4 * Copyright (C) 2001-2002 Sistina Software (UK) Limited.
6 * This file is released under the GPL.
9 #include <linux/blkdev.h>
10 #include <linux/device-mapper.h>
11 #include <linux/delay.h>
13 #include <linux/init.h>
14 #include <linux/kdev_t.h>
15 #include <linux/list.h>
16 #include <linux/mempool.h>
17 #include <linux/module.h>
18 #include <linux/slab.h>
19 #include <linux/vmalloc.h>
20 #include <linux/log2.h>
21 #include <linux/dm-kcopyd.h>
22 #include <linux/workqueue.h>
24 #include "dm-exception-store.h"
26 #define DM_MSG_PREFIX "snapshots"
29 * The percentage increment we will wake up users at
31 #define WAKE_UP_PERCENT 5
34 * kcopyd priority of snapshot operations
36 #define SNAPSHOT_COPY_PRIORITY 2
39 * Reserve 1MB for each snapshot initially (with minimum of 1 page).
41 #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1)
44 * The size of the mempool used to track chunks in use.
48 #define DM_TRACKED_CHUNK_HASH_SIZE 16
49 #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
50 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
52 struct dm_exception_table {
55 struct list_head *table;
59 struct rw_semaphore lock;
61 struct dm_dev *origin;
66 /* List of snapshots per Origin */
67 struct list_head list;
69 /* You can't use a snapshot if this is 0 (e.g. if full) */
72 /* Origin writes don't trigger exceptions until this is set */
75 /* Whether or not owning mapped_device is suspended */
78 mempool_t *pending_pool;
80 atomic_t pending_exceptions_count;
82 struct dm_exception_table pending;
83 struct dm_exception_table complete;
86 * pe_lock protects all pending_exception operations and access
87 * as well as the snapshot_bios list.
91 /* The on disk metadata handler */
92 struct dm_exception_store *store;
94 struct dm_kcopyd_client *kcopyd_client;
96 /* Queue of snapshot writes for ksnapd to flush */
97 struct bio_list queued_bios;
98 struct work_struct queued_bios_work;
100 /* Chunks with outstanding reads */
101 mempool_t *tracked_chunk_pool;
102 spinlock_t tracked_chunk_lock;
103 struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
106 struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
110 EXPORT_SYMBOL(dm_snap_cow);
112 static struct workqueue_struct *ksnapd;
113 static void flush_queued_bios(struct work_struct *work);
115 static sector_t chunk_to_sector(struct dm_exception_store *store,
118 return chunk << store->chunk_shift;
121 static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
124 * There is only ever one instance of a particular block
125 * device so we can compare pointers safely.
130 struct dm_snap_pending_exception {
131 struct dm_exception e;
134 * Origin buffers waiting for this to complete are held
137 struct bio_list origin_bios;
138 struct bio_list snapshot_bios;
141 * Short-term queue of pending exceptions prior to submission.
143 struct list_head list;
146 * The primary pending_exception is the one that holds
147 * the ref_count and the list of origin_bios for a
148 * group of pending_exceptions. It is always last to get freed.
149 * These fields get set up when writing to the origin.
151 struct dm_snap_pending_exception *primary_pe;
154 * Number of pending_exceptions processing this chunk.
155 * When this drops to zero we must complete the origin bios.
156 * If incrementing or decrementing this, hold pe->snap->lock for
157 * the sibling concerned and not pe->primary_pe->snap->lock unless
162 /* Pointer back to snapshot context */
163 struct dm_snapshot *snap;
166 * 1 indicates the exception has already been sent to
173 * Hash table mapping origin volumes to lists of snapshots and
174 * a lock to protect it
176 static struct kmem_cache *exception_cache;
177 static struct kmem_cache *pending_cache;
179 struct dm_snap_tracked_chunk {
180 struct hlist_node node;
184 static struct kmem_cache *tracked_chunk_cache;
186 static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s,
189 struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool,
195 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
196 hlist_add_head(&c->node,
197 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
198 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
203 static void stop_tracking_chunk(struct dm_snapshot *s,
204 struct dm_snap_tracked_chunk *c)
208 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
210 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
212 mempool_free(c, s->tracked_chunk_pool);
215 static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
217 struct dm_snap_tracked_chunk *c;
218 struct hlist_node *hn;
221 spin_lock_irq(&s->tracked_chunk_lock);
223 hlist_for_each_entry(c, hn,
224 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
225 if (c->chunk == chunk) {
231 spin_unlock_irq(&s->tracked_chunk_lock);
237 * This conflicting I/O is extremely improbable in the caller,
238 * so msleep(1) is sufficient and there is no need for a wait queue.
240 static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
242 while (__chunk_is_tracked(s, chunk))
247 * One of these per registered origin, held in the snapshot_origins hash
250 /* The origin device */
251 struct block_device *bdev;
253 struct list_head hash_list;
255 /* List of snapshots for this origin */
256 struct list_head snapshots;
260 * Size of the hash table for origin volumes. If we make this
261 * the size of the minors list then it should be nearly perfect
263 #define ORIGIN_HASH_SIZE 256
264 #define ORIGIN_MASK 0xFF
265 static struct list_head *_origins;
266 static struct rw_semaphore _origins_lock;
268 static int init_origin_hash(void)
272 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
275 DMERR("unable to allocate memory");
279 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
280 INIT_LIST_HEAD(_origins + i);
281 init_rwsem(&_origins_lock);
286 static void exit_origin_hash(void)
291 static unsigned origin_hash(struct block_device *bdev)
293 return bdev->bd_dev & ORIGIN_MASK;
296 static struct origin *__lookup_origin(struct block_device *origin)
298 struct list_head *ol;
301 ol = &_origins[origin_hash(origin)];
302 list_for_each_entry (o, ol, hash_list)
303 if (bdev_equal(o->bdev, origin))
309 static void __insert_origin(struct origin *o)
311 struct list_head *sl = &_origins[origin_hash(o->bdev)];
312 list_add_tail(&o->hash_list, sl);
316 * _origins_lock must be held when calling this function.
317 * Returns number of snapshots registered using the supplied cow device, plus:
318 * snap_src - a snapshot suitable for use as a source of exception handover
319 * snap_dest - a snapshot capable of receiving exception handover.
321 * Possible return values and states:
322 * 0: NULL, NULL - first new snapshot
323 * 1: snap_src, NULL - normal snapshot
324 * 2: snap_src, snap_dest - waiting for handover
325 * 2: snap_src, NULL - handed over, waiting for old to be deleted
326 * 1: NULL, snap_dest - source got destroyed without handover
328 static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
329 struct dm_snapshot **snap_src,
330 struct dm_snapshot **snap_dest)
332 struct dm_snapshot *s;
337 o = __lookup_origin(snap->origin->bdev);
341 list_for_each_entry(s, &o->snapshots, list) {
342 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
352 } else if (snap_dest)
363 * On success, returns 1 if this snapshot is a handover destination,
364 * otherwise returns 0.
366 static int __validate_exception_handover(struct dm_snapshot *snap)
368 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
370 /* Does snapshot need exceptions handed over to it? */
371 if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest) == 2) ||
373 snap->ti->error = "Snapshot cow pairing for exception "
374 "table handover failed";
379 * If no snap_src was found, snap cannot become a handover
388 static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
390 struct dm_snapshot *l;
392 /* Sort the list according to chunk size, largest-first smallest-last */
393 list_for_each_entry(l, &o->snapshots, list)
394 if (l->store->chunk_size < s->store->chunk_size)
396 list_add_tail(&s->list, &l->list);
400 * Make a note of the snapshot and its origin so we can look it
401 * up when the origin has a write on it.
403 * Also validate snapshot exception store handovers.
404 * On success, returns 1 if this registration is a handover destination,
405 * otherwise returns 0.
407 static int register_snapshot(struct dm_snapshot *snap)
409 struct origin *o, *new_o = NULL;
410 struct block_device *bdev = snap->origin->bdev;
413 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
417 down_write(&_origins_lock);
419 r = __validate_exception_handover(snap);
425 o = __lookup_origin(bdev);
432 /* Initialise the struct */
433 INIT_LIST_HEAD(&o->snapshots);
439 __insert_snapshot(o, snap);
442 up_write(&_origins_lock);
448 * Move snapshot to correct place in list according to chunk size.
450 static void reregister_snapshot(struct dm_snapshot *s)
452 struct block_device *bdev = s->origin->bdev;
454 down_write(&_origins_lock);
457 __insert_snapshot(__lookup_origin(bdev), s);
459 up_write(&_origins_lock);
462 static void unregister_snapshot(struct dm_snapshot *s)
466 down_write(&_origins_lock);
467 o = __lookup_origin(s->origin->bdev);
470 if (o && list_empty(&o->snapshots)) {
471 list_del(&o->hash_list);
475 up_write(&_origins_lock);
479 * Implementation of the exception hash tables.
480 * The lowest hash_shift bits of the chunk number are ignored, allowing
481 * some consecutive chunks to be grouped together.
483 static int dm_exception_table_init(struct dm_exception_table *et,
484 uint32_t size, unsigned hash_shift)
488 et->hash_shift = hash_shift;
489 et->hash_mask = size - 1;
490 et->table = dm_vcalloc(size, sizeof(struct list_head));
494 for (i = 0; i < size; i++)
495 INIT_LIST_HEAD(et->table + i);
500 static void dm_exception_table_exit(struct dm_exception_table *et,
501 struct kmem_cache *mem)
503 struct list_head *slot;
504 struct dm_exception *ex, *next;
507 size = et->hash_mask + 1;
508 for (i = 0; i < size; i++) {
509 slot = et->table + i;
511 list_for_each_entry_safe (ex, next, slot, hash_list)
512 kmem_cache_free(mem, ex);
518 static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
520 return (chunk >> et->hash_shift) & et->hash_mask;
523 static void dm_remove_exception(struct dm_exception *e)
525 list_del(&e->hash_list);
529 * Return the exception data for a sector, or NULL if not
532 static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
535 struct list_head *slot;
536 struct dm_exception *e;
538 slot = &et->table[exception_hash(et, chunk)];
539 list_for_each_entry (e, slot, hash_list)
540 if (chunk >= e->old_chunk &&
541 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
547 static struct dm_exception *alloc_completed_exception(void)
549 struct dm_exception *e;
551 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
553 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
558 static void free_completed_exception(struct dm_exception *e)
560 kmem_cache_free(exception_cache, e);
563 static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
565 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
568 atomic_inc(&s->pending_exceptions_count);
574 static void free_pending_exception(struct dm_snap_pending_exception *pe)
576 struct dm_snapshot *s = pe->snap;
578 mempool_free(pe, s->pending_pool);
579 smp_mb__before_atomic_dec();
580 atomic_dec(&s->pending_exceptions_count);
583 static void dm_insert_exception(struct dm_exception_table *eh,
584 struct dm_exception *new_e)
587 struct dm_exception *e = NULL;
589 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
591 /* Add immediately if this table doesn't support consecutive chunks */
595 /* List is ordered by old_chunk */
596 list_for_each_entry_reverse(e, l, hash_list) {
597 /* Insert after an existing chunk? */
598 if (new_e->old_chunk == (e->old_chunk +
599 dm_consecutive_chunk_count(e) + 1) &&
600 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
601 dm_consecutive_chunk_count(e) + 1)) {
602 dm_consecutive_chunk_count_inc(e);
603 free_completed_exception(new_e);
607 /* Insert before an existing chunk? */
608 if (new_e->old_chunk == (e->old_chunk - 1) &&
609 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
610 dm_consecutive_chunk_count_inc(e);
613 free_completed_exception(new_e);
617 if (new_e->old_chunk > e->old_chunk)
622 list_add(&new_e->hash_list, e ? &e->hash_list : l);
626 * Callback used by the exception stores to load exceptions when
629 static int dm_add_exception(void *context, chunk_t old, chunk_t new)
631 struct dm_snapshot *s = context;
632 struct dm_exception *e;
634 e = alloc_completed_exception();
640 /* Consecutive_count is implicitly initialised to zero */
643 dm_insert_exception(&s->complete, e);
648 #define min_not_zero(l, r) (((l) == 0) ? (r) : (((r) == 0) ? (l) : min(l, r)))
651 * Return a minimum chunk size of all snapshots that have the specified origin.
652 * Return zero if the origin has no snapshots.
654 static sector_t __minimum_chunk_size(struct origin *o)
656 struct dm_snapshot *snap;
657 unsigned chunk_size = 0;
660 list_for_each_entry(snap, &o->snapshots, list)
661 chunk_size = min_not_zero(chunk_size,
662 snap->store->chunk_size);
670 static int calc_max_buckets(void)
672 /* use a fixed size of 2MB */
673 unsigned long mem = 2 * 1024 * 1024;
674 mem /= sizeof(struct list_head);
680 * Allocate room for a suitable hash table.
682 static int init_hash_tables(struct dm_snapshot *s)
684 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
687 * Calculate based on the size of the original volume or
690 cow_dev_size = get_dev_size(s->cow->bdev);
691 origin_dev_size = get_dev_size(s->origin->bdev);
692 max_buckets = calc_max_buckets();
694 hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift;
695 hash_size = min(hash_size, max_buckets);
699 hash_size = rounddown_pow_of_two(hash_size);
700 if (dm_exception_table_init(&s->complete, hash_size,
701 DM_CHUNK_CONSECUTIVE_BITS))
705 * Allocate hash table for in-flight exceptions
706 * Make this smaller than the real hash table
712 if (dm_exception_table_init(&s->pending, hash_size, 0)) {
713 dm_exception_table_exit(&s->complete, exception_cache);
721 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
723 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
725 struct dm_snapshot *s;
728 char *origin_path, *cow_path;
732 ti->error = "requires exactly 4 arguments";
737 origin_path = argv[0];
741 s = kmalloc(sizeof(*s), GFP_KERNEL);
743 ti->error = "Cannot allocate snapshot context private "
753 r = dm_get_device(ti, cow_path, 0, 0,
754 FMODE_READ | FMODE_WRITE, &s->cow);
756 ti->error = "Cannot get COW device";
760 r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
762 ti->error = "Couldn't create exception store";
770 r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin);
772 ti->error = "Cannot get origin device";
780 atomic_set(&s->pending_exceptions_count, 0);
781 init_rwsem(&s->lock);
782 INIT_LIST_HEAD(&s->list);
783 spin_lock_init(&s->pe_lock);
785 /* Allocate hash table for COW data */
786 if (init_hash_tables(s)) {
787 ti->error = "Unable to allocate hash table space";
789 goto bad_hash_tables;
792 r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client);
794 ti->error = "Could not create kcopyd client";
798 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
799 if (!s->pending_pool) {
800 ti->error = "Could not allocate mempool for pending exceptions";
801 goto bad_pending_pool;
804 s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS,
805 tracked_chunk_cache);
806 if (!s->tracked_chunk_pool) {
807 ti->error = "Could not allocate tracked_chunk mempool for "
809 goto bad_tracked_chunk_pool;
812 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
813 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
815 spin_lock_init(&s->tracked_chunk_lock);
817 bio_list_init(&s->queued_bios);
818 INIT_WORK(&s->queued_bios_work, flush_queued_bios);
821 ti->num_flush_requests = 1;
823 /* Add snapshot to the list of snapshots for this origin */
824 /* Exceptions aren't triggered till snapshot_resume() is called */
825 r = register_snapshot(s);
827 ti->error = "Snapshot origin struct allocation failed";
828 goto bad_load_and_register;
830 /* invalid handover, register_snapshot has set ti->error */
831 goto bad_load_and_register;
835 * Metadata must only be loaded into one table at once, so skip this
836 * if metadata will be handed over during resume.
837 * Chunk size will be set during the handover - set it to zero to
838 * ensure it's ignored.
841 s->store->chunk_size = 0;
845 r = s->store->type->read_metadata(s->store, dm_add_exception,
848 ti->error = "Failed to read snapshot metadata";
849 goto bad_read_metadata;
852 DMWARN("Snapshot is marked invalid.");
855 if (!s->store->chunk_size) {
856 ti->error = "Chunk size not set";
857 goto bad_read_metadata;
859 ti->split_io = s->store->chunk_size;
864 unregister_snapshot(s);
866 bad_load_and_register:
867 mempool_destroy(s->tracked_chunk_pool);
869 bad_tracked_chunk_pool:
870 mempool_destroy(s->pending_pool);
873 dm_kcopyd_client_destroy(s->kcopyd_client);
876 dm_exception_table_exit(&s->pending, pending_cache);
877 dm_exception_table_exit(&s->complete, exception_cache);
880 dm_put_device(ti, s->origin);
883 dm_exception_store_destroy(s->store);
886 dm_put_device(ti, s->cow);
895 static void __free_exceptions(struct dm_snapshot *s)
897 dm_kcopyd_client_destroy(s->kcopyd_client);
898 s->kcopyd_client = NULL;
900 dm_exception_table_exit(&s->pending, pending_cache);
901 dm_exception_table_exit(&s->complete, exception_cache);
904 static void __handover_exceptions(struct dm_snapshot *snap_src,
905 struct dm_snapshot *snap_dest)
908 struct dm_exception_table table_swap;
909 struct dm_exception_store *store_swap;
913 * Swap all snapshot context information between the two instances.
915 u.table_swap = snap_dest->complete;
916 snap_dest->complete = snap_src->complete;
917 snap_src->complete = u.table_swap;
919 u.store_swap = snap_dest->store;
920 snap_dest->store = snap_src->store;
921 snap_src->store = u.store_swap;
923 snap_dest->store->snap = snap_dest;
924 snap_src->store->snap = snap_src;
926 snap_dest->ti->split_io = snap_dest->store->chunk_size;
927 snap_dest->valid = snap_src->valid;
930 * Set source invalid to ensure it receives no further I/O.
935 static void snapshot_dtr(struct dm_target *ti)
937 #ifdef CONFIG_DM_DEBUG
940 struct dm_snapshot *s = ti->private;
941 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
943 flush_workqueue(ksnapd);
945 down_read(&_origins_lock);
946 /* Check whether exception handover must be cancelled */
947 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest);
948 if (snap_src && snap_dest && (s == snap_src)) {
949 down_write(&snap_dest->lock);
950 snap_dest->valid = 0;
951 up_write(&snap_dest->lock);
952 DMERR("Cancelling snapshot handover.");
954 up_read(&_origins_lock);
956 /* Prevent further origin writes from using this snapshot. */
957 /* After this returns there can be no new kcopyd jobs. */
958 unregister_snapshot(s);
960 while (atomic_read(&s->pending_exceptions_count))
963 * Ensure instructions in mempool_destroy aren't reordered
964 * before atomic_read.
968 #ifdef CONFIG_DM_DEBUG
969 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
970 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
973 mempool_destroy(s->tracked_chunk_pool);
975 __free_exceptions(s);
977 mempool_destroy(s->pending_pool);
979 dm_put_device(ti, s->origin);
981 dm_exception_store_destroy(s->store);
983 dm_put_device(ti, s->cow);
989 * Flush a list of buffers.
991 static void flush_bios(struct bio *bio)
998 generic_make_request(bio);
1003 static void flush_queued_bios(struct work_struct *work)
1005 struct dm_snapshot *s =
1006 container_of(work, struct dm_snapshot, queued_bios_work);
1007 struct bio *queued_bios;
1008 unsigned long flags;
1010 spin_lock_irqsave(&s->pe_lock, flags);
1011 queued_bios = bio_list_get(&s->queued_bios);
1012 spin_unlock_irqrestore(&s->pe_lock, flags);
1014 flush_bios(queued_bios);
1018 * Error a list of buffers.
1020 static void error_bios(struct bio *bio)
1026 bio->bi_next = NULL;
1032 static void __invalidate_snapshot(struct dm_snapshot *s, int err)
1038 DMERR("Invalidating snapshot: Error reading/writing.");
1039 else if (err == -ENOMEM)
1040 DMERR("Invalidating snapshot: Unable to allocate exception.");
1042 if (s->store->type->drop_snapshot)
1043 s->store->type->drop_snapshot(s->store);
1047 dm_table_event(s->ti->table);
1050 static void get_pending_exception(struct dm_snap_pending_exception *pe)
1052 atomic_inc(&pe->ref_count);
1055 static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe)
1057 struct dm_snap_pending_exception *primary_pe;
1058 struct bio *origin_bios = NULL;
1060 primary_pe = pe->primary_pe;
1063 * If this pe is involved in a write to the origin and
1064 * it is the last sibling to complete then release
1065 * the bios for the original write to the origin.
1068 atomic_dec_and_test(&primary_pe->ref_count)) {
1069 origin_bios = bio_list_get(&primary_pe->origin_bios);
1070 free_pending_exception(primary_pe);
1074 * Free the pe if it's not linked to an origin write or if
1075 * it's not itself a primary pe.
1077 if (!primary_pe || primary_pe != pe)
1078 free_pending_exception(pe);
1083 static void pending_complete(struct dm_snap_pending_exception *pe, int success)
1085 struct dm_exception *e;
1086 struct dm_snapshot *s = pe->snap;
1087 struct bio *origin_bios = NULL;
1088 struct bio *snapshot_bios = NULL;
1092 /* Read/write error - snapshot is unusable */
1093 down_write(&s->lock);
1094 __invalidate_snapshot(s, -EIO);
1099 e = alloc_completed_exception();
1101 down_write(&s->lock);
1102 __invalidate_snapshot(s, -ENOMEM);
1108 down_write(&s->lock);
1110 free_completed_exception(e);
1115 /* Check for conflicting reads */
1116 __check_for_conflicting_io(s, pe->e.old_chunk);
1119 * Add a proper exception, and remove the
1120 * in-flight exception from the list.
1122 dm_insert_exception(&s->complete, e);
1125 dm_remove_exception(&pe->e);
1126 snapshot_bios = bio_list_get(&pe->snapshot_bios);
1127 origin_bios = put_pending_exception(pe);
1131 /* Submit any pending write bios */
1133 error_bios(snapshot_bios);
1135 flush_bios(snapshot_bios);
1137 flush_bios(origin_bios);
1140 static void commit_callback(void *context, int success)
1142 struct dm_snap_pending_exception *pe = context;
1144 pending_complete(pe, success);
1148 * Called when the copy I/O has finished. kcopyd actually runs
1149 * this code so don't block.
1151 static void copy_callback(int read_err, unsigned long write_err, void *context)
1153 struct dm_snap_pending_exception *pe = context;
1154 struct dm_snapshot *s = pe->snap;
1156 if (read_err || write_err)
1157 pending_complete(pe, 0);
1160 /* Update the metadata if we are persistent */
1161 s->store->type->commit_exception(s->store, &pe->e,
1162 commit_callback, pe);
1166 * Dispatches the copy operation to kcopyd.
1168 static void start_copy(struct dm_snap_pending_exception *pe)
1170 struct dm_snapshot *s = pe->snap;
1171 struct dm_io_region src, dest;
1172 struct block_device *bdev = s->origin->bdev;
1175 dev_size = get_dev_size(bdev);
1178 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
1179 src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
1181 dest.bdev = s->cow->bdev;
1182 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
1183 dest.count = src.count;
1185 /* Hand over to kcopyd */
1186 dm_kcopyd_copy(s->kcopyd_client,
1187 &src, 1, &dest, 0, copy_callback, pe);
1190 static struct dm_snap_pending_exception *
1191 __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1193 struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
1198 return container_of(e, struct dm_snap_pending_exception, e);
1202 * Looks to see if this snapshot already has a pending exception
1203 * for this chunk, otherwise it allocates a new one and inserts
1204 * it into the pending table.
1206 * NOTE: a write lock must be held on snap->lock before calling
1209 static struct dm_snap_pending_exception *
1210 __find_pending_exception(struct dm_snapshot *s,
1211 struct dm_snap_pending_exception *pe, chunk_t chunk)
1213 struct dm_snap_pending_exception *pe2;
1215 pe2 = __lookup_pending_exception(s, chunk);
1217 free_pending_exception(pe);
1221 pe->e.old_chunk = chunk;
1222 bio_list_init(&pe->origin_bios);
1223 bio_list_init(&pe->snapshot_bios);
1224 pe->primary_pe = NULL;
1225 atomic_set(&pe->ref_count, 0);
1228 if (s->store->type->prepare_exception(s->store, &pe->e)) {
1229 free_pending_exception(pe);
1233 get_pending_exception(pe);
1234 dm_insert_exception(&s->pending, &pe->e);
1239 static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
1240 struct bio *bio, chunk_t chunk)
1242 bio->bi_bdev = s->cow->bdev;
1243 bio->bi_sector = chunk_to_sector(s->store,
1244 dm_chunk_number(e->new_chunk) +
1245 (chunk - e->old_chunk)) +
1247 s->store->chunk_mask);
1250 static int snapshot_map(struct dm_target *ti, struct bio *bio,
1251 union map_info *map_context)
1253 struct dm_exception *e;
1254 struct dm_snapshot *s = ti->private;
1255 int r = DM_MAPIO_REMAPPED;
1257 struct dm_snap_pending_exception *pe = NULL;
1259 if (unlikely(bio_empty_barrier(bio))) {
1260 bio->bi_bdev = s->cow->bdev;
1261 return DM_MAPIO_REMAPPED;
1264 chunk = sector_to_chunk(s->store, bio->bi_sector);
1266 /* Full snapshots are not usable */
1267 /* To get here the table must be live so s->active is always set. */
1271 /* FIXME: should only take write lock if we need
1272 * to copy an exception */
1273 down_write(&s->lock);
1280 /* If the block is already remapped - use that, else remap it */
1281 e = dm_lookup_exception(&s->complete, chunk);
1283 remap_exception(s, e, bio, chunk);
1288 * Write to snapshot - higher level takes care of RW/RO
1289 * flags so we should only get this if we are
1292 if (bio_rw(bio) == WRITE) {
1293 pe = __lookup_pending_exception(s, chunk);
1296 pe = alloc_pending_exception(s);
1297 down_write(&s->lock);
1300 free_pending_exception(pe);
1305 e = dm_lookup_exception(&s->complete, chunk);
1307 free_pending_exception(pe);
1308 remap_exception(s, e, bio, chunk);
1312 pe = __find_pending_exception(s, pe, chunk);
1314 __invalidate_snapshot(s, -ENOMEM);
1320 remap_exception(s, &pe->e, bio, chunk);
1321 bio_list_add(&pe->snapshot_bios, bio);
1323 r = DM_MAPIO_SUBMITTED;
1326 /* this is protected by snap->lock */
1333 bio->bi_bdev = s->origin->bdev;
1334 map_context->ptr = track_chunk(s, chunk);
1343 static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1344 int error, union map_info *map_context)
1346 struct dm_snapshot *s = ti->private;
1347 struct dm_snap_tracked_chunk *c = map_context->ptr;
1350 stop_tracking_chunk(s, c);
1355 static void snapshot_postsuspend(struct dm_target *ti)
1357 struct dm_snapshot *s = ti->private;
1359 down_write(&s->lock);
1364 static int snapshot_preresume(struct dm_target *ti)
1367 struct dm_snapshot *s = ti->private;
1368 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1370 down_read(&_origins_lock);
1371 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest);
1372 if (snap_src && snap_dest) {
1373 down_read(&snap_src->lock);
1374 if (s == snap_src) {
1375 DMERR("Unable to resume snapshot source until "
1376 "handover completes.");
1378 } else if (!snap_src->suspended) {
1379 DMERR("Unable to perform snapshot handover until "
1380 "source is suspended.");
1383 up_read(&snap_src->lock);
1385 up_read(&_origins_lock);
1390 static void snapshot_resume(struct dm_target *ti)
1392 struct dm_snapshot *s = ti->private;
1393 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1395 down_read(&_origins_lock);
1396 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest);
1397 if (snap_src && snap_dest) {
1398 down_write(&snap_src->lock);
1399 down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1400 __handover_exceptions(snap_src, snap_dest);
1401 up_write(&snap_dest->lock);
1402 up_write(&snap_src->lock);
1404 up_read(&_origins_lock);
1406 /* Now we have correct chunk size, reregister */
1407 reregister_snapshot(s);
1409 down_write(&s->lock);
1415 static int snapshot_status(struct dm_target *ti, status_type_t type,
1416 char *result, unsigned int maxlen)
1419 struct dm_snapshot *snap = ti->private;
1422 case STATUSTYPE_INFO:
1424 down_write(&snap->lock);
1429 if (snap->store->type->usage) {
1430 sector_t total_sectors, sectors_allocated,
1432 snap->store->type->usage(snap->store,
1436 DMEMIT("%llu/%llu %llu",
1437 (unsigned long long)sectors_allocated,
1438 (unsigned long long)total_sectors,
1439 (unsigned long long)metadata_sectors);
1445 up_write(&snap->lock);
1449 case STATUSTYPE_TABLE:
1451 * kdevname returns a static pointer so we need
1452 * to make private copies if the output is to
1455 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
1456 snap->store->type->status(snap->store, type, result + sz,
1464 static int snapshot_iterate_devices(struct dm_target *ti,
1465 iterate_devices_callout_fn fn, void *data)
1467 struct dm_snapshot *snap = ti->private;
1469 return fn(ti, snap->origin, 0, ti->len, data);
1473 /*-----------------------------------------------------------------
1475 *---------------------------------------------------------------*/
1478 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
1479 * supplied bio was ignored. The caller may submit it immediately.
1480 * (No remapping actually occurs as the origin is always a direct linear
1483 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
1484 * and any supplied bio is added to a list to be submitted once all
1485 * the necessary exceptions exist.
1487 static int __origin_write(struct list_head *snapshots, sector_t sector,
1490 int r = DM_MAPIO_REMAPPED, first = 0;
1491 struct dm_snapshot *snap;
1492 struct dm_exception *e;
1493 struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL;
1495 LIST_HEAD(pe_queue);
1497 /* Do all the snapshots on this origin */
1498 list_for_each_entry (snap, snapshots, list) {
1500 down_write(&snap->lock);
1502 /* Only deal with valid and active snapshots */
1503 if (!snap->valid || !snap->active)
1506 /* Nothing to do if writing beyond end of snapshot */
1507 if (sector >= dm_table_get_size(snap->ti->table))
1511 * Remember, different snapshots can have
1512 * different chunk sizes.
1514 chunk = sector_to_chunk(snap->store, sector);
1517 * Check exception table to see if block
1518 * is already remapped in this snapshot
1519 * and trigger an exception if not.
1521 * ref_count is initialised to 1 so pending_complete()
1522 * won't destroy the primary_pe while we're inside this loop.
1524 e = dm_lookup_exception(&snap->complete, chunk);
1528 pe = __lookup_pending_exception(snap, chunk);
1530 up_write(&snap->lock);
1531 pe = alloc_pending_exception(snap);
1532 down_write(&snap->lock);
1535 free_pending_exception(pe);
1539 e = dm_lookup_exception(&snap->complete, chunk);
1541 free_pending_exception(pe);
1545 pe = __find_pending_exception(snap, pe, chunk);
1547 __invalidate_snapshot(snap, -ENOMEM);
1554 * Either every pe here has same
1555 * primary_pe or none has one yet.
1558 primary_pe = pe->primary_pe;
1565 bio_list_add(&primary_pe->origin_bios, bio);
1567 r = DM_MAPIO_SUBMITTED;
1570 if (!pe->primary_pe) {
1571 pe->primary_pe = primary_pe;
1572 get_pending_exception(primary_pe);
1577 list_add_tail(&pe->list, &pe_queue);
1581 up_write(&snap->lock);
1588 * If this is the first time we're processing this chunk and
1589 * ref_count is now 1 it means all the pending exceptions
1590 * got completed while we were in the loop above, so it falls to
1591 * us here to remove the primary_pe and submit any origin_bios.
1594 if (first && atomic_dec_and_test(&primary_pe->ref_count)) {
1595 flush_bios(bio_list_get(&primary_pe->origin_bios));
1596 free_pending_exception(primary_pe);
1597 /* If we got here, pe_queue is necessarily empty. */
1602 * Now that we have a complete pe list we can start the copying.
1604 list_for_each_entry_safe(pe, next_pe, &pe_queue, list)
1611 * Called on a write from the origin driver.
1613 static int do_origin(struct dm_dev *origin, struct bio *bio)
1616 int r = DM_MAPIO_REMAPPED;
1618 down_read(&_origins_lock);
1619 o = __lookup_origin(origin->bdev);
1621 r = __origin_write(&o->snapshots, bio->bi_sector, bio);
1622 up_read(&_origins_lock);
1628 * Origin: maps a linear range of a device, with hooks for snapshotting.
1632 * Construct an origin mapping: <dev_path>
1633 * The context for an origin is merely a 'struct dm_dev *'
1634 * pointing to the real device.
1636 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1642 ti->error = "origin: incorrect number of arguments";
1646 r = dm_get_device(ti, argv[0], 0, ti->len,
1647 dm_table_get_mode(ti->table), &dev);
1649 ti->error = "Cannot get target device";
1654 ti->num_flush_requests = 1;
1659 static void origin_dtr(struct dm_target *ti)
1661 struct dm_dev *dev = ti->private;
1662 dm_put_device(ti, dev);
1665 static int origin_map(struct dm_target *ti, struct bio *bio,
1666 union map_info *map_context)
1668 struct dm_dev *dev = ti->private;
1669 bio->bi_bdev = dev->bdev;
1671 if (unlikely(bio_empty_barrier(bio)))
1672 return DM_MAPIO_REMAPPED;
1674 /* Only tell snapshots if this is a write */
1675 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
1679 * Set the target "split_io" field to the minimum of all the snapshots'
1682 static void origin_resume(struct dm_target *ti)
1684 struct dm_dev *dev = ti->private;
1686 down_read(&_origins_lock);
1688 ti->split_io = __minimum_chunk_size(__lookup_origin(dev->bdev));
1690 up_read(&_origins_lock);
1693 static int origin_status(struct dm_target *ti, status_type_t type, char *result,
1694 unsigned int maxlen)
1696 struct dm_dev *dev = ti->private;
1699 case STATUSTYPE_INFO:
1703 case STATUSTYPE_TABLE:
1704 snprintf(result, maxlen, "%s", dev->name);
1711 static int origin_iterate_devices(struct dm_target *ti,
1712 iterate_devices_callout_fn fn, void *data)
1714 struct dm_dev *dev = ti->private;
1716 return fn(ti, dev, 0, ti->len, data);
1719 static struct target_type origin_target = {
1720 .name = "snapshot-origin",
1721 .version = {1, 7, 0},
1722 .module = THIS_MODULE,
1726 .resume = origin_resume,
1727 .status = origin_status,
1728 .iterate_devices = origin_iterate_devices,
1731 static struct target_type snapshot_target = {
1733 .version = {1, 9, 0},
1734 .module = THIS_MODULE,
1735 .ctr = snapshot_ctr,
1736 .dtr = snapshot_dtr,
1737 .map = snapshot_map,
1738 .end_io = snapshot_end_io,
1739 .postsuspend = snapshot_postsuspend,
1740 .preresume = snapshot_preresume,
1741 .resume = snapshot_resume,
1742 .status = snapshot_status,
1743 .iterate_devices = snapshot_iterate_devices,
1746 static int __init dm_snapshot_init(void)
1750 r = dm_exception_store_init();
1752 DMERR("Failed to initialize exception stores");
1756 r = dm_register_target(&snapshot_target);
1758 DMERR("snapshot target register failed %d", r);
1759 goto bad_register_snapshot_target;
1762 r = dm_register_target(&origin_target);
1764 DMERR("Origin target register failed %d", r);
1768 r = init_origin_hash();
1770 DMERR("init_origin_hash failed.");
1774 exception_cache = KMEM_CACHE(dm_exception, 0);
1775 if (!exception_cache) {
1776 DMERR("Couldn't create exception cache.");
1781 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
1782 if (!pending_cache) {
1783 DMERR("Couldn't create pending cache.");
1788 tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0);
1789 if (!tracked_chunk_cache) {
1790 DMERR("Couldn't create cache to track chunks in use.");
1795 ksnapd = create_singlethread_workqueue("ksnapd");
1797 DMERR("Failed to create ksnapd workqueue.");
1799 goto bad_pending_pool;
1805 kmem_cache_destroy(tracked_chunk_cache);
1807 kmem_cache_destroy(pending_cache);
1809 kmem_cache_destroy(exception_cache);
1813 dm_unregister_target(&origin_target);
1815 dm_unregister_target(&snapshot_target);
1817 bad_register_snapshot_target:
1818 dm_exception_store_exit();
1822 static void __exit dm_snapshot_exit(void)
1824 destroy_workqueue(ksnapd);
1826 dm_unregister_target(&snapshot_target);
1827 dm_unregister_target(&origin_target);
1830 kmem_cache_destroy(pending_cache);
1831 kmem_cache_destroy(exception_cache);
1832 kmem_cache_destroy(tracked_chunk_cache);
1834 dm_exception_store_exit();
1838 module_init(dm_snapshot_init);
1839 module_exit(dm_snapshot_exit);
1841 MODULE_DESCRIPTION(DM_NAME " snapshot target");
1842 MODULE_AUTHOR("Joe Thornber");
1843 MODULE_LICENSE("GPL");