuspend = snapshot_merge_presuspend,
2205 .postsuspend = snapshot_postsuspend,
2206 .preresume = snapshot_preresume,
2207 .resume = snapshot_merge_resume,
2208 .status = snapshot_status,
2209 .iterate_devices = snapshot_iterate_devices,
2212 static int __init dm_snapshot_init(void)
2216 r = dm_exception_store_init();
2218 DMERR("Failed to initialize exception stores");
2222 r = dm_register_target(&snapshot_target);
2224 DMERR("snapshot target register failed %d", r);
2225 goto bad_register_snapshot_target;
2228 r = dm_register_target(&origin_target);
2230 DMERR("Origin target register failed %d", r);
2231 goto bad_register_origin_target;
2234 r = dm_register_target(&merge_target);
2236 DMERR("Merge target register failed %d", r);
2237 goto bad_register_merge_target;
2240 r = init_origin_hash();
2242 DMERR("init_origin_hash failed.");
2243 goto bad_origin_hash;
2246 exception_cache = KMEM_CACHE(dm_exception, 0);
2247 if (!exception_cache) {
2248 DMERR("Couldn't create exception cache.");
2250 goto bad_exception_cache;
2253 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
2254 if (!pending_cache) {
2255 DMERR("Couldn't create pending cache.");
2257 goto bad_pending_cache;
2260 tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0);
2261 if (!tracked_chunk_cache) {
2262 DMERR("Couldn't create cache to track chunks in use.");
2264 goto bad_tracked_chunk_cache;
2269 bad_tracked_chunk_cache:
2270 kmem_cache_destroy(pending_cache);
2272 kmem_cache_destroy(exception_cache);
2273 bad_exception_cache:
2276 dm_unregister_target(&merge_target);
2277 bad_register_merge_target:
2278 dm_unregister_target(&origin_target);
2279 bad_register_origin_target:
2280 dm_unregister_target(&snapshot_target);
2281 bad_register_snapshot_target:
2282 dm_exception_store_exit();
2287 static void __exit dm_snapshot_exit(void)
2289 dm_unregister_target(&snapshot_target);
2290 dm_unregister_target(&origin_target);
2291 dm_unregister_target(&merge_target);
2294 kmem_cache_destroy(pending_cache);
2295 kmem_cache_destroy(exception_cache);
2296 kmem_cache_destroy(tracked_chunk_cache);
2298 dm_exception_store_exit();
2302 module_init(dm_snapshot_init);
2303 module_exit(dm_snapshot_exit);
2305 MODULE_DESCRIPTION(DM_NAME " snapshot target");
2306 MODULE_AUTHOR("Joe Thornber");
2307 MODULE_LICENSE("GPL");