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"
25 #include "dm-bio-list.h"
27 #define DM_MSG_PREFIX "snapshots"
30 * The percentage increment we will wake up users at
32 #define WAKE_UP_PERCENT 5
35 * kcopyd priority of snapshot operations
37 #define SNAPSHOT_COPY_PRIORITY 2
40 * Reserve 1MB for each snapshot initially (with minimum of 1 page).
42 #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1)
45 * The size of the mempool used to track chunks in use.
49 #define DM_TRACKED_CHUNK_HASH_SIZE 16
50 #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
51 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
53 struct exception_table {
56 struct list_head *table;
60 struct rw_semaphore lock;
62 struct dm_dev *origin;
64 /* List of snapshots per Origin */
65 struct list_head list;
67 /* You can't use a snapshot if this is 0 (e.g. if full) */
70 /* Origin writes don't trigger exceptions until this is set */
73 mempool_t *pending_pool;
75 atomic_t pending_exceptions_count;
77 struct exception_table pending;
78 struct exception_table complete;
81 * pe_lock protects all pending_exception operations and access
82 * as well as the snapshot_bios list.
86 /* The on disk metadata handler */
87 struct dm_exception_store *store;
89 struct dm_kcopyd_client *kcopyd_client;
91 /* Queue of snapshot writes for ksnapd to flush */
92 struct bio_list queued_bios;
93 struct work_struct queued_bios_work;
95 /* Chunks with outstanding reads */
96 mempool_t *tracked_chunk_pool;
97 spinlock_t tracked_chunk_lock;
98 struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
101 static struct workqueue_struct *ksnapd;
102 static void flush_queued_bios(struct work_struct *work);
104 static sector_t chunk_to_sector(struct dm_exception_store *store,
107 return chunk << store->chunk_shift;
110 static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
113 * There is only ever one instance of a particular block
114 * device so we can compare pointers safely.
119 struct dm_snap_pending_exception {
120 struct dm_snap_exception e;
123 * Origin buffers waiting for this to complete are held
126 struct bio_list origin_bios;
127 struct bio_list snapshot_bios;
130 * Short-term queue of pending exceptions prior to submission.
132 struct list_head list;
135 * The primary pending_exception is the one that holds
136 * the ref_count and the list of origin_bios for a
137 * group of pending_exceptions. It is always last to get freed.
138 * These fields get set up when writing to the origin.
140 struct dm_snap_pending_exception *primary_pe;
143 * Number of pending_exceptions processing this chunk.
144 * When this drops to zero we must complete the origin bios.
145 * If incrementing or decrementing this, hold pe->snap->lock for
146 * the sibling concerned and not pe->primary_pe->snap->lock unless
151 /* Pointer back to snapshot context */
152 struct dm_snapshot *snap;
155 * 1 indicates the exception has already been sent to
162 * Hash table mapping origin volumes to lists of snapshots and
163 * a lock to protect it
165 static struct kmem_cache *exception_cache;
166 static struct kmem_cache *pending_cache;
168 struct dm_snap_tracked_chunk {
169 struct hlist_node node;
173 static struct kmem_cache *tracked_chunk_cache;
175 static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s,
178 struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool,
184 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
185 hlist_add_head(&c->node,
186 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
187 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
192 static void stop_tracking_chunk(struct dm_snapshot *s,
193 struct dm_snap_tracked_chunk *c)
197 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
199 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
201 mempool_free(c, s->tracked_chunk_pool);
204 static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
206 struct dm_snap_tracked_chunk *c;
207 struct hlist_node *hn;
210 spin_lock_irq(&s->tracked_chunk_lock);
212 hlist_for_each_entry(c, hn,
213 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
214 if (c->chunk == chunk) {
220 spin_unlock_irq(&s->tracked_chunk_lock);
226 * One of these per registered origin, held in the snapshot_origins hash
229 /* The origin device */
230 struct block_device *bdev;
232 struct list_head hash_list;
234 /* List of snapshots for this origin */
235 struct list_head snapshots;
239 * Size of the hash table for origin volumes. If we make this
240 * the size of the minors list then it should be nearly perfect
242 #define ORIGIN_HASH_SIZE 256
243 #define ORIGIN_MASK 0xFF
244 static struct list_head *_origins;
245 static struct rw_semaphore _origins_lock;
247 static int init_origin_hash(void)
251 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
254 DMERR("unable to allocate memory");
258 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
259 INIT_LIST_HEAD(_origins + i);
260 init_rwsem(&_origins_lock);
265 static void exit_origin_hash(void)
270 static unsigned origin_hash(struct block_device *bdev)
272 return bdev->bd_dev & ORIGIN_MASK;
275 static struct origin *__lookup_origin(struct block_device *origin)
277 struct list_head *ol;
280 ol = &_origins[origin_hash(origin)];
281 list_for_each_entry (o, ol, hash_list)
282 if (bdev_equal(o->bdev, origin))
288 static void __insert_origin(struct origin *o)
290 struct list_head *sl = &_origins[origin_hash(o->bdev)];
291 list_add_tail(&o->hash_list, sl);
295 * Make a note of the snapshot and its origin so we can look it
296 * up when the origin has a write on it.
298 static int register_snapshot(struct dm_snapshot *snap)
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 list_add_tail(&snap->list, &o->snapshots);
325 up_write(&_origins_lock);
329 static void unregister_snapshot(struct dm_snapshot *s)
333 down_write(&_origins_lock);
334 o = __lookup_origin(s->origin->bdev);
337 if (list_empty(&o->snapshots)) {
338 list_del(&o->hash_list);
342 up_write(&_origins_lock);
346 * Implementation of the exception hash tables.
347 * The lowest hash_shift bits of the chunk number are ignored, allowing
348 * some consecutive chunks to be grouped together.
350 static int init_exception_table(struct exception_table *et, uint32_t size,
355 et->hash_shift = hash_shift;
356 et->hash_mask = size - 1;
357 et->table = dm_vcalloc(size, sizeof(struct list_head));
361 for (i = 0; i < size; i++)
362 INIT_LIST_HEAD(et->table + i);
367 static void exit_exception_table(struct exception_table *et, struct kmem_cache *mem)
369 struct list_head *slot;
370 struct dm_snap_exception *ex, *next;
373 size = et->hash_mask + 1;
374 for (i = 0; i < size; i++) {
375 slot = et->table + i;
377 list_for_each_entry_safe (ex, next, slot, hash_list)
378 kmem_cache_free(mem, ex);
384 static uint32_t exception_hash(struct exception_table *et, chunk_t chunk)
386 return (chunk >> et->hash_shift) & et->hash_mask;
389 static void insert_exception(struct exception_table *eh,
390 struct dm_snap_exception *e)
392 struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)];
393 list_add(&e->hash_list, l);
396 static void remove_exception(struct dm_snap_exception *e)
398 list_del(&e->hash_list);
402 * Return the exception data for a sector, or NULL if not
405 static struct dm_snap_exception *lookup_exception(struct exception_table *et,
408 struct list_head *slot;
409 struct dm_snap_exception *e;
411 slot = &et->table[exception_hash(et, chunk)];
412 list_for_each_entry (e, slot, hash_list)
413 if (chunk >= e->old_chunk &&
414 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
420 static struct dm_snap_exception *alloc_exception(void)
422 struct dm_snap_exception *e;
424 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
426 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
431 static void free_exception(struct dm_snap_exception *e)
433 kmem_cache_free(exception_cache, e);
436 static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
438 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
441 atomic_inc(&s->pending_exceptions_count);
447 static void free_pending_exception(struct dm_snap_pending_exception *pe)
449 struct dm_snapshot *s = pe->snap;
451 mempool_free(pe, s->pending_pool);
452 smp_mb__before_atomic_dec();
453 atomic_dec(&s->pending_exceptions_count);
456 static void insert_completed_exception(struct dm_snapshot *s,
457 struct dm_snap_exception *new_e)
459 struct exception_table *eh = &s->complete;
461 struct dm_snap_exception *e = NULL;
463 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
465 /* Add immediately if this table doesn't support consecutive chunks */
469 /* List is ordered by old_chunk */
470 list_for_each_entry_reverse(e, l, hash_list) {
471 /* Insert after an existing chunk? */
472 if (new_e->old_chunk == (e->old_chunk +
473 dm_consecutive_chunk_count(e) + 1) &&
474 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
475 dm_consecutive_chunk_count(e) + 1)) {
476 dm_consecutive_chunk_count_inc(e);
477 free_exception(new_e);
481 /* Insert before an existing chunk? */
482 if (new_e->old_chunk == (e->old_chunk - 1) &&
483 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
484 dm_consecutive_chunk_count_inc(e);
487 free_exception(new_e);
491 if (new_e->old_chunk > e->old_chunk)
496 list_add(&new_e->hash_list, e ? &e->hash_list : l);
500 * Callback used by the exception stores to load exceptions when
503 static int dm_add_exception(void *context, chunk_t old, chunk_t new)
505 struct dm_snapshot *s = context;
506 struct dm_snap_exception *e;
508 e = alloc_exception();
514 /* Consecutive_count is implicitly initialised to zero */
517 insert_completed_exception(s, e);
525 static int calc_max_buckets(void)
527 /* use a fixed size of 2MB */
528 unsigned long mem = 2 * 1024 * 1024;
529 mem /= sizeof(struct list_head);
535 * Allocate room for a suitable hash table.
537 static int init_hash_tables(struct dm_snapshot *s)
539 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
542 * Calculate based on the size of the original volume or
545 cow_dev_size = get_dev_size(s->store->cow->bdev);
546 origin_dev_size = get_dev_size(s->origin->bdev);
547 max_buckets = calc_max_buckets();
549 hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift;
550 hash_size = min(hash_size, max_buckets);
552 hash_size = rounddown_pow_of_two(hash_size);
553 if (init_exception_table(&s->complete, hash_size,
554 DM_CHUNK_CONSECUTIVE_BITS))
558 * Allocate hash table for in-flight exceptions
559 * Make this smaller than the real hash table
565 if (init_exception_table(&s->pending, hash_size, 0)) {
566 exit_exception_table(&s->complete, exception_cache);
574 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
576 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
578 struct dm_snapshot *s;
582 struct dm_exception_store *store;
586 ti->error = "requires exactly 4 arguments";
591 origin_path = argv[0];
595 r = dm_exception_store_create(ti, argc, argv, &args_used, &store);
597 ti->error = "Couldn't create exception store";
605 s = kmalloc(sizeof(*s), GFP_KERNEL);
607 ti->error = "Cannot allocate snapshot context private "
613 r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin);
615 ti->error = "Cannot get origin device";
622 atomic_set(&s->pending_exceptions_count, 0);
623 init_rwsem(&s->lock);
624 spin_lock_init(&s->pe_lock);
626 /* Allocate hash table for COW data */
627 if (init_hash_tables(s)) {
628 ti->error = "Unable to allocate hash table space";
630 goto bad_hash_tables;
633 r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client);
635 ti->error = "Could not create kcopyd client";
639 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
640 if (!s->pending_pool) {
641 ti->error = "Could not allocate mempool for pending exceptions";
642 goto bad_pending_pool;
645 s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS,
646 tracked_chunk_cache);
647 if (!s->tracked_chunk_pool) {
648 ti->error = "Could not allocate tracked_chunk mempool for "
650 goto bad_tracked_chunk_pool;
653 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
654 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
656 spin_lock_init(&s->tracked_chunk_lock);
658 /* Metadata must only be loaded into one table at once */
659 r = s->store->type->read_metadata(s->store, dm_add_exception,
662 ti->error = "Failed to read snapshot metadata";
663 goto bad_load_and_register;
666 DMWARN("Snapshot is marked invalid.");
669 bio_list_init(&s->queued_bios);
670 INIT_WORK(&s->queued_bios_work, flush_queued_bios);
672 /* Add snapshot to the list of snapshots for this origin */
673 /* Exceptions aren't triggered till snapshot_resume() is called */
674 if (register_snapshot(s)) {
676 ti->error = "Cannot register snapshot origin";
677 goto bad_load_and_register;
681 ti->split_io = s->store->chunk_size;
685 bad_load_and_register:
686 mempool_destroy(s->tracked_chunk_pool);
688 bad_tracked_chunk_pool:
689 mempool_destroy(s->pending_pool);
692 dm_kcopyd_client_destroy(s->kcopyd_client);
695 exit_exception_table(&s->pending, pending_cache);
696 exit_exception_table(&s->complete, exception_cache);
699 dm_put_device(ti, s->origin);
705 dm_exception_store_destroy(store);
711 static void __free_exceptions(struct dm_snapshot *s)
713 dm_kcopyd_client_destroy(s->kcopyd_client);
714 s->kcopyd_client = NULL;
716 exit_exception_table(&s->pending, pending_cache);
717 exit_exception_table(&s->complete, exception_cache);
720 static void snapshot_dtr(struct dm_target *ti)
722 #ifdef CONFIG_DM_DEBUG
725 struct dm_snapshot *s = ti->private;
727 flush_workqueue(ksnapd);
729 /* Prevent further origin writes from using this snapshot. */
730 /* After this returns there can be no new kcopyd jobs. */
731 unregister_snapshot(s);
733 while (atomic_read(&s->pending_exceptions_count))
736 * Ensure instructions in mempool_destroy aren't reordered
737 * before atomic_read.
741 #ifdef CONFIG_DM_DEBUG
742 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
743 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
746 mempool_destroy(s->tracked_chunk_pool);
748 __free_exceptions(s);
750 mempool_destroy(s->pending_pool);
752 dm_put_device(ti, s->origin);
754 dm_exception_store_destroy(s->store);
760 * Flush a list of buffers.
762 static void flush_bios(struct bio *bio)
769 generic_make_request(bio);
774 static void flush_queued_bios(struct work_struct *work)
776 struct dm_snapshot *s =
777 container_of(work, struct dm_snapshot, queued_bios_work);
778 struct bio *queued_bios;
781 spin_lock_irqsave(&s->pe_lock, flags);
782 queued_bios = bio_list_get(&s->queued_bios);
783 spin_unlock_irqrestore(&s->pe_lock, flags);
785 flush_bios(queued_bios);
789 * Error a list of buffers.
791 static void error_bios(struct bio *bio)
803 static void __invalidate_snapshot(struct dm_snapshot *s, int err)
809 DMERR("Invalidating snapshot: Error reading/writing.");
810 else if (err == -ENOMEM)
811 DMERR("Invalidating snapshot: Unable to allocate exception.");
813 if (s->store->type->drop_snapshot)
814 s->store->type->drop_snapshot(s->store);
818 dm_table_event(s->store->ti->table);
821 static void get_pending_exception(struct dm_snap_pending_exception *pe)
823 atomic_inc(&pe->ref_count);
826 static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe)
828 struct dm_snap_pending_exception *primary_pe;
829 struct bio *origin_bios = NULL;
831 primary_pe = pe->primary_pe;
834 * If this pe is involved in a write to the origin and
835 * it is the last sibling to complete then release
836 * the bios for the original write to the origin.
839 atomic_dec_and_test(&primary_pe->ref_count)) {
840 origin_bios = bio_list_get(&primary_pe->origin_bios);
841 free_pending_exception(primary_pe);
845 * Free the pe if it's not linked to an origin write or if
846 * it's not itself a primary pe.
848 if (!primary_pe || primary_pe != pe)
849 free_pending_exception(pe);
854 static void pending_complete(struct dm_snap_pending_exception *pe, int success)
856 struct dm_snap_exception *e;
857 struct dm_snapshot *s = pe->snap;
858 struct bio *origin_bios = NULL;
859 struct bio *snapshot_bios = NULL;
863 /* Read/write error - snapshot is unusable */
864 down_write(&s->lock);
865 __invalidate_snapshot(s, -EIO);
870 e = alloc_exception();
872 down_write(&s->lock);
873 __invalidate_snapshot(s, -ENOMEM);
879 down_write(&s->lock);
887 * Check for conflicting reads. This is extremely improbable,
888 * so msleep(1) is sufficient and there is no need for a wait queue.
890 while (__chunk_is_tracked(s, pe->e.old_chunk))
894 * Add a proper exception, and remove the
895 * in-flight exception from the list.
897 insert_completed_exception(s, e);
900 remove_exception(&pe->e);
901 snapshot_bios = bio_list_get(&pe->snapshot_bios);
902 origin_bios = put_pending_exception(pe);
906 /* Submit any pending write bios */
908 error_bios(snapshot_bios);
910 flush_bios(snapshot_bios);
912 flush_bios(origin_bios);
915 static void commit_callback(void *context, int success)
917 struct dm_snap_pending_exception *pe = context;
919 pending_complete(pe, success);
923 * Called when the copy I/O has finished. kcopyd actually runs
924 * this code so don't block.
926 static void copy_callback(int read_err, unsigned long write_err, void *context)
928 struct dm_snap_pending_exception *pe = context;
929 struct dm_snapshot *s = pe->snap;
931 if (read_err || write_err)
932 pending_complete(pe, 0);
935 /* Update the metadata if we are persistent */
936 s->store->type->commit_exception(s->store, &pe->e,
937 commit_callback, pe);
941 * Dispatches the copy operation to kcopyd.
943 static void start_copy(struct dm_snap_pending_exception *pe)
945 struct dm_snapshot *s = pe->snap;
946 struct dm_io_region src, dest;
947 struct block_device *bdev = s->origin->bdev;
950 dev_size = get_dev_size(bdev);
953 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
954 src.count = min(s->store->chunk_size, dev_size - src.sector);
956 dest.bdev = s->store->cow->bdev;
957 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
958 dest.count = src.count;
960 /* Hand over to kcopyd */
961 dm_kcopyd_copy(s->kcopyd_client,
962 &src, 1, &dest, 0, copy_callback, pe);
965 static struct dm_snap_pending_exception *
966 __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
968 struct dm_snap_exception *e = lookup_exception(&s->pending, chunk);
973 return container_of(e, struct dm_snap_pending_exception, e);
977 * Looks to see if this snapshot already has a pending exception
978 * for this chunk, otherwise it allocates a new one and inserts
979 * it into the pending table.
981 * NOTE: a write lock must be held on snap->lock before calling
984 static struct dm_snap_pending_exception *
985 __find_pending_exception(struct dm_snapshot *s,
986 struct dm_snap_pending_exception *pe, chunk_t chunk)
988 struct dm_snap_pending_exception *pe2;
990 pe2 = __lookup_pending_exception(s, chunk);
992 free_pending_exception(pe);
996 pe->e.old_chunk = chunk;
997 bio_list_init(&pe->origin_bios);
998 bio_list_init(&pe->snapshot_bios);
999 pe->primary_pe = NULL;
1000 atomic_set(&pe->ref_count, 0);
1003 if (s->store->type->prepare_exception(s->store, &pe->e)) {
1004 free_pending_exception(pe);
1008 get_pending_exception(pe);
1009 insert_exception(&s->pending, &pe->e);
1014 static void remap_exception(struct dm_snapshot *s, struct dm_snap_exception *e,
1015 struct bio *bio, chunk_t chunk)
1017 bio->bi_bdev = s->store->cow->bdev;
1018 bio->bi_sector = chunk_to_sector(s->store,
1019 dm_chunk_number(e->new_chunk) +
1020 (chunk - e->old_chunk)) +
1022 s->store->chunk_mask);
1025 static int snapshot_map(struct dm_target *ti, struct bio *bio,
1026 union map_info *map_context)
1028 struct dm_snap_exception *e;
1029 struct dm_snapshot *s = ti->private;
1030 int r = DM_MAPIO_REMAPPED;
1032 struct dm_snap_pending_exception *pe = NULL;
1034 chunk = sector_to_chunk(s->store, bio->bi_sector);
1036 /* Full snapshots are not usable */
1037 /* To get here the table must be live so s->active is always set. */
1041 /* FIXME: should only take write lock if we need
1042 * to copy an exception */
1043 down_write(&s->lock);
1050 /* If the block is already remapped - use that, else remap it */
1051 e = lookup_exception(&s->complete, chunk);
1053 remap_exception(s, e, bio, chunk);
1058 * Write to snapshot - higher level takes care of RW/RO
1059 * flags so we should only get this if we are
1062 if (bio_rw(bio) == WRITE) {
1063 pe = __lookup_pending_exception(s, chunk);
1066 pe = alloc_pending_exception(s);
1067 down_write(&s->lock);
1070 free_pending_exception(pe);
1075 e = lookup_exception(&s->complete, chunk);
1077 free_pending_exception(pe);
1078 remap_exception(s, e, bio, chunk);
1082 pe = __find_pending_exception(s, pe, chunk);
1084 __invalidate_snapshot(s, -ENOMEM);
1090 remap_exception(s, &pe->e, bio, chunk);
1091 bio_list_add(&pe->snapshot_bios, bio);
1093 r = DM_MAPIO_SUBMITTED;
1096 /* this is protected by snap->lock */
1103 bio->bi_bdev = s->origin->bdev;
1104 map_context->ptr = track_chunk(s, chunk);
1113 static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1114 int error, union map_info *map_context)
1116 struct dm_snapshot *s = ti->private;
1117 struct dm_snap_tracked_chunk *c = map_context->ptr;
1120 stop_tracking_chunk(s, c);
1125 static void snapshot_resume(struct dm_target *ti)
1127 struct dm_snapshot *s = ti->private;
1129 down_write(&s->lock);
1134 static int snapshot_status(struct dm_target *ti, status_type_t type,
1135 char *result, unsigned int maxlen)
1138 struct dm_snapshot *snap = ti->private;
1141 case STATUSTYPE_INFO:
1145 if (snap->store->type->fraction_full) {
1146 sector_t numerator, denominator;
1147 snap->store->type->fraction_full(snap->store,
1151 (unsigned long long)numerator,
1152 (unsigned long long)denominator);
1159 case STATUSTYPE_TABLE:
1161 * kdevname returns a static pointer so we need
1162 * to make private copies if the output is to
1165 DMEMIT("%s", snap->origin->name);
1166 snap->store->type->status(snap->store, type, result + sz,
1174 /*-----------------------------------------------------------------
1176 *---------------------------------------------------------------*/
1177 static int __origin_write(struct list_head *snapshots, struct bio *bio)
1179 int r = DM_MAPIO_REMAPPED, first = 0;
1180 struct dm_snapshot *snap;
1181 struct dm_snap_exception *e;
1182 struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL;
1184 LIST_HEAD(pe_queue);
1186 /* Do all the snapshots on this origin */
1187 list_for_each_entry (snap, snapshots, list) {
1189 down_write(&snap->lock);
1191 /* Only deal with valid and active snapshots */
1192 if (!snap->valid || !snap->active)
1195 /* Nothing to do if writing beyond end of snapshot */
1196 if (bio->bi_sector >= dm_table_get_size(snap->store->ti->table))
1200 * Remember, different snapshots can have
1201 * different chunk sizes.
1203 chunk = sector_to_chunk(snap->store, bio->bi_sector);
1206 * Check exception table to see if block
1207 * is already remapped in this snapshot
1208 * and trigger an exception if not.
1210 * ref_count is initialised to 1 so pending_complete()
1211 * won't destroy the primary_pe while we're inside this loop.
1213 e = lookup_exception(&snap->complete, chunk);
1217 pe = __lookup_pending_exception(snap, chunk);
1219 up_write(&snap->lock);
1220 pe = alloc_pending_exception(snap);
1221 down_write(&snap->lock);
1224 free_pending_exception(pe);
1228 e = lookup_exception(&snap->complete, chunk);
1230 free_pending_exception(pe);
1234 pe = __find_pending_exception(snap, pe, chunk);
1236 __invalidate_snapshot(snap, -ENOMEM);
1243 * Either every pe here has same
1244 * primary_pe or none has one yet.
1247 primary_pe = pe->primary_pe;
1253 bio_list_add(&primary_pe->origin_bios, bio);
1255 r = DM_MAPIO_SUBMITTED;
1258 if (!pe->primary_pe) {
1259 pe->primary_pe = primary_pe;
1260 get_pending_exception(primary_pe);
1265 list_add_tail(&pe->list, &pe_queue);
1269 up_write(&snap->lock);
1276 * If this is the first time we're processing this chunk and
1277 * ref_count is now 1 it means all the pending exceptions
1278 * got completed while we were in the loop above, so it falls to
1279 * us here to remove the primary_pe and submit any origin_bios.
1282 if (first && atomic_dec_and_test(&primary_pe->ref_count)) {
1283 flush_bios(bio_list_get(&primary_pe->origin_bios));
1284 free_pending_exception(primary_pe);
1285 /* If we got here, pe_queue is necessarily empty. */
1290 * Now that we have a complete pe list we can start the copying.
1292 list_for_each_entry_safe(pe, next_pe, &pe_queue, list)
1299 * Called on a write from the origin driver.
1301 static int do_origin(struct dm_dev *origin, struct bio *bio)
1304 int r = DM_MAPIO_REMAPPED;
1306 down_read(&_origins_lock);
1307 o = __lookup_origin(origin->bdev);
1309 r = __origin_write(&o->snapshots, bio);
1310 up_read(&_origins_lock);
1316 * Origin: maps a linear range of a device, with hooks for snapshotting.
1320 * Construct an origin mapping: <dev_path>
1321 * The context for an origin is merely a 'struct dm_dev *'
1322 * pointing to the real device.
1324 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1330 ti->error = "origin: incorrect number of arguments";
1334 r = dm_get_device(ti, argv[0], 0, ti->len,
1335 dm_table_get_mode(ti->table), &dev);
1337 ti->error = "Cannot get target device";
1345 static void origin_dtr(struct dm_target *ti)
1347 struct dm_dev *dev = ti->private;
1348 dm_put_device(ti, dev);
1351 static int origin_map(struct dm_target *ti, struct bio *bio,
1352 union map_info *map_context)
1354 struct dm_dev *dev = ti->private;
1355 bio->bi_bdev = dev->bdev;
1357 /* Only tell snapshots if this is a write */
1358 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
1361 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
1364 * Set the target "split_io" field to the minimum of all the snapshots'
1367 static void origin_resume(struct dm_target *ti)
1369 struct dm_dev *dev = ti->private;
1370 struct dm_snapshot *snap;
1372 chunk_t chunk_size = 0;
1374 down_read(&_origins_lock);
1375 o = __lookup_origin(dev->bdev);
1377 list_for_each_entry (snap, &o->snapshots, list)
1378 chunk_size = min_not_zero(chunk_size,
1379 snap->store->chunk_size);
1380 up_read(&_origins_lock);
1382 ti->split_io = chunk_size;
1385 static int origin_status(struct dm_target *ti, status_type_t type, char *result,
1386 unsigned int maxlen)
1388 struct dm_dev *dev = ti->private;
1391 case STATUSTYPE_INFO:
1395 case STATUSTYPE_TABLE:
1396 snprintf(result, maxlen, "%s", dev->name);
1403 static struct target_type origin_target = {
1404 .name = "snapshot-origin",
1405 .version = {1, 6, 0},
1406 .module = THIS_MODULE,
1410 .resume = origin_resume,
1411 .status = origin_status,
1414 static struct target_type snapshot_target = {
1416 .version = {1, 6, 0},
1417 .module = THIS_MODULE,
1418 .ctr = snapshot_ctr,
1419 .dtr = snapshot_dtr,
1420 .map = snapshot_map,
1421 .end_io = snapshot_end_io,
1422 .resume = snapshot_resume,
1423 .status = snapshot_status,
1426 static int __init dm_snapshot_init(void)
1430 r = dm_exception_store_init();
1432 DMERR("Failed to initialize exception stores");
1436 r = dm_register_target(&snapshot_target);
1438 DMERR("snapshot target register failed %d", r);
1442 r = dm_register_target(&origin_target);
1444 DMERR("Origin target register failed %d", r);
1448 r = init_origin_hash();
1450 DMERR("init_origin_hash failed.");
1454 exception_cache = KMEM_CACHE(dm_snap_exception, 0);
1455 if (!exception_cache) {
1456 DMERR("Couldn't create exception cache.");
1461 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
1462 if (!pending_cache) {
1463 DMERR("Couldn't create pending cache.");
1468 tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0);
1469 if (!tracked_chunk_cache) {
1470 DMERR("Couldn't create cache to track chunks in use.");
1475 ksnapd = create_singlethread_workqueue("ksnapd");
1477 DMERR("Failed to create ksnapd workqueue.");
1479 goto bad_pending_pool;
1485 kmem_cache_destroy(tracked_chunk_cache);
1487 kmem_cache_destroy(pending_cache);
1489 kmem_cache_destroy(exception_cache);
1493 dm_unregister_target(&origin_target);
1495 dm_unregister_target(&snapshot_target);
1499 static void __exit dm_snapshot_exit(void)
1501 destroy_workqueue(ksnapd);
1503 dm_unregister_target(&snapshot_target);
1504 dm_unregister_target(&origin_target);
1507 kmem_cache_destroy(pending_cache);
1508 kmem_cache_destroy(exception_cache);
1509 kmem_cache_destroy(tracked_chunk_cache);
1511 dm_exception_store_exit();
1515 module_init(dm_snapshot_init);
1516 module_exit(dm_snapshot_exit);
1518 MODULE_DESCRIPTION(DM_NAME " snapshot target");
1519 MODULE_AUTHOR("Joe Thornber");
1520 MODULE_LICENSE("GPL");