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)
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) >> s->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,
520 unsigned long chunk_size;
523 chunk_size = simple_strtoul(chunk_size_arg, &value, 10);
524 if (*chunk_size_arg == '\0' || *value != '\0') {
525 *error = "Invalid chunk size";
530 s->chunk_size = s->chunk_mask = s->chunk_shift = 0;
535 * Chunk size must be multiple of page size. Silently
536 * round up if it's not.
538 chunk_size = round_up(chunk_size, PAGE_SIZE >> 9);
540 /* Check chunk_size is a power of 2 */
541 if (!is_power_of_2(chunk_size)) {
542 *error = "Chunk size is not a power of 2";
546 /* Validate the chunk size against the device block size */
547 if (chunk_size % (bdev_hardsect_size(s->cow->bdev) >> 9)) {
548 *error = "Chunk size is not a multiple of device blocksize";
552 s->chunk_size = chunk_size;
553 s->chunk_mask = chunk_size - 1;
554 s->chunk_shift = ffs(chunk_size) - 1;
560 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
562 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
564 struct dm_snapshot *s;
572 ti->error = "requires exactly 4 arguments";
577 origin_path = argv[0];
579 persistent = toupper(*argv[2]);
581 if (persistent != 'P' && persistent != 'N') {
582 ti->error = "Persistent flag is not P or N";
587 s = kmalloc(sizeof(*s), GFP_KERNEL);
589 ti->error = "Cannot allocate snapshot context private "
595 r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin);
597 ti->error = "Cannot get origin device";
601 r = dm_get_device(ti, cow_path, 0, 0,
602 FMODE_READ | FMODE_WRITE, &s->cow);
604 dm_put_device(ti, s->origin);
605 ti->error = "Cannot get COW device";
609 r = set_chunk_size(s, argv[3], &ti->error);
615 atomic_set(&s->pending_exceptions_count, 0);
616 init_rwsem(&s->lock);
617 spin_lock_init(&s->pe_lock);
619 /* Allocate hash table for COW data */
620 if (init_hash_tables(s)) {
621 ti->error = "Unable to allocate hash table space";
626 r = dm_exception_store_create(argv[2], ti, &s->store);
628 ti->error = "Couldn't create exception store";
635 r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client);
637 ti->error = "Could not create kcopyd client";
641 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
642 if (!s->pending_pool) {
643 ti->error = "Could not allocate mempool for pending exceptions";
647 s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS,
648 tracked_chunk_cache);
649 if (!s->tracked_chunk_pool) {
650 ti->error = "Could not allocate tracked_chunk mempool for "
652 goto bad_tracked_chunk_pool;
655 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
656 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
658 spin_lock_init(&s->tracked_chunk_lock);
660 /* Metadata must only be loaded into one table at once */
661 r = s->store->type->read_metadata(s->store, dm_add_exception,
664 ti->error = "Failed to read snapshot metadata";
665 goto bad_load_and_register;
668 DMWARN("Snapshot is marked invalid.");
671 bio_list_init(&s->queued_bios);
672 INIT_WORK(&s->queued_bios_work, flush_queued_bios);
674 /* Add snapshot to the list of snapshots for this origin */
675 /* Exceptions aren't triggered till snapshot_resume() is called */
676 if (register_snapshot(s)) {
678 ti->error = "Cannot register snapshot origin";
679 goto bad_load_and_register;
683 ti->split_io = s->chunk_size;
687 bad_load_and_register:
688 mempool_destroy(s->tracked_chunk_pool);
690 bad_tracked_chunk_pool:
691 mempool_destroy(s->pending_pool);
694 dm_kcopyd_client_destroy(s->kcopyd_client);
697 s->store->type->dtr(s->store);
700 exit_exception_table(&s->pending, pending_cache);
701 exit_exception_table(&s->complete, exception_cache);
704 dm_put_device(ti, s->cow);
705 dm_put_device(ti, s->origin);
714 static void __free_exceptions(struct dm_snapshot *s)
716 dm_kcopyd_client_destroy(s->kcopyd_client);
717 s->kcopyd_client = NULL;
719 exit_exception_table(&s->pending, pending_cache);
720 exit_exception_table(&s->complete, exception_cache);
722 s->store->type->dtr(s->store);
725 static void snapshot_dtr(struct dm_target *ti)
727 #ifdef CONFIG_DM_DEBUG
730 struct dm_snapshot *s = ti->private;
732 flush_workqueue(ksnapd);
734 /* Prevent further origin writes from using this snapshot. */
735 /* After this returns there can be no new kcopyd jobs. */
736 unregister_snapshot(s);
738 while (atomic_read(&s->pending_exceptions_count))
741 * Ensure instructions in mempool_destroy aren't reordered
742 * before atomic_read.
746 #ifdef CONFIG_DM_DEBUG
747 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
748 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
751 mempool_destroy(s->tracked_chunk_pool);
753 __free_exceptions(s);
755 mempool_destroy(s->pending_pool);
757 dm_put_device(ti, s->origin);
758 dm_put_device(ti, s->cow);
764 * Flush a list of buffers.
766 static void flush_bios(struct bio *bio)
773 generic_make_request(bio);
778 static void flush_queued_bios(struct work_struct *work)
780 struct dm_snapshot *s =
781 container_of(work, struct dm_snapshot, queued_bios_work);
782 struct bio *queued_bios;
785 spin_lock_irqsave(&s->pe_lock, flags);
786 queued_bios = bio_list_get(&s->queued_bios);
787 spin_unlock_irqrestore(&s->pe_lock, flags);
789 flush_bios(queued_bios);
793 * Error a list of buffers.
795 static void error_bios(struct bio *bio)
807 static void __invalidate_snapshot(struct dm_snapshot *s, int err)
813 DMERR("Invalidating snapshot: Error reading/writing.");
814 else if (err == -ENOMEM)
815 DMERR("Invalidating snapshot: Unable to allocate exception.");
817 if (s->store->type->drop_snapshot)
818 s->store->type->drop_snapshot(s->store);
822 dm_table_event(s->store->ti->table);
825 static void get_pending_exception(struct dm_snap_pending_exception *pe)
827 atomic_inc(&pe->ref_count);
830 static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe)
832 struct dm_snap_pending_exception *primary_pe;
833 struct bio *origin_bios = NULL;
835 primary_pe = pe->primary_pe;
838 * If this pe is involved in a write to the origin and
839 * it is the last sibling to complete then release
840 * the bios for the original write to the origin.
843 atomic_dec_and_test(&primary_pe->ref_count)) {
844 origin_bios = bio_list_get(&primary_pe->origin_bios);
845 free_pending_exception(primary_pe);
849 * Free the pe if it's not linked to an origin write or if
850 * it's not itself a primary pe.
852 if (!primary_pe || primary_pe != pe)
853 free_pending_exception(pe);
858 static void pending_complete(struct dm_snap_pending_exception *pe, int success)
860 struct dm_snap_exception *e;
861 struct dm_snapshot *s = pe->snap;
862 struct bio *origin_bios = NULL;
863 struct bio *snapshot_bios = NULL;
867 /* Read/write error - snapshot is unusable */
868 down_write(&s->lock);
869 __invalidate_snapshot(s, -EIO);
874 e = alloc_exception();
876 down_write(&s->lock);
877 __invalidate_snapshot(s, -ENOMEM);
883 down_write(&s->lock);
891 * Check for conflicting reads. This is extremely improbable,
892 * so msleep(1) is sufficient and there is no need for a wait queue.
894 while (__chunk_is_tracked(s, pe->e.old_chunk))
898 * Add a proper exception, and remove the
899 * in-flight exception from the list.
901 insert_completed_exception(s, e);
904 remove_exception(&pe->e);
905 snapshot_bios = bio_list_get(&pe->snapshot_bios);
906 origin_bios = put_pending_exception(pe);
910 /* Submit any pending write bios */
912 error_bios(snapshot_bios);
914 flush_bios(snapshot_bios);
916 flush_bios(origin_bios);
919 static void commit_callback(void *context, int success)
921 struct dm_snap_pending_exception *pe = context;
923 pending_complete(pe, success);
927 * Called when the copy I/O has finished. kcopyd actually runs
928 * this code so don't block.
930 static void copy_callback(int read_err, unsigned long write_err, void *context)
932 struct dm_snap_pending_exception *pe = context;
933 struct dm_snapshot *s = pe->snap;
935 if (read_err || write_err)
936 pending_complete(pe, 0);
939 /* Update the metadata if we are persistent */
940 s->store->type->commit_exception(s->store, &pe->e,
941 commit_callback, pe);
945 * Dispatches the copy operation to kcopyd.
947 static void start_copy(struct dm_snap_pending_exception *pe)
949 struct dm_snapshot *s = pe->snap;
950 struct dm_io_region src, dest;
951 struct block_device *bdev = s->origin->bdev;
954 dev_size = get_dev_size(bdev);
957 src.sector = chunk_to_sector(s, pe->e.old_chunk);
958 src.count = min(s->chunk_size, dev_size - src.sector);
960 dest.bdev = s->cow->bdev;
961 dest.sector = chunk_to_sector(s, pe->e.new_chunk);
962 dest.count = src.count;
964 /* Hand over to kcopyd */
965 dm_kcopyd_copy(s->kcopyd_client,
966 &src, 1, &dest, 0, copy_callback, pe);
969 static struct dm_snap_pending_exception *
970 __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
972 struct dm_snap_exception *e = lookup_exception(&s->pending, chunk);
977 return container_of(e, struct dm_snap_pending_exception, e);
981 * Looks to see if this snapshot already has a pending exception
982 * for this chunk, otherwise it allocates a new one and inserts
983 * it into the pending table.
985 * NOTE: a write lock must be held on snap->lock before calling
988 static struct dm_snap_pending_exception *
989 __find_pending_exception(struct dm_snapshot *s,
990 struct dm_snap_pending_exception *pe, chunk_t chunk)
992 struct dm_snap_pending_exception *pe2;
994 pe2 = __lookup_pending_exception(s, chunk);
996 free_pending_exception(pe);
1000 pe->e.old_chunk = chunk;
1001 bio_list_init(&pe->origin_bios);
1002 bio_list_init(&pe->snapshot_bios);
1003 pe->primary_pe = NULL;
1004 atomic_set(&pe->ref_count, 0);
1007 if (s->store->type->prepare_exception(s->store, &pe->e)) {
1008 free_pending_exception(pe);
1012 get_pending_exception(pe);
1013 insert_exception(&s->pending, &pe->e);
1018 static void remap_exception(struct dm_snapshot *s, struct dm_snap_exception *e,
1019 struct bio *bio, chunk_t chunk)
1021 bio->bi_bdev = s->cow->bdev;
1022 bio->bi_sector = chunk_to_sector(s, dm_chunk_number(e->new_chunk) +
1023 (chunk - e->old_chunk)) +
1024 (bio->bi_sector & s->chunk_mask);
1027 static int snapshot_map(struct dm_target *ti, struct bio *bio,
1028 union map_info *map_context)
1030 struct dm_snap_exception *e;
1031 struct dm_snapshot *s = ti->private;
1032 int r = DM_MAPIO_REMAPPED;
1034 struct dm_snap_pending_exception *pe = NULL;
1036 chunk = sector_to_chunk(s, bio->bi_sector);
1038 /* Full snapshots are not usable */
1039 /* To get here the table must be live so s->active is always set. */
1043 /* FIXME: should only take write lock if we need
1044 * to copy an exception */
1045 down_write(&s->lock);
1052 /* If the block is already remapped - use that, else remap it */
1053 e = lookup_exception(&s->complete, chunk);
1055 remap_exception(s, e, bio, chunk);
1060 * Write to snapshot - higher level takes care of RW/RO
1061 * flags so we should only get this if we are
1064 if (bio_rw(bio) == WRITE) {
1065 pe = __lookup_pending_exception(s, chunk);
1068 pe = alloc_pending_exception(s);
1069 down_write(&s->lock);
1072 free_pending_exception(pe);
1077 e = lookup_exception(&s->complete, chunk);
1079 free_pending_exception(pe);
1080 remap_exception(s, e, bio, chunk);
1084 pe = __find_pending_exception(s, pe, chunk);
1086 __invalidate_snapshot(s, -ENOMEM);
1092 remap_exception(s, &pe->e, bio, chunk);
1093 bio_list_add(&pe->snapshot_bios, bio);
1095 r = DM_MAPIO_SUBMITTED;
1098 /* this is protected by snap->lock */
1105 bio->bi_bdev = s->origin->bdev;
1106 map_context->ptr = track_chunk(s, chunk);
1115 static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1116 int error, union map_info *map_context)
1118 struct dm_snapshot *s = ti->private;
1119 struct dm_snap_tracked_chunk *c = map_context->ptr;
1122 stop_tracking_chunk(s, c);
1127 static void snapshot_resume(struct dm_target *ti)
1129 struct dm_snapshot *s = ti->private;
1131 down_write(&s->lock);
1136 static int snapshot_status(struct dm_target *ti, status_type_t type,
1137 char *result, unsigned int maxlen)
1139 struct dm_snapshot *snap = ti->private;
1142 case STATUSTYPE_INFO:
1144 snprintf(result, maxlen, "Invalid");
1146 if (snap->store->type->fraction_full) {
1147 sector_t numerator, denominator;
1148 snap->store->type->fraction_full(snap->store,
1151 snprintf(result, maxlen, "%llu/%llu",
1152 (unsigned long long)numerator,
1153 (unsigned long long)denominator);
1156 snprintf(result, maxlen, "Unknown");
1160 case STATUSTYPE_TABLE:
1162 * kdevname returns a static pointer so we need
1163 * to make private copies if the output is to
1166 snprintf(result, maxlen, "%s %s %s %llu",
1167 snap->origin->name, snap->cow->name,
1168 snap->store->type->name,
1169 (unsigned long long)snap->chunk_size);
1176 /*-----------------------------------------------------------------
1178 *---------------------------------------------------------------*/
1179 static int __origin_write(struct list_head *snapshots, struct bio *bio)
1181 int r = DM_MAPIO_REMAPPED, first = 0;
1182 struct dm_snapshot *snap;
1183 struct dm_snap_exception *e;
1184 struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL;
1186 LIST_HEAD(pe_queue);
1188 /* Do all the snapshots on this origin */
1189 list_for_each_entry (snap, snapshots, list) {
1191 down_write(&snap->lock);
1193 /* Only deal with valid and active snapshots */
1194 if (!snap->valid || !snap->active)
1197 /* Nothing to do if writing beyond end of snapshot */
1198 if (bio->bi_sector >= dm_table_get_size(snap->store->ti->table))
1202 * Remember, different snapshots can have
1203 * different chunk sizes.
1205 chunk = sector_to_chunk(snap, bio->bi_sector);
1208 * Check exception table to see if block
1209 * is already remapped in this snapshot
1210 * and trigger an exception if not.
1212 * ref_count is initialised to 1 so pending_complete()
1213 * won't destroy the primary_pe while we're inside this loop.
1215 e = lookup_exception(&snap->complete, chunk);
1219 pe = __lookup_pending_exception(snap, chunk);
1221 up_write(&snap->lock);
1222 pe = alloc_pending_exception(snap);
1223 down_write(&snap->lock);
1226 free_pending_exception(pe);
1230 e = lookup_exception(&snap->complete, chunk);
1232 free_pending_exception(pe);
1236 pe = __find_pending_exception(snap, pe, chunk);
1238 __invalidate_snapshot(snap, -ENOMEM);
1245 * Either every pe here has same
1246 * primary_pe or none has one yet.
1249 primary_pe = pe->primary_pe;
1255 bio_list_add(&primary_pe->origin_bios, bio);
1257 r = DM_MAPIO_SUBMITTED;
1260 if (!pe->primary_pe) {
1261 pe->primary_pe = primary_pe;
1262 get_pending_exception(primary_pe);
1267 list_add_tail(&pe->list, &pe_queue);
1271 up_write(&snap->lock);
1278 * If this is the first time we're processing this chunk and
1279 * ref_count is now 1 it means all the pending exceptions
1280 * got completed while we were in the loop above, so it falls to
1281 * us here to remove the primary_pe and submit any origin_bios.
1284 if (first && atomic_dec_and_test(&primary_pe->ref_count)) {
1285 flush_bios(bio_list_get(&primary_pe->origin_bios));
1286 free_pending_exception(primary_pe);
1287 /* If we got here, pe_queue is necessarily empty. */
1292 * Now that we have a complete pe list we can start the copying.
1294 list_for_each_entry_safe(pe, next_pe, &pe_queue, list)
1301 * Called on a write from the origin driver.
1303 static int do_origin(struct dm_dev *origin, struct bio *bio)
1306 int r = DM_MAPIO_REMAPPED;
1308 down_read(&_origins_lock);
1309 o = __lookup_origin(origin->bdev);
1311 r = __origin_write(&o->snapshots, bio);
1312 up_read(&_origins_lock);
1318 * Origin: maps a linear range of a device, with hooks for snapshotting.
1322 * Construct an origin mapping: <dev_path>
1323 * The context for an origin is merely a 'struct dm_dev *'
1324 * pointing to the real device.
1326 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1332 ti->error = "origin: incorrect number of arguments";
1336 r = dm_get_device(ti, argv[0], 0, ti->len,
1337 dm_table_get_mode(ti->table), &dev);
1339 ti->error = "Cannot get target device";
1347 static void origin_dtr(struct dm_target *ti)
1349 struct dm_dev *dev = ti->private;
1350 dm_put_device(ti, dev);
1353 static int origin_map(struct dm_target *ti, struct bio *bio,
1354 union map_info *map_context)
1356 struct dm_dev *dev = ti->private;
1357 bio->bi_bdev = dev->bdev;
1359 /* Only tell snapshots if this is a write */
1360 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
1363 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
1366 * Set the target "split_io" field to the minimum of all the snapshots'
1369 static void origin_resume(struct dm_target *ti)
1371 struct dm_dev *dev = ti->private;
1372 struct dm_snapshot *snap;
1374 chunk_t chunk_size = 0;
1376 down_read(&_origins_lock);
1377 o = __lookup_origin(dev->bdev);
1379 list_for_each_entry (snap, &o->snapshots, list)
1380 chunk_size = min_not_zero(chunk_size, snap->chunk_size);
1381 up_read(&_origins_lock);
1383 ti->split_io = chunk_size;
1386 static int origin_status(struct dm_target *ti, status_type_t type, char *result,
1387 unsigned int maxlen)
1389 struct dm_dev *dev = ti->private;
1392 case STATUSTYPE_INFO:
1396 case STATUSTYPE_TABLE:
1397 snprintf(result, maxlen, "%s", dev->name);
1404 static struct target_type origin_target = {
1405 .name = "snapshot-origin",
1406 .version = {1, 6, 0},
1407 .module = THIS_MODULE,
1411 .resume = origin_resume,
1412 .status = origin_status,
1415 static struct target_type snapshot_target = {
1417 .version = {1, 6, 0},
1418 .module = THIS_MODULE,
1419 .ctr = snapshot_ctr,
1420 .dtr = snapshot_dtr,
1421 .map = snapshot_map,
1422 .end_io = snapshot_end_io,
1423 .resume = snapshot_resume,
1424 .status = snapshot_status,
1427 static int __init dm_snapshot_init(void)
1431 r = dm_exception_store_init();
1433 DMERR("Failed to initialize exception stores");
1437 r = dm_register_target(&snapshot_target);
1439 DMERR("snapshot target register failed %d", r);
1443 r = dm_register_target(&origin_target);
1445 DMERR("Origin target register failed %d", r);
1449 r = init_origin_hash();
1451 DMERR("init_origin_hash failed.");
1455 exception_cache = KMEM_CACHE(dm_snap_exception, 0);
1456 if (!exception_cache) {
1457 DMERR("Couldn't create exception cache.");
1462 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
1463 if (!pending_cache) {
1464 DMERR("Couldn't create pending cache.");
1469 tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0);
1470 if (!tracked_chunk_cache) {
1471 DMERR("Couldn't create cache to track chunks in use.");
1476 ksnapd = create_singlethread_workqueue("ksnapd");
1478 DMERR("Failed to create ksnapd workqueue.");
1480 goto bad_pending_pool;
1486 kmem_cache_destroy(tracked_chunk_cache);
1488 kmem_cache_destroy(pending_cache);
1490 kmem_cache_destroy(exception_cache);
1494 dm_unregister_target(&origin_target);
1496 dm_unregister_target(&snapshot_target);
1500 static void __exit dm_snapshot_exit(void)
1502 destroy_workqueue(ksnapd);
1504 dm_unregister_target(&snapshot_target);
1505 dm_unregister_target(&origin_target);
1508 kmem_cache_destroy(pending_cache);
1509 kmem_cache_destroy(exception_cache);
1510 kmem_cache_destroy(tracked_chunk_cache);
1512 dm_exception_store_exit();
1516 module_init(dm_snapshot_init);
1517 module_exit(dm_snapshot_exit);
1519 MODULE_DESCRIPTION(DM_NAME " snapshot target");
1520 MODULE_AUTHOR("Joe Thornber");
1521 MODULE_LICENSE("GPL");