4 * Copyright (C) 2001-2002 Sistina Software (UK) Limited.
6 * This file is released under the GPL.
9 #include <linux/blkdev.h>
10 #include <linux/device-mapper.h>
11 #include <linux/delay.h>
13 #include <linux/init.h>
14 #include <linux/kdev_t.h>
15 #include <linux/list.h>
16 #include <linux/mempool.h>
17 #include <linux/module.h>
18 #include <linux/slab.h>
19 #include <linux/vmalloc.h>
20 #include <linux/log2.h>
21 #include <linux/dm-kcopyd.h>
22 #include <linux/workqueue.h>
24 #include "dm-exception-store.h"
26 #define DM_MSG_PREFIX "snapshots"
29 * The percentage increment we will wake up users at
31 #define WAKE_UP_PERCENT 5
34 * kcopyd priority of snapshot operations
36 #define SNAPSHOT_COPY_PRIORITY 2
39 * Reserve 1MB for each snapshot initially (with minimum of 1 page).
41 #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1)
44 * The size of the mempool used to track chunks in use.
48 #define DM_TRACKED_CHUNK_HASH_SIZE 16
49 #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
50 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
52 struct exception_table {
55 struct list_head *table;
59 struct rw_semaphore lock;
61 struct dm_dev *origin;
63 /* List of snapshots per Origin */
64 struct list_head list;
66 /* You can't use a snapshot if this is 0 (e.g. if full) */
69 /* Origin writes don't trigger exceptions until this is set */
72 mempool_t *pending_pool;
74 atomic_t pending_exceptions_count;
76 struct exception_table pending;
77 struct exception_table complete;
80 * pe_lock protects all pending_exception operations and access
81 * as well as the snapshot_bios list.
85 /* The on disk metadata handler */
86 struct dm_exception_store *store;
88 struct dm_kcopyd_client *kcopyd_client;
90 /* Queue of snapshot writes for ksnapd to flush */
91 struct bio_list queued_bios;
92 struct work_struct queued_bios_work;
94 /* Chunks with outstanding reads */
95 mempool_t *tracked_chunk_pool;
96 spinlock_t tracked_chunk_lock;
97 struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
100 static struct workqueue_struct *ksnapd;
101 static void flush_queued_bios(struct work_struct *work);
103 static sector_t chunk_to_sector(struct dm_exception_store *store,
106 return chunk << store->chunk_shift;
109 static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
112 * There is only ever one instance of a particular block
113 * device so we can compare pointers safely.
118 struct dm_snap_pending_exception {
119 struct dm_snap_exception e;
122 * Origin buffers waiting for this to complete are held
125 struct bio_list origin_bios;
126 struct bio_list snapshot_bios;
129 * Short-term queue of pending exceptions prior to submission.
131 struct list_head list;
134 * The primary pending_exception is the one that holds
135 * the ref_count and the list of origin_bios for a
136 * group of pending_exceptions. It is always last to get freed.
137 * These fields get set up when writing to the origin.
139 struct dm_snap_pending_exception *primary_pe;
142 * Number of pending_exceptions processing this chunk.
143 * When this drops to zero we must complete the origin bios.
144 * If incrementing or decrementing this, hold pe->snap->lock for
145 * the sibling concerned and not pe->primary_pe->snap->lock unless
150 /* Pointer back to snapshot context */
151 struct dm_snapshot *snap;
154 * 1 indicates the exception has already been sent to
161 * Hash table mapping origin volumes to lists of snapshots and
162 * a lock to protect it
164 static struct kmem_cache *exception_cache;
165 static struct kmem_cache *pending_cache;
167 struct dm_snap_tracked_chunk {
168 struct hlist_node node;
172 static struct kmem_cache *tracked_chunk_cache;
174 static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s,
177 struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool,
183 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
184 hlist_add_head(&c->node,
185 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
186 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
191 static void stop_tracking_chunk(struct dm_snapshot *s,
192 struct dm_snap_tracked_chunk *c)
196 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
198 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
200 mempool_free(c, s->tracked_chunk_pool);
203 static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
205 struct dm_snap_tracked_chunk *c;
206 struct hlist_node *hn;
209 spin_lock_irq(&s->tracked_chunk_lock);
211 hlist_for_each_entry(c, hn,
212 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
213 if (c->chunk == chunk) {
219 spin_unlock_irq(&s->tracked_chunk_lock);
225 * One of these per registered origin, held in the snapshot_origins hash
228 /* The origin device */
229 struct block_device *bdev;
231 struct list_head hash_list;
233 /* List of snapshots for this origin */
234 struct list_head snapshots;
238 * Size of the hash table for origin volumes. If we make this
239 * the size of the minors list then it should be nearly perfect
241 #define ORIGIN_HASH_SIZE 256
242 #define ORIGIN_MASK 0xFF
243 static struct list_head *_origins;
244 static struct rw_semaphore _origins_lock;
246 static int init_origin_hash(void)
250 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
253 DMERR("unable to allocate memory");
257 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
258 INIT_LIST_HEAD(_origins + i);
259 init_rwsem(&_origins_lock);
264 static void exit_origin_hash(void)
269 static unsigned origin_hash(struct block_device *bdev)
271 return bdev->bd_dev & ORIGIN_MASK;
274 static struct origin *__lookup_origin(struct block_device *origin)
276 struct list_head *ol;
279 ol = &_origins[origin_hash(origin)];
280 list_for_each_entry (o, ol, hash_list)
281 if (bdev_equal(o->bdev, origin))
287 static void __insert_origin(struct origin *o)
289 struct list_head *sl = &_origins[origin_hash(o->bdev)];
290 list_add_tail(&o->hash_list, sl);
294 * Make a note of the snapshot and its origin so we can look it
295 * up when the origin has a write on it.
297 static int register_snapshot(struct dm_snapshot *snap)
299 struct dm_snapshot *l;
300 struct origin *o, *new_o;
301 struct block_device *bdev = snap->origin->bdev;
303 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
307 down_write(&_origins_lock);
308 o = __lookup_origin(bdev);
316 /* Initialise the struct */
317 INIT_LIST_HEAD(&o->snapshots);
323 /* Sort the list according to chunk size, largest-first smallest-last */
324 list_for_each_entry(l, &o->snapshots, list)
325 if (l->store->chunk_size < snap->store->chunk_size)
327 list_add_tail(&snap->list, &l->list);
329 up_write(&_origins_lock);
333 static void unregister_snapshot(struct dm_snapshot *s)
337 down_write(&_origins_lock);
338 o = __lookup_origin(s->origin->bdev);
341 if (list_empty(&o->snapshots)) {
342 list_del(&o->hash_list);
346 up_write(&_origins_lock);
350 * Implementation of the exception hash tables.
351 * The lowest hash_shift bits of the chunk number are ignored, allowing
352 * some consecutive chunks to be grouped together.
354 static int init_exception_table(struct exception_table *et, uint32_t size,
359 et->hash_shift = hash_shift;
360 et->hash_mask = size - 1;
361 et->table = dm_vcalloc(size, sizeof(struct list_head));
365 for (i = 0; i < size; i++)
366 INIT_LIST_HEAD(et->table + i);
371 static void exit_exception_table(struct exception_table *et, struct kmem_cache *mem)
373 struct list_head *slot;
374 struct dm_snap_exception *ex, *next;
377 size = et->hash_mask + 1;
378 for (i = 0; i < size; i++) {
379 slot = et->table + i;
381 list_for_each_entry_safe (ex, next, slot, hash_list)
382 kmem_cache_free(mem, ex);
388 static uint32_t exception_hash(struct exception_table *et, chunk_t chunk)
390 return (chunk >> et->hash_shift) & et->hash_mask;
393 static void insert_exception(struct exception_table *eh,
394 struct dm_snap_exception *e)
396 struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)];
397 list_add(&e->hash_list, l);
400 static void remove_exception(struct dm_snap_exception *e)
402 list_del(&e->hash_list);
406 * Return the exception data for a sector, or NULL if not
409 static struct dm_snap_exception *lookup_exception(struct exception_table *et,
412 struct list_head *slot;
413 struct dm_snap_exception *e;
415 slot = &et->table[exception_hash(et, chunk)];
416 list_for_each_entry (e, slot, hash_list)
417 if (chunk >= e->old_chunk &&
418 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
424 static struct dm_snap_exception *alloc_exception(void)
426 struct dm_snap_exception *e;
428 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
430 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
435 static void free_exception(struct dm_snap_exception *e)
437 kmem_cache_free(exception_cache, e);
440 static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
442 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
445 atomic_inc(&s->pending_exceptions_count);
451 static void free_pending_exception(struct dm_snap_pending_exception *pe)
453 struct dm_snapshot *s = pe->snap;
455 mempool_free(pe, s->pending_pool);
456 smp_mb__before_atomic_dec();
457 atomic_dec(&s->pending_exceptions_count);
460 static void insert_completed_exception(struct dm_snapshot *s,
461 struct dm_snap_exception *new_e)
463 struct exception_table *eh = &s->complete;
465 struct dm_snap_exception *e = NULL;
467 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
469 /* Add immediately if this table doesn't support consecutive chunks */
473 /* List is ordered by old_chunk */
474 list_for_each_entry_reverse(e, l, hash_list) {
475 /* Insert after an existing chunk? */
476 if (new_e->old_chunk == (e->old_chunk +
477 dm_consecutive_chunk_count(e) + 1) &&
478 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
479 dm_consecutive_chunk_count(e) + 1)) {
480 dm_consecutive_chunk_count_inc(e);
481 free_exception(new_e);
485 /* Insert before an existing chunk? */
486 if (new_e->old_chunk == (e->old_chunk - 1) &&
487 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
488 dm_consecutive_chunk_count_inc(e);
491 free_exception(new_e);
495 if (new_e->old_chunk > e->old_chunk)
500 list_add(&new_e->hash_list, e ? &e->hash_list : l);
504 * Callback used by the exception stores to load exceptions when
507 static int dm_add_exception(void *context, chunk_t old, chunk_t new)
509 struct dm_snapshot *s = context;
510 struct dm_snap_exception *e;
512 e = alloc_exception();
518 /* Consecutive_count is implicitly initialised to zero */
521 insert_completed_exception(s, e);
526 #define min_not_zero(l, r) (((l) == 0) ? (r) : (((r) == 0) ? (l) : min(l, r)))
529 * Return a minimum chunk size of all snapshots that have the specified origin.
530 * Return zero if the origin has no snapshots.
532 static sector_t __minimum_chunk_size(struct origin *o)
534 struct dm_snapshot *snap;
535 unsigned chunk_size = 0;
538 list_for_each_entry(snap, &o->snapshots, list)
539 chunk_size = min_not_zero(chunk_size,
540 snap->store->chunk_size);
548 static int calc_max_buckets(void)
550 /* use a fixed size of 2MB */
551 unsigned long mem = 2 * 1024 * 1024;
552 mem /= sizeof(struct list_head);
558 * Allocate room for a suitable hash table.
560 static int init_hash_tables(struct dm_snapshot *s)
562 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
565 * Calculate based on the size of the original volume or
568 cow_dev_size = get_dev_size(s->store->cow->bdev);
569 origin_dev_size = get_dev_size(s->origin->bdev);
570 max_buckets = calc_max_buckets();
572 hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift;
573 hash_size = min(hash_size, max_buckets);
577 hash_size = rounddown_pow_of_two(hash_size);
578 if (init_exception_table(&s->complete, hash_size,
579 DM_CHUNK_CONSECUTIVE_BITS))
583 * Allocate hash table for in-flight exceptions
584 * Make this smaller than the real hash table
590 if (init_exception_table(&s->pending, hash_size, 0)) {
591 exit_exception_table(&s->complete, exception_cache);
599 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
601 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
603 struct dm_snapshot *s;
607 struct dm_exception_store *store;
611 ti->error = "requires exactly 4 arguments";
616 origin_path = argv[0];
620 r = dm_exception_store_create(ti, argc, argv, &args_used, &store);
622 ti->error = "Couldn't create exception store";
630 s = kmalloc(sizeof(*s), GFP_KERNEL);
632 ti->error = "Cannot allocate snapshot context private "
638 r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin);
640 ti->error = "Cannot get origin device";
647 atomic_set(&s->pending_exceptions_count, 0);
648 init_rwsem(&s->lock);
649 spin_lock_init(&s->pe_lock);
651 /* Allocate hash table for COW data */
652 if (init_hash_tables(s)) {
653 ti->error = "Unable to allocate hash table space";
655 goto bad_hash_tables;
658 r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client);
660 ti->error = "Could not create kcopyd client";
664 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
665 if (!s->pending_pool) {
666 ti->error = "Could not allocate mempool for pending exceptions";
667 goto bad_pending_pool;
670 s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS,
671 tracked_chunk_cache);
672 if (!s->tracked_chunk_pool) {
673 ti->error = "Could not allocate tracked_chunk mempool for "
675 goto bad_tracked_chunk_pool;
678 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
679 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
681 spin_lock_init(&s->tracked_chunk_lock);
683 /* Metadata must only be loaded into one table at once */
684 r = s->store->type->read_metadata(s->store, dm_add_exception,
687 ti->error = "Failed to read snapshot metadata";
688 goto bad_load_and_register;
691 DMWARN("Snapshot is marked invalid.");
694 bio_list_init(&s->queued_bios);
695 INIT_WORK(&s->queued_bios_work, flush_queued_bios);
697 if (!s->store->chunk_size) {
698 ti->error = "Chunk size not set";
699 goto bad_load_and_register;
702 /* Add snapshot to the list of snapshots for this origin */
703 /* Exceptions aren't triggered till snapshot_resume() is called */
704 if (register_snapshot(s)) {
706 ti->error = "Cannot register snapshot origin";
707 goto bad_load_and_register;
711 ti->split_io = s->store->chunk_size;
712 ti->num_flush_requests = 1;
716 bad_load_and_register:
717 mempool_destroy(s->tracked_chunk_pool);
719 bad_tracked_chunk_pool:
720 mempool_destroy(s->pending_pool);
723 dm_kcopyd_client_destroy(s->kcopyd_client);
726 exit_exception_table(&s->pending, pending_cache);
727 exit_exception_table(&s->complete, exception_cache);
730 dm_put_device(ti, s->origin);
736 dm_exception_store_destroy(store);
742 static void __free_exceptions(struct dm_snapshot *s)
744 dm_kcopyd_client_destroy(s->kcopyd_client);
745 s->kcopyd_client = NULL;
747 exit_exception_table(&s->pending, pending_cache);
748 exit_exception_table(&s->complete, exception_cache);
751 static void snapshot_dtr(struct dm_target *ti)
753 #ifdef CONFIG_DM_DEBUG
756 struct dm_snapshot *s = ti->private;
758 flush_workqueue(ksnapd);
760 /* Prevent further origin writes from using this snapshot. */
761 /* After this returns there can be no new kcopyd jobs. */
762 unregister_snapshot(s);
764 while (atomic_read(&s->pending_exceptions_count))
767 * Ensure instructions in mempool_destroy aren't reordered
768 * before atomic_read.
772 #ifdef CONFIG_DM_DEBUG
773 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
774 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
777 mempool_destroy(s->tracked_chunk_pool);
779 __free_exceptions(s);
781 mempool_destroy(s->pending_pool);
783 dm_put_device(ti, s->origin);
785 dm_exception_store_destroy(s->store);
791 * Flush a list of buffers.
793 static void flush_bios(struct bio *bio)
800 generic_make_request(bio);
805 static void flush_queued_bios(struct work_struct *work)
807 struct dm_snapshot *s =
808 container_of(work, struct dm_snapshot, queued_bios_work);
809 struct bio *queued_bios;
812 spin_lock_irqsave(&s->pe_lock, flags);
813 queued_bios = bio_list_get(&s->queued_bios);
814 spin_unlock_irqrestore(&s->pe_lock, flags);
816 flush_bios(queued_bios);
820 * Error a list of buffers.
822 static void error_bios(struct bio *bio)
834 static void __invalidate_snapshot(struct dm_snapshot *s, int err)
840 DMERR("Invalidating snapshot: Error reading/writing.");
841 else if (err == -ENOMEM)
842 DMERR("Invalidating snapshot: Unable to allocate exception.");
844 if (s->store->type->drop_snapshot)
845 s->store->type->drop_snapshot(s->store);
849 dm_table_event(s->store->ti->table);
852 static void get_pending_exception(struct dm_snap_pending_exception *pe)
854 atomic_inc(&pe->ref_count);
857 static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe)
859 struct dm_snap_pending_exception *primary_pe;
860 struct bio *origin_bios = NULL;
862 primary_pe = pe->primary_pe;
865 * If this pe is involved in a write to the origin and
866 * it is the last sibling to complete then release
867 * the bios for the original write to the origin.
870 atomic_dec_and_test(&primary_pe->ref_count)) {
871 origin_bios = bio_list_get(&primary_pe->origin_bios);
872 free_pending_exception(primary_pe);
876 * Free the pe if it's not linked to an origin write or if
877 * it's not itself a primary pe.
879 if (!primary_pe || primary_pe != pe)
880 free_pending_exception(pe);
885 static void pending_complete(struct dm_snap_pending_exception *pe, int success)
887 struct dm_snap_exception *e;
888 struct dm_snapshot *s = pe->snap;
889 struct bio *origin_bios = NULL;
890 struct bio *snapshot_bios = NULL;
894 /* Read/write error - snapshot is unusable */
895 down_write(&s->lock);
896 __invalidate_snapshot(s, -EIO);
901 e = alloc_exception();
903 down_write(&s->lock);
904 __invalidate_snapshot(s, -ENOMEM);
910 down_write(&s->lock);
918 * Check for conflicting reads. This is extremely improbable,
919 * so msleep(1) is sufficient and there is no need for a wait queue.
921 while (__chunk_is_tracked(s, pe->e.old_chunk))
925 * Add a proper exception, and remove the
926 * in-flight exception from the list.
928 insert_completed_exception(s, e);
931 remove_exception(&pe->e);
932 snapshot_bios = bio_list_get(&pe->snapshot_bios);
933 origin_bios = put_pending_exception(pe);
937 /* Submit any pending write bios */
939 error_bios(snapshot_bios);
941 flush_bios(snapshot_bios);
943 flush_bios(origin_bios);
946 static void commit_callback(void *context, int success)
948 struct dm_snap_pending_exception *pe = context;
950 pending_complete(pe, success);
954 * Called when the copy I/O has finished. kcopyd actually runs
955 * this code so don't block.
957 static void copy_callback(int read_err, unsigned long write_err, void *context)
959 struct dm_snap_pending_exception *pe = context;
960 struct dm_snapshot *s = pe->snap;
962 if (read_err || write_err)
963 pending_complete(pe, 0);
966 /* Update the metadata if we are persistent */
967 s->store->type->commit_exception(s->store, &pe->e,
968 commit_callback, pe);
972 * Dispatches the copy operation to kcopyd.
974 static void start_copy(struct dm_snap_pending_exception *pe)
976 struct dm_snapshot *s = pe->snap;
977 struct dm_io_region src, dest;
978 struct block_device *bdev = s->origin->bdev;
981 dev_size = get_dev_size(bdev);
984 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
985 src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
987 dest.bdev = s->store->cow->bdev;
988 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
989 dest.count = src.count;
991 /* Hand over to kcopyd */
992 dm_kcopyd_copy(s->kcopyd_client,
993 &src, 1, &dest, 0, copy_callback, pe);
996 static struct dm_snap_pending_exception *
997 __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
999 struct dm_snap_exception *e = lookup_exception(&s->pending, chunk);
1004 return container_of(e, struct dm_snap_pending_exception, e);
1008 * Looks to see if this snapshot already has a pending exception
1009 * for this chunk, otherwise it allocates a new one and inserts
1010 * it into the pending table.
1012 * NOTE: a write lock must be held on snap->lock before calling
1015 static struct dm_snap_pending_exception *
1016 __find_pending_exception(struct dm_snapshot *s,
1017 struct dm_snap_pending_exception *pe, chunk_t chunk)
1019 struct dm_snap_pending_exception *pe2;
1021 pe2 = __lookup_pending_exception(s, chunk);
1023 free_pending_exception(pe);
1027 pe->e.old_chunk = chunk;
1028 bio_list_init(&pe->origin_bios);
1029 bio_list_init(&pe->snapshot_bios);
1030 pe->primary_pe = NULL;
1031 atomic_set(&pe->ref_count, 0);
1034 if (s->store->type->prepare_exception(s->store, &pe->e)) {
1035 free_pending_exception(pe);
1039 get_pending_exception(pe);
1040 insert_exception(&s->pending, &pe->e);
1045 static void remap_exception(struct dm_snapshot *s, struct dm_snap_exception *e,
1046 struct bio *bio, chunk_t chunk)
1048 bio->bi_bdev = s->store->cow->bdev;
1049 bio->bi_sector = chunk_to_sector(s->store,
1050 dm_chunk_number(e->new_chunk) +
1051 (chunk - e->old_chunk)) +
1053 s->store->chunk_mask);
1056 static int snapshot_map(struct dm_target *ti, struct bio *bio,
1057 union map_info *map_context)
1059 struct dm_snap_exception *e;
1060 struct dm_snapshot *s = ti->private;
1061 int r = DM_MAPIO_REMAPPED;
1063 struct dm_snap_pending_exception *pe = NULL;
1065 if (unlikely(bio_empty_barrier(bio))) {
1066 bio->bi_bdev = s->store->cow->bdev;
1067 return DM_MAPIO_REMAPPED;
1070 chunk = sector_to_chunk(s->store, bio->bi_sector);
1072 /* Full snapshots are not usable */
1073 /* To get here the table must be live so s->active is always set. */
1077 /* FIXME: should only take write lock if we need
1078 * to copy an exception */
1079 down_write(&s->lock);
1086 /* If the block is already remapped - use that, else remap it */
1087 e = lookup_exception(&s->complete, chunk);
1089 remap_exception(s, e, bio, chunk);
1094 * Write to snapshot - higher level takes care of RW/RO
1095 * flags so we should only get this if we are
1098 if (bio_rw(bio) == WRITE) {
1099 pe = __lookup_pending_exception(s, chunk);
1102 pe = alloc_pending_exception(s);
1103 down_write(&s->lock);
1106 free_pending_exception(pe);
1111 e = lookup_exception(&s->complete, chunk);
1113 free_pending_exception(pe);
1114 remap_exception(s, e, bio, chunk);
1118 pe = __find_pending_exception(s, pe, chunk);
1120 __invalidate_snapshot(s, -ENOMEM);
1126 remap_exception(s, &pe->e, bio, chunk);
1127 bio_list_add(&pe->snapshot_bios, bio);
1129 r = DM_MAPIO_SUBMITTED;
1132 /* this is protected by snap->lock */
1139 bio->bi_bdev = s->origin->bdev;
1140 map_context->ptr = track_chunk(s, chunk);
1149 static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1150 int error, union map_info *map_context)
1152 struct dm_snapshot *s = ti->private;
1153 struct dm_snap_tracked_chunk *c = map_context->ptr;
1156 stop_tracking_chunk(s, c);
1161 static void snapshot_resume(struct dm_target *ti)
1163 struct dm_snapshot *s = ti->private;
1165 down_write(&s->lock);
1170 static int snapshot_status(struct dm_target *ti, status_type_t type,
1171 char *result, unsigned int maxlen)
1174 struct dm_snapshot *snap = ti->private;
1177 case STATUSTYPE_INFO:
1179 down_write(&snap->lock);
1184 if (snap->store->type->fraction_full) {
1185 sector_t numerator, denominator;
1186 snap->store->type->fraction_full(snap->store,
1190 (unsigned long long)numerator,
1191 (unsigned long long)denominator);
1197 up_write(&snap->lock);
1201 case STATUSTYPE_TABLE:
1203 * kdevname returns a static pointer so we need
1204 * to make private copies if the output is to
1207 DMEMIT("%s", snap->origin->name);
1208 snap->store->type->status(snap->store, type, result + sz,
1216 static int snapshot_iterate_devices(struct dm_target *ti,
1217 iterate_devices_callout_fn fn, void *data)
1219 struct dm_snapshot *snap = ti->private;
1221 return fn(ti, snap->origin, 0, ti->len, data);
1225 /*-----------------------------------------------------------------
1227 *---------------------------------------------------------------*/
1228 static int __origin_write(struct list_head *snapshots, struct bio *bio)
1230 int r = DM_MAPIO_REMAPPED, first = 0;
1231 struct dm_snapshot *snap;
1232 struct dm_snap_exception *e;
1233 struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL;
1235 LIST_HEAD(pe_queue);
1237 /* Do all the snapshots on this origin */
1238 list_for_each_entry (snap, snapshots, list) {
1240 down_write(&snap->lock);
1242 /* Only deal with valid and active snapshots */
1243 if (!snap->valid || !snap->active)
1246 /* Nothing to do if writing beyond end of snapshot */
1247 if (bio->bi_sector >= dm_table_get_size(snap->store->ti->table))
1251 * Remember, different snapshots can have
1252 * different chunk sizes.
1254 chunk = sector_to_chunk(snap->store, bio->bi_sector);
1257 * Check exception table to see if block
1258 * is already remapped in this snapshot
1259 * and trigger an exception if not.
1261 * ref_count is initialised to 1 so pending_complete()
1262 * won't destroy the primary_pe while we're inside this loop.
1264 e = lookup_exception(&snap->complete, chunk);
1268 pe = __lookup_pending_exception(snap, chunk);
1270 up_write(&snap->lock);
1271 pe = alloc_pending_exception(snap);
1272 down_write(&snap->lock);
1275 free_pending_exception(pe);
1279 e = lookup_exception(&snap->complete, chunk);
1281 free_pending_exception(pe);
1285 pe = __find_pending_exception(snap, pe, chunk);
1287 __invalidate_snapshot(snap, -ENOMEM);
1294 * Either every pe here has same
1295 * primary_pe or none has one yet.
1298 primary_pe = pe->primary_pe;
1304 bio_list_add(&primary_pe->origin_bios, bio);
1306 r = DM_MAPIO_SUBMITTED;
1309 if (!pe->primary_pe) {
1310 pe->primary_pe = primary_pe;
1311 get_pending_exception(primary_pe);
1316 list_add_tail(&pe->list, &pe_queue);
1320 up_write(&snap->lock);
1327 * If this is the first time we're processing this chunk and
1328 * ref_count is now 1 it means all the pending exceptions
1329 * got completed while we were in the loop above, so it falls to
1330 * us here to remove the primary_pe and submit any origin_bios.
1333 if (first && atomic_dec_and_test(&primary_pe->ref_count)) {
1334 flush_bios(bio_list_get(&primary_pe->origin_bios));
1335 free_pending_exception(primary_pe);
1336 /* If we got here, pe_queue is necessarily empty. */
1341 * Now that we have a complete pe list we can start the copying.
1343 list_for_each_entry_safe(pe, next_pe, &pe_queue, list)
1350 * Called on a write from the origin driver.
1352 static int do_origin(struct dm_dev *origin, struct bio *bio)
1355 int r = DM_MAPIO_REMAPPED;
1357 down_read(&_origins_lock);
1358 o = __lookup_origin(origin->bdev);
1360 r = __origin_write(&o->snapshots, bio);
1361 up_read(&_origins_lock);
1367 * Origin: maps a linear range of a device, with hooks for snapshotting.
1371 * Construct an origin mapping: <dev_path>
1372 * The context for an origin is merely a 'struct dm_dev *'
1373 * pointing to the real device.
1375 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1381 ti->error = "origin: incorrect number of arguments";
1385 r = dm_get_device(ti, argv[0], 0, ti->len,
1386 dm_table_get_mode(ti->table), &dev);
1388 ti->error = "Cannot get target device";
1393 ti->num_flush_requests = 1;
1398 static void origin_dtr(struct dm_target *ti)
1400 struct dm_dev *dev = ti->private;
1401 dm_put_device(ti, dev);
1404 static int origin_map(struct dm_target *ti, struct bio *bio,
1405 union map_info *map_context)
1407 struct dm_dev *dev = ti->private;
1408 bio->bi_bdev = dev->bdev;
1410 if (unlikely(bio_empty_barrier(bio)))
1411 return DM_MAPIO_REMAPPED;
1413 /* Only tell snapshots if this is a write */
1414 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
1418 * Set the target "split_io" field to the minimum of all the snapshots'
1421 static void origin_resume(struct dm_target *ti)
1423 struct dm_dev *dev = ti->private;
1425 down_read(&_origins_lock);
1427 ti->split_io = __minimum_chunk_size(__lookup_origin(dev->bdev));
1429 up_read(&_origins_lock);
1432 static int origin_status(struct dm_target *ti, status_type_t type, char *result,
1433 unsigned int maxlen)
1435 struct dm_dev *dev = ti->private;
1438 case STATUSTYPE_INFO:
1442 case STATUSTYPE_TABLE:
1443 snprintf(result, maxlen, "%s", dev->name);
1450 static int origin_iterate_devices(struct dm_target *ti,
1451 iterate_devices_callout_fn fn, void *data)
1453 struct dm_dev *dev = ti->private;
1455 return fn(ti, dev, 0, ti->len, data);
1458 static struct target_type origin_target = {
1459 .name = "snapshot-origin",
1460 .version = {1, 7, 0},
1461 .module = THIS_MODULE,
1465 .resume = origin_resume,
1466 .status = origin_status,
1467 .iterate_devices = origin_iterate_devices,
1470 static struct target_type snapshot_target = {
1472 .version = {1, 7, 0},
1473 .module = THIS_MODULE,
1474 .ctr = snapshot_ctr,
1475 .dtr = snapshot_dtr,
1476 .map = snapshot_map,
1477 .end_io = snapshot_end_io,
1478 .resume = snapshot_resume,
1479 .status = snapshot_status,
1480 .iterate_devices = snapshot_iterate_devices,
1483 static int __init dm_snapshot_init(void)
1487 r = dm_exception_store_init();
1489 DMERR("Failed to initialize exception stores");
1493 r = dm_register_target(&snapshot_target);
1495 DMERR("snapshot target register failed %d", r);
1496 goto bad_register_snapshot_target;
1499 r = dm_register_target(&origin_target);
1501 DMERR("Origin target register failed %d", r);
1505 r = init_origin_hash();
1507 DMERR("init_origin_hash failed.");
1511 exception_cache = KMEM_CACHE(dm_snap_exception, 0);
1512 if (!exception_cache) {
1513 DMERR("Couldn't create exception cache.");
1518 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
1519 if (!pending_cache) {
1520 DMERR("Couldn't create pending cache.");
1525 tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0);
1526 if (!tracked_chunk_cache) {
1527 DMERR("Couldn't create cache to track chunks in use.");
1532 ksnapd = create_singlethread_workqueue("ksnapd");
1534 DMERR("Failed to create ksnapd workqueue.");
1536 goto bad_pending_pool;
1542 kmem_cache_destroy(tracked_chunk_cache);
1544 kmem_cache_destroy(pending_cache);
1546 kmem_cache_destroy(exception_cache);
1550 dm_unregister_target(&origin_target);
1552 dm_unregister_target(&snapshot_target);
1554 bad_register_snapshot_target:
1555 dm_exception_store_exit();
1559 static void __exit dm_snapshot_exit(void)
1561 destroy_workqueue(ksnapd);
1563 dm_unregister_target(&snapshot_target);
1564 dm_unregister_target(&origin_target);
1567 kmem_cache_destroy(pending_cache);
1568 kmem_cache_destroy(exception_cache);
1569 kmem_cache_destroy(tracked_chunk_cache);
1571 dm_exception_store_exit();
1575 module_init(dm_snapshot_init);
1576 module_exit(dm_snapshot_exit);
1578 MODULE_DESCRIPTION(DM_NAME " snapshot target");
1579 MODULE_AUTHOR("Joe Thornber");
1580 MODULE_LICENSE("GPL");