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 * One of these per registered origin, held in the snapshot_origins hash
240 /* The origin device */
241 struct block_device *bdev;
243 struct list_head hash_list;
245 /* List of snapshots for this origin */
246 struct list_head snapshots;
250 * Size of the hash table for origin volumes. If we make this
251 * the size of the minors list then it should be nearly perfect
253 #define ORIGIN_HASH_SIZE 256
254 #define ORIGIN_MASK 0xFF
255 static struct list_head *_origins;
256 static struct rw_semaphore _origins_lock;
258 static int init_origin_hash(void)
262 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
265 DMERR("unable to allocate memory");
269 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
270 INIT_LIST_HEAD(_origins + i);
271 init_rwsem(&_origins_lock);
276 static void exit_origin_hash(void)
281 static unsigned origin_hash(struct block_device *bdev)
283 return bdev->bd_dev & ORIGIN_MASK;
286 static struct origin *__lookup_origin(struct block_device *origin)
288 struct list_head *ol;
291 ol = &_origins[origin_hash(origin)];
292 list_for_each_entry (o, ol, hash_list)
293 if (bdev_equal(o->bdev, origin))
299 static void __insert_origin(struct origin *o)
301 struct list_head *sl = &_origins[origin_hash(o->bdev)];
302 list_add_tail(&o->hash_list, sl);
306 * _origins_lock must be held when calling this function.
307 * Returns number of snapshots registered using the supplied cow device, plus:
308 * snap_src - a snapshot suitable for use as a source of exception handover
309 * snap_dest - a snapshot capable of receiving exception handover.
311 * Possible return values and states:
312 * 0: NULL, NULL - first new snapshot
313 * 1: snap_src, NULL - normal snapshot
314 * 2: snap_src, snap_dest - waiting for handover
315 * 2: snap_src, NULL - handed over, waiting for old to be deleted
316 * 1: NULL, snap_dest - source got destroyed without handover
318 static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
319 struct dm_snapshot **snap_src,
320 struct dm_snapshot **snap_dest)
322 struct dm_snapshot *s;
327 o = __lookup_origin(snap->origin->bdev);
331 list_for_each_entry(s, &o->snapshots, list) {
332 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
342 } else if (snap_dest)
353 * On success, returns 1 if this snapshot is a handover destination,
354 * otherwise returns 0.
356 static int __validate_exception_handover(struct dm_snapshot *snap)
358 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
360 /* Does snapshot need exceptions handed over to it? */
361 if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest) == 2) ||
363 snap->ti->error = "Snapshot cow pairing for exception "
364 "table handover failed";
369 * If no snap_src was found, snap cannot become a handover
378 static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
380 struct dm_snapshot *l;
382 /* Sort the list according to chunk size, largest-first smallest-last */
383 list_for_each_entry(l, &o->snapshots, list)
384 if (l->store->chunk_size < s->store->chunk_size)
386 list_add_tail(&s->list, &l->list);
390 * Make a note of the snapshot and its origin so we can look it
391 * up when the origin has a write on it.
393 * Also validate snapshot exception store handovers.
394 * On success, returns 1 if this registration is a handover destination,
395 * otherwise returns 0.
397 static int register_snapshot(struct dm_snapshot *snap)
399 struct origin *o, *new_o = NULL;
400 struct block_device *bdev = snap->origin->bdev;
403 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
407 down_write(&_origins_lock);
409 r = __validate_exception_handover(snap);
415 o = __lookup_origin(bdev);
422 /* Initialise the struct */
423 INIT_LIST_HEAD(&o->snapshots);
429 __insert_snapshot(o, snap);
432 up_write(&_origins_lock);
438 * Move snapshot to correct place in list according to chunk size.
440 static void reregister_snapshot(struct dm_snapshot *s)
442 struct block_device *bdev = s->origin->bdev;
444 down_write(&_origins_lock);
447 __insert_snapshot(__lookup_origin(bdev), s);
449 up_write(&_origins_lock);
452 static void unregister_snapshot(struct dm_snapshot *s)
456 down_write(&_origins_lock);
457 o = __lookup_origin(s->origin->bdev);
460 if (o && list_empty(&o->snapshots)) {
461 list_del(&o->hash_list);
465 up_write(&_origins_lock);
469 * Implementation of the exception hash tables.
470 * The lowest hash_shift bits of the chunk number are ignored, allowing
471 * some consecutive chunks to be grouped together.
473 static int dm_exception_table_init(struct dm_exception_table *et,
474 uint32_t size, unsigned hash_shift)
478 et->hash_shift = hash_shift;
479 et->hash_mask = size - 1;
480 et->table = dm_vcalloc(size, sizeof(struct list_head));
484 for (i = 0; i < size; i++)
485 INIT_LIST_HEAD(et->table + i);
490 static void dm_exception_table_exit(struct dm_exception_table *et,
491 struct kmem_cache *mem)
493 struct list_head *slot;
494 struct dm_exception *ex, *next;
497 size = et->hash_mask + 1;
498 for (i = 0; i < size; i++) {
499 slot = et->table + i;
501 list_for_each_entry_safe (ex, next, slot, hash_list)
502 kmem_cache_free(mem, ex);
508 static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
510 return (chunk >> et->hash_shift) & et->hash_mask;
513 static void dm_remove_exception(struct dm_exception *e)
515 list_del(&e->hash_list);
519 * Return the exception data for a sector, or NULL if not
522 static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
525 struct list_head *slot;
526 struct dm_exception *e;
528 slot = &et->table[exception_hash(et, chunk)];
529 list_for_each_entry (e, slot, hash_list)
530 if (chunk >= e->old_chunk &&
531 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
537 static struct dm_exception *alloc_completed_exception(void)
539 struct dm_exception *e;
541 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
543 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
548 static void free_completed_exception(struct dm_exception *e)
550 kmem_cache_free(exception_cache, e);
553 static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
555 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
558 atomic_inc(&s->pending_exceptions_count);
564 static void free_pending_exception(struct dm_snap_pending_exception *pe)
566 struct dm_snapshot *s = pe->snap;
568 mempool_free(pe, s->pending_pool);
569 smp_mb__before_atomic_dec();
570 atomic_dec(&s->pending_exceptions_count);
573 static void dm_insert_exception(struct dm_exception_table *eh,
574 struct dm_exception *new_e)
577 struct dm_exception *e = NULL;
579 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
581 /* Add immediately if this table doesn't support consecutive chunks */
585 /* List is ordered by old_chunk */
586 list_for_each_entry_reverse(e, l, hash_list) {
587 /* Insert after an existing chunk? */
588 if (new_e->old_chunk == (e->old_chunk +
589 dm_consecutive_chunk_count(e) + 1) &&
590 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
591 dm_consecutive_chunk_count(e) + 1)) {
592 dm_consecutive_chunk_count_inc(e);
593 free_completed_exception(new_e);
597 /* Insert before an existing chunk? */
598 if (new_e->old_chunk == (e->old_chunk - 1) &&
599 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
600 dm_consecutive_chunk_count_inc(e);
603 free_completed_exception(new_e);
607 if (new_e->old_chunk > e->old_chunk)
612 list_add(&new_e->hash_list, e ? &e->hash_list : l);
616 * Callback used by the exception stores to load exceptions when
619 static int dm_add_exception(void *context, chunk_t old, chunk_t new)
621 struct dm_snapshot *s = context;
622 struct dm_exception *e;
624 e = alloc_completed_exception();
630 /* Consecutive_count is implicitly initialised to zero */
633 dm_insert_exception(&s->complete, e);
638 #define min_not_zero(l, r) (((l) == 0) ? (r) : (((r) == 0) ? (l) : min(l, r)))
641 * Return a minimum chunk size of all snapshots that have the specified origin.
642 * Return zero if the origin has no snapshots.
644 static sector_t __minimum_chunk_size(struct origin *o)
646 struct dm_snapshot *snap;
647 unsigned chunk_size = 0;
650 list_for_each_entry(snap, &o->snapshots, list)
651 chunk_size = min_not_zero(chunk_size,
652 snap->store->chunk_size);
660 static int calc_max_buckets(void)
662 /* use a fixed size of 2MB */
663 unsigned long mem = 2 * 1024 * 1024;
664 mem /= sizeof(struct list_head);
670 * Allocate room for a suitable hash table.
672 static int init_hash_tables(struct dm_snapshot *s)
674 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
677 * Calculate based on the size of the original volume or
680 cow_dev_size = get_dev_size(s->cow->bdev);
681 origin_dev_size = get_dev_size(s->origin->bdev);
682 max_buckets = calc_max_buckets();
684 hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift;
685 hash_size = min(hash_size, max_buckets);
689 hash_size = rounddown_pow_of_two(hash_size);
690 if (dm_exception_table_init(&s->complete, hash_size,
691 DM_CHUNK_CONSECUTIVE_BITS))
695 * Allocate hash table for in-flight exceptions
696 * Make this smaller than the real hash table
702 if (dm_exception_table_init(&s->pending, hash_size, 0)) {
703 dm_exception_table_exit(&s->complete, exception_cache);
711 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
713 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
715 struct dm_snapshot *s;
718 char *origin_path, *cow_path;
722 ti->error = "requires exactly 4 arguments";
727 origin_path = argv[0];
731 s = kmalloc(sizeof(*s), GFP_KERNEL);
733 ti->error = "Cannot allocate snapshot context private "
743 r = dm_get_device(ti, cow_path, 0, 0,
744 FMODE_READ | FMODE_WRITE, &s->cow);
746 ti->error = "Cannot get COW device";
750 r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
752 ti->error = "Couldn't create exception store";
760 r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin);
762 ti->error = "Cannot get origin device";
770 atomic_set(&s->pending_exceptions_count, 0);
771 init_rwsem(&s->lock);
772 INIT_LIST_HEAD(&s->list);
773 spin_lock_init(&s->pe_lock);
775 /* Allocate hash table for COW data */
776 if (init_hash_tables(s)) {
777 ti->error = "Unable to allocate hash table space";
779 goto bad_hash_tables;
782 r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client);
784 ti->error = "Could not create kcopyd client";
788 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
789 if (!s->pending_pool) {
790 ti->error = "Could not allocate mempool for pending exceptions";
791 goto bad_pending_pool;
794 s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS,
795 tracked_chunk_cache);
796 if (!s->tracked_chunk_pool) {
797 ti->error = "Could not allocate tracked_chunk mempool for "
799 goto bad_tracked_chunk_pool;
802 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
803 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
805 spin_lock_init(&s->tracked_chunk_lock);
807 bio_list_init(&s->queued_bios);
808 INIT_WORK(&s->queued_bios_work, flush_queued_bios);
811 ti->num_flush_requests = 1;
813 /* Add snapshot to the list of snapshots for this origin */
814 /* Exceptions aren't triggered till snapshot_resume() is called */
815 r = register_snapshot(s);
817 ti->error = "Snapshot origin struct allocation failed";
818 goto bad_load_and_register;
820 /* invalid handover, register_snapshot has set ti->error */
821 goto bad_load_and_register;
825 * Metadata must only be loaded into one table at once, so skip this
826 * if metadata will be handed over during resume.
827 * Chunk size will be set during the handover - set it to zero to
828 * ensure it's ignored.
831 s->store->chunk_size = 0;
835 r = s->store->type->read_metadata(s->store, dm_add_exception,
838 ti->error = "Failed to read snapshot metadata";
839 goto bad_read_metadata;
842 DMWARN("Snapshot is marked invalid.");
845 if (!s->store->chunk_size) {
846 ti->error = "Chunk size not set";
847 goto bad_read_metadata;
849 ti->split_io = s->store->chunk_size;
854 unregister_snapshot(s);
856 bad_load_and_register:
857 mempool_destroy(s->tracked_chunk_pool);
859 bad_tracked_chunk_pool:
860 mempool_destroy(s->pending_pool);
863 dm_kcopyd_client_destroy(s->kcopyd_client);
866 dm_exception_table_exit(&s->pending, pending_cache);
867 dm_exception_table_exit(&s->complete, exception_cache);
870 dm_put_device(ti, s->origin);
873 dm_exception_store_destroy(s->store);
876 dm_put_device(ti, s->cow);
885 static void __free_exceptions(struct dm_snapshot *s)
887 dm_kcopyd_client_destroy(s->kcopyd_client);
888 s->kcopyd_client = NULL;
890 dm_exception_table_exit(&s->pending, pending_cache);
891 dm_exception_table_exit(&s->complete, exception_cache);
894 static void __handover_exceptions(struct dm_snapshot *snap_src,
895 struct dm_snapshot *snap_dest)
898 struct dm_exception_table table_swap;
899 struct dm_exception_store *store_swap;
903 * Swap all snapshot context information between the two instances.
905 u.table_swap = snap_dest->complete;
906 snap_dest->complete = snap_src->complete;
907 snap_src->complete = u.table_swap;
909 u.store_swap = snap_dest->store;
910 snap_dest->store = snap_src->store;
911 snap_src->store = u.store_swap;
913 snap_dest->store->snap = snap_dest;
914 snap_src->store->snap = snap_src;
916 snap_dest->ti->split_io = snap_dest->store->chunk_size;
917 snap_dest->valid = snap_src->valid;
920 * Set source invalid to ensure it receives no further I/O.
925 static void snapshot_dtr(struct dm_target *ti)
927 #ifdef CONFIG_DM_DEBUG
930 struct dm_snapshot *s = ti->private;
931 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
933 flush_workqueue(ksnapd);
935 down_read(&_origins_lock);
936 /* Check whether exception handover must be cancelled */
937 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest);
938 if (snap_src && snap_dest && (s == snap_src)) {
939 down_write(&snap_dest->lock);
940 snap_dest->valid = 0;
941 up_write(&snap_dest->lock);
942 DMERR("Cancelling snapshot handover.");
944 up_read(&_origins_lock);
946 /* Prevent further origin writes from using this snapshot. */
947 /* After this returns there can be no new kcopyd jobs. */
948 unregister_snapshot(s);
950 while (atomic_read(&s->pending_exceptions_count))
953 * Ensure instructions in mempool_destroy aren't reordered
954 * before atomic_read.
958 #ifdef CONFIG_DM_DEBUG
959 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
960 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
963 mempool_destroy(s->tracked_chunk_pool);
965 __free_exceptions(s);
967 mempool_destroy(s->pending_pool);
969 dm_put_device(ti, s->origin);
971 dm_exception_store_destroy(s->store);
973 dm_put_device(ti, s->cow);
979 * Flush a list of buffers.
981 static void flush_bios(struct bio *bio)
988 generic_make_request(bio);
993 static void flush_queued_bios(struct work_struct *work)
995 struct dm_snapshot *s =
996 container_of(work, struct dm_snapshot, queued_bios_work);
997 struct bio *queued_bios;
1000 spin_lock_irqsave(&s->pe_lock, flags);
1001 queued_bios = bio_list_get(&s->queued_bios);
1002 spin_unlock_irqrestore(&s->pe_lock, flags);
1004 flush_bios(queued_bios);
1008 * Error a list of buffers.
1010 static void error_bios(struct bio *bio)
1016 bio->bi_next = NULL;
1022 static void __invalidate_snapshot(struct dm_snapshot *s, int err)
1028 DMERR("Invalidating snapshot: Error reading/writing.");
1029 else if (err == -ENOMEM)
1030 DMERR("Invalidating snapshot: Unable to allocate exception.");
1032 if (s->store->type->drop_snapshot)
1033 s->store->type->drop_snapshot(s->store);
1037 dm_table_event(s->ti->table);
1040 static void get_pending_exception(struct dm_snap_pending_exception *pe)
1042 atomic_inc(&pe->ref_count);
1045 static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe)
1047 struct dm_snap_pending_exception *primary_pe;
1048 struct bio *origin_bios = NULL;
1050 primary_pe = pe->primary_pe;
1053 * If this pe is involved in a write to the origin and
1054 * it is the last sibling to complete then release
1055 * the bios for the original write to the origin.
1058 atomic_dec_and_test(&primary_pe->ref_count)) {
1059 origin_bios = bio_list_get(&primary_pe->origin_bios);
1060 free_pending_exception(primary_pe);
1064 * Free the pe if it's not linked to an origin write or if
1065 * it's not itself a primary pe.
1067 if (!primary_pe || primary_pe != pe)
1068 free_pending_exception(pe);
1073 static void pending_complete(struct dm_snap_pending_exception *pe, int success)
1075 struct dm_exception *e;
1076 struct dm_snapshot *s = pe->snap;
1077 struct bio *origin_bios = NULL;
1078 struct bio *snapshot_bios = NULL;
1082 /* Read/write error - snapshot is unusable */
1083 down_write(&s->lock);
1084 __invalidate_snapshot(s, -EIO);
1089 e = alloc_completed_exception();
1091 down_write(&s->lock);
1092 __invalidate_snapshot(s, -ENOMEM);
1098 down_write(&s->lock);
1100 free_completed_exception(e);
1106 * Check for conflicting reads. This is extremely improbable,
1107 * so msleep(1) is sufficient and there is no need for a wait queue.
1109 while (__chunk_is_tracked(s, pe->e.old_chunk))
1113 * Add a proper exception, and remove the
1114 * in-flight exception from the list.
1116 dm_insert_exception(&s->complete, e);
1119 dm_remove_exception(&pe->e);
1120 snapshot_bios = bio_list_get(&pe->snapshot_bios);
1121 origin_bios = put_pending_exception(pe);
1125 /* Submit any pending write bios */
1127 error_bios(snapshot_bios);
1129 flush_bios(snapshot_bios);
1131 flush_bios(origin_bios);
1134 static void commit_callback(void *context, int success)
1136 struct dm_snap_pending_exception *pe = context;
1138 pending_complete(pe, success);
1142 * Called when the copy I/O has finished. kcopyd actually runs
1143 * this code so don't block.
1145 static void copy_callback(int read_err, unsigned long write_err, void *context)
1147 struct dm_snap_pending_exception *pe = context;
1148 struct dm_snapshot *s = pe->snap;
1150 if (read_err || write_err)
1151 pending_complete(pe, 0);
1154 /* Update the metadata if we are persistent */
1155 s->store->type->commit_exception(s->store, &pe->e,
1156 commit_callback, pe);
1160 * Dispatches the copy operation to kcopyd.
1162 static void start_copy(struct dm_snap_pending_exception *pe)
1164 struct dm_snapshot *s = pe->snap;
1165 struct dm_io_region src, dest;
1166 struct block_device *bdev = s->origin->bdev;
1169 dev_size = get_dev_size(bdev);
1172 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
1173 src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
1175 dest.bdev = s->cow->bdev;
1176 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
1177 dest.count = src.count;
1179 /* Hand over to kcopyd */
1180 dm_kcopyd_copy(s->kcopyd_client,
1181 &src, 1, &dest, 0, copy_callback, pe);
1184 static struct dm_snap_pending_exception *
1185 __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1187 struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
1192 return container_of(e, struct dm_snap_pending_exception, e);
1196 * Looks to see if this snapshot already has a pending exception
1197 * for this chunk, otherwise it allocates a new one and inserts
1198 * it into the pending table.
1200 * NOTE: a write lock must be held on snap->lock before calling
1203 static struct dm_snap_pending_exception *
1204 __find_pending_exception(struct dm_snapshot *s,
1205 struct dm_snap_pending_exception *pe, chunk_t chunk)
1207 struct dm_snap_pending_exception *pe2;
1209 pe2 = __lookup_pending_exception(s, chunk);
1211 free_pending_exception(pe);
1215 pe->e.old_chunk = chunk;
1216 bio_list_init(&pe->origin_bios);
1217 bio_list_init(&pe->snapshot_bios);
1218 pe->primary_pe = NULL;
1219 atomic_set(&pe->ref_count, 0);
1222 if (s->store->type->prepare_exception(s->store, &pe->e)) {
1223 free_pending_exception(pe);
1227 get_pending_exception(pe);
1228 dm_insert_exception(&s->pending, &pe->e);
1233 static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
1234 struct bio *bio, chunk_t chunk)
1236 bio->bi_bdev = s->cow->bdev;
1237 bio->bi_sector = chunk_to_sector(s->store,
1238 dm_chunk_number(e->new_chunk) +
1239 (chunk - e->old_chunk)) +
1241 s->store->chunk_mask);
1244 static int snapshot_map(struct dm_target *ti, struct bio *bio,
1245 union map_info *map_context)
1247 struct dm_exception *e;
1248 struct dm_snapshot *s = ti->private;
1249 int r = DM_MAPIO_REMAPPED;
1251 struct dm_snap_pending_exception *pe = NULL;
1253 if (unlikely(bio_empty_barrier(bio))) {
1254 bio->bi_bdev = s->cow->bdev;
1255 return DM_MAPIO_REMAPPED;
1258 chunk = sector_to_chunk(s->store, bio->bi_sector);
1260 /* Full snapshots are not usable */
1261 /* To get here the table must be live so s->active is always set. */
1265 /* FIXME: should only take write lock if we need
1266 * to copy an exception */
1267 down_write(&s->lock);
1274 /* If the block is already remapped - use that, else remap it */
1275 e = dm_lookup_exception(&s->complete, chunk);
1277 remap_exception(s, e, bio, chunk);
1282 * Write to snapshot - higher level takes care of RW/RO
1283 * flags so we should only get this if we are
1286 if (bio_rw(bio) == WRITE) {
1287 pe = __lookup_pending_exception(s, chunk);
1290 pe = alloc_pending_exception(s);
1291 down_write(&s->lock);
1294 free_pending_exception(pe);
1299 e = dm_lookup_exception(&s->complete, chunk);
1301 free_pending_exception(pe);
1302 remap_exception(s, e, bio, chunk);
1306 pe = __find_pending_exception(s, pe, chunk);
1308 __invalidate_snapshot(s, -ENOMEM);
1314 remap_exception(s, &pe->e, bio, chunk);
1315 bio_list_add(&pe->snapshot_bios, bio);
1317 r = DM_MAPIO_SUBMITTED;
1320 /* this is protected by snap->lock */
1327 bio->bi_bdev = s->origin->bdev;
1328 map_context->ptr = track_chunk(s, chunk);
1337 static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1338 int error, union map_info *map_context)
1340 struct dm_snapshot *s = ti->private;
1341 struct dm_snap_tracked_chunk *c = map_context->ptr;
1344 stop_tracking_chunk(s, c);
1349 static void snapshot_postsuspend(struct dm_target *ti)
1351 struct dm_snapshot *s = ti->private;
1353 down_write(&s->lock);
1358 static int snapshot_preresume(struct dm_target *ti)
1361 struct dm_snapshot *s = ti->private;
1362 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1364 down_read(&_origins_lock);
1365 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest);
1366 if (snap_src && snap_dest) {
1367 down_read(&snap_src->lock);
1368 if (s == snap_src) {
1369 DMERR("Unable to resume snapshot source until "
1370 "handover completes.");
1372 } else if (!snap_src->suspended) {
1373 DMERR("Unable to perform snapshot handover until "
1374 "source is suspended.");
1377 up_read(&snap_src->lock);
1379 up_read(&_origins_lock);
1384 static void snapshot_resume(struct dm_target *ti)
1386 struct dm_snapshot *s = ti->private;
1387 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1389 down_read(&_origins_lock);
1390 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest);
1391 if (snap_src && snap_dest) {
1392 down_write(&snap_src->lock);
1393 down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1394 __handover_exceptions(snap_src, snap_dest);
1395 up_write(&snap_dest->lock);
1396 up_write(&snap_src->lock);
1398 up_read(&_origins_lock);
1400 /* Now we have correct chunk size, reregister */
1401 reregister_snapshot(s);
1403 down_write(&s->lock);
1409 static int snapshot_status(struct dm_target *ti, status_type_t type,
1410 char *result, unsigned int maxlen)
1413 struct dm_snapshot *snap = ti->private;
1416 case STATUSTYPE_INFO:
1418 down_write(&snap->lock);
1423 if (snap->store->type->usage) {
1424 sector_t total_sectors, sectors_allocated,
1426 snap->store->type->usage(snap->store,
1430 DMEMIT("%llu/%llu %llu",
1431 (unsigned long long)sectors_allocated,
1432 (unsigned long long)total_sectors,
1433 (unsigned long long)metadata_sectors);
1439 up_write(&snap->lock);
1443 case STATUSTYPE_TABLE:
1445 * kdevname returns a static pointer so we need
1446 * to make private copies if the output is to
1449 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
1450 snap->store->type->status(snap->store, type, result + sz,
1458 static int snapshot_iterate_devices(struct dm_target *ti,
1459 iterate_devices_callout_fn fn, void *data)
1461 struct dm_snapshot *snap = ti->private;
1463 return fn(ti, snap->origin, 0, ti->len, data);
1467 /*-----------------------------------------------------------------
1469 *---------------------------------------------------------------*/
1470 static int __origin_write(struct list_head *snapshots, struct bio *bio)
1472 int r = DM_MAPIO_REMAPPED, first = 0;
1473 struct dm_snapshot *snap;
1474 struct dm_exception *e;
1475 struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL;
1477 LIST_HEAD(pe_queue);
1479 /* Do all the snapshots on this origin */
1480 list_for_each_entry (snap, snapshots, list) {
1482 down_write(&snap->lock);
1484 /* Only deal with valid and active snapshots */
1485 if (!snap->valid || !snap->active)
1488 /* Nothing to do if writing beyond end of snapshot */
1489 if (bio->bi_sector >= dm_table_get_size(snap->ti->table))
1493 * Remember, different snapshots can have
1494 * different chunk sizes.
1496 chunk = sector_to_chunk(snap->store, bio->bi_sector);
1499 * Check exception table to see if block
1500 * is already remapped in this snapshot
1501 * and trigger an exception if not.
1503 * ref_count is initialised to 1 so pending_complete()
1504 * won't destroy the primary_pe while we're inside this loop.
1506 e = dm_lookup_exception(&snap->complete, chunk);
1510 pe = __lookup_pending_exception(snap, chunk);
1512 up_write(&snap->lock);
1513 pe = alloc_pending_exception(snap);
1514 down_write(&snap->lock);
1517 free_pending_exception(pe);
1521 e = dm_lookup_exception(&snap->complete, chunk);
1523 free_pending_exception(pe);
1527 pe = __find_pending_exception(snap, pe, chunk);
1529 __invalidate_snapshot(snap, -ENOMEM);
1536 * Either every pe here has same
1537 * primary_pe or none has one yet.
1540 primary_pe = pe->primary_pe;
1546 bio_list_add(&primary_pe->origin_bios, bio);
1548 r = DM_MAPIO_SUBMITTED;
1551 if (!pe->primary_pe) {
1552 pe->primary_pe = primary_pe;
1553 get_pending_exception(primary_pe);
1558 list_add_tail(&pe->list, &pe_queue);
1562 up_write(&snap->lock);
1569 * If this is the first time we're processing this chunk and
1570 * ref_count is now 1 it means all the pending exceptions
1571 * got completed while we were in the loop above, so it falls to
1572 * us here to remove the primary_pe and submit any origin_bios.
1575 if (first && atomic_dec_and_test(&primary_pe->ref_count)) {
1576 flush_bios(bio_list_get(&primary_pe->origin_bios));
1577 free_pending_exception(primary_pe);
1578 /* If we got here, pe_queue is necessarily empty. */
1583 * Now that we have a complete pe list we can start the copying.
1585 list_for_each_entry_safe(pe, next_pe, &pe_queue, list)
1592 * Called on a write from the origin driver.
1594 static int do_origin(struct dm_dev *origin, struct bio *bio)
1597 int r = DM_MAPIO_REMAPPED;
1599 down_read(&_origins_lock);
1600 o = __lookup_origin(origin->bdev);
1602 r = __origin_write(&o->snapshots, bio);
1603 up_read(&_origins_lock);
1609 * Origin: maps a linear range of a device, with hooks for snapshotting.
1613 * Construct an origin mapping: <dev_path>
1614 * The context for an origin is merely a 'struct dm_dev *'
1615 * pointing to the real device.
1617 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1623 ti->error = "origin: incorrect number of arguments";
1627 r = dm_get_device(ti, argv[0], 0, ti->len,
1628 dm_table_get_mode(ti->table), &dev);
1630 ti->error = "Cannot get target device";
1635 ti->num_flush_requests = 1;
1640 static void origin_dtr(struct dm_target *ti)
1642 struct dm_dev *dev = ti->private;
1643 dm_put_device(ti, dev);
1646 static int origin_map(struct dm_target *ti, struct bio *bio,
1647 union map_info *map_context)
1649 struct dm_dev *dev = ti->private;
1650 bio->bi_bdev = dev->bdev;
1652 if (unlikely(bio_empty_barrier(bio)))
1653 return DM_MAPIO_REMAPPED;
1655 /* Only tell snapshots if this is a write */
1656 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
1660 * Set the target "split_io" field to the minimum of all the snapshots'
1663 static void origin_resume(struct dm_target *ti)
1665 struct dm_dev *dev = ti->private;
1667 down_read(&_origins_lock);
1669 ti->split_io = __minimum_chunk_size(__lookup_origin(dev->bdev));
1671 up_read(&_origins_lock);
1674 static int origin_status(struct dm_target *ti, status_type_t type, char *result,
1675 unsigned int maxlen)
1677 struct dm_dev *dev = ti->private;
1680 case STATUSTYPE_INFO:
1684 case STATUSTYPE_TABLE:
1685 snprintf(result, maxlen, "%s", dev->name);
1692 static int origin_iterate_devices(struct dm_target *ti,
1693 iterate_devices_callout_fn fn, void *data)
1695 struct dm_dev *dev = ti->private;
1697 return fn(ti, dev, 0, ti->len, data);
1700 static struct target_type origin_target = {
1701 .name = "snapshot-origin",
1702 .version = {1, 7, 0},
1703 .module = THIS_MODULE,
1707 .resume = origin_resume,
1708 .status = origin_status,
1709 .iterate_devices = origin_iterate_devices,
1712 static struct target_type snapshot_target = {
1714 .version = {1, 9, 0},
1715 .module = THIS_MODULE,
1716 .ctr = snapshot_ctr,
1717 .dtr = snapshot_dtr,
1718 .map = snapshot_map,
1719 .end_io = snapshot_end_io,
1720 .postsuspend = snapshot_postsuspend,
1721 .preresume = snapshot_preresume,
1722 .resume = snapshot_resume,
1723 .status = snapshot_status,
1724 .iterate_devices = snapshot_iterate_devices,
1727 static int __init dm_snapshot_init(void)
1731 r = dm_exception_store_init();
1733 DMERR("Failed to initialize exception stores");
1737 r = dm_register_target(&snapshot_target);
1739 DMERR("snapshot target register failed %d", r);
1740 goto bad_register_snapshot_target;
1743 r = dm_register_target(&origin_target);
1745 DMERR("Origin target register failed %d", r);
1749 r = init_origin_hash();
1751 DMERR("init_origin_hash failed.");
1755 exception_cache = KMEM_CACHE(dm_exception, 0);
1756 if (!exception_cache) {
1757 DMERR("Couldn't create exception cache.");
1762 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
1763 if (!pending_cache) {
1764 DMERR("Couldn't create pending cache.");
1769 tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0);
1770 if (!tracked_chunk_cache) {
1771 DMERR("Couldn't create cache to track chunks in use.");
1776 ksnapd = create_singlethread_workqueue("ksnapd");
1778 DMERR("Failed to create ksnapd workqueue.");
1780 goto bad_pending_pool;
1786 kmem_cache_destroy(tracked_chunk_cache);
1788 kmem_cache_destroy(pending_cache);
1790 kmem_cache_destroy(exception_cache);
1794 dm_unregister_target(&origin_target);
1796 dm_unregister_target(&snapshot_target);
1798 bad_register_snapshot_target:
1799 dm_exception_store_exit();
1803 static void __exit dm_snapshot_exit(void)
1805 destroy_workqueue(ksnapd);
1807 dm_unregister_target(&snapshot_target);
1808 dm_unregister_target(&origin_target);
1811 kmem_cache_destroy(pending_cache);
1812 kmem_cache_destroy(exception_cache);
1813 kmem_cache_destroy(tracked_chunk_cache);
1815 dm_exception_store_exit();
1819 module_init(dm_snapshot_init);
1820 module_exit(dm_snapshot_exit);
1822 MODULE_DESCRIPTION(DM_NAME " snapshot target");
1823 MODULE_AUTHOR("Joe Thornber");
1824 MODULE_LICENSE("GPL");