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/ctype.h>
11 #include <linux/device-mapper.h>
12 #include <linux/delay.h>
14 #include <linux/init.h>
15 #include <linux/kdev_t.h>
16 #include <linux/list.h>
17 #include <linux/mempool.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
20 #include <linux/vmalloc.h>
21 #include <linux/log2.h>
22 #include <linux/dm-kcopyd.h>
24 #include "dm-exception-store.h"
26 #include "dm-bio-list.h"
28 #define DM_MSG_PREFIX "snapshots"
31 * The percentage increment we will wake up users at
33 #define WAKE_UP_PERCENT 5
36 * kcopyd priority of snapshot operations
38 #define SNAPSHOT_COPY_PRIORITY 2
41 * Reserve 1MB for each snapshot initially (with minimum of 1 page).
43 #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1)
46 * The size of the mempool used to track chunks in use.
50 static struct workqueue_struct *ksnapd;
51 static void flush_queued_bios(struct work_struct *work);
53 struct dm_snap_pending_exception {
54 struct dm_snap_exception e;
57 * Origin buffers waiting for this to complete are held
60 struct bio_list origin_bios;
61 struct bio_list snapshot_bios;
64 * Short-term queue of pending exceptions prior to submission.
66 struct list_head list;
69 * The primary pending_exception is the one that holds
70 * the ref_count and the list of origin_bios for a
71 * group of pending_exceptions. It is always last to get freed.
72 * These fields get set up when writing to the origin.
74 struct dm_snap_pending_exception *primary_pe;
77 * Number of pending_exceptions processing this chunk.
78 * When this drops to zero we must complete the origin bios.
79 * If incrementing or decrementing this, hold pe->snap->lock for
80 * the sibling concerned and not pe->primary_pe->snap->lock unless
85 /* Pointer back to snapshot context */
86 struct dm_snapshot *snap;
89 * 1 indicates the exception has already been sent to
96 * Hash table mapping origin volumes to lists of snapshots and
97 * a lock to protect it
99 static struct kmem_cache *exception_cache;
100 static struct kmem_cache *pending_cache;
102 struct dm_snap_tracked_chunk {
103 struct hlist_node node;
107 static struct kmem_cache *tracked_chunk_cache;
109 static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s,
112 struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool,
118 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
119 hlist_add_head(&c->node,
120 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
121 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
126 static void stop_tracking_chunk(struct dm_snapshot *s,
127 struct dm_snap_tracked_chunk *c)
131 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
133 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
135 mempool_free(c, s->tracked_chunk_pool);
138 static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
140 struct dm_snap_tracked_chunk *c;
141 struct hlist_node *hn;
144 spin_lock_irq(&s->tracked_chunk_lock);
146 hlist_for_each_entry(c, hn,
147 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
148 if (c->chunk == chunk) {
154 spin_unlock_irq(&s->tracked_chunk_lock);
160 * One of these per registered origin, held in the snapshot_origins hash
163 /* The origin device */
164 struct block_device *bdev;
166 struct list_head hash_list;
168 /* List of snapshots for this origin */
169 struct list_head snapshots;
173 * Size of the hash table for origin volumes. If we make this
174 * the size of the minors list then it should be nearly perfect
176 #define ORIGIN_HASH_SIZE 256
177 #define ORIGIN_MASK 0xFF
178 static struct list_head *_origins;
179 static struct rw_semaphore _origins_lock;
181 static int init_origin_hash(void)
185 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
188 DMERR("unable to allocate memory");
192 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
193 INIT_LIST_HEAD(_origins + i);
194 init_rwsem(&_origins_lock);
199 static void exit_origin_hash(void)
204 static unsigned origin_hash(struct block_device *bdev)
206 return bdev->bd_dev & ORIGIN_MASK;
209 static struct origin *__lookup_origin(struct block_device *origin)
211 struct list_head *ol;
214 ol = &_origins[origin_hash(origin)];
215 list_for_each_entry (o, ol, hash_list)
216 if (bdev_equal(o->bdev, origin))
222 static void __insert_origin(struct origin *o)
224 struct list_head *sl = &_origins[origin_hash(o->bdev)];
225 list_add_tail(&o->hash_list, sl);
229 * Make a note of the snapshot and its origin so we can look it
230 * up when the origin has a write on it.
232 static int register_snapshot(struct dm_snapshot *snap)
234 struct origin *o, *new_o;
235 struct block_device *bdev = snap->origin->bdev;
237 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
241 down_write(&_origins_lock);
242 o = __lookup_origin(bdev);
250 /* Initialise the struct */
251 INIT_LIST_HEAD(&o->snapshots);
257 list_add_tail(&snap->list, &o->snapshots);
259 up_write(&_origins_lock);
263 static void unregister_snapshot(struct dm_snapshot *s)
267 down_write(&_origins_lock);
268 o = __lookup_origin(s->origin->bdev);
271 if (list_empty(&o->snapshots)) {
272 list_del(&o->hash_list);
276 up_write(&_origins_lock);
280 * Implementation of the exception hash tables.
281 * The lowest hash_shift bits of the chunk number are ignored, allowing
282 * some consecutive chunks to be grouped together.
284 static int init_exception_table(struct exception_table *et, uint32_t size,
289 et->hash_shift = hash_shift;
290 et->hash_mask = size - 1;
291 et->table = dm_vcalloc(size, sizeof(struct list_head));
295 for (i = 0; i < size; i++)
296 INIT_LIST_HEAD(et->table + i);
301 static void exit_exception_table(struct exception_table *et, struct kmem_cache *mem)
303 struct list_head *slot;
304 struct dm_snap_exception *ex, *next;
307 size = et->hash_mask + 1;
308 for (i = 0; i < size; i++) {
309 slot = et->table + i;
311 list_for_each_entry_safe (ex, next, slot, hash_list)
312 kmem_cache_free(mem, ex);
318 static uint32_t exception_hash(struct exception_table *et, chunk_t chunk)
320 return (chunk >> et->hash_shift) & et->hash_mask;
323 static void insert_exception(struct exception_table *eh,
324 struct dm_snap_exception *e)
326 struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)];
327 list_add(&e->hash_list, l);
330 static void remove_exception(struct dm_snap_exception *e)
332 list_del(&e->hash_list);
336 * Return the exception data for a sector, or NULL if not
339 static struct dm_snap_exception *lookup_exception(struct exception_table *et,
342 struct list_head *slot;
343 struct dm_snap_exception *e;
345 slot = &et->table[exception_hash(et, chunk)];
346 list_for_each_entry (e, slot, hash_list)
347 if (chunk >= e->old_chunk &&
348 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
354 static struct dm_snap_exception *alloc_exception(void)
356 struct dm_snap_exception *e;
358 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
360 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
365 static void free_exception(struct dm_snap_exception *e)
367 kmem_cache_free(exception_cache, e);
370 static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
372 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
375 atomic_inc(&s->pending_exceptions_count);
381 static void free_pending_exception(struct dm_snap_pending_exception *pe)
383 struct dm_snapshot *s = pe->snap;
385 mempool_free(pe, s->pending_pool);
386 smp_mb__before_atomic_dec();
387 atomic_dec(&s->pending_exceptions_count);
390 static void insert_completed_exception(struct dm_snapshot *s,
391 struct dm_snap_exception *new_e)
393 struct exception_table *eh = &s->complete;
395 struct dm_snap_exception *e = NULL;
397 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
399 /* Add immediately if this table doesn't support consecutive chunks */
403 /* List is ordered by old_chunk */
404 list_for_each_entry_reverse(e, l, hash_list) {
405 /* Insert after an existing chunk? */
406 if (new_e->old_chunk == (e->old_chunk +
407 dm_consecutive_chunk_count(e) + 1) &&
408 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
409 dm_consecutive_chunk_count(e) + 1)) {
410 dm_consecutive_chunk_count_inc(e);
411 free_exception(new_e);
415 /* Insert before an existing chunk? */
416 if (new_e->old_chunk == (e->old_chunk - 1) &&
417 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
418 dm_consecutive_chunk_count_inc(e);
421 free_exception(new_e);
425 if (new_e->old_chunk > e->old_chunk)
430 list_add(&new_e->hash_list, e ? &e->hash_list : l);
434 * Callback used by the exception stores to load exceptions when
437 static int dm_add_exception(void *context, chunk_t old, chunk_t new)
439 struct dm_snapshot *s = context;
440 struct dm_snap_exception *e;
442 e = alloc_exception();
448 /* Consecutive_count is implicitly initialised to zero */
451 insert_completed_exception(s, e);
459 static int calc_max_buckets(void)
461 /* use a fixed size of 2MB */
462 unsigned long mem = 2 * 1024 * 1024;
463 mem /= sizeof(struct list_head);
469 * Allocate room for a suitable hash table.
471 static int init_hash_tables(struct dm_snapshot *s, chunk_t chunk_shift)
473 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
476 * Calculate based on the size of the original volume or
479 cow_dev_size = get_dev_size(s->cow->bdev);
480 origin_dev_size = get_dev_size(s->origin->bdev);
481 max_buckets = calc_max_buckets();
483 hash_size = min(origin_dev_size, cow_dev_size) >> chunk_shift;
484 hash_size = min(hash_size, max_buckets);
486 hash_size = rounddown_pow_of_two(hash_size);
487 if (init_exception_table(&s->complete, hash_size,
488 DM_CHUNK_CONSECUTIVE_BITS))
492 * Allocate hash table for in-flight exceptions
493 * Make this smaller than the real hash table
499 if (init_exception_table(&s->pending, hash_size, 0)) {
500 exit_exception_table(&s->complete, exception_cache);
508 * Round a number up to the nearest 'size' boundary. size must
511 static ulong round_up(ulong n, ulong size)
514 return (n + size) & ~size;
517 static int set_chunk_size(struct dm_snapshot *s, const char *chunk_size_arg,
518 chunk_t *chunk_size, chunk_t *chunk_mask,
519 chunk_t *chunk_shift, char **error)
521 unsigned long chunk_size_ulong;
524 chunk_size_ulong = simple_strtoul(chunk_size_arg, &value, 10);
525 if (*chunk_size_arg == '\0' || *value != '\0') {
526 *error = "Invalid chunk size";
530 if (!chunk_size_ulong) {
531 *chunk_size = *chunk_mask = *chunk_shift = 0;
536 * Chunk size must be multiple of page size. Silently
537 * round up if it's not.
539 chunk_size_ulong = round_up(chunk_size_ulong, PAGE_SIZE >> 9);
541 /* Check chunk_size is a power of 2 */
542 if (!is_power_of_2(chunk_size_ulong)) {
543 *error = "Chunk size is not a power of 2";
547 /* Validate the chunk size against the device block size */
548 if (chunk_size_ulong % (bdev_hardsect_size(s->cow->bdev) >> 9)) {
549 *error = "Chunk size is not a multiple of device blocksize";
553 *chunk_size = chunk_size_ulong;
554 *chunk_mask = chunk_size_ulong - 1;
555 *chunk_shift = ffs(chunk_size_ulong) - 1;
561 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
563 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
565 struct dm_snapshot *s;
571 chunk_t chunk_size, chunk_mask, chunk_shift;
574 ti->error = "requires exactly 4 arguments";
579 origin_path = argv[0];
581 persistent = toupper(*argv[2]);
583 if (persistent != 'P' && persistent != 'N') {
584 ti->error = "Persistent flag is not P or N";
589 s = kmalloc(sizeof(*s), GFP_KERNEL);
591 ti->error = "Cannot allocate snapshot context private "
597 r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin);
599 ti->error = "Cannot get origin device";
603 r = dm_get_device(ti, cow_path, 0, 0,
604 FMODE_READ | FMODE_WRITE, &s->cow);
606 dm_put_device(ti, s->origin);
607 ti->error = "Cannot get COW device";
611 r = set_chunk_size(s, argv[3], &chunk_size, &chunk_mask, &chunk_shift,
618 atomic_set(&s->pending_exceptions_count, 0);
619 init_rwsem(&s->lock);
620 spin_lock_init(&s->pe_lock);
622 /* Allocate hash table for COW data */
623 if (init_hash_tables(s, chunk_shift)) {
624 ti->error = "Unable to allocate hash table space";
629 r = dm_exception_store_create(argv[2], ti, chunk_size, chunk_mask,
630 chunk_shift, &s->store);
632 ti->error = "Couldn't create exception store";
639 r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client);
641 ti->error = "Could not create kcopyd client";
645 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
646 if (!s->pending_pool) {
647 ti->error = "Could not allocate mempool for pending exceptions";
651 s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS,
652 tracked_chunk_cache);
653 if (!s->tracked_chunk_pool) {
654 ti->error = "Could not allocate tracked_chunk mempool for "
656 goto bad_tracked_chunk_pool;
659 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
660 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
662 spin_lock_init(&s->tracked_chunk_lock);
664 /* Metadata must only be loaded into one table at once */
665 r = s->store->type->read_metadata(s->store, dm_add_exception,
668 ti->error = "Failed to read snapshot metadata";
669 goto bad_load_and_register;
672 DMWARN("Snapshot is marked invalid.");
675 bio_list_init(&s->queued_bios);
676 INIT_WORK(&s->queued_bios_work, flush_queued_bios);
678 /* Add snapshot to the list of snapshots for this origin */
679 /* Exceptions aren't triggered till snapshot_resume() is called */
680 if (register_snapshot(s)) {
682 ti->error = "Cannot register snapshot origin";
683 goto bad_load_and_register;
687 ti->split_io = s->store->chunk_size;
691 bad_load_and_register:
692 mempool_destroy(s->tracked_chunk_pool);
694 bad_tracked_chunk_pool:
695 mempool_destroy(s->pending_pool);
698 dm_kcopyd_client_destroy(s->kcopyd_client);
701 s->store->type->dtr(s->store);
704 exit_exception_table(&s->pending, pending_cache);
705 exit_exception_table(&s->complete, exception_cache);
708 dm_put_device(ti, s->cow);
709 dm_put_device(ti, s->origin);
718 static void __free_exceptions(struct dm_snapshot *s)
720 dm_kcopyd_client_destroy(s->kcopyd_client);
721 s->kcopyd_client = NULL;
723 exit_exception_table(&s->pending, pending_cache);
724 exit_exception_table(&s->complete, exception_cache);
726 s->store->type->dtr(s->store);
729 static void snapshot_dtr(struct dm_target *ti)
731 #ifdef CONFIG_DM_DEBUG
734 struct dm_snapshot *s = ti->private;
736 flush_workqueue(ksnapd);
738 /* Prevent further origin writes from using this snapshot. */
739 /* After this returns there can be no new kcopyd jobs. */
740 unregister_snapshot(s);
742 while (atomic_read(&s->pending_exceptions_count))
745 * Ensure instructions in mempool_destroy aren't reordered
746 * before atomic_read.
750 #ifdef CONFIG_DM_DEBUG
751 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
752 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
755 mempool_destroy(s->tracked_chunk_pool);
757 __free_exceptions(s);
759 mempool_destroy(s->pending_pool);
761 dm_put_device(ti, s->origin);
762 dm_put_device(ti, s->cow);
768 * Flush a list of buffers.
770 static void flush_bios(struct bio *bio)
777 generic_make_request(bio);
782 static void flush_queued_bios(struct work_struct *work)
784 struct dm_snapshot *s =
785 container_of(work, struct dm_snapshot, queued_bios_work);
786 struct bio *queued_bios;
789 spin_lock_irqsave(&s->pe_lock, flags);
790 queued_bios = bio_list_get(&s->queued_bios);
791 spin_unlock_irqrestore(&s->pe_lock, flags);
793 flush_bios(queued_bios);
797 * Error a list of buffers.
799 static void error_bios(struct bio *bio)
811 static void __invalidate_snapshot(struct dm_snapshot *s, int err)
817 DMERR("Invalidating snapshot: Error reading/writing.");
818 else if (err == -ENOMEM)
819 DMERR("Invalidating snapshot: Unable to allocate exception.");
821 if (s->store->type->drop_snapshot)
822 s->store->type->drop_snapshot(s->store);
826 dm_table_event(s->store->ti->table);
829 static void get_pending_exception(struct dm_snap_pending_exception *pe)
831 atomic_inc(&pe->ref_count);
834 static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe)
836 struct dm_snap_pending_exception *primary_pe;
837 struct bio *origin_bios = NULL;
839 primary_pe = pe->primary_pe;
842 * If this pe is involved in a write to the origin and
843 * it is the last sibling to complete then release
844 * the bios for the original write to the origin.
847 atomic_dec_and_test(&primary_pe->ref_count)) {
848 origin_bios = bio_list_get(&primary_pe->origin_bios);
849 free_pending_exception(primary_pe);
853 * Free the pe if it's not linked to an origin write or if
854 * it's not itself a primary pe.
856 if (!primary_pe || primary_pe != pe)
857 free_pending_exception(pe);
862 static void pending_complete(struct dm_snap_pending_exception *pe, int success)
864 struct dm_snap_exception *e;
865 struct dm_snapshot *s = pe->snap;
866 struct bio *origin_bios = NULL;
867 struct bio *snapshot_bios = NULL;
871 /* Read/write error - snapshot is unusable */
872 down_write(&s->lock);
873 __invalidate_snapshot(s, -EIO);
878 e = alloc_exception();
880 down_write(&s->lock);
881 __invalidate_snapshot(s, -ENOMEM);
887 down_write(&s->lock);
895 * Check for conflicting reads. This is extremely improbable,
896 * so msleep(1) is sufficient and there is no need for a wait queue.
898 while (__chunk_is_tracked(s, pe->e.old_chunk))
902 * Add a proper exception, and remove the
903 * in-flight exception from the list.
905 insert_completed_exception(s, e);
908 remove_exception(&pe->e);
909 snapshot_bios = bio_list_get(&pe->snapshot_bios);
910 origin_bios = put_pending_exception(pe);
914 /* Submit any pending write bios */
916 error_bios(snapshot_bios);
918 flush_bios(snapshot_bios);
920 flush_bios(origin_bios);
923 static void commit_callback(void *context, int success)
925 struct dm_snap_pending_exception *pe = context;
927 pending_complete(pe, success);
931 * Called when the copy I/O has finished. kcopyd actually runs
932 * this code so don't block.
934 static void copy_callback(int read_err, unsigned long write_err, void *context)
936 struct dm_snap_pending_exception *pe = context;
937 struct dm_snapshot *s = pe->snap;
939 if (read_err || write_err)
940 pending_complete(pe, 0);
943 /* Update the metadata if we are persistent */
944 s->store->type->commit_exception(s->store, &pe->e,
945 commit_callback, pe);
949 * Dispatches the copy operation to kcopyd.
951 static void start_copy(struct dm_snap_pending_exception *pe)
953 struct dm_snapshot *s = pe->snap;
954 struct dm_io_region src, dest;
955 struct block_device *bdev = s->origin->bdev;
958 dev_size = get_dev_size(bdev);
961 src.sector = chunk_to_sector(s, pe->e.old_chunk);
962 src.count = min(s->store->chunk_size, dev_size - src.sector);
964 dest.bdev = s->cow->bdev;
965 dest.sector = chunk_to_sector(s, pe->e.new_chunk);
966 dest.count = src.count;
968 /* Hand over to kcopyd */
969 dm_kcopyd_copy(s->kcopyd_client,
970 &src, 1, &dest, 0, copy_callback, pe);
973 static struct dm_snap_pending_exception *
974 __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
976 struct dm_snap_exception *e = lookup_exception(&s->pending, chunk);
981 return container_of(e, struct dm_snap_pending_exception, e);
985 * Looks to see if this snapshot already has a pending exception
986 * for this chunk, otherwise it allocates a new one and inserts
987 * it into the pending table.
989 * NOTE: a write lock must be held on snap->lock before calling
992 static struct dm_snap_pending_exception *
993 __find_pending_exception(struct dm_snapshot *s,
994 struct dm_snap_pending_exception *pe, chunk_t chunk)
996 struct dm_snap_pending_exception *pe2;
998 pe2 = __lookup_pending_exception(s, chunk);
1000 free_pending_exception(pe);
1004 pe->e.old_chunk = chunk;
1005 bio_list_init(&pe->origin_bios);
1006 bio_list_init(&pe->snapshot_bios);
1007 pe->primary_pe = NULL;
1008 atomic_set(&pe->ref_count, 0);
1011 if (s->store->type->prepare_exception(s->store, &pe->e)) {
1012 free_pending_exception(pe);
1016 get_pending_exception(pe);
1017 insert_exception(&s->pending, &pe->e);
1022 static void remap_exception(struct dm_snapshot *s, struct dm_snap_exception *e,
1023 struct bio *bio, chunk_t chunk)
1025 bio->bi_bdev = s->cow->bdev;
1026 bio->bi_sector = chunk_to_sector(s, dm_chunk_number(e->new_chunk) +
1027 (chunk - e->old_chunk)) +
1028 (bio->bi_sector & s->store->chunk_mask);
1031 static int snapshot_map(struct dm_target *ti, struct bio *bio,
1032 union map_info *map_context)
1034 struct dm_snap_exception *e;
1035 struct dm_snapshot *s = ti->private;
1036 int r = DM_MAPIO_REMAPPED;
1038 struct dm_snap_pending_exception *pe = NULL;
1040 chunk = sector_to_chunk(s, bio->bi_sector);
1042 /* Full snapshots are not usable */
1043 /* To get here the table must be live so s->active is always set. */
1047 /* FIXME: should only take write lock if we need
1048 * to copy an exception */
1049 down_write(&s->lock);
1056 /* If the block is already remapped - use that, else remap it */
1057 e = lookup_exception(&s->complete, chunk);
1059 remap_exception(s, e, bio, chunk);
1064 * Write to snapshot - higher level takes care of RW/RO
1065 * flags so we should only get this if we are
1068 if (bio_rw(bio) == WRITE) {
1069 pe = __lookup_pending_exception(s, chunk);
1072 pe = alloc_pending_exception(s);
1073 down_write(&s->lock);
1076 free_pending_exception(pe);
1081 e = lookup_exception(&s->complete, chunk);
1083 free_pending_exception(pe);
1084 remap_exception(s, e, bio, chunk);
1088 pe = __find_pending_exception(s, pe, chunk);
1090 __invalidate_snapshot(s, -ENOMEM);
1096 remap_exception(s, &pe->e, bio, chunk);
1097 bio_list_add(&pe->snapshot_bios, bio);
1099 r = DM_MAPIO_SUBMITTED;
1102 /* this is protected by snap->lock */
1109 bio->bi_bdev = s->origin->bdev;
1110 map_context->ptr = track_chunk(s, chunk);
1119 static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1120 int error, union map_info *map_context)
1122 struct dm_snapshot *s = ti->private;
1123 struct dm_snap_tracked_chunk *c = map_context->ptr;
1126 stop_tracking_chunk(s, c);
1131 static void snapshot_resume(struct dm_target *ti)
1133 struct dm_snapshot *s = ti->private;
1135 down_write(&s->lock);
1140 static int snapshot_status(struct dm_target *ti, status_type_t type,
1141 char *result, unsigned int maxlen)
1143 struct dm_snapshot *snap = ti->private;
1146 case STATUSTYPE_INFO:
1148 snprintf(result, maxlen, "Invalid");
1150 if (snap->store->type->fraction_full) {
1151 sector_t numerator, denominator;
1152 snap->store->type->fraction_full(snap->store,
1155 snprintf(result, maxlen, "%llu/%llu",
1156 (unsigned long long)numerator,
1157 (unsigned long long)denominator);
1160 snprintf(result, maxlen, "Unknown");
1164 case STATUSTYPE_TABLE:
1166 * kdevname returns a static pointer so we need
1167 * to make private copies if the output is to
1170 snprintf(result, maxlen, "%s %s %s %llu",
1171 snap->origin->name, snap->cow->name,
1172 snap->store->type->name,
1173 (unsigned long long)snap->store->chunk_size);
1180 /*-----------------------------------------------------------------
1182 *---------------------------------------------------------------*/
1183 static int __origin_write(struct list_head *snapshots, struct bio *bio)
1185 int r = DM_MAPIO_REMAPPED, first = 0;
1186 struct dm_snapshot *snap;
1187 struct dm_snap_exception *e;
1188 struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL;
1190 LIST_HEAD(pe_queue);
1192 /* Do all the snapshots on this origin */
1193 list_for_each_entry (snap, snapshots, list) {
1195 down_write(&snap->lock);
1197 /* Only deal with valid and active snapshots */
1198 if (!snap->valid || !snap->active)
1201 /* Nothing to do if writing beyond end of snapshot */
1202 if (bio->bi_sector >= dm_table_get_size(snap->store->ti->table))
1206 * Remember, different snapshots can have
1207 * different chunk sizes.
1209 chunk = sector_to_chunk(snap, bio->bi_sector);
1212 * Check exception table to see if block
1213 * is already remapped in this snapshot
1214 * and trigger an exception if not.
1216 * ref_count is initialised to 1 so pending_complete()
1217 * won't destroy the primary_pe while we're inside this loop.
1219 e = lookup_exception(&snap->complete, chunk);
1223 pe = __lookup_pending_exception(snap, chunk);
1225 up_write(&snap->lock);
1226 pe = alloc_pending_exception(snap);
1227 down_write(&snap->lock);
1230 free_pending_exception(pe);
1234 e = lookup_exception(&snap->complete, chunk);
1236 free_pending_exception(pe);
1240 pe = __find_pending_exception(snap, pe, chunk);
1242 __invalidate_snapshot(snap, -ENOMEM);
1249 * Either every pe here has same
1250 * primary_pe or none has one yet.
1253 primary_pe = pe->primary_pe;
1259 bio_list_add(&primary_pe->origin_bios, bio);
1261 r = DM_MAPIO_SUBMITTED;
1264 if (!pe->primary_pe) {
1265 pe->primary_pe = primary_pe;
1266 get_pending_exception(primary_pe);
1271 list_add_tail(&pe->list, &pe_queue);
1275 up_write(&snap->lock);
1282 * If this is the first time we're processing this chunk and
1283 * ref_count is now 1 it means all the pending exceptions
1284 * got completed while we were in the loop above, so it falls to
1285 * us here to remove the primary_pe and submit any origin_bios.
1288 if (first && atomic_dec_and_test(&primary_pe->ref_count)) {
1289 flush_bios(bio_list_get(&primary_pe->origin_bios));
1290 free_pending_exception(primary_pe);
1291 /* If we got here, pe_queue is necessarily empty. */
1296 * Now that we have a complete pe list we can start the copying.
1298 list_for_each_entry_safe(pe, next_pe, &pe_queue, list)
1305 * Called on a write from the origin driver.
1307 static int do_origin(struct dm_dev *origin, struct bio *bio)
1310 int r = DM_MAPIO_REMAPPED;
1312 down_read(&_origins_lock);
1313 o = __lookup_origin(origin->bdev);
1315 r = __origin_write(&o->snapshots, bio);
1316 up_read(&_origins_lock);
1322 * Origin: maps a linear range of a device, with hooks for snapshotting.
1326 * Construct an origin mapping: <dev_path>
1327 * The context for an origin is merely a 'struct dm_dev *'
1328 * pointing to the real device.
1330 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1336 ti->error = "origin: incorrect number of arguments";
1340 r = dm_get_device(ti, argv[0], 0, ti->len,
1341 dm_table_get_mode(ti->table), &dev);
1343 ti->error = "Cannot get target device";
1351 static void origin_dtr(struct dm_target *ti)
1353 struct dm_dev *dev = ti->private;
1354 dm_put_device(ti, dev);
1357 static int origin_map(struct dm_target *ti, struct bio *bio,
1358 union map_info *map_context)
1360 struct dm_dev *dev = ti->private;
1361 bio->bi_bdev = dev->bdev;
1363 /* Only tell snapshots if this is a write */
1364 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
1367 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
1370 * Set the target "split_io" field to the minimum of all the snapshots'
1373 static void origin_resume(struct dm_target *ti)
1375 struct dm_dev *dev = ti->private;
1376 struct dm_snapshot *snap;
1378 chunk_t chunk_size = 0;
1380 down_read(&_origins_lock);
1381 o = __lookup_origin(dev->bdev);
1383 list_for_each_entry (snap, &o->snapshots, list)
1384 chunk_size = min_not_zero(chunk_size,
1385 snap->store->chunk_size);
1386 up_read(&_origins_lock);
1388 ti->split_io = chunk_size;
1391 static int origin_status(struct dm_target *ti, status_type_t type, char *result,
1392 unsigned int maxlen)
1394 struct dm_dev *dev = ti->private;
1397 case STATUSTYPE_INFO:
1401 case STATUSTYPE_TABLE:
1402 snprintf(result, maxlen, "%s", dev->name);
1409 static struct target_type origin_target = {
1410 .name = "snapshot-origin",
1411 .version = {1, 6, 0},
1412 .module = THIS_MODULE,
1416 .resume = origin_resume,
1417 .status = origin_status,
1420 static struct target_type snapshot_target = {
1422 .version = {1, 6, 0},
1423 .module = THIS_MODULE,
1424 .ctr = snapshot_ctr,
1425 .dtr = snapshot_dtr,
1426 .map = snapshot_map,
1427 .end_io = snapshot_end_io,
1428 .resume = snapshot_resume,
1429 .status = snapshot_status,
1432 static int __init dm_snapshot_init(void)
1436 r = dm_exception_store_init();
1438 DMERR("Failed to initialize exception stores");
1442 r = dm_register_target(&snapshot_target);
1444 DMERR("snapshot target register failed %d", r);
1448 r = dm_register_target(&origin_target);
1450 DMERR("Origin target register failed %d", r);
1454 r = init_origin_hash();
1456 DMERR("init_origin_hash failed.");
1460 exception_cache = KMEM_CACHE(dm_snap_exception, 0);
1461 if (!exception_cache) {
1462 DMERR("Couldn't create exception cache.");
1467 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
1468 if (!pending_cache) {
1469 DMERR("Couldn't create pending cache.");
1474 tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0);
1475 if (!tracked_chunk_cache) {
1476 DMERR("Couldn't create cache to track chunks in use.");
1481 ksnapd = create_singlethread_workqueue("ksnapd");
1483 DMERR("Failed to create ksnapd workqueue.");
1485 goto bad_pending_pool;
1491 kmem_cache_destroy(tracked_chunk_cache);
1493 kmem_cache_destroy(pending_cache);
1495 kmem_cache_destroy(exception_cache);
1499 dm_unregister_target(&origin_target);
1501 dm_unregister_target(&snapshot_target);
1505 static void __exit dm_snapshot_exit(void)
1507 destroy_workqueue(ksnapd);
1509 dm_unregister_target(&snapshot_target);
1510 dm_unregister_target(&origin_target);
1513 kmem_cache_destroy(pending_cache);
1514 kmem_cache_destroy(exception_cache);
1515 kmem_cache_destroy(tracked_chunk_cache);
1517 dm_exception_store_exit();
1521 module_init(dm_snapshot_init);
1522 module_exit(dm_snapshot_exit);
1524 MODULE_DESCRIPTION(DM_NAME " snapshot target");
1525 MODULE_AUTHOR("Joe Thornber");
1526 MODULE_LICENSE("GPL");