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,
474 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
477 * Calculate based on the size of the original volume or
480 cow_dev_size = get_dev_size(cow->bdev);
481 origin_dev_size = get_dev_size(s->origin->bdev);
482 max_buckets = calc_max_buckets();
484 hash_size = min(origin_dev_size, cow_dev_size) >> chunk_shift;
485 hash_size = min(hash_size, max_buckets);
487 hash_size = rounddown_pow_of_two(hash_size);
488 if (init_exception_table(&s->complete, hash_size,
489 DM_CHUNK_CONSECUTIVE_BITS))
493 * Allocate hash table for in-flight exceptions
494 * Make this smaller than the real hash table
500 if (init_exception_table(&s->pending, hash_size, 0)) {
501 exit_exception_table(&s->complete, exception_cache);
509 * Round a number up to the nearest 'size' boundary. size must
512 static ulong round_up(ulong n, ulong size)
515 return (n + size) & ~size;
518 static int set_chunk_size(struct dm_snapshot *s, const char *chunk_size_arg,
519 chunk_t *chunk_size, chunk_t *chunk_mask,
520 chunk_t *chunk_shift, struct dm_dev *cow,
523 unsigned long chunk_size_ulong;
526 chunk_size_ulong = simple_strtoul(chunk_size_arg, &value, 10);
527 if (*chunk_size_arg == '\0' || *value != '\0') {
528 *error = "Invalid chunk size";
532 if (!chunk_size_ulong) {
533 *chunk_size = *chunk_mask = *chunk_shift = 0;
538 * Chunk size must be multiple of page size. Silently
539 * round up if it's not.
541 chunk_size_ulong = round_up(chunk_size_ulong, PAGE_SIZE >> 9);
543 /* Check chunk_size is a power of 2 */
544 if (!is_power_of_2(chunk_size_ulong)) {
545 *error = "Chunk size is not a power of 2";
549 /* Validate the chunk size against the device block size */
550 if (chunk_size_ulong % (bdev_hardsect_size(cow->bdev) >> 9)) {
551 *error = "Chunk size is not a multiple of device blocksize";
555 *chunk_size = chunk_size_ulong;
556 *chunk_mask = chunk_size_ulong - 1;
557 *chunk_shift = ffs(chunk_size_ulong) - 1;
563 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
565 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
567 struct dm_snapshot *s;
573 chunk_t chunk_size, chunk_mask, chunk_shift;
577 ti->error = "requires exactly 4 arguments";
582 origin_path = argv[0];
584 persistent = toupper(*argv[2]);
586 if (persistent != 'P' && persistent != 'N') {
587 ti->error = "Persistent flag is not P or N";
592 s = kmalloc(sizeof(*s), GFP_KERNEL);
594 ti->error = "Cannot allocate snapshot context private "
600 r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin);
602 ti->error = "Cannot get origin device";
606 r = dm_get_device(ti, cow_path, 0, 0,
607 FMODE_READ | FMODE_WRITE, &cow);
609 dm_put_device(ti, s->origin);
610 ti->error = "Cannot get COW device";
614 r = set_chunk_size(s, argv[3], &chunk_size, &chunk_mask, &chunk_shift,
621 atomic_set(&s->pending_exceptions_count, 0);
622 init_rwsem(&s->lock);
623 spin_lock_init(&s->pe_lock);
625 /* Allocate hash table for COW data */
626 if (init_hash_tables(s, chunk_shift, cow)) {
627 ti->error = "Unable to allocate hash table space";
632 r = dm_exception_store_create(argv[2], ti, chunk_size, chunk_mask,
633 chunk_shift, cow, &s->store);
635 ti->error = "Couldn't create exception store";
642 r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client);
644 ti->error = "Could not create kcopyd client";
648 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
649 if (!s->pending_pool) {
650 ti->error = "Could not allocate mempool for pending exceptions";
654 s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS,
655 tracked_chunk_cache);
656 if (!s->tracked_chunk_pool) {
657 ti->error = "Could not allocate tracked_chunk mempool for "
659 goto bad_tracked_chunk_pool;
662 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
663 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
665 spin_lock_init(&s->tracked_chunk_lock);
667 /* Metadata must only be loaded into one table at once */
668 r = s->store->type->read_metadata(s->store, dm_add_exception,
671 ti->error = "Failed to read snapshot metadata";
672 goto bad_load_and_register;
675 DMWARN("Snapshot is marked invalid.");
678 bio_list_init(&s->queued_bios);
679 INIT_WORK(&s->queued_bios_work, flush_queued_bios);
681 /* Add snapshot to the list of snapshots for this origin */
682 /* Exceptions aren't triggered till snapshot_resume() is called */
683 if (register_snapshot(s)) {
685 ti->error = "Cannot register snapshot origin";
686 goto bad_load_and_register;
690 ti->split_io = s->store->chunk_size;
694 bad_load_and_register:
695 mempool_destroy(s->tracked_chunk_pool);
697 bad_tracked_chunk_pool:
698 mempool_destroy(s->pending_pool);
701 dm_kcopyd_client_destroy(s->kcopyd_client);
704 s->store->type->dtr(s->store);
707 exit_exception_table(&s->pending, pending_cache);
708 exit_exception_table(&s->complete, exception_cache);
711 dm_put_device(ti, cow);
712 dm_put_device(ti, s->origin);
721 static void __free_exceptions(struct dm_snapshot *s)
723 dm_kcopyd_client_destroy(s->kcopyd_client);
724 s->kcopyd_client = NULL;
726 exit_exception_table(&s->pending, pending_cache);
727 exit_exception_table(&s->complete, exception_cache);
729 s->store->type->dtr(s->store);
732 static void snapshot_dtr(struct dm_target *ti)
734 #ifdef CONFIG_DM_DEBUG
737 struct dm_snapshot *s = ti->private;
738 struct dm_dev *cow = s->store->cow;
740 flush_workqueue(ksnapd);
742 /* Prevent further origin writes from using this snapshot. */
743 /* After this returns there can be no new kcopyd jobs. */
744 unregister_snapshot(s);
746 while (atomic_read(&s->pending_exceptions_count))
749 * Ensure instructions in mempool_destroy aren't reordered
750 * before atomic_read.
754 #ifdef CONFIG_DM_DEBUG
755 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
756 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
759 mempool_destroy(s->tracked_chunk_pool);
761 __free_exceptions(s);
763 mempool_destroy(s->pending_pool);
765 dm_put_device(ti, s->origin);
766 dm_put_device(ti, cow);
772 * Flush a list of buffers.
774 static void flush_bios(struct bio *bio)
781 generic_make_request(bio);
786 static void flush_queued_bios(struct work_struct *work)
788 struct dm_snapshot *s =
789 container_of(work, struct dm_snapshot, queued_bios_work);
790 struct bio *queued_bios;
793 spin_lock_irqsave(&s->pe_lock, flags);
794 queued_bios = bio_list_get(&s->queued_bios);
795 spin_unlock_irqrestore(&s->pe_lock, flags);
797 flush_bios(queued_bios);
801 * Error a list of buffers.
803 static void error_bios(struct bio *bio)
815 static void __invalidate_snapshot(struct dm_snapshot *s, int err)
821 DMERR("Invalidating snapshot: Error reading/writing.");
822 else if (err == -ENOMEM)
823 DMERR("Invalidating snapshot: Unable to allocate exception.");
825 if (s->store->type->drop_snapshot)
826 s->store->type->drop_snapshot(s->store);
830 dm_table_event(s->store->ti->table);
833 static void get_pending_exception(struct dm_snap_pending_exception *pe)
835 atomic_inc(&pe->ref_count);
838 static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe)
840 struct dm_snap_pending_exception *primary_pe;
841 struct bio *origin_bios = NULL;
843 primary_pe = pe->primary_pe;
846 * If this pe is involved in a write to the origin and
847 * it is the last sibling to complete then release
848 * the bios for the original write to the origin.
851 atomic_dec_and_test(&primary_pe->ref_count)) {
852 origin_bios = bio_list_get(&primary_pe->origin_bios);
853 free_pending_exception(primary_pe);
857 * Free the pe if it's not linked to an origin write or if
858 * it's not itself a primary pe.
860 if (!primary_pe || primary_pe != pe)
861 free_pending_exception(pe);
866 static void pending_complete(struct dm_snap_pending_exception *pe, int success)
868 struct dm_snap_exception *e;
869 struct dm_snapshot *s = pe->snap;
870 struct bio *origin_bios = NULL;
871 struct bio *snapshot_bios = NULL;
875 /* Read/write error - snapshot is unusable */
876 down_write(&s->lock);
877 __invalidate_snapshot(s, -EIO);
882 e = alloc_exception();
884 down_write(&s->lock);
885 __invalidate_snapshot(s, -ENOMEM);
891 down_write(&s->lock);
899 * Check for conflicting reads. This is extremely improbable,
900 * so msleep(1) is sufficient and there is no need for a wait queue.
902 while (__chunk_is_tracked(s, pe->e.old_chunk))
906 * Add a proper exception, and remove the
907 * in-flight exception from the list.
909 insert_completed_exception(s, e);
912 remove_exception(&pe->e);
913 snapshot_bios = bio_list_get(&pe->snapshot_bios);
914 origin_bios = put_pending_exception(pe);
918 /* Submit any pending write bios */
920 error_bios(snapshot_bios);
922 flush_bios(snapshot_bios);
924 flush_bios(origin_bios);
927 static void commit_callback(void *context, int success)
929 struct dm_snap_pending_exception *pe = context;
931 pending_complete(pe, success);
935 * Called when the copy I/O has finished. kcopyd actually runs
936 * this code so don't block.
938 static void copy_callback(int read_err, unsigned long write_err, void *context)
940 struct dm_snap_pending_exception *pe = context;
941 struct dm_snapshot *s = pe->snap;
943 if (read_err || write_err)
944 pending_complete(pe, 0);
947 /* Update the metadata if we are persistent */
948 s->store->type->commit_exception(s->store, &pe->e,
949 commit_callback, pe);
953 * Dispatches the copy operation to kcopyd.
955 static void start_copy(struct dm_snap_pending_exception *pe)
957 struct dm_snapshot *s = pe->snap;
958 struct dm_io_region src, dest;
959 struct block_device *bdev = s->origin->bdev;
962 dev_size = get_dev_size(bdev);
965 src.sector = chunk_to_sector(s, pe->e.old_chunk);
966 src.count = min(s->store->chunk_size, dev_size - src.sector);
968 dest.bdev = s->store->cow->bdev;
969 dest.sector = chunk_to_sector(s, pe->e.new_chunk);
970 dest.count = src.count;
972 /* Hand over to kcopyd */
973 dm_kcopyd_copy(s->kcopyd_client,
974 &src, 1, &dest, 0, copy_callback, pe);
977 static struct dm_snap_pending_exception *
978 __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
980 struct dm_snap_exception *e = lookup_exception(&s->pending, chunk);
985 return container_of(e, struct dm_snap_pending_exception, e);
989 * Looks to see if this snapshot already has a pending exception
990 * for this chunk, otherwise it allocates a new one and inserts
991 * it into the pending table.
993 * NOTE: a write lock must be held on snap->lock before calling
996 static struct dm_snap_pending_exception *
997 __find_pending_exception(struct dm_snapshot *s,
998 struct dm_snap_pending_exception *pe, chunk_t chunk)
1000 struct dm_snap_pending_exception *pe2;
1002 pe2 = __lookup_pending_exception(s, chunk);
1004 free_pending_exception(pe);
1008 pe->e.old_chunk = chunk;
1009 bio_list_init(&pe->origin_bios);
1010 bio_list_init(&pe->snapshot_bios);
1011 pe->primary_pe = NULL;
1012 atomic_set(&pe->ref_count, 0);
1015 if (s->store->type->prepare_exception(s->store, &pe->e)) {
1016 free_pending_exception(pe);
1020 get_pending_exception(pe);
1021 insert_exception(&s->pending, &pe->e);
1026 static void remap_exception(struct dm_snapshot *s, struct dm_snap_exception *e,
1027 struct bio *bio, chunk_t chunk)
1029 bio->bi_bdev = s->store->cow->bdev;
1030 bio->bi_sector = chunk_to_sector(s, dm_chunk_number(e->new_chunk) +
1031 (chunk - e->old_chunk)) +
1032 (bio->bi_sector & s->store->chunk_mask);
1035 static int snapshot_map(struct dm_target *ti, struct bio *bio,
1036 union map_info *map_context)
1038 struct dm_snap_exception *e;
1039 struct dm_snapshot *s = ti->private;
1040 int r = DM_MAPIO_REMAPPED;
1042 struct dm_snap_pending_exception *pe = NULL;
1044 chunk = sector_to_chunk(s, bio->bi_sector);
1046 /* Full snapshots are not usable */
1047 /* To get here the table must be live so s->active is always set. */
1051 /* FIXME: should only take write lock if we need
1052 * to copy an exception */
1053 down_write(&s->lock);
1060 /* If the block is already remapped - use that, else remap it */
1061 e = lookup_exception(&s->complete, chunk);
1063 remap_exception(s, e, bio, chunk);
1068 * Write to snapshot - higher level takes care of RW/RO
1069 * flags so we should only get this if we are
1072 if (bio_rw(bio) == WRITE) {
1073 pe = __lookup_pending_exception(s, chunk);
1076 pe = alloc_pending_exception(s);
1077 down_write(&s->lock);
1080 free_pending_exception(pe);
1085 e = lookup_exception(&s->complete, chunk);
1087 free_pending_exception(pe);
1088 remap_exception(s, e, bio, chunk);
1092 pe = __find_pending_exception(s, pe, chunk);
1094 __invalidate_snapshot(s, -ENOMEM);
1100 remap_exception(s, &pe->e, bio, chunk);
1101 bio_list_add(&pe->snapshot_bios, bio);
1103 r = DM_MAPIO_SUBMITTED;
1106 /* this is protected by snap->lock */
1113 bio->bi_bdev = s->origin->bdev;
1114 map_context->ptr = track_chunk(s, chunk);
1123 static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1124 int error, union map_info *map_context)
1126 struct dm_snapshot *s = ti->private;
1127 struct dm_snap_tracked_chunk *c = map_context->ptr;
1130 stop_tracking_chunk(s, c);
1135 static void snapshot_resume(struct dm_target *ti)
1137 struct dm_snapshot *s = ti->private;
1139 down_write(&s->lock);
1144 static int snapshot_status(struct dm_target *ti, status_type_t type,
1145 char *result, unsigned int maxlen)
1147 struct dm_snapshot *snap = ti->private;
1150 case STATUSTYPE_INFO:
1152 snprintf(result, maxlen, "Invalid");
1154 if (snap->store->type->fraction_full) {
1155 sector_t numerator, denominator;
1156 snap->store->type->fraction_full(snap->store,
1159 snprintf(result, maxlen, "%llu/%llu",
1160 (unsigned long long)numerator,
1161 (unsigned long long)denominator);
1164 snprintf(result, maxlen, "Unknown");
1168 case STATUSTYPE_TABLE:
1170 * kdevname returns a static pointer so we need
1171 * to make private copies if the output is to
1174 snprintf(result, maxlen, "%s %s %s %llu",
1175 snap->origin->name, snap->store->cow->name,
1176 snap->store->type->name,
1177 (unsigned long long)snap->store->chunk_size);
1184 /*-----------------------------------------------------------------
1186 *---------------------------------------------------------------*/
1187 static int __origin_write(struct list_head *snapshots, struct bio *bio)
1189 int r = DM_MAPIO_REMAPPED, first = 0;
1190 struct dm_snapshot *snap;
1191 struct dm_snap_exception *e;
1192 struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL;
1194 LIST_HEAD(pe_queue);
1196 /* Do all the snapshots on this origin */
1197 list_for_each_entry (snap, snapshots, list) {
1199 down_write(&snap->lock);
1201 /* Only deal with valid and active snapshots */
1202 if (!snap->valid || !snap->active)
1205 /* Nothing to do if writing beyond end of snapshot */
1206 if (bio->bi_sector >= dm_table_get_size(snap->store->ti->table))
1210 * Remember, different snapshots can have
1211 * different chunk sizes.
1213 chunk = sector_to_chunk(snap, bio->bi_sector);
1216 * Check exception table to see if block
1217 * is already remapped in this snapshot
1218 * and trigger an exception if not.
1220 * ref_count is initialised to 1 so pending_complete()
1221 * won't destroy the primary_pe while we're inside this loop.
1223 e = lookup_exception(&snap->complete, chunk);
1227 pe = __lookup_pending_exception(snap, chunk);
1229 up_write(&snap->lock);
1230 pe = alloc_pending_exception(snap);
1231 down_write(&snap->lock);
1234 free_pending_exception(pe);
1238 e = lookup_exception(&snap->complete, chunk);
1240 free_pending_exception(pe);
1244 pe = __find_pending_exception(snap, pe, chunk);
1246 __invalidate_snapshot(snap, -ENOMEM);
1253 * Either every pe here has same
1254 * primary_pe or none has one yet.
1257 primary_pe = pe->primary_pe;
1263 bio_list_add(&primary_pe->origin_bios, bio);
1265 r = DM_MAPIO_SUBMITTED;
1268 if (!pe->primary_pe) {
1269 pe->primary_pe = primary_pe;
1270 get_pending_exception(primary_pe);
1275 list_add_tail(&pe->list, &pe_queue);
1279 up_write(&snap->lock);
1286 * If this is the first time we're processing this chunk and
1287 * ref_count is now 1 it means all the pending exceptions
1288 * got completed while we were in the loop above, so it falls to
1289 * us here to remove the primary_pe and submit any origin_bios.
1292 if (first && atomic_dec_and_test(&primary_pe->ref_count)) {
1293 flush_bios(bio_list_get(&primary_pe->origin_bios));
1294 free_pending_exception(primary_pe);
1295 /* If we got here, pe_queue is necessarily empty. */
1300 * Now that we have a complete pe list we can start the copying.
1302 list_for_each_entry_safe(pe, next_pe, &pe_queue, list)
1309 * Called on a write from the origin driver.
1311 static int do_origin(struct dm_dev *origin, struct bio *bio)
1314 int r = DM_MAPIO_REMAPPED;
1316 down_read(&_origins_lock);
1317 o = __lookup_origin(origin->bdev);
1319 r = __origin_write(&o->snapshots, bio);
1320 up_read(&_origins_lock);
1326 * Origin: maps a linear range of a device, with hooks for snapshotting.
1330 * Construct an origin mapping: <dev_path>
1331 * The context for an origin is merely a 'struct dm_dev *'
1332 * pointing to the real device.
1334 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1340 ti->error = "origin: incorrect number of arguments";
1344 r = dm_get_device(ti, argv[0], 0, ti->len,
1345 dm_table_get_mode(ti->table), &dev);
1347 ti->error = "Cannot get target device";
1355 static void origin_dtr(struct dm_target *ti)
1357 struct dm_dev *dev = ti->private;
1358 dm_put_device(ti, dev);
1361 static int origin_map(struct dm_target *ti, struct bio *bio,
1362 union map_info *map_context)
1364 struct dm_dev *dev = ti->private;
1365 bio->bi_bdev = dev->bdev;
1367 /* Only tell snapshots if this is a write */
1368 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
1371 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
1374 * Set the target "split_io" field to the minimum of all the snapshots'
1377 static void origin_resume(struct dm_target *ti)
1379 struct dm_dev *dev = ti->private;
1380 struct dm_snapshot *snap;
1382 chunk_t chunk_size = 0;
1384 down_read(&_origins_lock);
1385 o = __lookup_origin(dev->bdev);
1387 list_for_each_entry (snap, &o->snapshots, list)
1388 chunk_size = min_not_zero(chunk_size,
1389 snap->store->chunk_size);
1390 up_read(&_origins_lock);
1392 ti->split_io = chunk_size;
1395 static int origin_status(struct dm_target *ti, status_type_t type, char *result,
1396 unsigned int maxlen)
1398 struct dm_dev *dev = ti->private;
1401 case STATUSTYPE_INFO:
1405 case STATUSTYPE_TABLE:
1406 snprintf(result, maxlen, "%s", dev->name);
1413 static struct target_type origin_target = {
1414 .name = "snapshot-origin",
1415 .version = {1, 6, 0},
1416 .module = THIS_MODULE,
1420 .resume = origin_resume,
1421 .status = origin_status,
1424 static struct target_type snapshot_target = {
1426 .version = {1, 6, 0},
1427 .module = THIS_MODULE,
1428 .ctr = snapshot_ctr,
1429 .dtr = snapshot_dtr,
1430 .map = snapshot_map,
1431 .end_io = snapshot_end_io,
1432 .resume = snapshot_resume,
1433 .status = snapshot_status,
1436 static int __init dm_snapshot_init(void)
1440 r = dm_exception_store_init();
1442 DMERR("Failed to initialize exception stores");
1446 r = dm_register_target(&snapshot_target);
1448 DMERR("snapshot target register failed %d", r);
1452 r = dm_register_target(&origin_target);
1454 DMERR("Origin target register failed %d", r);
1458 r = init_origin_hash();
1460 DMERR("init_origin_hash failed.");
1464 exception_cache = KMEM_CACHE(dm_snap_exception, 0);
1465 if (!exception_cache) {
1466 DMERR("Couldn't create exception cache.");
1471 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
1472 if (!pending_cache) {
1473 DMERR("Couldn't create pending cache.");
1478 tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0);
1479 if (!tracked_chunk_cache) {
1480 DMERR("Couldn't create cache to track chunks in use.");
1485 ksnapd = create_singlethread_workqueue("ksnapd");
1487 DMERR("Failed to create ksnapd workqueue.");
1489 goto bad_pending_pool;
1495 kmem_cache_destroy(tracked_chunk_cache);
1497 kmem_cache_destroy(pending_cache);
1499 kmem_cache_destroy(exception_cache);
1503 dm_unregister_target(&origin_target);
1505 dm_unregister_target(&snapshot_target);
1509 static void __exit dm_snapshot_exit(void)
1511 destroy_workqueue(ksnapd);
1513 dm_unregister_target(&snapshot_target);
1514 dm_unregister_target(&origin_target);
1517 kmem_cache_destroy(pending_cache);
1518 kmem_cache_destroy(exception_cache);
1519 kmem_cache_destroy(tracked_chunk_cache);
1521 dm_exception_store_exit();
1525 module_init(dm_snapshot_init);
1526 module_exit(dm_snapshot_exit);
1528 MODULE_DESCRIPTION(DM_NAME " snapshot target");
1529 MODULE_AUTHOR("Joe Thornber");
1530 MODULE_LICENSE("GPL");