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 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
128 "A percentage of time allocated for copy on write");
130 struct dm_dev *dm_snap_origin(struct dm_snapshot *s)
134 EXPORT_SYMBOL(dm_snap_origin);
136 struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
140 EXPORT_SYMBOL(dm_snap_cow);
142 static sector_t chunk_to_sector(struct dm_exception_store *store,
145 return chunk << store->chunk_shift;
148 static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
151 * There is only ever one instance of a particular block
152 * device so we can compare pointers safely.
157 struct dm_snap_pending_exception {
158 struct dm_exception e;
161 * Origin buffers waiting for this to complete are held
164 struct bio_list origin_bios;
165 struct bio_list snapshot_bios;
167 /* Pointer back to snapshot context */
168 struct dm_snapshot *snap;
171 * 1 indicates the exception has already been sent to
177 * For writing a complete chunk, bypassing the copy.
179 struct bio *full_bio;
180 bio_end_io_t *full_bio_end_io;
181 void *full_bio_private;
185 * Hash table mapping origin volumes to lists of snapshots and
186 * a lock to protect it
188 static struct kmem_cache *exception_cache;
189 static struct kmem_cache *pending_cache;
191 struct dm_snap_tracked_chunk {
192 struct hlist_node node;
196 static void init_tracked_chunk(struct bio *bio)
198 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
199 INIT_HLIST_NODE(&c->node);
202 static bool is_bio_tracked(struct bio *bio)
204 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
205 return !hlist_unhashed(&c->node);
208 static void track_chunk(struct dm_snapshot *s, struct bio *bio, chunk_t chunk)
210 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
214 spin_lock_irq(&s->tracked_chunk_lock);
215 hlist_add_head(&c->node,
216 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
217 spin_unlock_irq(&s->tracked_chunk_lock);
220 static void stop_tracking_chunk(struct dm_snapshot *s, struct bio *bio)
222 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
225 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
227 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
230 static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
232 struct dm_snap_tracked_chunk *c;
235 spin_lock_irq(&s->tracked_chunk_lock);
237 hlist_for_each_entry(c,
238 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
239 if (c->chunk == chunk) {
245 spin_unlock_irq(&s->tracked_chunk_lock);
251 * This conflicting I/O is extremely improbable in the caller,
252 * so msleep(1) is sufficient and there is no need for a wait queue.
254 static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
256 while (__chunk_is_tracked(s, chunk))
261 * One of these per registered origin, held in the snapshot_origins hash
264 /* The origin device */
265 struct block_device *bdev;
267 struct list_head hash_list;
269 /* List of snapshots for this origin */
270 struct list_head snapshots;
274 * Size of the hash table for origin volumes. If we make this
275 * the size of the minors list then it should be nearly perfect
277 #define ORIGIN_HASH_SIZE 256
278 #define ORIGIN_MASK 0xFF
279 static struct list_head *_origins;
280 static struct rw_semaphore _origins_lock;
282 static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
283 static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
284 static uint64_t _pending_exceptions_done_count;
286 static int init_origin_hash(void)
290 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
293 DMERR("unable to allocate memory");
297 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
298 INIT_LIST_HEAD(_origins + i);
299 init_rwsem(&_origins_lock);
304 static void exit_origin_hash(void)
309 static unsigned origin_hash(struct block_device *bdev)
311 return bdev->bd_dev & ORIGIN_MASK;
314 static struct origin *__lookup_origin(struct block_device *origin)
316 struct list_head *ol;
319 ol = &_origins[origin_hash(origin)];
320 list_for_each_entry (o, ol, hash_list)
321 if (bdev_equal(o->bdev, origin))
327 static void __insert_origin(struct origin *o)
329 struct list_head *sl = &_origins[origin_hash(o->bdev)];
330 list_add_tail(&o->hash_list, sl);
334 * _origins_lock must be held when calling this function.
335 * Returns number of snapshots registered using the supplied cow device, plus:
336 * snap_src - a snapshot suitable for use as a source of exception handover
337 * snap_dest - a snapshot capable of receiving exception handover.
338 * snap_merge - an existing snapshot-merge target linked to the same origin.
339 * There can be at most one snapshot-merge target. The parameter is optional.
341 * Possible return values and states of snap_src and snap_dest.
342 * 0: NULL, NULL - first new snapshot
343 * 1: snap_src, NULL - normal snapshot
344 * 2: snap_src, snap_dest - waiting for handover
345 * 2: snap_src, NULL - handed over, waiting for old to be deleted
346 * 1: NULL, snap_dest - source got destroyed without handover
348 static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
349 struct dm_snapshot **snap_src,
350 struct dm_snapshot **snap_dest,
351 struct dm_snapshot **snap_merge)
353 struct dm_snapshot *s;
358 o = __lookup_origin(snap->origin->bdev);
362 list_for_each_entry(s, &o->snapshots, list) {
363 if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
365 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
375 } else if (snap_dest)
386 * On success, returns 1 if this snapshot is a handover destination,
387 * otherwise returns 0.
389 static int __validate_exception_handover(struct dm_snapshot *snap)
391 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
392 struct dm_snapshot *snap_merge = NULL;
394 /* Does snapshot need exceptions handed over to it? */
395 if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
396 &snap_merge) == 2) ||
398 snap->ti->error = "Snapshot cow pairing for exception "
399 "table handover failed";
404 * If no snap_src was found, snap cannot become a handover
411 * Non-snapshot-merge handover?
413 if (!dm_target_is_snapshot_merge(snap->ti))
417 * Do not allow more than one merging snapshot.
420 snap->ti->error = "A snapshot is already merging.";
424 if (!snap_src->store->type->prepare_merge ||
425 !snap_src->store->type->commit_merge) {
426 snap->ti->error = "Snapshot exception store does not "
427 "support snapshot-merge.";
434 static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
436 struct dm_snapshot *l;
438 /* Sort the list according to chunk size, largest-first smallest-last */
439 list_for_each_entry(l, &o->snapshots, list)
440 if (l->store->chunk_size < s->store->chunk_size)
442 list_add_tail(&s->list, &l->list);
446 * Make a note of the snapshot and its origin so we can look it
447 * up when the origin has a write on it.
449 * Also validate snapshot exception store handovers.
450 * On success, returns 1 if this registration is a handover destination,
451 * otherwise returns 0.
453 static int register_snapshot(struct dm_snapshot *snap)
455 struct origin *o, *new_o = NULL;
456 struct block_device *bdev = snap->origin->bdev;
459 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
463 down_write(&_origins_lock);
465 r = __validate_exception_handover(snap);
471 o = __lookup_origin(bdev);
478 /* Initialise the struct */
479 INIT_LIST_HEAD(&o->snapshots);
485 __insert_snapshot(o, snap);
488 up_write(&_origins_lock);
494 * Move snapshot to correct place in list according to chunk size.
496 static void reregister_snapshot(struct dm_snapshot *s)
498 struct block_device *bdev = s->origin->bdev;
500 down_write(&_origins_lock);
503 __insert_snapshot(__lookup_origin(bdev), s);
505 up_write(&_origins_lock);
508 static void unregister_snapshot(struct dm_snapshot *s)
512 down_write(&_origins_lock);
513 o = __lookup_origin(s->origin->bdev);
516 if (o && list_empty(&o->snapshots)) {
517 list_del(&o->hash_list);
521 up_write(&_origins_lock);
525 * Implementation of the exception hash tables.
526 * The lowest hash_shift bits of the chunk number are ignored, allowing
527 * some consecutive chunks to be grouped together.
529 static int dm_exception_table_init(struct dm_exception_table *et,
530 uint32_t size, unsigned hash_shift)
534 et->hash_shift = hash_shift;
535 et->hash_mask = size - 1;
536 et->table = dm_vcalloc(size, sizeof(struct list_head));
540 for (i = 0; i < size; i++)
541 INIT_LIST_HEAD(et->table + i);
546 static void dm_exception_table_exit(struct dm_exception_table *et,
547 struct kmem_cache *mem)
549 struct list_head *slot;
550 struct dm_exception *ex, *next;
553 size = et->hash_mask + 1;
554 for (i = 0; i < size; i++) {
555 slot = et->table + i;
557 list_for_each_entry_safe (ex, next, slot, hash_list)
558 kmem_cache_free(mem, ex);
564 static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
566 return (chunk >> et->hash_shift) & et->hash_mask;
569 static void dm_remove_exception(struct dm_exception *e)
571 list_del(&e->hash_list);
575 * Return the exception data for a sector, or NULL if not
578 static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
581 struct list_head *slot;
582 struct dm_exception *e;
584 slot = &et->table[exception_hash(et, chunk)];
585 list_for_each_entry (e, slot, hash_list)
586 if (chunk >= e->old_chunk &&
587 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
593 static struct dm_exception *alloc_completed_exception(void)
595 struct dm_exception *e;
597 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
599 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
604 static void free_completed_exception(struct dm_exception *e)
606 kmem_cache_free(exception_cache, e);
609 static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
611 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
614 atomic_inc(&s->pending_exceptions_count);
620 static void free_pending_exception(struct dm_snap_pending_exception *pe)
622 struct dm_snapshot *s = pe->snap;
624 mempool_free(pe, s->pending_pool);
625 smp_mb__before_atomic_dec();
626 atomic_dec(&s->pending_exceptions_count);
629 static void dm_insert_exception(struct dm_exception_table *eh,
630 struct dm_exception *new_e)
633 struct dm_exception *e = NULL;
635 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
637 /* Add immediately if this table doesn't support consecutive chunks */
641 /* List is ordered by old_chunk */
642 list_for_each_entry_reverse(e, l, hash_list) {
643 /* Insert after an existing chunk? */
644 if (new_e->old_chunk == (e->old_chunk +
645 dm_consecutive_chunk_count(e) + 1) &&
646 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
647 dm_consecutive_chunk_count(e) + 1)) {
648 dm_consecutive_chunk_count_inc(e);
649 free_completed_exception(new_e);
653 /* Insert before an existing chunk? */
654 if (new_e->old_chunk == (e->old_chunk - 1) &&
655 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
656 dm_consecutive_chunk_count_inc(e);
659 free_completed_exception(new_e);
663 if (new_e->old_chunk > e->old_chunk)
668 list_add(&new_e->hash_list, e ? &e->hash_list : l);
672 * Callback used by the exception stores to load exceptions when
675 static int dm_add_exception(void *context, chunk_t old, chunk_t new)
677 struct dm_snapshot *s = context;
678 struct dm_exception *e;
680 e = alloc_completed_exception();
686 /* Consecutive_count is implicitly initialised to zero */
689 dm_insert_exception(&s->complete, e);
695 * Return a minimum chunk size of all snapshots that have the specified origin.
696 * Return zero if the origin has no snapshots.
698 static uint32_t __minimum_chunk_size(struct origin *o)
700 struct dm_snapshot *snap;
701 unsigned chunk_size = 0;
704 list_for_each_entry(snap, &o->snapshots, list)
705 chunk_size = min_not_zero(chunk_size,
706 snap->store->chunk_size);
708 return (uint32_t) chunk_size;
714 static int calc_max_buckets(void)
716 /* use a fixed size of 2MB */
717 unsigned long mem = 2 * 1024 * 1024;
718 mem /= sizeof(struct list_head);
724 * Allocate room for a suitable hash table.
726 static int init_hash_tables(struct dm_snapshot *s)
728 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
731 * Calculate based on the size of the original volume or
734 cow_dev_size = get_dev_size(s->cow->bdev);
735 origin_dev_size = get_dev_size(s->origin->bdev);
736 max_buckets = calc_max_buckets();
738 hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift;
739 hash_size = min(hash_size, max_buckets);
743 hash_size = rounddown_pow_of_two(hash_size);
744 if (dm_exception_table_init(&s->complete, hash_size,
745 DM_CHUNK_CONSECUTIVE_BITS))
749 * Allocate hash table for in-flight exceptions
750 * Make this smaller than the real hash table
756 if (dm_exception_table_init(&s->pending, hash_size, 0)) {
757 dm_exception_table_exit(&s->complete, exception_cache);
764 static void merge_shutdown(struct dm_snapshot *s)
766 clear_bit_unlock(RUNNING_MERGE, &s->state_bits);
767 smp_mb__after_clear_bit();
768 wake_up_bit(&s->state_bits, RUNNING_MERGE);
771 static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s)
773 s->first_merging_chunk = 0;
774 s->num_merging_chunks = 0;
776 return bio_list_get(&s->bios_queued_during_merge);
780 * Remove one chunk from the index of completed exceptions.
782 static int __remove_single_exception_chunk(struct dm_snapshot *s,
785 struct dm_exception *e;
787 e = dm_lookup_exception(&s->complete, old_chunk);
789 DMERR("Corruption detected: exception for block %llu is "
790 "on disk but not in memory",
791 (unsigned long long)old_chunk);
796 * If this is the only chunk using this exception, remove exception.
798 if (!dm_consecutive_chunk_count(e)) {
799 dm_remove_exception(e);
800 free_completed_exception(e);
805 * The chunk may be either at the beginning or the end of a
806 * group of consecutive chunks - never in the middle. We are
807 * removing chunks in the opposite order to that in which they
808 * were added, so this should always be true.
809 * Decrement the consecutive chunk counter and adjust the
810 * starting point if necessary.
812 if (old_chunk == e->old_chunk) {
815 } else if (old_chunk != e->old_chunk +
816 dm_consecutive_chunk_count(e)) {
817 DMERR("Attempt to merge block %llu from the "
818 "middle of a chunk range [%llu - %llu]",
819 (unsigned long long)old_chunk,
820 (unsigned long long)e->old_chunk,
822 e->old_chunk + dm_consecutive_chunk_count(e));
826 dm_consecutive_chunk_count_dec(e);
831 static void flush_bios(struct bio *bio);
833 static int remove_single_exception_chunk(struct dm_snapshot *s)
835 struct bio *b = NULL;
837 chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1;
839 down_write(&s->lock);
842 * Process chunks (and associated exceptions) in reverse order
843 * so that dm_consecutive_chunk_count_dec() accounting works.
846 r = __remove_single_exception_chunk(s, old_chunk);
849 } while (old_chunk-- > s->first_merging_chunk);
851 b = __release_queued_bios_after_merge(s);
861 static int origin_write_extent(struct dm_snapshot *merging_snap,
862 sector_t sector, unsigned chunk_size);
864 static void merge_callback(int read_err, unsigned long write_err,
867 static uint64_t read_pending_exceptions_done_count(void)
869 uint64_t pending_exceptions_done;
871 spin_lock(&_pending_exceptions_done_spinlock);
872 pending_exceptions_done = _pending_exceptions_done_count;
873 spin_unlock(&_pending_exceptions_done_spinlock);
875 return pending_exceptions_done;
878 static void increment_pending_exceptions_done_count(void)
880 spin_lock(&_pending_exceptions_done_spinlock);
881 _pending_exceptions_done_count++;
882 spin_unlock(&_pending_exceptions_done_spinlock);
884 wake_up_all(&_pending_exceptions_done);
887 static void snapshot_merge_next_chunks(struct dm_snapshot *s)
889 int i, linear_chunks;
890 chunk_t old_chunk, new_chunk;
891 struct dm_io_region src, dest;
893 uint64_t previous_count;
895 BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
896 if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
900 * valid flag never changes during merge, so no lock required.
903 DMERR("Snapshot is invalid: can't merge");
907 linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
909 if (linear_chunks <= 0) {
910 if (linear_chunks < 0) {
911 DMERR("Read error in exception store: "
912 "shutting down merge");
913 down_write(&s->lock);
920 /* Adjust old_chunk and new_chunk to reflect start of linear region */
921 old_chunk = old_chunk + 1 - linear_chunks;
922 new_chunk = new_chunk + 1 - linear_chunks;
925 * Use one (potentially large) I/O to copy all 'linear_chunks'
926 * from the exception store to the origin
928 io_size = linear_chunks * s->store->chunk_size;
930 dest.bdev = s->origin->bdev;
931 dest.sector = chunk_to_sector(s->store, old_chunk);
932 dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector);
934 src.bdev = s->cow->bdev;
935 src.sector = chunk_to_sector(s->store, new_chunk);
936 src.count = dest.count;
939 * Reallocate any exceptions needed in other snapshots then
940 * wait for the pending exceptions to complete.
941 * Each time any pending exception (globally on the system)
942 * completes we are woken and repeat the process to find out
943 * if we can proceed. While this may not seem a particularly
944 * efficient algorithm, it is not expected to have any
945 * significant impact on performance.
947 previous_count = read_pending_exceptions_done_count();
948 while (origin_write_extent(s, dest.sector, io_size)) {
949 wait_event(_pending_exceptions_done,
950 (read_pending_exceptions_done_count() !=
952 /* Retry after the wait, until all exceptions are done. */
953 previous_count = read_pending_exceptions_done_count();
956 down_write(&s->lock);
957 s->first_merging_chunk = old_chunk;
958 s->num_merging_chunks = linear_chunks;
961 /* Wait until writes to all 'linear_chunks' drain */
962 for (i = 0; i < linear_chunks; i++)
963 __check_for_conflicting_io(s, old_chunk + i);
965 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s);
972 static void error_bios(struct bio *bio);
974 static void merge_callback(int read_err, unsigned long write_err, void *context)
976 struct dm_snapshot *s = context;
977 struct bio *b = NULL;
979 if (read_err || write_err) {
981 DMERR("Read error: shutting down merge.");
983 DMERR("Write error: shutting down merge.");
987 if (s->store->type->commit_merge(s->store,
988 s->num_merging_chunks) < 0) {
989 DMERR("Write error in exception store: shutting down merge");
993 if (remove_single_exception_chunk(s) < 0)
996 snapshot_merge_next_chunks(s);
1001 down_write(&s->lock);
1002 s->merge_failed = 1;
1003 b = __release_queued_bios_after_merge(s);
1010 static void start_merge(struct dm_snapshot *s)
1012 if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits))
1013 snapshot_merge_next_chunks(s);
1016 static int wait_schedule(void *ptr)
1024 * Stop the merging process and wait until it finishes.
1026 static void stop_merge(struct dm_snapshot *s)
1028 set_bit(SHUTDOWN_MERGE, &s->state_bits);
1029 wait_on_bit(&s->state_bits, RUNNING_MERGE, wait_schedule,
1030 TASK_UNINTERRUPTIBLE);
1031 clear_bit(SHUTDOWN_MERGE, &s->state_bits);
1035 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
1037 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1039 struct dm_snapshot *s;
1042 char *origin_path, *cow_path;
1043 unsigned args_used, num_flush_bios = 1;
1044 fmode_t origin_mode = FMODE_READ;
1047 ti->error = "requires exactly 4 arguments";
1052 if (dm_target_is_snapshot_merge(ti)) {
1054 origin_mode = FMODE_WRITE;
1057 s = kmalloc(sizeof(*s), GFP_KERNEL);
1059 ti->error = "Cannot allocate private snapshot structure";
1064 origin_path = argv[0];
1068 r = dm_get_device(ti, origin_path, origin_mode, &s->origin);
1070 ti->error = "Cannot get origin device";
1078 r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow);
1080 ti->error = "Cannot get COW device";
1084 r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
1086 ti->error = "Couldn't create exception store";
1097 atomic_set(&s->pending_exceptions_count, 0);
1098 init_rwsem(&s->lock);
1099 INIT_LIST_HEAD(&s->list);
1100 spin_lock_init(&s->pe_lock);
1102 s->merge_failed = 0;
1103 s->first_merging_chunk = 0;
1104 s->num_merging_chunks = 0;
1105 bio_list_init(&s->bios_queued_during_merge);
1107 /* Allocate hash table for COW data */
1108 if (init_hash_tables(s)) {
1109 ti->error = "Unable to allocate hash table space";
1111 goto bad_hash_tables;
1114 s->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle);
1115 if (IS_ERR(s->kcopyd_client)) {
1116 r = PTR_ERR(s->kcopyd_client);
1117 ti->error = "Could not create kcopyd client";
1121 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
1122 if (!s->pending_pool) {
1123 ti->error = "Could not allocate mempool for pending exceptions";
1125 goto bad_pending_pool;
1128 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1129 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
1131 spin_lock_init(&s->tracked_chunk_lock);
1134 ti->num_flush_bios = num_flush_bios;
1135 ti->per_bio_data_size = sizeof(struct dm_snap_tracked_chunk);
1137 /* Add snapshot to the list of snapshots for this origin */
1138 /* Exceptions aren't triggered till snapshot_resume() is called */
1139 r = register_snapshot(s);
1141 ti->error = "Snapshot origin struct allocation failed";
1142 goto bad_load_and_register;
1144 /* invalid handover, register_snapshot has set ti->error */
1145 goto bad_load_and_register;
1149 * Metadata must only be loaded into one table at once, so skip this
1150 * if metadata will be handed over during resume.
1151 * Chunk size will be set during the handover - set it to zero to
1152 * ensure it's ignored.
1155 s->store->chunk_size = 0;
1159 r = s->store->type->read_metadata(s->store, dm_add_exception,
1162 ti->error = "Failed to read snapshot metadata";
1163 goto bad_read_metadata;
1166 DMWARN("Snapshot is marked invalid.");
1169 if (!s->store->chunk_size) {
1170 ti->error = "Chunk size not set";
1171 goto bad_read_metadata;
1174 r = dm_set_target_max_io_len(ti, s->store->chunk_size);
1176 goto bad_read_metadata;
1181 unregister_snapshot(s);
1183 bad_load_and_register:
1184 mempool_destroy(s->pending_pool);
1187 dm_kcopyd_client_destroy(s->kcopyd_client);
1190 dm_exception_table_exit(&s->pending, pending_cache);
1191 dm_exception_table_exit(&s->complete, exception_cache);
1194 dm_exception_store_destroy(s->store);
1197 dm_put_device(ti, s->cow);
1200 dm_put_device(ti, s->origin);
1209 static void __free_exceptions(struct dm_snapshot *s)
1211 dm_kcopyd_client_destroy(s->kcopyd_client);
1212 s->kcopyd_client = NULL;
1214 dm_exception_table_exit(&s->pending, pending_cache);
1215 dm_exception_table_exit(&s->complete, exception_cache);
1218 static void __handover_exceptions(struct dm_snapshot *snap_src,
1219 struct dm_snapshot *snap_dest)
1222 struct dm_exception_table table_swap;
1223 struct dm_exception_store *store_swap;
1227 * Swap all snapshot context information between the two instances.
1229 u.table_swap = snap_dest->complete;
1230 snap_dest->complete = snap_src->complete;
1231 snap_src->complete = u.table_swap;
1233 u.store_swap = snap_dest->store;
1234 snap_dest->store = snap_src->store;
1235 snap_src->store = u.store_swap;
1237 snap_dest->store->snap = snap_dest;
1238 snap_src->store->snap = snap_src;
1240 snap_dest->ti->max_io_len = snap_dest->store->chunk_size;
1241 snap_dest->valid = snap_src->valid;
1244 * Set source invalid to ensure it receives no further I/O.
1246 snap_src->valid = 0;
1249 static void snapshot_dtr(struct dm_target *ti)
1251 #ifdef CONFIG_DM_DEBUG
1254 struct dm_snapshot *s = ti->private;
1255 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1257 down_read(&_origins_lock);
1258 /* Check whether exception handover must be cancelled */
1259 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1260 if (snap_src && snap_dest && (s == snap_src)) {
1261 down_write(&snap_dest->lock);
1262 snap_dest->valid = 0;
1263 up_write(&snap_dest->lock);
1264 DMERR("Cancelling snapshot handover.");
1266 up_read(&_origins_lock);
1268 if (dm_target_is_snapshot_merge(ti))
1271 /* Prevent further origin writes from using this snapshot. */
1272 /* After this returns there can be no new kcopyd jobs. */
1273 unregister_snapshot(s);
1275 while (atomic_read(&s->pending_exceptions_count))
1278 * Ensure instructions in mempool_destroy aren't reordered
1279 * before atomic_read.
1283 #ifdef CONFIG_DM_DEBUG
1284 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1285 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
1288 __free_exceptions(s);
1290 mempool_destroy(s->pending_pool);
1292 dm_exception_store_destroy(s->store);
1294 dm_put_device(ti, s->cow);
1296 dm_put_device(ti, s->origin);
1302 * Flush a list of buffers.
1304 static void flush_bios(struct bio *bio)
1310 bio->bi_next = NULL;
1311 generic_make_request(bio);
1316 static int do_origin(struct dm_dev *origin, struct bio *bio);
1319 * Flush a list of buffers.
1321 static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
1328 bio->bi_next = NULL;
1329 r = do_origin(s->origin, bio);
1330 if (r == DM_MAPIO_REMAPPED)
1331 generic_make_request(bio);
1337 * Error a list of buffers.
1339 static void error_bios(struct bio *bio)
1345 bio->bi_next = NULL;
1351 static void __invalidate_snapshot(struct dm_snapshot *s, int err)
1357 DMERR("Invalidating snapshot: Error reading/writing.");
1358 else if (err == -ENOMEM)
1359 DMERR("Invalidating snapshot: Unable to allocate exception.");
1361 if (s->store->type->drop_snapshot)
1362 s->store->type->drop_snapshot(s->store);
1366 dm_table_event(s->ti->table);
1369 static void pending_complete(struct dm_snap_pending_exception *pe, int success)
1371 struct dm_exception *e;
1372 struct dm_snapshot *s = pe->snap;
1373 struct bio *origin_bios = NULL;
1374 struct bio *snapshot_bios = NULL;
1375 struct bio *full_bio = NULL;
1379 /* Read/write error - snapshot is unusable */
1380 down_write(&s->lock);
1381 __invalidate_snapshot(s, -EIO);
1386 e = alloc_completed_exception();
1388 down_write(&s->lock);
1389 __invalidate_snapshot(s, -ENOMEM);
1395 down_write(&s->lock);
1397 free_completed_exception(e);
1402 /* Check for conflicting reads */
1403 __check_for_conflicting_io(s, pe->e.old_chunk);
1406 * Add a proper exception, and remove the
1407 * in-flight exception from the list.
1409 dm_insert_exception(&s->complete, e);
1412 dm_remove_exception(&pe->e);
1413 snapshot_bios = bio_list_get(&pe->snapshot_bios);
1414 origin_bios = bio_list_get(&pe->origin_bios);
1415 full_bio = pe->full_bio;
1417 full_bio->bi_end_io = pe->full_bio_end_io;
1418 full_bio->bi_private = pe->full_bio_private;
1420 free_pending_exception(pe);
1422 increment_pending_exceptions_done_count();
1426 /* Submit any pending write bios */
1429 bio_io_error(full_bio);
1430 error_bios(snapshot_bios);
1433 bio_endio(full_bio, 0);
1434 flush_bios(snapshot_bios);
1437 retry_origin_bios(s, origin_bios);
1440 static void commit_callback(void *context, int success)
1442 struct dm_snap_pending_exception *pe = context;
1444 pending_complete(pe, success);
1448 * Called when the copy I/O has finished. kcopyd actually runs
1449 * this code so don't block.
1451 static void copy_callback(int read_err, unsigned long write_err, void *context)
1453 struct dm_snap_pending_exception *pe = context;
1454 struct dm_snapshot *s = pe->snap;
1456 if (read_err || write_err)
1457 pending_complete(pe, 0);
1460 /* Update the metadata if we are persistent */
1461 s->store->type->commit_exception(s->store, &pe->e,
1462 commit_callback, pe);
1466 * Dispatches the copy operation to kcopyd.
1468 static void start_copy(struct dm_snap_pending_exception *pe)
1470 struct dm_snapshot *s = pe->snap;
1471 struct dm_io_region src, dest;
1472 struct block_device *bdev = s->origin->bdev;
1475 dev_size = get_dev_size(bdev);
1478 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
1479 src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
1481 dest.bdev = s->cow->bdev;
1482 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
1483 dest.count = src.count;
1485 /* Hand over to kcopyd */
1486 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe);
1489 static void full_bio_end_io(struct bio *bio, int error)
1491 void *callback_data = bio->bi_private;
1493 dm_kcopyd_do_callback(callback_data, 0, error ? 1 : 0);
1496 static void start_full_bio(struct dm_snap_pending_exception *pe,
1499 struct dm_snapshot *s = pe->snap;
1500 void *callback_data;
1503 pe->full_bio_end_io = bio->bi_end_io;
1504 pe->full_bio_private = bio->bi_private;
1506 callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client,
1509 bio->bi_end_io = full_bio_end_io;
1510 bio->bi_private = callback_data;
1512 generic_make_request(bio);
1515 static struct dm_snap_pending_exception *
1516 __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1518 struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
1523 return container_of(e, struct dm_snap_pending_exception, e);
1527 * Looks to see if this snapshot already has a pending exception
1528 * for this chunk, otherwise it allocates a new one and inserts
1529 * it into the pending table.
1531 * NOTE: a write lock must be held on snap->lock before calling
1534 static struct dm_snap_pending_exception *
1535 __find_pending_exception(struct dm_snapshot *s,
1536 struct dm_snap_pending_exception *pe, chunk_t chunk)
1538 struct dm_snap_pending_exception *pe2;
1540 pe2 = __lookup_pending_exception(s, chunk);
1542 free_pending_exception(pe);
1546 pe->e.old_chunk = chunk;
1547 bio_list_init(&pe->origin_bios);
1548 bio_list_init(&pe->snapshot_bios);
1550 pe->full_bio = NULL;
1552 if (s->store->type->prepare_exception(s->store, &pe->e)) {
1553 free_pending_exception(pe);
1557 dm_insert_exception(&s->pending, &pe->e);
1562 static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
1563 struct bio *bio, chunk_t chunk)
1565 bio->bi_bdev = s->cow->bdev;
1566 bio->bi_sector = chunk_to_sector(s->store,
1567 dm_chunk_number(e->new_chunk) +
1568 (chunk - e->old_chunk)) +
1570 s->store->chunk_mask);
1573 static int snapshot_map(struct dm_target *ti, struct bio *bio)
1575 struct dm_exception *e;
1576 struct dm_snapshot *s = ti->private;
1577 int r = DM_MAPIO_REMAPPED;
1579 struct dm_snap_pending_exception *pe = NULL;
1581 init_tracked_chunk(bio);
1583 if (bio->bi_rw & REQ_FLUSH) {
1584 bio->bi_bdev = s->cow->bdev;
1585 return DM_MAPIO_REMAPPED;
1588 chunk = sector_to_chunk(s->store, bio->bi_sector);
1590 /* Full snapshots are not usable */
1591 /* To get here the table must be live so s->active is always set. */
1595 /* FIXME: should only take write lock if we need
1596 * to copy an exception */
1597 down_write(&s->lock);
1604 /* If the block is already remapped - use that, else remap it */
1605 e = dm_lookup_exception(&s->complete, chunk);
1607 remap_exception(s, e, bio, chunk);
1612 * Write to snapshot - higher level takes care of RW/RO
1613 * flags so we should only get this if we are
1616 if (bio_rw(bio) == WRITE) {
1617 pe = __lookup_pending_exception(s, chunk);
1620 pe = alloc_pending_exception(s);
1621 down_write(&s->lock);
1624 free_pending_exception(pe);
1629 e = dm_lookup_exception(&s->complete, chunk);
1631 free_pending_exception(pe);
1632 remap_exception(s, e, bio, chunk);
1636 pe = __find_pending_exception(s, pe, chunk);
1638 __invalidate_snapshot(s, -ENOMEM);
1644 remap_exception(s, &pe->e, bio, chunk);
1646 r = DM_MAPIO_SUBMITTED;
1649 bio->bi_size == (s->store->chunk_size << SECTOR_SHIFT)) {
1652 start_full_bio(pe, bio);
1656 bio_list_add(&pe->snapshot_bios, bio);
1659 /* this is protected by snap->lock */
1666 bio->bi_bdev = s->origin->bdev;
1667 track_chunk(s, bio, chunk);
1677 * A snapshot-merge target behaves like a combination of a snapshot
1678 * target and a snapshot-origin target. It only generates new
1679 * exceptions in other snapshots and not in the one that is being
1682 * For each chunk, if there is an existing exception, it is used to
1683 * redirect I/O to the cow device. Otherwise I/O is sent to the origin,
1684 * which in turn might generate exceptions in other snapshots.
1685 * If merging is currently taking place on the chunk in question, the
1686 * I/O is deferred by adding it to s->bios_queued_during_merge.
1688 static int snapshot_merge_map(struct dm_target *ti, struct bio *bio)
1690 struct dm_exception *e;
1691 struct dm_snapshot *s = ti->private;
1692 int r = DM_MAPIO_REMAPPED;
1695 init_tracked_chunk(bio);
1697 if (bio->bi_rw & REQ_FLUSH) {
1698 if (!dm_bio_get_target_bio_nr(bio))
1699 bio->bi_bdev = s->origin->bdev;
1701 bio->bi_bdev = s->cow->bdev;
1702 return DM_MAPIO_REMAPPED;
1705 chunk = sector_to_chunk(s->store, bio->bi_sector);
1707 down_write(&s->lock);
1709 /* Full merging snapshots are redirected to the origin */
1711 goto redirect_to_origin;
1713 /* If the block is already remapped - use that */
1714 e = dm_lookup_exception(&s->complete, chunk);
1716 /* Queue writes overlapping with chunks being merged */
1717 if (bio_rw(bio) == WRITE &&
1718 chunk >= s->first_merging_chunk &&
1719 chunk < (s->first_merging_chunk +
1720 s->num_merging_chunks)) {
1721 bio->bi_bdev = s->origin->bdev;
1722 bio_list_add(&s->bios_queued_during_merge, bio);
1723 r = DM_MAPIO_SUBMITTED;
1727 remap_exception(s, e, bio, chunk);
1729 if (bio_rw(bio) == WRITE)
1730 track_chunk(s, bio, chunk);
1735 bio->bi_bdev = s->origin->bdev;
1737 if (bio_rw(bio) == WRITE) {
1739 return do_origin(s->origin, bio);
1748 static int snapshot_end_io(struct dm_target *ti, struct bio *bio, int error)
1750 struct dm_snapshot *s = ti->private;
1752 if (is_bio_tracked(bio))
1753 stop_tracking_chunk(s, bio);
1758 static void snapshot_merge_presuspend(struct dm_target *ti)
1760 struct dm_snapshot *s = ti->private;
1765 static int snapshot_preresume(struct dm_target *ti)
1768 struct dm_snapshot *s = ti->private;
1769 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1771 down_read(&_origins_lock);
1772 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1773 if (snap_src && snap_dest) {
1774 down_read(&snap_src->lock);
1775 if (s == snap_src) {
1776 DMERR("Unable to resume snapshot source until "
1777 "handover completes.");
1779 } else if (!dm_suspended(snap_src->ti)) {
1780 DMERR("Unable to perform snapshot handover until "
1781 "source is suspended.");
1784 up_read(&snap_src->lock);
1786 up_read(&_origins_lock);
1791 static void snapshot_resume(struct dm_target *ti)
1793 struct dm_snapshot *s = ti->private;
1794 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1796 down_read(&_origins_lock);
1797 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1798 if (snap_src && snap_dest) {
1799 down_write(&snap_src->lock);
1800 down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1801 __handover_exceptions(snap_src, snap_dest);
1802 up_write(&snap_dest->lock);
1803 up_write(&snap_src->lock);
1805 up_read(&_origins_lock);
1807 /* Now we have correct chunk size, reregister */
1808 reregister_snapshot(s);
1810 down_write(&s->lock);
1815 static uint32_t get_origin_minimum_chunksize(struct block_device *bdev)
1817 uint32_t min_chunksize;
1819 down_read(&_origins_lock);
1820 min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
1821 up_read(&_origins_lock);
1823 return min_chunksize;
1826 static void snapshot_merge_resume(struct dm_target *ti)
1828 struct dm_snapshot *s = ti->private;
1831 * Handover exceptions from existing snapshot.
1833 snapshot_resume(ti);
1836 * snapshot-merge acts as an origin, so set ti->max_io_len
1838 ti->max_io_len = get_origin_minimum_chunksize(s->origin->bdev);
1843 static void snapshot_status(struct dm_target *ti, status_type_t type,
1844 unsigned status_flags, char *result, unsigned maxlen)
1847 struct dm_snapshot *snap = ti->private;
1850 case STATUSTYPE_INFO:
1852 down_write(&snap->lock);
1856 else if (snap->merge_failed)
1857 DMEMIT("Merge failed");
1859 if (snap->store->type->usage) {
1860 sector_t total_sectors, sectors_allocated,
1862 snap->store->type->usage(snap->store,
1866 DMEMIT("%llu/%llu %llu",
1867 (unsigned long long)sectors_allocated,
1868 (unsigned long long)total_sectors,
1869 (unsigned long long)metadata_sectors);
1875 up_write(&snap->lock);
1879 case STATUSTYPE_TABLE:
1881 * kdevname returns a static pointer so we need
1882 * to make private copies if the output is to
1885 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
1886 snap->store->type->status(snap->store, type, result + sz,
1892 static int snapshot_iterate_devices(struct dm_target *ti,
1893 iterate_devices_callout_fn fn, void *data)
1895 struct dm_snapshot *snap = ti->private;
1898 r = fn(ti, snap->origin, 0, ti->len, data);
1901 r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data);
1907 /*-----------------------------------------------------------------
1909 *---------------------------------------------------------------*/
1912 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
1913 * supplied bio was ignored. The caller may submit it immediately.
1914 * (No remapping actually occurs as the origin is always a direct linear
1917 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
1918 * and any supplied bio is added to a list to be submitted once all
1919 * the necessary exceptions exist.
1921 static int __origin_write(struct list_head *snapshots, sector_t sector,
1924 int r = DM_MAPIO_REMAPPED;
1925 struct dm_snapshot *snap;
1926 struct dm_exception *e;
1927 struct dm_snap_pending_exception *pe;
1928 struct dm_snap_pending_exception *pe_to_start_now = NULL;
1929 struct dm_snap_pending_exception *pe_to_start_last = NULL;
1932 /* Do all the snapshots on this origin */
1933 list_for_each_entry (snap, snapshots, list) {
1935 * Don't make new exceptions in a merging snapshot
1936 * because it has effectively been deleted
1938 if (dm_target_is_snapshot_merge(snap->ti))
1941 down_write(&snap->lock);
1943 /* Only deal with valid and active snapshots */
1944 if (!snap->valid || !snap->active)
1947 /* Nothing to do if writing beyond end of snapshot */
1948 if (sector >= dm_table_get_size(snap->ti->table))
1952 * Remember, different snapshots can have
1953 * different chunk sizes.
1955 chunk = sector_to_chunk(snap->store, sector);
1958 * Check exception table to see if block
1959 * is already remapped in this snapshot
1960 * and trigger an exception if not.
1962 e = dm_lookup_exception(&snap->complete, chunk);
1966 pe = __lookup_pending_exception(snap, chunk);
1968 up_write(&snap->lock);
1969 pe = alloc_pending_exception(snap);
1970 down_write(&snap->lock);
1973 free_pending_exception(pe);
1977 e = dm_lookup_exception(&snap->complete, chunk);
1979 free_pending_exception(pe);
1983 pe = __find_pending_exception(snap, pe, chunk);
1985 __invalidate_snapshot(snap, -ENOMEM);
1990 r = DM_MAPIO_SUBMITTED;
1993 * If an origin bio was supplied, queue it to wait for the
1994 * completion of this exception, and start this one last,
1995 * at the end of the function.
1998 bio_list_add(&pe->origin_bios, bio);
2003 pe_to_start_last = pe;
2009 pe_to_start_now = pe;
2013 up_write(&snap->lock);
2015 if (pe_to_start_now) {
2016 start_copy(pe_to_start_now);
2017 pe_to_start_now = NULL;
2022 * Submit the exception against which the bio is queued last,
2023 * to give the other exceptions a head start.
2025 if (pe_to_start_last)
2026 start_copy(pe_to_start_last);
2032 * Called on a write from the origin driver.
2034 static int do_origin(struct dm_dev *origin, struct bio *bio)
2037 int r = DM_MAPIO_REMAPPED;
2039 down_read(&_origins_lock);
2040 o = __lookup_origin(origin->bdev);
2042 r = __origin_write(&o->snapshots, bio->bi_sector, bio);
2043 up_read(&_origins_lock);
2049 * Trigger exceptions in all non-merging snapshots.
2051 * The chunk size of the merging snapshot may be larger than the chunk
2052 * size of some other snapshot so we may need to reallocate multiple
2053 * chunks in other snapshots.
2055 * We scan all the overlapping exceptions in the other snapshots.
2056 * Returns 1 if anything was reallocated and must be waited for,
2057 * otherwise returns 0.
2059 * size must be a multiple of merging_snap's chunk_size.
2061 static int origin_write_extent(struct dm_snapshot *merging_snap,
2062 sector_t sector, unsigned size)
2069 * The origin's __minimum_chunk_size() got stored in max_io_len
2070 * by snapshot_merge_resume().
2072 down_read(&_origins_lock);
2073 o = __lookup_origin(merging_snap->origin->bdev);
2074 for (n = 0; n < size; n += merging_snap->ti->max_io_len)
2075 if (__origin_write(&o->snapshots, sector + n, NULL) ==
2078 up_read(&_origins_lock);
2084 * Origin: maps a linear range of a device, with hooks for snapshotting.
2088 * Construct an origin mapping: <dev_path>
2089 * The context for an origin is merely a 'struct dm_dev *'
2090 * pointing to the real device.
2092 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2098 ti->error = "origin: incorrect number of arguments";
2102 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev);
2104 ti->error = "Cannot get target device";
2109 ti->num_flush_bios = 1;
2114 static void origin_dtr(struct dm_target *ti)
2116 struct dm_dev *dev = ti->private;
2117 dm_put_device(ti, dev);
2120 static int origin_map(struct dm_target *ti, struct bio *bio)
2122 struct dm_dev *dev = ti->private;
2123 bio->bi_bdev = dev->bdev;
2125 if (bio->bi_rw & REQ_FLUSH)
2126 return DM_MAPIO_REMAPPED;
2128 /* Only tell snapshots if this is a write */
2129 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
2133 * Set the target "max_io_len" field to the minimum of all the snapshots'
2136 static void origin_resume(struct dm_target *ti)
2138 struct dm_dev *dev = ti->private;
2140 ti->max_io_len = get_origin_minimum_chunksize(dev->bdev);
2143 static void origin_status(struct dm_target *ti, status_type_t type,
2144 unsigned status_flags, char *result, unsigned maxlen)
2146 struct dm_dev *dev = ti->private;
2149 case STATUSTYPE_INFO:
2153 case STATUSTYPE_TABLE:
2154 snprintf(result, maxlen, "%s", dev->name);
2159 static int origin_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2160 struct bio_vec *biovec, int max_size)
2162 struct dm_dev *dev = ti->private;
2163 struct request_queue *q = bdev_get_queue(dev->bdev);
2165 if (!q->merge_bvec_fn)
2168 bvm->bi_bdev = dev->bdev;
2170 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2173 static int origin_iterate_devices(struct dm_target *ti,
2174 iterate_devices_callout_fn fn, void *data)
2176 struct dm_dev *dev = ti->private;
2178 return fn(ti, dev, 0, ti->len, data);
2181 static struct target_type origin_target = {
2182 .name = "snapshot-origin",
2183 .version = {1, 8, 1},
2184 .module = THIS_MODULE,
2188 .resume = origin_resume,
2189 .status = origin_status,
2190 .merge = origin_merge,
2191 .iterate_devices = origin_iterate_devices,
2194 static struct target_type snapshot_target = {
2196 .version = {1, 11, 1},
2197 .module = THIS_MODULE,
2198 .ctr = snapshot_ctr,
2199 .dtr = snapshot_dtr,
2200 .map = snapshot_map,
2201 .end_io = snapshot_end_io,
2202 .preresume = snapshot_preresume,
2203 .resume = snapshot_resume,
2204 .status = snapshot_status,
2205 .iterate_devices = snapshot_iterate_devices,
2208 static struct target_type merge_target = {
2209 .name = dm_snapshot_merge_target_name,
2210 .version = {1, 2, 0},
2211 .module = THIS_MODULE,
2212 .ctr = snapshot_ctr,
2213 .dtr = snapshot_dtr,
2214 .map = snapshot_merge_map,
2215 .end_io = snapshot_end_io,
2216 .presuspend = snapshot_merge_presuspend,
2217 .preresume = snapshot_preresume,
2218 .resume = snapshot_merge_resume,
2219 .status = snapshot_status,
2220 .iterate_devices = snapshot_iterate_devices,
2223 static int __init dm_snapshot_init(void)
2227 r = dm_exception_store_init();
2229 DMERR("Failed to initialize exception stores");
2233 r = dm_register_target(&snapshot_target);
2235 DMERR("snapshot target register failed %d", r);
2236 goto bad_register_snapshot_target;
2239 r = dm_register_target(&origin_target);
2241 DMERR("Origin target register failed %d", r);
2242 goto bad_register_origin_target;
2245 r = dm_register_target(&merge_target);
2247 DMERR("Merge target register failed %d", r);
2248 goto bad_register_merge_target;
2251 r = init_origin_hash();
2253 DMERR("init_origin_hash failed.");
2254 goto bad_origin_hash;
2257 exception_cache = KMEM_CACHE(dm_exception, 0);
2258 if (!exception_cache) {
2259 DMERR("Couldn't create exception cache.");
2261 goto bad_exception_cache;
2264 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
2265 if (!pending_cache) {
2266 DMERR("Couldn't create pending cache.");
2268 goto bad_pending_cache;
2274 kmem_cache_destroy(exception_cache);
2275 bad_exception_cache:
2278 dm_unregister_target(&merge_target);
2279 bad_register_merge_target:
2280 dm_unregister_target(&origin_target);
2281 bad_register_origin_target:
2282 dm_unregister_target(&snapshot_target);
2283 bad_register_snapshot_target:
2284 dm_exception_store_exit();
2289 static void __exit dm_snapshot_exit(void)
2291 dm_unregister_target(&snapshot_target);
2292 dm_unregister_target(&origin_target);
2293 dm_unregister_target(&merge_target);
2296 kmem_cache_destroy(pending_cache);
2297 kmem_cache_destroy(exception_cache);
2299 dm_exception_store_exit();
2303 module_init(dm_snapshot_init);
2304 module_exit(dm_snapshot_exit);
2306 MODULE_DESCRIPTION(DM_NAME " snapshot target");
2307 MODULE_AUTHOR("Joe Thornber");
2308 MODULE_LICENSE("GPL");
2309 MODULE_ALIAS("dm-snapshot-origin");
2310 MODULE_ALIAS("dm-snapshot-merge");