4 * Copyright (C) 2001-2002 Sistina Software (UK) Limited.
6 * This file is released under the GPL.
9 #include <linux/blkdev.h>
10 #include <linux/device-mapper.h>
11 #include <linux/delay.h>
13 #include <linux/init.h>
14 #include <linux/kdev_t.h>
15 #include <linux/list.h>
16 #include <linux/mempool.h>
17 #include <linux/module.h>
18 #include <linux/slab.h>
19 #include <linux/vmalloc.h>
20 #include <linux/log2.h>
21 #include <linux/dm-kcopyd.h>
23 #include "dm-exception-store.h"
25 #define DM_MSG_PREFIX "snapshots"
27 static const char dm_snapshot_merge_target_name[] = "snapshot-merge";
29 #define dm_target_is_snapshot_merge(ti) \
30 ((ti)->type->name == dm_snapshot_merge_target_name)
33 * The size of the mempool used to track chunks in use.
37 #define DM_TRACKED_CHUNK_HASH_SIZE 16
38 #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
39 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
41 struct dm_exception_table {
44 struct list_head *table;
48 struct rw_semaphore lock;
50 struct dm_dev *origin;
55 /* List of snapshots per Origin */
56 struct list_head list;
59 * You can't use a snapshot if this is 0 (e.g. if full).
60 * A snapshot-merge target never clears this.
64 /* Origin writes don't trigger exceptions until this is set */
67 atomic_t pending_exceptions_count;
69 mempool_t *pending_pool;
71 struct dm_exception_table pending;
72 struct dm_exception_table complete;
75 * pe_lock protects all pending_exception operations and access
76 * as well as the snapshot_bios list.
80 /* Chunks with outstanding reads */
81 spinlock_t tracked_chunk_lock;
82 struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
84 /* The on disk metadata handler */
85 struct dm_exception_store *store;
87 struct dm_kcopyd_client *kcopyd_client;
89 /* Wait for events based on state_bits */
90 unsigned long state_bits;
92 /* Range of chunks currently being merged. */
93 chunk_t first_merging_chunk;
94 int num_merging_chunks;
97 * The merge operation failed if this flag is set.
98 * Failure modes are handled as follows:
99 * - I/O error reading the header
100 * => don't load the target; abort.
101 * - Header does not have "valid" flag set
102 * => use the origin; forget about the snapshot.
103 * - I/O error when reading exceptions
104 * => don't load the target; abort.
105 * (We can't use the intermediate origin state.)
106 * - I/O error while merging
107 * => stop merging; set merge_failed; process I/O normally.
112 * Incoming bios that overlap with chunks being merged must wait
113 * for them to be committed.
115 struct bio_list bios_queued_during_merge;
120 * RUNNING_MERGE - Merge operation is in progress.
121 * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped;
122 * cleared afterwards.
124 #define RUNNING_MERGE 0
125 #define SHUTDOWN_MERGE 1
127 struct dm_dev *dm_snap_origin(struct dm_snapshot *s)
131 EXPORT_SYMBOL(dm_snap_origin);
133 struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
137 EXPORT_SYMBOL(dm_snap_cow);
139 static sector_t chunk_to_sector(struct dm_exception_store *store,
142 return chunk << store->chunk_shift;
145 static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
148 * There is only ever one instance of a particular block
149 * device so we can compare pointers safely.
154 struct dm_snap_pending_exception {
155 struct dm_exception e;
158 * Origin buffers waiting for this to complete are held
161 struct bio_list origin_bios;
162 struct bio_list snapshot_bios;
164 /* Pointer back to snapshot context */
165 struct dm_snapshot *snap;
168 * 1 indicates the exception has already been sent to
174 * For writing a complete chunk, bypassing the copy.
176 struct bio *full_bio;
177 bio_end_io_t *full_bio_end_io;
178 void *full_bio_private;
182 * Hash table mapping origin volumes to lists of snapshots and
183 * a lock to protect it
185 static struct kmem_cache *exception_cache;
186 static struct kmem_cache *pending_cache;
188 struct dm_snap_tracked_chunk {
189 struct hlist_node node;
193 static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s,
197 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
201 spin_lock_irq(&s->tracked_chunk_lock);
202 hlist_add_head(&c->node,
203 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
204 spin_unlock_irq(&s->tracked_chunk_lock);
209 static void stop_tracking_chunk(struct dm_snapshot *s,
210 struct dm_snap_tracked_chunk *c)
214 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
216 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
219 static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
221 struct dm_snap_tracked_chunk *c;
222 struct hlist_node *hn;
225 spin_lock_irq(&s->tracked_chunk_lock);
227 hlist_for_each_entry(c, hn,
228 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
229 if (c->chunk == chunk) {
235 spin_unlock_irq(&s->tracked_chunk_lock);
241 * This conflicting I/O is extremely improbable in the caller,
242 * so msleep(1) is sufficient and there is no need for a wait queue.
244 static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
246 while (__chunk_is_tracked(s, chunk))
251 * One of these per registered origin, held in the snapshot_origins hash
254 /* The origin device */
255 struct block_device *bdev;
257 struct list_head hash_list;
259 /* List of snapshots for this origin */
260 struct list_head snapshots;
264 * Size of the hash table for origin volumes. If we make this
265 * the size of the minors list then it should be nearly perfect
267 #define ORIGIN_HASH_SIZE 256
268 #define ORIGIN_MASK 0xFF
269 static struct list_head *_origins;
270 static struct rw_semaphore _origins_lock;
272 static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
273 static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
274 static uint64_t _pending_exceptions_done_count;
276 static int init_origin_hash(void)
280 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
283 DMERR("unable to allocate memory");
287 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
288 INIT_LIST_HEAD(_origins + i);
289 init_rwsem(&_origins_lock);
294 static void exit_origin_hash(void)
299 static unsigned origin_hash(struct block_device *bdev)
301 return bdev->bd_dev & ORIGIN_MASK;
304 static struct origin *__lookup_origin(struct block_device *origin)
306 struct list_head *ol;
309 ol = &_origins[origin_hash(origin)];
310 list_for_each_entry (o, ol, hash_list)
311 if (bdev_equal(o->bdev, origin))
317 static void __insert_origin(struct origin *o)
319 struct list_head *sl = &_origins[origin_hash(o->bdev)];
320 list_add_tail(&o->hash_list, sl);
324 * _origins_lock must be held when calling this function.
325 * Returns number of snapshots registered using the supplied cow device, plus:
326 * snap_src - a snapshot suitable for use as a source of exception handover
327 * snap_dest - a snapshot capable of receiving exception handover.
328 * snap_merge - an existing snapshot-merge target linked to the same origin.
329 * There can be at most one snapshot-merge target. The parameter is optional.
331 * Possible return values and states of snap_src and snap_dest.
332 * 0: NULL, NULL - first new snapshot
333 * 1: snap_src, NULL - normal snapshot
334 * 2: snap_src, snap_dest - waiting for handover
335 * 2: snap_src, NULL - handed over, waiting for old to be deleted
336 * 1: NULL, snap_dest - source got destroyed without handover
338 static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
339 struct dm_snapshot **snap_src,
340 struct dm_snapshot **snap_dest,
341 struct dm_snapshot **snap_merge)
343 struct dm_snapshot *s;
348 o = __lookup_origin(snap->origin->bdev);
352 list_for_each_entry(s, &o->snapshots, list) {
353 if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
355 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
365 } else if (snap_dest)
376 * On success, returns 1 if this snapshot is a handover destination,
377 * otherwise returns 0.
379 static int __validate_exception_handover(struct dm_snapshot *snap)
381 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
382 struct dm_snapshot *snap_merge = NULL;
384 /* Does snapshot need exceptions handed over to it? */
385 if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
386 &snap_merge) == 2) ||
388 snap->ti->error = "Snapshot cow pairing for exception "
389 "table handover failed";
394 * If no snap_src was found, snap cannot become a handover
401 * Non-snapshot-merge handover?
403 if (!dm_target_is_snapshot_merge(snap->ti))
407 * Do not allow more than one merging snapshot.
410 snap->ti->error = "A snapshot is already merging.";
414 if (!snap_src->store->type->prepare_merge ||
415 !snap_src->store->type->commit_merge) {
416 snap->ti->error = "Snapshot exception store does not "
417 "support snapshot-merge.";
424 static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
426 struct dm_snapshot *l;
428 /* Sort the list according to chunk size, largest-first smallest-last */
429 list_for_each_entry(l, &o->snapshots, list)
430 if (l->store->chunk_size < s->store->chunk_size)
432 list_add_tail(&s->list, &l->list);
436 * Make a note of the snapshot and its origin so we can look it
437 * up when the origin has a write on it.
439 * Also validate snapshot exception store handovers.
440 * On success, returns 1 if this registration is a handover destination,
441 * otherwise returns 0.
443 static int register_snapshot(struct dm_snapshot *snap)
445 struct origin *o, *new_o = NULL;
446 struct block_device *bdev = snap->origin->bdev;
449 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
453 down_write(&_origins_lock);
455 r = __validate_exception_handover(snap);
461 o = __lookup_origin(bdev);
468 /* Initialise the struct */
469 INIT_LIST_HEAD(&o->snapshots);
475 __insert_snapshot(o, snap);
478 up_write(&_origins_lock);
484 * Move snapshot to correct place in list according to chunk size.
486 static void reregister_snapshot(struct dm_snapshot *s)
488 struct block_device *bdev = s->origin->bdev;
490 down_write(&_origins_lock);
493 __insert_snapshot(__lookup_origin(bdev), s);
495 up_write(&_origins_lock);
498 static void unregister_snapshot(struct dm_snapshot *s)
502 down_write(&_origins_lock);
503 o = __lookup_origin(s->origin->bdev);
506 if (o && list_empty(&o->snapshots)) {
507 list_del(&o->hash_list);
511 up_write(&_origins_lock);
515 * Implementation of the exception hash tables.
516 * The lowest hash_shift bits of the chunk number are ignored, allowing
517 * some consecutive chunks to be grouped together.
519 static int dm_exception_table_init(struct dm_exception_table *et,
520 uint32_t size, unsigned hash_shift)
524 et->hash_shift = hash_shift;
525 et->hash_mask = size - 1;
526 et->table = dm_vcalloc(size, sizeof(struct list_head));
530 for (i = 0; i < size; i++)
531 INIT_LIST_HEAD(et->table + i);
536 static void dm_exception_table_exit(struct dm_exception_table *et,
537 struct kmem_cache *mem)
539 struct list_head *slot;
540 struct dm_exception *ex, *next;
543 size = et->hash_mask + 1;
544 for (i = 0; i < size; i++) {
545 slot = et->table + i;
547 list_for_each_entry_safe (ex, next, slot, hash_list)
548 kmem_cache_free(mem, ex);
554 static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
556 return (chunk >> et->hash_shift) & et->hash_mask;
559 static void dm_remove_exception(struct dm_exception *e)
561 list_del(&e->hash_list);
565 * Return the exception data for a sector, or NULL if not
568 static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
571 struct list_head *slot;
572 struct dm_exception *e;
574 slot = &et->table[exception_hash(et, chunk)];
575 list_for_each_entry (e, slot, hash_list)
576 if (chunk >= e->old_chunk &&
577 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
583 static struct dm_exception *alloc_completed_exception(void)
585 struct dm_exception *e;
587 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
589 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
594 static void free_completed_exception(struct dm_exception *e)
596 kmem_cache_free(exception_cache, e);
599 static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
601 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
604 atomic_inc(&s->pending_exceptions_count);
610 static void free_pending_exception(struct dm_snap_pending_exception *pe)
612 struct dm_snapshot *s = pe->snap;
614 mempool_free(pe, s->pending_pool);
615 smp_mb__before_atomic_dec();
616 atomic_dec(&s->pending_exceptions_count);
619 static void dm_insert_exception(struct dm_exception_table *eh,
620 struct dm_exception *new_e)
623 struct dm_exception *e = NULL;
625 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
627 /* Add immediately if this table doesn't support consecutive chunks */
631 /* List is ordered by old_chunk */
632 list_for_each_entry_reverse(e, l, hash_list) {
633 /* Insert after an existing chunk? */
634 if (new_e->old_chunk == (e->old_chunk +
635 dm_consecutive_chunk_count(e) + 1) &&
636 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
637 dm_consecutive_chunk_count(e) + 1)) {
638 dm_consecutive_chunk_count_inc(e);
639 free_completed_exception(new_e);
643 /* Insert before an existing chunk? */
644 if (new_e->old_chunk == (e->old_chunk - 1) &&
645 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
646 dm_consecutive_chunk_count_inc(e);
649 free_completed_exception(new_e);
653 if (new_e->old_chunk > e->old_chunk)
658 list_add(&new_e->hash_list, e ? &e->hash_list : l);
662 * Callback used by the exception stores to load exceptions when
665 static int dm_add_exception(void *context, chunk_t old, chunk_t new)
667 struct dm_snapshot *s = context;
668 struct dm_exception *e;
670 e = alloc_completed_exception();
676 /* Consecutive_count is implicitly initialised to zero */
679 dm_insert_exception(&s->complete, e);
685 * Return a minimum chunk size of all snapshots that have the specified origin.
686 * Return zero if the origin has no snapshots.
688 static uint32_t __minimum_chunk_size(struct origin *o)
690 struct dm_snapshot *snap;
691 unsigned chunk_size = 0;
694 list_for_each_entry(snap, &o->snapshots, list)
695 chunk_size = min_not_zero(chunk_size,
696 snap->store->chunk_size);
698 return (uint32_t) chunk_size;
704 static int calc_max_buckets(void)
706 /* use a fixed size of 2MB */
707 unsigned long mem = 2 * 1024 * 1024;
708 mem /= sizeof(struct list_head);
714 * Allocate room for a suitable hash table.
716 static int init_hash_tables(struct dm_snapshot *s)
718 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
721 * Calculate based on the size of the original volume or
724 cow_dev_size = get_dev_size(s->cow->bdev);
725 origin_dev_size = get_dev_size(s->origin->bdev);
726 max_buckets = calc_max_buckets();
728 hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift;
729 hash_size = min(hash_size, max_buckets);
733 hash_size = rounddown_pow_of_two(hash_size);
734 if (dm_exception_table_init(&s->complete, hash_size,
735 DM_CHUNK_CONSECUTIVE_BITS))
739 * Allocate hash table for in-flight exceptions
740 * Make this smaller than the real hash table
746 if (dm_exception_table_init(&s->pending, hash_size, 0)) {
747 dm_exception_table_exit(&s->complete, exception_cache);
754 static void merge_shutdown(struct dm_snapshot *s)
756 clear_bit_unlock(RUNNING_MERGE, &s->state_bits);
757 smp_mb__after_clear_bit();
758 wake_up_bit(&s->state_bits, RUNNING_MERGE);
761 static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s)
763 s->first_merging_chunk = 0;
764 s->num_merging_chunks = 0;
766 return bio_list_get(&s->bios_queued_during_merge);
770 * Remove one chunk from the index of completed exceptions.
772 static int __remove_single_exception_chunk(struct dm_snapshot *s,
775 struct dm_exception *e;
777 e = dm_lookup_exception(&s->complete, old_chunk);
779 DMERR("Corruption detected: exception for block %llu is "
780 "on disk but not in memory",
781 (unsigned long long)old_chunk);
786 * If this is the only chunk using this exception, remove exception.
788 if (!dm_consecutive_chunk_count(e)) {
789 dm_remove_exception(e);
790 free_completed_exception(e);
795 * The chunk may be either at the beginning or the end of a
796 * group of consecutive chunks - never in the middle. We are
797 * removing chunks in the opposite order to that in which they
798 * were added, so this should always be true.
799 * Decrement the consecutive chunk counter and adjust the
800 * starting point if necessary.
802 if (old_chunk == e->old_chunk) {
805 } else if (old_chunk != e->old_chunk +
806 dm_consecutive_chunk_count(e)) {
807 DMERR("Attempt to merge block %llu from the "
808 "middle of a chunk range [%llu - %llu]",
809 (unsigned long long)old_chunk,
810 (unsigned long long)e->old_chunk,
812 e->old_chunk + dm_consecutive_chunk_count(e));
816 dm_consecutive_chunk_count_dec(e);
821 static void flush_bios(struct bio *bio);
823 static int remove_single_exception_chunk(struct dm_snapshot *s)
825 struct bio *b = NULL;
827 chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1;
829 down_write(&s->lock);
832 * Process chunks (and associated exceptions) in reverse order
833 * so that dm_consecutive_chunk_count_dec() accounting works.
836 r = __remove_single_exception_chunk(s, old_chunk);
839 } while (old_chunk-- > s->first_merging_chunk);
841 b = __release_queued_bios_after_merge(s);
851 static int origin_write_extent(struct dm_snapshot *merging_snap,
852 sector_t sector, unsigned chunk_size);
854 static void merge_callback(int read_err, unsigned long write_err,
857 static uint64_t read_pending_exceptions_done_count(void)
859 uint64_t pending_exceptions_done;
861 spin_lock(&_pending_exceptions_done_spinlock);
862 pending_exceptions_done = _pending_exceptions_done_count;
863 spin_unlock(&_pending_exceptions_done_spinlock);
865 return pending_exceptions_done;
868 static void increment_pending_exceptions_done_count(void)
870 spin_lock(&_pending_exceptions_done_spinlock);
871 _pending_exceptions_done_count++;
872 spin_unlock(&_pending_exceptions_done_spinlock);
874 wake_up_all(&_pending_exceptions_done);
877 static void snapshot_merge_next_chunks(struct dm_snapshot *s)
879 int i, linear_chunks;
880 chunk_t old_chunk, new_chunk;
881 struct dm_io_region src, dest;
883 uint64_t previous_count;
885 BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
886 if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
890 * valid flag never changes during merge, so no lock required.
893 DMERR("Snapshot is invalid: can't merge");
897 linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
899 if (linear_chunks <= 0) {
900 if (linear_chunks < 0) {
901 DMERR("Read error in exception store: "
902 "shutting down merge");
903 down_write(&s->lock);
910 /* Adjust old_chunk and new_chunk to reflect start of linear region */
911 old_chunk = old_chunk + 1 - linear_chunks;
912 new_chunk = new_chunk + 1 - linear_chunks;
915 * Use one (potentially large) I/O to copy all 'linear_chunks'
916 * from the exception store to the origin
918 io_size = linear_chunks * s->store->chunk_size;
920 dest.bdev = s->origin->bdev;
921 dest.sector = chunk_to_sector(s->store, old_chunk);
922 dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector);
924 src.bdev = s->cow->bdev;
925 src.sector = chunk_to_sector(s->store, new_chunk);
926 src.count = dest.count;
929 * Reallocate any exceptions needed in other snapshots then
930 * wait for the pending exceptions to complete.
931 * Each time any pending exception (globally on the system)
932 * completes we are woken and repeat the process to find out
933 * if we can proceed. While this may not seem a particularly
934 * efficient algorithm, it is not expected to have any
935 * significant impact on performance.
937 previous_count = read_pending_exceptions_done_count();
938 while (origin_write_extent(s, dest.sector, io_size)) {
939 wait_event(_pending_exceptions_done,
940 (read_pending_exceptions_done_count() !=
942 /* Retry after the wait, until all exceptions are done. */
943 previous_count = read_pending_exceptions_done_count();
946 down_write(&s->lock);
947 s->first_merging_chunk = old_chunk;
948 s->num_merging_chunks = linear_chunks;
951 /* Wait until writes to all 'linear_chunks' drain */
952 for (i = 0; i < linear_chunks; i++)
953 __check_for_conflicting_io(s, old_chunk + i);
955 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s);
962 static void error_bios(struct bio *bio);
964 static void merge_callback(int read_err, unsigned long write_err, void *context)
966 struct dm_snapshot *s = context;
967 struct bio *b = NULL;
969 if (read_err || write_err) {
971 DMERR("Read error: shutting down merge.");
973 DMERR("Write error: shutting down merge.");
977 if (s->store->type->commit_merge(s->store,
978 s->num_merging_chunks) < 0) {
979 DMERR("Write error in exception store: shutting down merge");
983 if (remove_single_exception_chunk(s) < 0)
986 snapshot_merge_next_chunks(s);
991 down_write(&s->lock);
993 b = __release_queued_bios_after_merge(s);
1000 static void start_merge(struct dm_snapshot *s)
1002 if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits))
1003 snapshot_merge_next_chunks(s);
1006 static int wait_schedule(void *ptr)
1014 * Stop the merging process and wait until it finishes.
1016 static void stop_merge(struct dm_snapshot *s)
1018 set_bit(SHUTDOWN_MERGE, &s->state_bits);
1019 wait_on_bit(&s->state_bits, RUNNING_MERGE, wait_schedule,
1020 TASK_UNINTERRUPTIBLE);
1021 clear_bit(SHUTDOWN_MERGE, &s->state_bits);
1025 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
1027 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1029 struct dm_snapshot *s;
1032 char *origin_path, *cow_path;
1033 unsigned args_used, num_flush_requests = 1;
1034 fmode_t origin_mode = FMODE_READ;
1037 ti->error = "requires exactly 4 arguments";
1042 if (dm_target_is_snapshot_merge(ti)) {
1043 num_flush_requests = 2;
1044 origin_mode = FMODE_WRITE;
1047 s = kmalloc(sizeof(*s), GFP_KERNEL);
1049 ti->error = "Cannot allocate private snapshot structure";
1054 origin_path = argv[0];
1058 r = dm_get_device(ti, origin_path, origin_mode, &s->origin);
1060 ti->error = "Cannot get origin device";
1068 r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow);
1070 ti->error = "Cannot get COW device";
1074 r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
1076 ti->error = "Couldn't create exception store";
1087 atomic_set(&s->pending_exceptions_count, 0);
1088 init_rwsem(&s->lock);
1089 INIT_LIST_HEAD(&s->list);
1090 spin_lock_init(&s->pe_lock);
1092 s->merge_failed = 0;
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 s->kcopyd_client = dm_kcopyd_client_create();
1105 if (IS_ERR(s->kcopyd_client)) {
1106 r = PTR_ERR(s->kcopyd_client);
1107 ti->error = "Could not create kcopyd client";
1111 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
1112 if (!s->pending_pool) {
1113 ti->error = "Could not allocate mempool for pending exceptions";
1114 goto bad_pending_pool;
1117 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1118 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
1120 spin_lock_init(&s->tracked_chunk_lock);
1123 ti->num_flush_requests = num_flush_requests;
1124 ti->per_bio_data_size = sizeof(struct dm_snap_tracked_chunk);
1126 /* Add snapshot to the list of snapshots for this origin */
1127 /* Exceptions aren't triggered till snapshot_resume() is called */
1128 r = register_snapshot(s);
1130 ti->error = "Snapshot origin struct allocation failed";
1131 goto bad_load_and_register;
1133 /* invalid handover, register_snapshot has set ti->error */
1134 goto bad_load_and_register;
1138 * Metadata must only be loaded into one table at once, so skip this
1139 * if metadata will be handed over during resume.
1140 * Chunk size will be set during the handover - set it to zero to
1141 * ensure it's ignored.
1144 s->store->chunk_size = 0;
1148 r = s->store->type->read_metadata(s->store, dm_add_exception,
1151 ti->error = "Failed to read snapshot metadata";
1152 goto bad_read_metadata;
1155 DMWARN("Snapshot is marked invalid.");
1158 if (!s->store->chunk_size) {
1159 ti->error = "Chunk size not set";
1160 goto bad_read_metadata;
1163 r = dm_set_target_max_io_len(ti, s->store->chunk_size);
1165 goto bad_read_metadata;
1170 unregister_snapshot(s);
1172 bad_load_and_register:
1173 mempool_destroy(s->pending_pool);
1176 dm_kcopyd_client_destroy(s->kcopyd_client);
1179 dm_exception_table_exit(&s->pending, pending_cache);
1180 dm_exception_table_exit(&s->complete, exception_cache);
1183 dm_exception_store_destroy(s->store);
1186 dm_put_device(ti, s->cow);
1189 dm_put_device(ti, s->origin);
1198 static void __free_exceptions(struct dm_snapshot *s)
1200 dm_kcopyd_client_destroy(s->kcopyd_client);
1201 s->kcopyd_client = NULL;
1203 dm_exception_table_exit(&s->pending, pending_cache);
1204 dm_exception_table_exit(&s->complete, exception_cache);
1207 static void __handover_exceptions(struct dm_snapshot *snap_src,
1208 struct dm_snapshot *snap_dest)
1211 struct dm_exception_table table_swap;
1212 struct dm_exception_store *store_swap;
1216 * Swap all snapshot context information between the two instances.
1218 u.table_swap = snap_dest->complete;
1219 snap_dest->complete = snap_src->complete;
1220 snap_src->complete = u.table_swap;
1222 u.store_swap = snap_dest->store;
1223 snap_dest->store = snap_src->store;
1224 snap_src->store = u.store_swap;
1226 snap_dest->store->snap = snap_dest;
1227 snap_src->store->snap = snap_src;
1229 snap_dest->ti->max_io_len = snap_dest->store->chunk_size;
1230 snap_dest->valid = snap_src->valid;
1233 * Set source invalid to ensure it receives no further I/O.
1235 snap_src->valid = 0;
1238 static void snapshot_dtr(struct dm_target *ti)
1240 #ifdef CONFIG_DM_DEBUG
1243 struct dm_snapshot *s = ti->private;
1244 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1246 down_read(&_origins_lock);
1247 /* Check whether exception handover must be cancelled */
1248 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1249 if (snap_src && snap_dest && (s == snap_src)) {
1250 down_write(&snap_dest->lock);
1251 snap_dest->valid = 0;
1252 up_write(&snap_dest->lock);
1253 DMERR("Cancelling snapshot handover.");
1255 up_read(&_origins_lock);
1257 if (dm_target_is_snapshot_merge(ti))
1260 /* Prevent further origin writes from using this snapshot. */
1261 /* After this returns there can be no new kcopyd jobs. */
1262 unregister_snapshot(s);
1264 while (atomic_read(&s->pending_exceptions_count))
1267 * Ensure instructions in mempool_destroy aren't reordered
1268 * before atomic_read.
1272 #ifdef CONFIG_DM_DEBUG
1273 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1274 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
1277 __free_exceptions(s);
1279 mempool_destroy(s->pending_pool);
1281 dm_exception_store_destroy(s->store);
1283 dm_put_device(ti, s->cow);
1285 dm_put_device(ti, s->origin);
1291 * Flush a list of buffers.
1293 static void flush_bios(struct bio *bio)
1299 bio->bi_next = NULL;
1300 generic_make_request(bio);
1305 static int do_origin(struct dm_dev *origin, struct bio *bio);
1308 * Flush a list of buffers.
1310 static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
1317 bio->bi_next = NULL;
1318 r = do_origin(s->origin, bio);
1319 if (r == DM_MAPIO_REMAPPED)
1320 generic_make_request(bio);
1326 * Error a list of buffers.
1328 static void error_bios(struct bio *bio)
1334 bio->bi_next = NULL;
1340 static void __invalidate_snapshot(struct dm_snapshot *s, int err)
1346 DMERR("Invalidating snapshot: Error reading/writing.");
1347 else if (err == -ENOMEM)
1348 DMERR("Invalidating snapshot: Unable to allocate exception.");
1350 if (s->store->type->drop_snapshot)
1351 s->store->type->drop_snapshot(s->store);
1355 dm_table_event(s->ti->table);
1358 static void pending_complete(struct dm_snap_pending_exception *pe, int success)
1360 struct dm_exception *e;
1361 struct dm_snapshot *s = pe->snap;
1362 struct bio *origin_bios = NULL;
1363 struct bio *snapshot_bios = NULL;
1364 struct bio *full_bio = NULL;
1368 /* Read/write error - snapshot is unusable */
1369 down_write(&s->lock);
1370 __invalidate_snapshot(s, -EIO);
1375 e = alloc_completed_exception();
1377 down_write(&s->lock);
1378 __invalidate_snapshot(s, -ENOMEM);
1384 down_write(&s->lock);
1386 free_completed_exception(e);
1391 /* Check for conflicting reads */
1392 __check_for_conflicting_io(s, pe->e.old_chunk);
1395 * Add a proper exception, and remove the
1396 * in-flight exception from the list.
1398 dm_insert_exception(&s->complete, e);
1401 dm_remove_exception(&pe->e);
1402 snapshot_bios = bio_list_get(&pe->snapshot_bios);
1403 origin_bios = bio_list_get(&pe->origin_bios);
1404 full_bio = pe->full_bio;
1406 full_bio->bi_end_io = pe->full_bio_end_io;
1407 full_bio->bi_private = pe->full_bio_private;
1409 free_pending_exception(pe);
1411 increment_pending_exceptions_done_count();
1415 /* Submit any pending write bios */
1418 bio_io_error(full_bio);
1419 error_bios(snapshot_bios);
1422 bio_endio(full_bio, 0);
1423 flush_bios(snapshot_bios);
1426 retry_origin_bios(s, origin_bios);
1429 static void commit_callback(void *context, int success)
1431 struct dm_snap_pending_exception *pe = context;
1433 pending_complete(pe, success);
1437 * Called when the copy I/O has finished. kcopyd actually runs
1438 * this code so don't block.
1440 static void copy_callback(int read_err, unsigned long write_err, void *context)
1442 struct dm_snap_pending_exception *pe = context;
1443 struct dm_snapshot *s = pe->snap;
1445 if (read_err || write_err)
1446 pending_complete(pe, 0);
1449 /* Update the metadata if we are persistent */
1450 s->store->type->commit_exception(s->store, &pe->e,
1451 commit_callback, pe);
1455 * Dispatches the copy operation to kcopyd.
1457 static void start_copy(struct dm_snap_pending_exception *pe)
1459 struct dm_snapshot *s = pe->snap;
1460 struct dm_io_region src, dest;
1461 struct block_device *bdev = s->origin->bdev;
1464 dev_size = get_dev_size(bdev);
1467 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
1468 src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
1470 dest.bdev = s->cow->bdev;
1471 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
1472 dest.count = src.count;
1474 /* Hand over to kcopyd */
1475 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe);
1478 static void full_bio_end_io(struct bio *bio, int error)
1480 void *callback_data = bio->bi_private;
1482 dm_kcopyd_do_callback(callback_data, 0, error ? 1 : 0);
1485 static void start_full_bio(struct dm_snap_pending_exception *pe,
1488 struct dm_snapshot *s = pe->snap;
1489 void *callback_data;
1492 pe->full_bio_end_io = bio->bi_end_io;
1493 pe->full_bio_private = bio->bi_private;
1495 callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client,
1498 bio->bi_end_io = full_bio_end_io;
1499 bio->bi_private = callback_data;
1501 generic_make_request(bio);
1504 static struct dm_snap_pending_exception *
1505 __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1507 struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
1512 return container_of(e, struct dm_snap_pending_exception, e);
1516 * Looks to see if this snapshot already has a pending exception
1517 * for this chunk, otherwise it allocates a new one and inserts
1518 * it into the pending table.
1520 * NOTE: a write lock must be held on snap->lock before calling
1523 static struct dm_snap_pending_exception *
1524 __find_pending_exception(struct dm_snapshot *s,
1525 struct dm_snap_pending_exception *pe, chunk_t chunk)
1527 struct dm_snap_pending_exception *pe2;
1529 pe2 = __lookup_pending_exception(s, chunk);
1531 free_pending_exception(pe);
1535 pe->e.old_chunk = chunk;
1536 bio_list_init(&pe->origin_bios);
1537 bio_list_init(&pe->snapshot_bios);
1539 pe->full_bio = NULL;
1541 if (s->store->type->prepare_exception(s->store, &pe->e)) {
1542 free_pending_exception(pe);
1546 dm_insert_exception(&s->pending, &pe->e);
1551 static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
1552 struct bio *bio, chunk_t chunk)
1554 bio->bi_bdev = s->cow->bdev;
1555 bio->bi_sector = chunk_to_sector(s->store,
1556 dm_chunk_number(e->new_chunk) +
1557 (chunk - e->old_chunk)) +
1559 s->store->chunk_mask);
1562 static int snapshot_map(struct dm_target *ti, struct bio *bio,
1563 union map_info *map_context)
1565 struct dm_exception *e;
1566 struct dm_snapshot *s = ti->private;
1567 int r = DM_MAPIO_REMAPPED;
1569 struct dm_snap_pending_exception *pe = NULL;
1571 if (bio->bi_rw & REQ_FLUSH) {
1572 bio->bi_bdev = s->cow->bdev;
1573 return DM_MAPIO_REMAPPED;
1576 chunk = sector_to_chunk(s->store, bio->bi_sector);
1578 /* Full snapshots are not usable */
1579 /* To get here the table must be live so s->active is always set. */
1583 /* FIXME: should only take write lock if we need
1584 * to copy an exception */
1585 down_write(&s->lock);
1592 /* If the block is already remapped - use that, else remap it */
1593 e = dm_lookup_exception(&s->complete, chunk);
1595 remap_exception(s, e, bio, chunk);
1600 * Write to snapshot - higher level takes care of RW/RO
1601 * flags so we should only get this if we are
1604 if (bio_rw(bio) == WRITE) {
1605 pe = __lookup_pending_exception(s, chunk);
1608 pe = alloc_pending_exception(s);
1609 down_write(&s->lock);
1612 free_pending_exception(pe);
1617 e = dm_lookup_exception(&s->complete, chunk);
1619 free_pending_exception(pe);
1620 remap_exception(s, e, bio, chunk);
1624 pe = __find_pending_exception(s, pe, chunk);
1626 __invalidate_snapshot(s, -ENOMEM);
1632 remap_exception(s, &pe->e, bio, chunk);
1634 r = DM_MAPIO_SUBMITTED;
1637 bio->bi_size == (s->store->chunk_size << SECTOR_SHIFT)) {
1640 start_full_bio(pe, bio);
1644 bio_list_add(&pe->snapshot_bios, bio);
1647 /* this is protected by snap->lock */
1654 bio->bi_bdev = s->origin->bdev;
1655 map_context->ptr = track_chunk(s, bio, chunk);
1665 * A snapshot-merge target behaves like a combination of a snapshot
1666 * target and a snapshot-origin target. It only generates new
1667 * exceptions in other snapshots and not in the one that is being
1670 * For each chunk, if there is an existing exception, it is used to
1671 * redirect I/O to the cow device. Otherwise I/O is sent to the origin,
1672 * which in turn might generate exceptions in other snapshots.
1673 * If merging is currently taking place on the chunk in question, the
1674 * I/O is deferred by adding it to s->bios_queued_during_merge.
1676 static int snapshot_merge_map(struct dm_target *ti, struct bio *bio,
1677 union map_info *map_context)
1679 struct dm_exception *e;
1680 struct dm_snapshot *s = ti->private;
1681 int r = DM_MAPIO_REMAPPED;
1684 if (bio->bi_rw & REQ_FLUSH) {
1685 if (!map_context->target_request_nr)
1686 bio->bi_bdev = s->origin->bdev;
1688 bio->bi_bdev = s->cow->bdev;
1689 map_context->ptr = NULL;
1690 return DM_MAPIO_REMAPPED;
1693 chunk = sector_to_chunk(s->store, bio->bi_sector);
1695 down_write(&s->lock);
1697 /* Full merging snapshots are redirected to the origin */
1699 goto redirect_to_origin;
1701 /* If the block is already remapped - use that */
1702 e = dm_lookup_exception(&s->complete, chunk);
1704 /* Queue writes overlapping with chunks being merged */
1705 if (bio_rw(bio) == WRITE &&
1706 chunk >= s->first_merging_chunk &&
1707 chunk < (s->first_merging_chunk +
1708 s->num_merging_chunks)) {
1709 bio->bi_bdev = s->origin->bdev;
1710 bio_list_add(&s->bios_queued_during_merge, bio);
1711 r = DM_MAPIO_SUBMITTED;
1715 remap_exception(s, e, bio, chunk);
1717 if (bio_rw(bio) == WRITE)
1718 map_context->ptr = track_chunk(s, bio, chunk);
1723 bio->bi_bdev = s->origin->bdev;
1725 if (bio_rw(bio) == WRITE) {
1727 return do_origin(s->origin, bio);
1736 static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1737 int error, union map_info *map_context)
1739 struct dm_snapshot *s = ti->private;
1740 struct dm_snap_tracked_chunk *c = map_context->ptr;
1743 stop_tracking_chunk(s, c);
1748 static void snapshot_merge_presuspend(struct dm_target *ti)
1750 struct dm_snapshot *s = ti->private;
1755 static int snapshot_preresume(struct dm_target *ti)
1758 struct dm_snapshot *s = ti->private;
1759 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1761 down_read(&_origins_lock);
1762 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1763 if (snap_src && snap_dest) {
1764 down_read(&snap_src->lock);
1765 if (s == snap_src) {
1766 DMERR("Unable to resume snapshot source until "
1767 "handover completes.");
1769 } else if (!dm_suspended(snap_src->ti)) {
1770 DMERR("Unable to perform snapshot handover until "
1771 "source is suspended.");
1774 up_read(&snap_src->lock);
1776 up_read(&_origins_lock);
1781 static void snapshot_resume(struct dm_target *ti)
1783 struct dm_snapshot *s = ti->private;
1784 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1786 down_read(&_origins_lock);
1787 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1788 if (snap_src && snap_dest) {
1789 down_write(&snap_src->lock);
1790 down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1791 __handover_exceptions(snap_src, snap_dest);
1792 up_write(&snap_dest->lock);
1793 up_write(&snap_src->lock);
1795 up_read(&_origins_lock);
1797 /* Now we have correct chunk size, reregister */
1798 reregister_snapshot(s);
1800 down_write(&s->lock);
1805 static uint32_t get_origin_minimum_chunksize(struct block_device *bdev)
1807 uint32_t min_chunksize;
1809 down_read(&_origins_lock);
1810 min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
1811 up_read(&_origins_lock);
1813 return min_chunksize;
1816 static void snapshot_merge_resume(struct dm_target *ti)
1818 struct dm_snapshot *s = ti->private;
1821 * Handover exceptions from existing snapshot.
1823 snapshot_resume(ti);
1826 * snapshot-merge acts as an origin, so set ti->max_io_len
1828 ti->max_io_len = get_origin_minimum_chunksize(s->origin->bdev);
1833 static int snapshot_status(struct dm_target *ti, status_type_t type,
1834 unsigned status_flags, char *result, unsigned maxlen)
1837 struct dm_snapshot *snap = ti->private;
1840 case STATUSTYPE_INFO:
1842 down_write(&snap->lock);
1846 else if (snap->merge_failed)
1847 DMEMIT("Merge failed");
1849 if (snap->store->type->usage) {
1850 sector_t total_sectors, sectors_allocated,
1852 snap->store->type->usage(snap->store,
1856 DMEMIT("%llu/%llu %llu",
1857 (unsigned long long)sectors_allocated,
1858 (unsigned long long)total_sectors,
1859 (unsigned long long)metadata_sectors);
1865 up_write(&snap->lock);
1869 case STATUSTYPE_TABLE:
1871 * kdevname returns a static pointer so we need
1872 * to make private copies if the output is to
1875 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
1876 snap->store->type->status(snap->store, type, result + sz,
1884 static int snapshot_iterate_devices(struct dm_target *ti,
1885 iterate_devices_callout_fn fn, void *data)
1887 struct dm_snapshot *snap = ti->private;
1890 r = fn(ti, snap->origin, 0, ti->len, data);
1893 r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data);
1899 /*-----------------------------------------------------------------
1901 *---------------------------------------------------------------*/
1904 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
1905 * supplied bio was ignored. The caller may submit it immediately.
1906 * (No remapping actually occurs as the origin is always a direct linear
1909 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
1910 * and any supplied bio is added to a list to be submitted once all
1911 * the necessary exceptions exist.
1913 static int __origin_write(struct list_head *snapshots, sector_t sector,
1916 int r = DM_MAPIO_REMAPPED;
1917 struct dm_snapshot *snap;
1918 struct dm_exception *e;
1919 struct dm_snap_pending_exception *pe;
1920 struct dm_snap_pending_exception *pe_to_start_now = NULL;
1921 struct dm_snap_pending_exception *pe_to_start_last = NULL;
1924 /* Do all the snapshots on this origin */
1925 list_for_each_entry (snap, snapshots, list) {
1927 * Don't make new exceptions in a merging snapshot
1928 * because it has effectively been deleted
1930 if (dm_target_is_snapshot_merge(snap->ti))
1933 down_write(&snap->lock);
1935 /* Only deal with valid and active snapshots */
1936 if (!snap->valid || !snap->active)
1939 /* Nothing to do if writing beyond end of snapshot */
1940 if (sector >= dm_table_get_size(snap->ti->table))
1944 * Remember, different snapshots can have
1945 * different chunk sizes.
1947 chunk = sector_to_chunk(snap->store, sector);
1950 * Check exception table to see if block
1951 * is already remapped in this snapshot
1952 * and trigger an exception if not.
1954 e = dm_lookup_exception(&snap->complete, chunk);
1958 pe = __lookup_pending_exception(snap, chunk);
1960 up_write(&snap->lock);
1961 pe = alloc_pending_exception(snap);
1962 down_write(&snap->lock);
1965 free_pending_exception(pe);
1969 e = dm_lookup_exception(&snap->complete, chunk);
1971 free_pending_exception(pe);
1975 pe = __find_pending_exception(snap, pe, chunk);
1977 __invalidate_snapshot(snap, -ENOMEM);
1982 r = DM_MAPIO_SUBMITTED;
1985 * If an origin bio was supplied, queue it to wait for the
1986 * completion of this exception, and start this one last,
1987 * at the end of the function.
1990 bio_list_add(&pe->origin_bios, bio);
1995 pe_to_start_last = pe;
2001 pe_to_start_now = pe;
2005 up_write(&snap->lock);
2007 if (pe_to_start_now) {
2008 start_copy(pe_to_start_now);
2009 pe_to_start_now = NULL;
2014 * Submit the exception against which the bio is queued last,
2015 * to give the other exceptions a head start.
2017 if (pe_to_start_last)
2018 start_copy(pe_to_start_last);
2024 * Called on a write from the origin driver.
2026 static int do_origin(struct dm_dev *origin, struct bio *bio)
2029 int r = DM_MAPIO_REMAPPED;
2031 down_read(&_origins_lock);
2032 o = __lookup_origin(origin->bdev);
2034 r = __origin_write(&o->snapshots, bio->bi_sector, bio);
2035 up_read(&_origins_lock);
2041 * Trigger exceptions in all non-merging snapshots.
2043 * The chunk size of the merging snapshot may be larger than the chunk
2044 * size of some other snapshot so we may need to reallocate multiple
2045 * chunks in other snapshots.
2047 * We scan all the overlapping exceptions in the other snapshots.
2048 * Returns 1 if anything was reallocated and must be waited for,
2049 * otherwise returns 0.
2051 * size must be a multiple of merging_snap's chunk_size.
2053 static int origin_write_extent(struct dm_snapshot *merging_snap,
2054 sector_t sector, unsigned size)
2061 * The origin's __minimum_chunk_size() got stored in max_io_len
2062 * by snapshot_merge_resume().
2064 down_read(&_origins_lock);
2065 o = __lookup_origin(merging_snap->origin->bdev);
2066 for (n = 0; n < size; n += merging_snap->ti->max_io_len)
2067 if (__origin_write(&o->snapshots, sector + n, NULL) ==
2070 up_read(&_origins_lock);
2076 * Origin: maps a linear range of a device, with hooks for snapshotting.
2080 * Construct an origin mapping: <dev_path>
2081 * The context for an origin is merely a 'struct dm_dev *'
2082 * pointing to the real device.
2084 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2090 ti->error = "origin: incorrect number of arguments";
2094 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev);
2096 ti->error = "Cannot get target device";
2101 ti->num_flush_requests = 1;
2106 static void origin_dtr(struct dm_target *ti)
2108 struct dm_dev *dev = ti->private;
2109 dm_put_device(ti, dev);
2112 static int origin_map(struct dm_target *ti, struct bio *bio,
2113 union map_info *map_context)
2115 struct dm_dev *dev = ti->private;
2116 bio->bi_bdev = dev->bdev;
2118 if (bio->bi_rw & REQ_FLUSH)
2119 return DM_MAPIO_REMAPPED;
2121 /* Only tell snapshots if this is a write */
2122 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
2126 * Set the target "max_io_len" field to the minimum of all the snapshots'
2129 static void origin_resume(struct dm_target *ti)
2131 struct dm_dev *dev = ti->private;
2133 ti->max_io_len = get_origin_minimum_chunksize(dev->bdev);
2136 static int origin_status(struct dm_target *ti, status_type_t type,
2137 unsigned status_flags, char *result, unsigned maxlen)
2139 struct dm_dev *dev = ti->private;
2142 case STATUSTYPE_INFO:
2146 case STATUSTYPE_TABLE:
2147 snprintf(result, maxlen, "%s", dev->name);
2154 static int origin_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2155 struct bio_vec *biovec, int max_size)
2157 struct dm_dev *dev = ti->private;
2158 struct request_queue *q = bdev_get_queue(dev->bdev);
2160 if (!q->merge_bvec_fn)
2163 bvm->bi_bdev = dev->bdev;
2165 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2168 static int origin_iterate_devices(struct dm_target *ti,
2169 iterate_devices_callout_fn fn, void *data)
2171 struct dm_dev *dev = ti->private;
2173 return fn(ti, dev, 0, ti->len, data);
2176 static struct target_type origin_target = {
2177 .name = "snapshot-origin",
2178 .version = {1, 8, 0},
2179 .module = THIS_MODULE,
2183 .resume = origin_resume,
2184 .status = origin_status,
2185 .merge = origin_merge,
2186 .iterate_devices = origin_iterate_devices,
2189 static struct target_type snapshot_target = {
2191 .version = {1, 11, 0},
2192 .module = THIS_MODULE,
2193 .ctr = snapshot_ctr,
2194 .dtr = snapshot_dtr,
2195 .map = snapshot_map,
2196 .end_io = snapshot_end_io,
2197 .preresume = snapshot_preresume,
2198 .resume = snapshot_resume,
2199 .status = snapshot_status,
2200 .iterate_devices = snapshot_iterate_devices,
2203 static struct target_type merge_target = {
2204 .name = dm_snapshot_merge_target_name,
2205 .version = {1, 2, 0},
2206 .module = THIS_MODULE,
2207 .ctr = snapshot_ctr,
2208 .dtr = snapshot_dtr,
2209 .map = snapshot_merge_map,
2210 .end_io = snapshot_end_io,
2211 .presuspend = snapshot_merge_presuspend,
2212 .preresume = snapshot_preresume,
2213 .resume = snapshot_merge_resume,
2214 .status = snapshot_status,
2215 .iterate_devices = snapshot_iterate_devices,
2218 static int __init dm_snapshot_init(void)
2222 r = dm_exception_store_init();
2224 DMERR("Failed to initialize exception stores");
2228 r = dm_register_target(&snapshot_target);
2230 DMERR("snapshot target register failed %d", r);
2231 goto bad_register_snapshot_target;
2234 r = dm_register_target(&origin_target);
2236 DMERR("Origin target register failed %d", r);
2237 goto bad_register_origin_target;
2240 r = dm_register_target(&merge_target);
2242 DMERR("Merge target register failed %d", r);
2243 goto bad_register_merge_target;
2246 r = init_origin_hash();
2248 DMERR("init_origin_hash failed.");
2249 goto bad_origin_hash;
2252 exception_cache = KMEM_CACHE(dm_exception, 0);
2253 if (!exception_cache) {
2254 DMERR("Couldn't create exception cache.");
2256 goto bad_exception_cache;
2259 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
2260 if (!pending_cache) {
2261 DMERR("Couldn't create pending cache.");
2263 goto bad_pending_cache;
2269 kmem_cache_destroy(exception_cache);
2270 bad_exception_cache:
2273 dm_unregister_target(&merge_target);
2274 bad_register_merge_target:
2275 dm_unregister_target(&origin_target);
2276 bad_register_origin_target:
2277 dm_unregister_target(&snapshot_target);
2278 bad_register_snapshot_target:
2279 dm_exception_store_exit();
2284 static void __exit dm_snapshot_exit(void)
2286 dm_unregister_target(&snapshot_target);
2287 dm_unregister_target(&origin_target);
2288 dm_unregister_target(&merge_target);
2291 kmem_cache_destroy(pending_cache);
2292 kmem_cache_destroy(exception_cache);
2294 dm_exception_store_exit();
2298 module_init(dm_snapshot_init);
2299 module_exit(dm_snapshot_exit);
2301 MODULE_DESCRIPTION(DM_NAME " snapshot target");
2302 MODULE_AUTHOR("Joe Thornber");
2303 MODULE_LICENSE("GPL");