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>
25 #include "dm-exception-store.h"
27 #define DM_MSG_PREFIX "snapshots"
29 static const char dm_snapshot_merge_target_name[] = "snapshot-merge";
31 #define dm_target_is_snapshot_merge(ti) \
32 ((ti)->type->name == dm_snapshot_merge_target_name)
35 * The size of the mempool used to track chunks in use.
39 #define DM_TRACKED_CHUNK_HASH_SIZE 16
40 #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
41 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
43 struct dm_exception_table {
46 struct list_head *table;
50 struct rw_semaphore lock;
52 struct dm_dev *origin;
57 /* List of snapshots per Origin */
58 struct list_head list;
61 * You can't use a snapshot if this is 0 (e.g. if full).
62 * A snapshot-merge target never clears this.
66 /* Origin writes don't trigger exceptions until this is set */
69 atomic_t pending_exceptions_count;
71 /* Protected by "lock" */
72 sector_t exception_start_sequence;
74 /* Protected by kcopyd single-threaded callback */
75 sector_t exception_complete_sequence;
78 * A list of pending exceptions that completed out of order.
79 * Protected by kcopyd single-threaded callback.
81 struct list_head out_of_order_list;
83 mempool_t *pending_pool;
85 struct dm_exception_table pending;
86 struct dm_exception_table complete;
89 * pe_lock protects all pending_exception operations and access
90 * as well as the snapshot_bios list.
94 /* Chunks with outstanding reads */
95 spinlock_t tracked_chunk_lock;
96 struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
98 /* The on disk metadata handler */
99 struct dm_exception_store *store;
101 struct dm_kcopyd_client *kcopyd_client;
103 /* Wait for events based on state_bits */
104 unsigned long state_bits;
106 /* Range of chunks currently being merged. */
107 chunk_t first_merging_chunk;
108 int num_merging_chunks;
111 * The merge operation failed if this flag is set.
112 * Failure modes are handled as follows:
113 * - I/O error reading the header
114 * => don't load the target; abort.
115 * - Header does not have "valid" flag set
116 * => use the origin; forget about the snapshot.
117 * - I/O error when reading exceptions
118 * => don't load the target; abort.
119 * (We can't use the intermediate origin state.)
120 * - I/O error while merging
121 * => stop merging; set merge_failed; process I/O normally.
126 * Incoming bios that overlap with chunks being merged must wait
127 * for them to be committed.
129 struct bio_list bios_queued_during_merge;
134 * RUNNING_MERGE - Merge operation is in progress.
135 * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped;
136 * cleared afterwards.
138 #define RUNNING_MERGE 0
139 #define SHUTDOWN_MERGE 1
141 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
142 "A percentage of time allocated for copy on write");
144 struct dm_dev *dm_snap_origin(struct dm_snapshot *s)
148 EXPORT_SYMBOL(dm_snap_origin);
150 struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
154 EXPORT_SYMBOL(dm_snap_cow);
156 static sector_t chunk_to_sector(struct dm_exception_store *store,
159 return chunk << store->chunk_shift;
162 static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
165 * There is only ever one instance of a particular block
166 * device so we can compare pointers safely.
171 struct dm_snap_pending_exception {
172 struct dm_exception e;
175 * Origin buffers waiting for this to complete are held
178 struct bio_list origin_bios;
179 struct bio_list snapshot_bios;
181 /* Pointer back to snapshot context */
182 struct dm_snapshot *snap;
185 * 1 indicates the exception has already been sent to
190 /* There was copying error. */
193 /* A sequence number, it is used for in-order completion. */
194 sector_t exception_sequence;
196 struct list_head out_of_order_entry;
199 * For writing a complete chunk, bypassing the copy.
201 struct bio *full_bio;
202 bio_end_io_t *full_bio_end_io;
203 void *full_bio_private;
207 * Hash table mapping origin volumes to lists of snapshots and
208 * a lock to protect it
210 static struct kmem_cache *exception_cache;
211 static struct kmem_cache *pending_cache;
213 struct dm_snap_tracked_chunk {
214 struct hlist_node node;
218 static void init_tracked_chunk(struct bio *bio)
220 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
221 INIT_HLIST_NODE(&c->node);
224 static bool is_bio_tracked(struct bio *bio)
226 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
227 return !hlist_unhashed(&c->node);
230 static void track_chunk(struct dm_snapshot *s, struct bio *bio, chunk_t chunk)
232 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
236 spin_lock_irq(&s->tracked_chunk_lock);
237 hlist_add_head(&c->node,
238 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
239 spin_unlock_irq(&s->tracked_chunk_lock);
242 static void stop_tracking_chunk(struct dm_snapshot *s, struct bio *bio)
244 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
247 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
249 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
252 static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
254 struct dm_snap_tracked_chunk *c;
257 spin_lock_irq(&s->tracked_chunk_lock);
259 hlist_for_each_entry(c,
260 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
261 if (c->chunk == chunk) {
267 spin_unlock_irq(&s->tracked_chunk_lock);
273 * This conflicting I/O is extremely improbable in the caller,
274 * so msleep(1) is sufficient and there is no need for a wait queue.
276 static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
278 while (__chunk_is_tracked(s, chunk))
283 * One of these per registered origin, held in the snapshot_origins hash
286 /* The origin device */
287 struct block_device *bdev;
289 struct list_head hash_list;
291 /* List of snapshots for this origin */
292 struct list_head snapshots;
296 * This structure is allocated for each origin target
300 struct dm_target *ti;
301 unsigned split_boundary;
302 struct list_head hash_list;
306 * Size of the hash table for origin volumes. If we make this
307 * the size of the minors list then it should be nearly perfect
309 #define ORIGIN_HASH_SIZE 256
310 #define ORIGIN_MASK 0xFF
311 static struct list_head *_origins;
312 static struct list_head *_dm_origins;
313 static struct rw_semaphore _origins_lock;
315 static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
316 static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
317 static uint64_t _pending_exceptions_done_count;
319 static int init_origin_hash(void)
323 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
326 DMERR("unable to allocate memory for _origins");
329 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
330 INIT_LIST_HEAD(_origins + i);
332 _dm_origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
335 DMERR("unable to allocate memory for _dm_origins");
339 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
340 INIT_LIST_HEAD(_dm_origins + i);
342 init_rwsem(&_origins_lock);
347 static void exit_origin_hash(void)
353 static unsigned origin_hash(struct block_device *bdev)
355 return bdev->bd_dev & ORIGIN_MASK;
358 static struct origin *__lookup_origin(struct block_device *origin)
360 struct list_head *ol;
363 ol = &_origins[origin_hash(origin)];
364 list_for_each_entry (o, ol, hash_list)
365 if (bdev_equal(o->bdev, origin))
371 static void __insert_origin(struct origin *o)
373 struct list_head *sl = &_origins[origin_hash(o->bdev)];
374 list_add_tail(&o->hash_list, sl);
377 static struct dm_origin *__lookup_dm_origin(struct block_device *origin)
379 struct list_head *ol;
382 ol = &_dm_origins[origin_hash(origin)];
383 list_for_each_entry (o, ol, hash_list)
384 if (bdev_equal(o->dev->bdev, origin))
390 static void __insert_dm_origin(struct dm_origin *o)
392 struct list_head *sl = &_dm_origins[origin_hash(o->dev->bdev)];
393 list_add_tail(&o->hash_list, sl);
396 static void __remove_dm_origin(struct dm_origin *o)
398 list_del(&o->hash_list);
402 * _origins_lock must be held when calling this function.
403 * Returns number of snapshots registered using the supplied cow device, plus:
404 * snap_src - a snapshot suitable for use as a source of exception handover
405 * snap_dest - a snapshot capable of receiving exception handover.
406 * snap_merge - an existing snapshot-merge target linked to the same origin.
407 * There can be at most one snapshot-merge target. The parameter is optional.
409 * Possible return values and states of snap_src and snap_dest.
410 * 0: NULL, NULL - first new snapshot
411 * 1: snap_src, NULL - normal snapshot
412 * 2: snap_src, snap_dest - waiting for handover
413 * 2: snap_src, NULL - handed over, waiting for old to be deleted
414 * 1: NULL, snap_dest - source got destroyed without handover
416 static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
417 struct dm_snapshot **snap_src,
418 struct dm_snapshot **snap_dest,
419 struct dm_snapshot **snap_merge)
421 struct dm_snapshot *s;
426 o = __lookup_origin(snap->origin->bdev);
430 list_for_each_entry(s, &o->snapshots, list) {
431 if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
433 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
443 } else if (snap_dest)
454 * On success, returns 1 if this snapshot is a handover destination,
455 * otherwise returns 0.
457 static int __validate_exception_handover(struct dm_snapshot *snap)
459 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
460 struct dm_snapshot *snap_merge = NULL;
462 /* Does snapshot need exceptions handed over to it? */
463 if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
464 &snap_merge) == 2) ||
466 snap->ti->error = "Snapshot cow pairing for exception "
467 "table handover failed";
472 * If no snap_src was found, snap cannot become a handover
479 * Non-snapshot-merge handover?
481 if (!dm_target_is_snapshot_merge(snap->ti))
485 * Do not allow more than one merging snapshot.
488 snap->ti->error = "A snapshot is already merging.";
492 if (!snap_src->store->type->prepare_merge ||
493 !snap_src->store->type->commit_merge) {
494 snap->ti->error = "Snapshot exception store does not "
495 "support snapshot-merge.";
502 static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
504 struct dm_snapshot *l;
506 /* Sort the list according to chunk size, largest-first smallest-last */
507 list_for_each_entry(l, &o->snapshots, list)
508 if (l->store->chunk_size < s->store->chunk_size)
510 list_add_tail(&s->list, &l->list);
514 * Make a note of the snapshot and its origin so we can look it
515 * up when the origin has a write on it.
517 * Also validate snapshot exception store handovers.
518 * On success, returns 1 if this registration is a handover destination,
519 * otherwise returns 0.
521 static int register_snapshot(struct dm_snapshot *snap)
523 struct origin *o, *new_o = NULL;
524 struct block_device *bdev = snap->origin->bdev;
527 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
531 down_write(&_origins_lock);
533 r = __validate_exception_handover(snap);
539 o = __lookup_origin(bdev);
546 /* Initialise the struct */
547 INIT_LIST_HEAD(&o->snapshots);
553 __insert_snapshot(o, snap);
556 up_write(&_origins_lock);
562 * Move snapshot to correct place in list according to chunk size.
564 static void reregister_snapshot(struct dm_snapshot *s)
566 struct block_device *bdev = s->origin->bdev;
568 down_write(&_origins_lock);
571 __insert_snapshot(__lookup_origin(bdev), s);
573 up_write(&_origins_lock);
576 static void unregister_snapshot(struct dm_snapshot *s)
580 down_write(&_origins_lock);
581 o = __lookup_origin(s->origin->bdev);
584 if (o && list_empty(&o->snapshots)) {
585 list_del(&o->hash_list);
589 up_write(&_origins_lock);
593 * Implementation of the exception hash tables.
594 * The lowest hash_shift bits of the chunk number are ignored, allowing
595 * some consecutive chunks to be grouped together.
597 static int dm_exception_table_init(struct dm_exception_table *et,
598 uint32_t size, unsigned hash_shift)
602 et->hash_shift = hash_shift;
603 et->hash_mask = size - 1;
604 et->table = dm_vcalloc(size, sizeof(struct list_head));
608 for (i = 0; i < size; i++)
609 INIT_LIST_HEAD(et->table + i);
614 static void dm_exception_table_exit(struct dm_exception_table *et,
615 struct kmem_cache *mem)
617 struct list_head *slot;
618 struct dm_exception *ex, *next;
621 size = et->hash_mask + 1;
622 for (i = 0; i < size; i++) {
623 slot = et->table + i;
625 list_for_each_entry_safe (ex, next, slot, hash_list)
626 kmem_cache_free(mem, ex);
632 static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
634 return (chunk >> et->hash_shift) & et->hash_mask;
637 static void dm_remove_exception(struct dm_exception *e)
639 list_del(&e->hash_list);
643 * Return the exception data for a sector, or NULL if not
646 static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
649 struct list_head *slot;
650 struct dm_exception *e;
652 slot = &et->table[exception_hash(et, chunk)];
653 list_for_each_entry (e, slot, hash_list)
654 if (chunk >= e->old_chunk &&
655 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
661 static struct dm_exception *alloc_completed_exception(gfp_t gfp)
663 struct dm_exception *e;
665 e = kmem_cache_alloc(exception_cache, gfp);
666 if (!e && gfp == GFP_NOIO)
667 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
672 static void free_completed_exception(struct dm_exception *e)
674 kmem_cache_free(exception_cache, e);
677 static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
679 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
682 atomic_inc(&s->pending_exceptions_count);
688 static void free_pending_exception(struct dm_snap_pending_exception *pe)
690 struct dm_snapshot *s = pe->snap;
692 mempool_free(pe, s->pending_pool);
693 smp_mb__before_atomic();
694 atomic_dec(&s->pending_exceptions_count);
697 static void dm_insert_exception(struct dm_exception_table *eh,
698 struct dm_exception *new_e)
701 struct dm_exception *e = NULL;
703 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
705 /* Add immediately if this table doesn't support consecutive chunks */
709 /* List is ordered by old_chunk */
710 list_for_each_entry_reverse(e, l, hash_list) {
711 /* Insert after an existing chunk? */
712 if (new_e->old_chunk == (e->old_chunk +
713 dm_consecutive_chunk_count(e) + 1) &&
714 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
715 dm_consecutive_chunk_count(e) + 1)) {
716 dm_consecutive_chunk_count_inc(e);
717 free_completed_exception(new_e);
721 /* Insert before an existing chunk? */
722 if (new_e->old_chunk == (e->old_chunk - 1) &&
723 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
724 dm_consecutive_chunk_count_inc(e);
727 free_completed_exception(new_e);
731 if (new_e->old_chunk > e->old_chunk)
736 list_add(&new_e->hash_list, e ? &e->hash_list : l);
740 * Callback used by the exception stores to load exceptions when
743 static int dm_add_exception(void *context, chunk_t old, chunk_t new)
745 struct dm_snapshot *s = context;
746 struct dm_exception *e;
748 e = alloc_completed_exception(GFP_KERNEL);
754 /* Consecutive_count is implicitly initialised to zero */
757 dm_insert_exception(&s->complete, e);
763 * Return a minimum chunk size of all snapshots that have the specified origin.
764 * Return zero if the origin has no snapshots.
766 static uint32_t __minimum_chunk_size(struct origin *o)
768 struct dm_snapshot *snap;
769 unsigned chunk_size = 0;
772 list_for_each_entry(snap, &o->snapshots, list)
773 chunk_size = min_not_zero(chunk_size,
774 snap->store->chunk_size);
776 return (uint32_t) chunk_size;
782 static int calc_max_buckets(void)
784 /* use a fixed size of 2MB */
785 unsigned long mem = 2 * 1024 * 1024;
786 mem /= sizeof(struct list_head);
792 * Allocate room for a suitable hash table.
794 static int init_hash_tables(struct dm_snapshot *s)
796 sector_t hash_size, cow_dev_size, max_buckets;
799 * Calculate based on the size of the original volume or
802 cow_dev_size = get_dev_size(s->cow->bdev);
803 max_buckets = calc_max_buckets();
805 hash_size = cow_dev_size >> s->store->chunk_shift;
806 hash_size = min(hash_size, max_buckets);
810 hash_size = rounddown_pow_of_two(hash_size);
811 if (dm_exception_table_init(&s->complete, hash_size,
812 DM_CHUNK_CONSECUTIVE_BITS))
816 * Allocate hash table for in-flight exceptions
817 * Make this smaller than the real hash table
823 if (dm_exception_table_init(&s->pending, hash_size, 0)) {
824 dm_exception_table_exit(&s->complete, exception_cache);
831 static void merge_shutdown(struct dm_snapshot *s)
833 clear_bit_unlock(RUNNING_MERGE, &s->state_bits);
834 smp_mb__after_atomic();
835 wake_up_bit(&s->state_bits, RUNNING_MERGE);
838 static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s)
840 s->first_merging_chunk = 0;
841 s->num_merging_chunks = 0;
843 return bio_list_get(&s->bios_queued_during_merge);
847 * Remove one chunk from the index of completed exceptions.
849 static int __remove_single_exception_chunk(struct dm_snapshot *s,
852 struct dm_exception *e;
854 e = dm_lookup_exception(&s->complete, old_chunk);
856 DMERR("Corruption detected: exception for block %llu is "
857 "on disk but not in memory",
858 (unsigned long long)old_chunk);
863 * If this is the only chunk using this exception, remove exception.
865 if (!dm_consecutive_chunk_count(e)) {
866 dm_remove_exception(e);
867 free_completed_exception(e);
872 * The chunk may be either at the beginning or the end of a
873 * group of consecutive chunks - never in the middle. We are
874 * removing chunks in the opposite order to that in which they
875 * were added, so this should always be true.
876 * Decrement the consecutive chunk counter and adjust the
877 * starting point if necessary.
879 if (old_chunk == e->old_chunk) {
882 } else if (old_chunk != e->old_chunk +
883 dm_consecutive_chunk_count(e)) {
884 DMERR("Attempt to merge block %llu from the "
885 "middle of a chunk range [%llu - %llu]",
886 (unsigned long long)old_chunk,
887 (unsigned long long)e->old_chunk,
889 e->old_chunk + dm_consecutive_chunk_count(e));
893 dm_consecutive_chunk_count_dec(e);
898 static void flush_bios(struct bio *bio);
900 static int remove_single_exception_chunk(struct dm_snapshot *s)
902 struct bio *b = NULL;
904 chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1;
906 down_write(&s->lock);
909 * Process chunks (and associated exceptions) in reverse order
910 * so that dm_consecutive_chunk_count_dec() accounting works.
913 r = __remove_single_exception_chunk(s, old_chunk);
916 } while (old_chunk-- > s->first_merging_chunk);
918 b = __release_queued_bios_after_merge(s);
928 static int origin_write_extent(struct dm_snapshot *merging_snap,
929 sector_t sector, unsigned chunk_size);
931 static void merge_callback(int read_err, unsigned long write_err,
934 static uint64_t read_pending_exceptions_done_count(void)
936 uint64_t pending_exceptions_done;
938 spin_lock(&_pending_exceptions_done_spinlock);
939 pending_exceptions_done = _pending_exceptions_done_count;
940 spin_unlock(&_pending_exceptions_done_spinlock);
942 return pending_exceptions_done;
945 static void increment_pending_exceptions_done_count(void)
947 spin_lock(&_pending_exceptions_done_spinlock);
948 _pending_exceptions_done_count++;
949 spin_unlock(&_pending_exceptions_done_spinlock);
951 wake_up_all(&_pending_exceptions_done);
954 static void snapshot_merge_next_chunks(struct dm_snapshot *s)
956 int i, linear_chunks;
957 chunk_t old_chunk, new_chunk;
958 struct dm_io_region src, dest;
960 uint64_t previous_count;
962 BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
963 if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
967 * valid flag never changes during merge, so no lock required.
970 DMERR("Snapshot is invalid: can't merge");
974 linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
976 if (linear_chunks <= 0) {
977 if (linear_chunks < 0) {
978 DMERR("Read error in exception store: "
979 "shutting down merge");
980 down_write(&s->lock);
987 /* Adjust old_chunk and new_chunk to reflect start of linear region */
988 old_chunk = old_chunk + 1 - linear_chunks;
989 new_chunk = new_chunk + 1 - linear_chunks;
992 * Use one (potentially large) I/O to copy all 'linear_chunks'
993 * from the exception store to the origin
995 io_size = linear_chunks * s->store->chunk_size;
997 dest.bdev = s->origin->bdev;
998 dest.sector = chunk_to_sector(s->store, old_chunk);
999 dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector);
1001 src.bdev = s->cow->bdev;
1002 src.sector = chunk_to_sector(s->store, new_chunk);
1003 src.count = dest.count;
1006 * Reallocate any exceptions needed in other snapshots then
1007 * wait for the pending exceptions to complete.
1008 * Each time any pending exception (globally on the system)
1009 * completes we are woken and repeat the process to find out
1010 * if we can proceed. While this may not seem a particularly
1011 * efficient algorithm, it is not expected to have any
1012 * significant impact on performance.
1014 previous_count = read_pending_exceptions_done_count();
1015 while (origin_write_extent(s, dest.sector, io_size)) {
1016 wait_event(_pending_exceptions_done,
1017 (read_pending_exceptions_done_count() !=
1019 /* Retry after the wait, until all exceptions are done. */
1020 previous_count = read_pending_exceptions_done_count();
1023 down_write(&s->lock);
1024 s->first_merging_chunk = old_chunk;
1025 s->num_merging_chunks = linear_chunks;
1028 /* Wait until writes to all 'linear_chunks' drain */
1029 for (i = 0; i < linear_chunks; i++)
1030 __check_for_conflicting_io(s, old_chunk + i);
1032 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s);
1039 static void error_bios(struct bio *bio);
1041 static void merge_callback(int read_err, unsigned long write_err, void *context)
1043 struct dm_snapshot *s = context;
1044 struct bio *b = NULL;
1046 if (read_err || write_err) {
1048 DMERR("Read error: shutting down merge.");
1050 DMERR("Write error: shutting down merge.");
1054 if (s->store->type->commit_merge(s->store,
1055 s->num_merging_chunks) < 0) {
1056 DMERR("Write error in exception store: shutting down merge");
1060 if (remove_single_exception_chunk(s) < 0)
1063 snapshot_merge_next_chunks(s);
1068 down_write(&s->lock);
1069 s->merge_failed = 1;
1070 b = __release_queued_bios_after_merge(s);
1077 static void start_merge(struct dm_snapshot *s)
1079 if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits))
1080 snapshot_merge_next_chunks(s);
1084 * Stop the merging process and wait until it finishes.
1086 static void stop_merge(struct dm_snapshot *s)
1088 set_bit(SHUTDOWN_MERGE, &s->state_bits);
1089 wait_on_bit(&s->state_bits, RUNNING_MERGE, TASK_UNINTERRUPTIBLE);
1090 clear_bit(SHUTDOWN_MERGE, &s->state_bits);
1094 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
1096 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1098 struct dm_snapshot *s;
1101 char *origin_path, *cow_path;
1102 unsigned args_used, num_flush_bios = 1;
1103 fmode_t origin_mode = FMODE_READ;
1106 ti->error = "requires exactly 4 arguments";
1111 if (dm_target_is_snapshot_merge(ti)) {
1113 origin_mode = FMODE_WRITE;
1116 s = kmalloc(sizeof(*s), GFP_KERNEL);
1118 ti->error = "Cannot allocate private snapshot structure";
1123 origin_path = argv[0];
1127 r = dm_get_device(ti, origin_path, origin_mode, &s->origin);
1129 ti->error = "Cannot get origin device";
1137 r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow);
1139 ti->error = "Cannot get COW device";
1143 r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
1145 ti->error = "Couldn't create exception store";
1156 atomic_set(&s->pending_exceptions_count, 0);
1157 s->exception_start_sequence = 0;
1158 s->exception_complete_sequence = 0;
1159 INIT_LIST_HEAD(&s->out_of_order_list);
1160 init_rwsem(&s->lock);
1161 INIT_LIST_HEAD(&s->list);
1162 spin_lock_init(&s->pe_lock);
1164 s->merge_failed = 0;
1165 s->first_merging_chunk = 0;
1166 s->num_merging_chunks = 0;
1167 bio_list_init(&s->bios_queued_during_merge);
1169 /* Allocate hash table for COW data */
1170 if (init_hash_tables(s)) {
1171 ti->error = "Unable to allocate hash table space";
1173 goto bad_hash_tables;
1176 s->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle);
1177 if (IS_ERR(s->kcopyd_client)) {
1178 r = PTR_ERR(s->kcopyd_client);
1179 ti->error = "Could not create kcopyd client";
1183 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
1184 if (!s->pending_pool) {
1185 ti->error = "Could not allocate mempool for pending exceptions";
1187 goto bad_pending_pool;
1190 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1191 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
1193 spin_lock_init(&s->tracked_chunk_lock);
1196 ti->num_flush_bios = num_flush_bios;
1197 ti->per_bio_data_size = sizeof(struct dm_snap_tracked_chunk);
1199 /* Add snapshot to the list of snapshots for this origin */
1200 /* Exceptions aren't triggered till snapshot_resume() is called */
1201 r = register_snapshot(s);
1203 ti->error = "Snapshot origin struct allocation failed";
1204 goto bad_load_and_register;
1206 /* invalid handover, register_snapshot has set ti->error */
1207 goto bad_load_and_register;
1211 * Metadata must only be loaded into one table at once, so skip this
1212 * if metadata will be handed over during resume.
1213 * Chunk size will be set during the handover - set it to zero to
1214 * ensure it's ignored.
1217 s->store->chunk_size = 0;
1221 r = s->store->type->read_metadata(s->store, dm_add_exception,
1224 ti->error = "Failed to read snapshot metadata";
1225 goto bad_read_metadata;
1228 DMWARN("Snapshot is marked invalid.");
1231 if (!s->store->chunk_size) {
1232 ti->error = "Chunk size not set";
1233 goto bad_read_metadata;
1236 r = dm_set_target_max_io_len(ti, s->store->chunk_size);
1238 goto bad_read_metadata;
1243 unregister_snapshot(s);
1245 bad_load_and_register:
1246 mempool_destroy(s->pending_pool);
1249 dm_kcopyd_client_destroy(s->kcopyd_client);
1252 dm_exception_table_exit(&s->pending, pending_cache);
1253 dm_exception_table_exit(&s->complete, exception_cache);
1256 dm_exception_store_destroy(s->store);
1259 dm_put_device(ti, s->cow);
1262 dm_put_device(ti, s->origin);
1271 static void __free_exceptions(struct dm_snapshot *s)
1273 dm_kcopyd_client_destroy(s->kcopyd_client);
1274 s->kcopyd_client = NULL;
1276 dm_exception_table_exit(&s->pending, pending_cache);
1277 dm_exception_table_exit(&s->complete, exception_cache);
1280 static void __handover_exceptions(struct dm_snapshot *snap_src,
1281 struct dm_snapshot *snap_dest)
1284 struct dm_exception_table table_swap;
1285 struct dm_exception_store *store_swap;
1289 * Swap all snapshot context information between the two instances.
1291 u.table_swap = snap_dest->complete;
1292 snap_dest->complete = snap_src->complete;
1293 snap_src->complete = u.table_swap;
1295 u.store_swap = snap_dest->store;
1296 snap_dest->store = snap_src->store;
1297 snap_src->store = u.store_swap;
1299 snap_dest->store->snap = snap_dest;
1300 snap_src->store->snap = snap_src;
1302 snap_dest->ti->max_io_len = snap_dest->store->chunk_size;
1303 snap_dest->valid = snap_src->valid;
1306 * Set source invalid to ensure it receives no further I/O.
1308 snap_src->valid = 0;
1311 static void snapshot_dtr(struct dm_target *ti)
1313 #ifdef CONFIG_DM_DEBUG
1316 struct dm_snapshot *s = ti->private;
1317 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1319 down_read(&_origins_lock);
1320 /* Check whether exception handover must be cancelled */
1321 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1322 if (snap_src && snap_dest && (s == snap_src)) {
1323 down_write(&snap_dest->lock);
1324 snap_dest->valid = 0;
1325 up_write(&snap_dest->lock);
1326 DMERR("Cancelling snapshot handover.");
1328 up_read(&_origins_lock);
1330 if (dm_target_is_snapshot_merge(ti))
1333 /* Prevent further origin writes from using this snapshot. */
1334 /* After this returns there can be no new kcopyd jobs. */
1335 unregister_snapshot(s);
1337 while (atomic_read(&s->pending_exceptions_count))
1340 * Ensure instructions in mempool_destroy aren't reordered
1341 * before atomic_read.
1345 #ifdef CONFIG_DM_DEBUG
1346 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1347 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
1350 __free_exceptions(s);
1352 mempool_destroy(s->pending_pool);
1354 dm_exception_store_destroy(s->store);
1356 dm_put_device(ti, s->cow);
1358 dm_put_device(ti, s->origin);
1364 * Flush a list of buffers.
1366 static void flush_bios(struct bio *bio)
1372 bio->bi_next = NULL;
1373 generic_make_request(bio);
1378 static int do_origin(struct dm_dev *origin, struct bio *bio);
1381 * Flush a list of buffers.
1383 static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
1390 bio->bi_next = NULL;
1391 r = do_origin(s->origin, bio);
1392 if (r == DM_MAPIO_REMAPPED)
1393 generic_make_request(bio);
1399 * Error a list of buffers.
1401 static void error_bios(struct bio *bio)
1407 bio->bi_next = NULL;
1413 static void __invalidate_snapshot(struct dm_snapshot *s, int err)
1419 DMERR("Invalidating snapshot: Error reading/writing.");
1420 else if (err == -ENOMEM)
1421 DMERR("Invalidating snapshot: Unable to allocate exception.");
1423 if (s->store->type->drop_snapshot)
1424 s->store->type->drop_snapshot(s->store);
1428 dm_table_event(s->ti->table);
1431 static void pending_complete(struct dm_snap_pending_exception *pe, int success)
1433 struct dm_exception *e;
1434 struct dm_snapshot *s = pe->snap;
1435 struct bio *origin_bios = NULL;
1436 struct bio *snapshot_bios = NULL;
1437 struct bio *full_bio = NULL;
1441 /* Read/write error - snapshot is unusable */
1442 down_write(&s->lock);
1443 __invalidate_snapshot(s, -EIO);
1448 e = alloc_completed_exception(GFP_NOIO);
1450 down_write(&s->lock);
1451 __invalidate_snapshot(s, -ENOMEM);
1457 down_write(&s->lock);
1459 free_completed_exception(e);
1464 /* Check for conflicting reads */
1465 __check_for_conflicting_io(s, pe->e.old_chunk);
1468 * Add a proper exception, and remove the
1469 * in-flight exception from the list.
1471 dm_insert_exception(&s->complete, e);
1474 dm_remove_exception(&pe->e);
1475 snapshot_bios = bio_list_get(&pe->snapshot_bios);
1476 origin_bios = bio_list_get(&pe->origin_bios);
1477 full_bio = pe->full_bio;
1479 full_bio->bi_end_io = pe->full_bio_end_io;
1480 full_bio->bi_private = pe->full_bio_private;
1482 increment_pending_exceptions_done_count();
1486 /* Submit any pending write bios */
1489 bio_io_error(full_bio);
1490 error_bios(snapshot_bios);
1493 bio_endio(full_bio, 0);
1494 flush_bios(snapshot_bios);
1497 retry_origin_bios(s, origin_bios);
1499 free_pending_exception(pe);
1502 static void commit_callback(void *context, int success)
1504 struct dm_snap_pending_exception *pe = context;
1506 pending_complete(pe, success);
1509 static void complete_exception(struct dm_snap_pending_exception *pe)
1511 struct dm_snapshot *s = pe->snap;
1513 if (unlikely(pe->copy_error))
1514 pending_complete(pe, 0);
1517 /* Update the metadata if we are persistent */
1518 s->store->type->commit_exception(s->store, &pe->e,
1519 commit_callback, pe);
1523 * Called when the copy I/O has finished. kcopyd actually runs
1524 * this code so don't block.
1526 static void copy_callback(int read_err, unsigned long write_err, void *context)
1528 struct dm_snap_pending_exception *pe = context;
1529 struct dm_snapshot *s = pe->snap;
1531 pe->copy_error = read_err || write_err;
1533 if (pe->exception_sequence == s->exception_complete_sequence) {
1534 s->exception_complete_sequence++;
1535 complete_exception(pe);
1537 while (!list_empty(&s->out_of_order_list)) {
1538 pe = list_entry(s->out_of_order_list.next,
1539 struct dm_snap_pending_exception, out_of_order_entry);
1540 if (pe->exception_sequence != s->exception_complete_sequence)
1542 s->exception_complete_sequence++;
1543 list_del(&pe->out_of_order_entry);
1544 complete_exception(pe);
1547 struct list_head *lh;
1548 struct dm_snap_pending_exception *pe2;
1550 list_for_each_prev(lh, &s->out_of_order_list) {
1551 pe2 = list_entry(lh, struct dm_snap_pending_exception, out_of_order_entry);
1552 if (pe2->exception_sequence < pe->exception_sequence)
1555 list_add(&pe->out_of_order_entry, lh);
1560 * Dispatches the copy operation to kcopyd.
1562 static void start_copy(struct dm_snap_pending_exception *pe)
1564 struct dm_snapshot *s = pe->snap;
1565 struct dm_io_region src, dest;
1566 struct block_device *bdev = s->origin->bdev;
1569 dev_size = get_dev_size(bdev);
1572 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
1573 src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
1575 dest.bdev = s->cow->bdev;
1576 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
1577 dest.count = src.count;
1579 /* Hand over to kcopyd */
1580 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe);
1583 static void full_bio_end_io(struct bio *bio, int error)
1585 void *callback_data = bio->bi_private;
1587 dm_kcopyd_do_callback(callback_data, 0, error ? 1 : 0);
1590 static void start_full_bio(struct dm_snap_pending_exception *pe,
1593 struct dm_snapshot *s = pe->snap;
1594 void *callback_data;
1597 pe->full_bio_end_io = bio->bi_end_io;
1598 pe->full_bio_private = bio->bi_private;
1600 callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client,
1603 bio->bi_end_io = full_bio_end_io;
1604 bio->bi_private = callback_data;
1606 generic_make_request(bio);
1609 static struct dm_snap_pending_exception *
1610 __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1612 struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
1617 return container_of(e, struct dm_snap_pending_exception, e);
1621 * Looks to see if this snapshot already has a pending exception
1622 * for this chunk, otherwise it allocates a new one and inserts
1623 * it into the pending table.
1625 * NOTE: a write lock must be held on snap->lock before calling
1628 static struct dm_snap_pending_exception *
1629 __find_pending_exception(struct dm_snapshot *s,
1630 struct dm_snap_pending_exception *pe, chunk_t chunk)
1632 struct dm_snap_pending_exception *pe2;
1634 pe2 = __lookup_pending_exception(s, chunk);
1636 free_pending_exception(pe);
1640 pe->e.old_chunk = chunk;
1641 bio_list_init(&pe->origin_bios);
1642 bio_list_init(&pe->snapshot_bios);
1644 pe->full_bio = NULL;
1646 if (s->store->type->prepare_exception(s->store, &pe->e)) {
1647 free_pending_exception(pe);
1651 pe->exception_sequence = s->exception_start_sequence++;
1653 dm_insert_exception(&s->pending, &pe->e);
1658 static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
1659 struct bio *bio, chunk_t chunk)
1661 bio->bi_bdev = s->cow->bdev;
1662 bio->bi_iter.bi_sector =
1663 chunk_to_sector(s->store, dm_chunk_number(e->new_chunk) +
1664 (chunk - e->old_chunk)) +
1665 (bio->bi_iter.bi_sector & s->store->chunk_mask);
1668 static int snapshot_map(struct dm_target *ti, struct bio *bio)
1670 struct dm_exception *e;
1671 struct dm_snapshot *s = ti->private;
1672 int r = DM_MAPIO_REMAPPED;
1674 struct dm_snap_pending_exception *pe = NULL;
1676 init_tracked_chunk(bio);
1678 if (bio->bi_rw & REQ_FLUSH) {
1679 bio->bi_bdev = s->cow->bdev;
1680 return DM_MAPIO_REMAPPED;
1683 chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector);
1685 /* Full snapshots are not usable */
1686 /* To get here the table must be live so s->active is always set. */
1690 /* FIXME: should only take write lock if we need
1691 * to copy an exception */
1692 down_write(&s->lock);
1699 /* If the block is already remapped - use that, else remap it */
1700 e = dm_lookup_exception(&s->complete, chunk);
1702 remap_exception(s, e, bio, chunk);
1707 * Write to snapshot - higher level takes care of RW/RO
1708 * flags so we should only get this if we are
1711 if (bio_rw(bio) == WRITE) {
1712 pe = __lookup_pending_exception(s, chunk);
1715 pe = alloc_pending_exception(s);
1716 down_write(&s->lock);
1719 free_pending_exception(pe);
1724 e = dm_lookup_exception(&s->complete, chunk);
1726 free_pending_exception(pe);
1727 remap_exception(s, e, bio, chunk);
1731 pe = __find_pending_exception(s, pe, chunk);
1733 __invalidate_snapshot(s, -ENOMEM);
1739 remap_exception(s, &pe->e, bio, chunk);
1741 r = DM_MAPIO_SUBMITTED;
1744 bio->bi_iter.bi_size ==
1745 (s->store->chunk_size << SECTOR_SHIFT)) {
1748 start_full_bio(pe, bio);
1752 bio_list_add(&pe->snapshot_bios, bio);
1755 /* this is protected by snap->lock */
1762 bio->bi_bdev = s->origin->bdev;
1763 track_chunk(s, bio, chunk);
1773 * A snapshot-merge target behaves like a combination of a snapshot
1774 * target and a snapshot-origin target. It only generates new
1775 * exceptions in other snapshots and not in the one that is being
1778 * For each chunk, if there is an existing exception, it is used to
1779 * redirect I/O to the cow device. Otherwise I/O is sent to the origin,
1780 * which in turn might generate exceptions in other snapshots.
1781 * If merging is currently taking place on the chunk in question, the
1782 * I/O is deferred by adding it to s->bios_queued_during_merge.
1784 static int snapshot_merge_map(struct dm_target *ti, struct bio *bio)
1786 struct dm_exception *e;
1787 struct dm_snapshot *s = ti->private;
1788 int r = DM_MAPIO_REMAPPED;
1791 init_tracked_chunk(bio);
1793 if (bio->bi_rw & REQ_FLUSH) {
1794 if (!dm_bio_get_target_bio_nr(bio))
1795 bio->bi_bdev = s->origin->bdev;
1797 bio->bi_bdev = s->cow->bdev;
1798 return DM_MAPIO_REMAPPED;
1801 chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector);
1803 down_write(&s->lock);
1805 /* Full merging snapshots are redirected to the origin */
1807 goto redirect_to_origin;
1809 /* If the block is already remapped - use that */
1810 e = dm_lookup_exception(&s->complete, chunk);
1812 /* Queue writes overlapping with chunks being merged */
1813 if (bio_rw(bio) == WRITE &&
1814 chunk >= s->first_merging_chunk &&
1815 chunk < (s->first_merging_chunk +
1816 s->num_merging_chunks)) {
1817 bio->bi_bdev = s->origin->bdev;
1818 bio_list_add(&s->bios_queued_during_merge, bio);
1819 r = DM_MAPIO_SUBMITTED;
1823 remap_exception(s, e, bio, chunk);
1825 if (bio_rw(bio) == WRITE)
1826 track_chunk(s, bio, chunk);
1831 bio->bi_bdev = s->origin->bdev;
1833 if (bio_rw(bio) == WRITE) {
1835 return do_origin(s->origin, bio);
1844 static int snapshot_end_io(struct dm_target *ti, struct bio *bio, int error)
1846 struct dm_snapshot *s = ti->private;
1848 if (is_bio_tracked(bio))
1849 stop_tracking_chunk(s, bio);
1854 static void snapshot_merge_presuspend(struct dm_target *ti)
1856 struct dm_snapshot *s = ti->private;
1861 static int snapshot_preresume(struct dm_target *ti)
1864 struct dm_snapshot *s = ti->private;
1865 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1867 down_read(&_origins_lock);
1868 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1869 if (snap_src && snap_dest) {
1870 down_read(&snap_src->lock);
1871 if (s == snap_src) {
1872 DMERR("Unable to resume snapshot source until "
1873 "handover completes.");
1875 } else if (!dm_suspended(snap_src->ti)) {
1876 DMERR("Unable to perform snapshot handover until "
1877 "source is suspended.");
1880 up_read(&snap_src->lock);
1882 up_read(&_origins_lock);
1887 static void snapshot_resume(struct dm_target *ti)
1889 struct dm_snapshot *s = ti->private;
1890 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL, *snap_merging = NULL;
1891 struct dm_origin *o;
1892 struct mapped_device *origin_md = NULL;
1893 bool must_restart_merging = false;
1895 down_read(&_origins_lock);
1897 o = __lookup_dm_origin(s->origin->bdev);
1899 origin_md = dm_table_get_md(o->ti->table);
1901 (void) __find_snapshots_sharing_cow(s, NULL, NULL, &snap_merging);
1903 origin_md = dm_table_get_md(snap_merging->ti->table);
1905 if (origin_md == dm_table_get_md(ti->table))
1908 if (dm_hold(origin_md))
1912 up_read(&_origins_lock);
1915 dm_internal_suspend_fast(origin_md);
1916 if (snap_merging && test_bit(RUNNING_MERGE, &snap_merging->state_bits)) {
1917 must_restart_merging = true;
1918 stop_merge(snap_merging);
1922 down_read(&_origins_lock);
1924 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1925 if (snap_src && snap_dest) {
1926 down_write(&snap_src->lock);
1927 down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1928 __handover_exceptions(snap_src, snap_dest);
1929 up_write(&snap_dest->lock);
1930 up_write(&snap_src->lock);
1933 up_read(&_origins_lock);
1936 if (must_restart_merging)
1937 start_merge(snap_merging);
1938 dm_internal_resume_fast(origin_md);
1942 /* Now we have correct chunk size, reregister */
1943 reregister_snapshot(s);
1945 down_write(&s->lock);
1950 static uint32_t get_origin_minimum_chunksize(struct block_device *bdev)
1952 uint32_t min_chunksize;
1954 down_read(&_origins_lock);
1955 min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
1956 up_read(&_origins_lock);
1958 return min_chunksize;
1961 static void snapshot_merge_resume(struct dm_target *ti)
1963 struct dm_snapshot *s = ti->private;
1966 * Handover exceptions from existing snapshot.
1968 snapshot_resume(ti);
1971 * snapshot-merge acts as an origin, so set ti->max_io_len
1973 ti->max_io_len = get_origin_minimum_chunksize(s->origin->bdev);
1978 static void snapshot_status(struct dm_target *ti, status_type_t type,
1979 unsigned status_flags, char *result, unsigned maxlen)
1982 struct dm_snapshot *snap = ti->private;
1985 case STATUSTYPE_INFO:
1987 down_write(&snap->lock);
1991 else if (snap->merge_failed)
1992 DMEMIT("Merge failed");
1994 if (snap->store->type->usage) {
1995 sector_t total_sectors, sectors_allocated,
1997 snap->store->type->usage(snap->store,
2001 DMEMIT("%llu/%llu %llu",
2002 (unsigned long long)sectors_allocated,
2003 (unsigned long long)total_sectors,
2004 (unsigned long long)metadata_sectors);
2010 up_write(&snap->lock);
2014 case STATUSTYPE_TABLE:
2016 * kdevname returns a static pointer so we need
2017 * to make private copies if the output is to
2020 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
2021 snap->store->type->status(snap->store, type, result + sz,
2027 static int snapshot_iterate_devices(struct dm_target *ti,
2028 iterate_devices_callout_fn fn, void *data)
2030 struct dm_snapshot *snap = ti->private;
2033 r = fn(ti, snap->origin, 0, ti->len, data);
2036 r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data);
2042 /*-----------------------------------------------------------------
2044 *---------------------------------------------------------------*/
2047 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
2048 * supplied bio was ignored. The caller may submit it immediately.
2049 * (No remapping actually occurs as the origin is always a direct linear
2052 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
2053 * and any supplied bio is added to a list to be submitted once all
2054 * the necessary exceptions exist.
2056 static int __origin_write(struct list_head *snapshots, sector_t sector,
2059 int r = DM_MAPIO_REMAPPED;
2060 struct dm_snapshot *snap;
2061 struct dm_exception *e;
2062 struct dm_snap_pending_exception *pe;
2063 struct dm_snap_pending_exception *pe_to_start_now = NULL;
2064 struct dm_snap_pending_exception *pe_to_start_last = NULL;
2067 /* Do all the snapshots on this origin */
2068 list_for_each_entry (snap, snapshots, list) {
2070 * Don't make new exceptions in a merging snapshot
2071 * because it has effectively been deleted
2073 if (dm_target_is_snapshot_merge(snap->ti))
2076 down_write(&snap->lock);
2078 /* Only deal with valid and active snapshots */
2079 if (!snap->valid || !snap->active)
2082 /* Nothing to do if writing beyond end of snapshot */
2083 if (sector >= dm_table_get_size(snap->ti->table))
2087 * Remember, different snapshots can have
2088 * different chunk sizes.
2090 chunk = sector_to_chunk(snap->store, sector);
2093 * Check exception table to see if block
2094 * is already remapped in this snapshot
2095 * and trigger an exception if not.
2097 e = dm_lookup_exception(&snap->complete, chunk);
2101 pe = __lookup_pending_exception(snap, chunk);
2103 up_write(&snap->lock);
2104 pe = alloc_pending_exception(snap);
2105 down_write(&snap->lock);
2108 free_pending_exception(pe);
2112 e = dm_lookup_exception(&snap->complete, chunk);
2114 free_pending_exception(pe);
2118 pe = __find_pending_exception(snap, pe, chunk);
2120 __invalidate_snapshot(snap, -ENOMEM);
2125 r = DM_MAPIO_SUBMITTED;
2128 * If an origin bio was supplied, queue it to wait for the
2129 * completion of this exception, and start this one last,
2130 * at the end of the function.
2133 bio_list_add(&pe->origin_bios, bio);
2138 pe_to_start_last = pe;
2144 pe_to_start_now = pe;
2148 up_write(&snap->lock);
2150 if (pe_to_start_now) {
2151 start_copy(pe_to_start_now);
2152 pe_to_start_now = NULL;
2157 * Submit the exception against which the bio is queued last,
2158 * to give the other exceptions a head start.
2160 if (pe_to_start_last)
2161 start_copy(pe_to_start_last);
2167 * Called on a write from the origin driver.
2169 static int do_origin(struct dm_dev *origin, struct bio *bio)
2172 int r = DM_MAPIO_REMAPPED;
2174 down_read(&_origins_lock);
2175 o = __lookup_origin(origin->bdev);
2177 r = __origin_write(&o->snapshots, bio->bi_iter.bi_sector, bio);
2178 up_read(&_origins_lock);
2184 * Trigger exceptions in all non-merging snapshots.
2186 * The chunk size of the merging snapshot may be larger than the chunk
2187 * size of some other snapshot so we may need to reallocate multiple
2188 * chunks in other snapshots.
2190 * We scan all the overlapping exceptions in the other snapshots.
2191 * Returns 1 if anything was reallocated and must be waited for,
2192 * otherwise returns 0.
2194 * size must be a multiple of merging_snap's chunk_size.
2196 static int origin_write_extent(struct dm_snapshot *merging_snap,
2197 sector_t sector, unsigned size)
2204 * The origin's __minimum_chunk_size() got stored in max_io_len
2205 * by snapshot_merge_resume().
2207 down_read(&_origins_lock);
2208 o = __lookup_origin(merging_snap->origin->bdev);
2209 for (n = 0; n < size; n += merging_snap->ti->max_io_len)
2210 if (__origin_write(&o->snapshots, sector + n, NULL) ==
2213 up_read(&_origins_lock);
2219 * Origin: maps a linear range of a device, with hooks for snapshotting.
2223 * Construct an origin mapping: <dev_path>
2224 * The context for an origin is merely a 'struct dm_dev *'
2225 * pointing to the real device.
2227 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2230 struct dm_origin *o;
2233 ti->error = "origin: incorrect number of arguments";
2237 o = kmalloc(sizeof(struct dm_origin), GFP_KERNEL);
2239 ti->error = "Cannot allocate private origin structure";
2244 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &o->dev);
2246 ti->error = "Cannot get target device";
2252 ti->num_flush_bios = 1;
2262 static void origin_dtr(struct dm_target *ti)
2264 struct dm_origin *o = ti->private;
2266 dm_put_device(ti, o->dev);
2270 static int origin_map(struct dm_target *ti, struct bio *bio)
2272 struct dm_origin *o = ti->private;
2273 unsigned available_sectors;
2275 bio->bi_bdev = o->dev->bdev;
2277 if (unlikely(bio->bi_rw & REQ_FLUSH))
2278 return DM_MAPIO_REMAPPED;
2280 if (bio_rw(bio) != WRITE)
2281 return DM_MAPIO_REMAPPED;
2283 available_sectors = o->split_boundary -
2284 ((unsigned)bio->bi_iter.bi_sector & (o->split_boundary - 1));
2286 if (bio_sectors(bio) > available_sectors)
2287 dm_accept_partial_bio(bio, available_sectors);
2289 /* Only tell snapshots if this is a write */
2290 return do_origin(o->dev, bio);
2294 * Set the target "max_io_len" field to the minimum of all the snapshots'
2297 static void origin_resume(struct dm_target *ti)
2299 struct dm_origin *o = ti->private;
2301 o->split_boundary = get_origin_minimum_chunksize(o->dev->bdev);
2303 down_write(&_origins_lock);
2304 __insert_dm_origin(o);
2305 up_write(&_origins_lock);
2308 static void origin_postsuspend(struct dm_target *ti)
2310 struct dm_origin *o = ti->private;
2312 down_write(&_origins_lock);
2313 __remove_dm_origin(o);
2314 up_write(&_origins_lock);
2317 static void origin_status(struct dm_target *ti, status_type_t type,
2318 unsigned status_flags, char *result, unsigned maxlen)
2320 struct dm_origin *o = ti->private;
2323 case STATUSTYPE_INFO:
2327 case STATUSTYPE_TABLE:
2328 snprintf(result, maxlen, "%s", o->dev->name);
2333 static int origin_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2334 struct bio_vec *biovec, int max_size)
2336 struct dm_origin *o = ti->private;
2337 struct request_queue *q = bdev_get_queue(o->dev->bdev);
2339 if (!q->merge_bvec_fn)
2342 bvm->bi_bdev = o->dev->bdev;
2344 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2347 static int origin_iterate_devices(struct dm_target *ti,
2348 iterate_devices_callout_fn fn, void *data)
2350 struct dm_origin *o = ti->private;
2352 return fn(ti, o->dev, 0, ti->len, data);
2355 static struct target_type origin_target = {
2356 .name = "snapshot-origin",
2357 .version = {1, 9, 0},
2358 .module = THIS_MODULE,
2362 .resume = origin_resume,
2363 .postsuspend = origin_postsuspend,
2364 .status = origin_status,
2365 .merge = origin_merge,
2366 .iterate_devices = origin_iterate_devices,
2369 static struct target_type snapshot_target = {
2371 .version = {1, 13, 0},
2372 .module = THIS_MODULE,
2373 .ctr = snapshot_ctr,
2374 .dtr = snapshot_dtr,
2375 .map = snapshot_map,
2376 .end_io = snapshot_end_io,
2377 .preresume = snapshot_preresume,
2378 .resume = snapshot_resume,
2379 .status = snapshot_status,
2380 .iterate_devices = snapshot_iterate_devices,
2383 static struct target_type merge_target = {
2384 .name = dm_snapshot_merge_target_name,
2385 .version = {1, 3, 0},
2386 .module = THIS_MODULE,
2387 .ctr = snapshot_ctr,
2388 .dtr = snapshot_dtr,
2389 .map = snapshot_merge_map,
2390 .end_io = snapshot_end_io,
2391 .presuspend = snapshot_merge_presuspend,
2392 .preresume = snapshot_preresume,
2393 .resume = snapshot_merge_resume,
2394 .status = snapshot_status,
2395 .iterate_devices = snapshot_iterate_devices,
2398 static int __init dm_snapshot_init(void)
2402 r = dm_exception_store_init();
2404 DMERR("Failed to initialize exception stores");
2408 r = dm_register_target(&snapshot_target);
2410 DMERR("snapshot target register failed %d", r);
2411 goto bad_register_snapshot_target;
2414 r = dm_register_target(&origin_target);
2416 DMERR("Origin target register failed %d", r);
2417 goto bad_register_origin_target;
2420 r = dm_register_target(&merge_target);
2422 DMERR("Merge target register failed %d", r);
2423 goto bad_register_merge_target;
2426 r = init_origin_hash();
2428 DMERR("init_origin_hash failed.");
2429 goto bad_origin_hash;
2432 exception_cache = KMEM_CACHE(dm_exception, 0);
2433 if (!exception_cache) {
2434 DMERR("Couldn't create exception cache.");
2436 goto bad_exception_cache;
2439 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
2440 if (!pending_cache) {
2441 DMERR("Couldn't create pending cache.");
2443 goto bad_pending_cache;
2449 kmem_cache_destroy(exception_cache);
2450 bad_exception_cache:
2453 dm_unregister_target(&merge_target);
2454 bad_register_merge_target:
2455 dm_unregister_target(&origin_target);
2456 bad_register_origin_target:
2457 dm_unregister_target(&snapshot_target);
2458 bad_register_snapshot_target:
2459 dm_exception_store_exit();
2464 static void __exit dm_snapshot_exit(void)
2466 dm_unregister_target(&snapshot_target);
2467 dm_unregister_target(&origin_target);
2468 dm_unregister_target(&merge_target);
2471 kmem_cache_destroy(pending_cache);
2472 kmem_cache_destroy(exception_cache);
2474 dm_exception_store_exit();
2478 module_init(dm_snapshot_init);
2479 module_exit(dm_snapshot_exit);
2481 MODULE_DESCRIPTION(DM_NAME " snapshot target");
2482 MODULE_AUTHOR("Joe Thornber");
2483 MODULE_LICENSE("GPL");
2484 MODULE_ALIAS("dm-snapshot-origin");
2485 MODULE_ALIAS("dm-snapshot-merge");