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"
28 static const char dm_snapshot_merge_target_name[] = "snapshot-merge";
30 #define dm_target_is_snapshot_merge(ti) \
31 ((ti)->type->name == dm_snapshot_merge_target_name)
34 * The percentage increment we will wake up users at
36 #define WAKE_UP_PERCENT 5
39 * kcopyd priority of snapshot operations
41 #define SNAPSHOT_COPY_PRIORITY 2
44 * Reserve 1MB for each snapshot initially (with minimum of 1 page).
46 #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1)
49 * The size of the mempool used to track chunks in use.
53 #define DM_TRACKED_CHUNK_HASH_SIZE 16
54 #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
55 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
57 struct dm_exception_table {
60 struct list_head *table;
64 struct rw_semaphore lock;
66 struct dm_dev *origin;
71 /* List of snapshots per Origin */
72 struct list_head list;
74 /* You can't use a snapshot if this is 0 (e.g. if full) */
77 /* Origin writes don't trigger exceptions until this is set */
80 /* Whether or not owning mapped_device is suspended */
83 mempool_t *pending_pool;
85 atomic_t pending_exceptions_count;
87 struct dm_exception_table pending;
88 struct dm_exception_table complete;
91 * pe_lock protects all pending_exception operations and access
92 * as well as the snapshot_bios list.
96 /* The on disk metadata handler */
97 struct dm_exception_store *store;
99 struct dm_kcopyd_client *kcopyd_client;
101 /* Queue of snapshot writes for ksnapd to flush */
102 struct bio_list queued_bios;
103 struct work_struct queued_bios_work;
105 /* Chunks with outstanding reads */
106 mempool_t *tracked_chunk_pool;
107 spinlock_t tracked_chunk_lock;
108 struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
111 struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
115 EXPORT_SYMBOL(dm_snap_cow);
117 static struct workqueue_struct *ksnapd;
118 static void flush_queued_bios(struct work_struct *work);
120 static sector_t chunk_to_sector(struct dm_exception_store *store,
123 return chunk << store->chunk_shift;
126 static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
129 * There is only ever one instance of a particular block
130 * device so we can compare pointers safely.
135 struct dm_snap_pending_exception {
136 struct dm_exception e;
139 * Origin buffers waiting for this to complete are held
142 struct bio_list origin_bios;
143 struct bio_list snapshot_bios;
145 /* Pointer back to snapshot context */
146 struct dm_snapshot *snap;
149 * 1 indicates the exception has already been sent to
156 * Hash table mapping origin volumes to lists of snapshots and
157 * a lock to protect it
159 static struct kmem_cache *exception_cache;
160 static struct kmem_cache *pending_cache;
162 struct dm_snap_tracked_chunk {
163 struct hlist_node node;
167 static struct kmem_cache *tracked_chunk_cache;
169 static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s,
172 struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool,
178 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
179 hlist_add_head(&c->node,
180 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
181 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
186 static void stop_tracking_chunk(struct dm_snapshot *s,
187 struct dm_snap_tracked_chunk *c)
191 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
193 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
195 mempool_free(c, s->tracked_chunk_pool);
198 static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
200 struct dm_snap_tracked_chunk *c;
201 struct hlist_node *hn;
204 spin_lock_irq(&s->tracked_chunk_lock);
206 hlist_for_each_entry(c, hn,
207 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
208 if (c->chunk == chunk) {
214 spin_unlock_irq(&s->tracked_chunk_lock);
220 * This conflicting I/O is extremely improbable in the caller,
221 * so msleep(1) is sufficient and there is no need for a wait queue.
223 static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
225 while (__chunk_is_tracked(s, chunk))
230 * One of these per registered origin, held in the snapshot_origins hash
233 /* The origin device */
234 struct block_device *bdev;
236 struct list_head hash_list;
238 /* List of snapshots for this origin */
239 struct list_head snapshots;
243 * Size of the hash table for origin volumes. If we make this
244 * the size of the minors list then it should be nearly perfect
246 #define ORIGIN_HASH_SIZE 256
247 #define ORIGIN_MASK 0xFF
248 static struct list_head *_origins;
249 static struct rw_semaphore _origins_lock;
251 static int init_origin_hash(void)
255 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
258 DMERR("unable to allocate memory");
262 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
263 INIT_LIST_HEAD(_origins + i);
264 init_rwsem(&_origins_lock);
269 static void exit_origin_hash(void)
274 static unsigned origin_hash(struct block_device *bdev)
276 return bdev->bd_dev & ORIGIN_MASK;
279 static struct origin *__lookup_origin(struct block_device *origin)
281 struct list_head *ol;
284 ol = &_origins[origin_hash(origin)];
285 list_for_each_entry (o, ol, hash_list)
286 if (bdev_equal(o->bdev, origin))
292 static void __insert_origin(struct origin *o)
294 struct list_head *sl = &_origins[origin_hash(o->bdev)];
295 list_add_tail(&o->hash_list, sl);
299 * _origins_lock must be held when calling this function.
300 * Returns number of snapshots registered using the supplied cow device, plus:
301 * snap_src - a snapshot suitable for use as a source of exception handover
302 * snap_dest - a snapshot capable of receiving exception handover.
303 * snap_merge - an existing snapshot-merge target linked to the same origin.
304 * There can be at most one snapshot-merge target. The parameter is optional.
306 * Possible return values and states of snap_src and snap_dest.
307 * 0: NULL, NULL - first new snapshot
308 * 1: snap_src, NULL - normal snapshot
309 * 2: snap_src, snap_dest - waiting for handover
310 * 2: snap_src, NULL - handed over, waiting for old to be deleted
311 * 1: NULL, snap_dest - source got destroyed without handover
313 static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
314 struct dm_snapshot **snap_src,
315 struct dm_snapshot **snap_dest,
316 struct dm_snapshot **snap_merge)
318 struct dm_snapshot *s;
323 o = __lookup_origin(snap->origin->bdev);
327 list_for_each_entry(s, &o->snapshots, list) {
328 if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
330 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
340 } else if (snap_dest)
351 * On success, returns 1 if this snapshot is a handover destination,
352 * otherwise returns 0.
354 static int __validate_exception_handover(struct dm_snapshot *snap)
356 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
357 struct dm_snapshot *snap_merge = NULL;
359 /* Does snapshot need exceptions handed over to it? */
360 if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
361 &snap_merge) == 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
376 * Non-snapshot-merge handover?
378 if (!dm_target_is_snapshot_merge(snap->ti))
382 * Do not allow more than one merging snapshot.
385 snap->ti->error = "A snapshot is already merging.";
392 static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
394 struct dm_snapshot *l;
396 /* Sort the list according to chunk size, largest-first smallest-last */
397 list_for_each_entry(l, &o->snapshots, list)
398 if (l->store->chunk_size < s->store->chunk_size)
400 list_add_tail(&s->list, &l->list);
404 * Make a note of the snapshot and its origin so we can look it
405 * up when the origin has a write on it.
407 * Also validate snapshot exception store handovers.
408 * On success, returns 1 if this registration is a handover destination,
409 * otherwise returns 0.
411 static int register_snapshot(struct dm_snapshot *snap)
413 struct origin *o, *new_o = NULL;
414 struct block_device *bdev = snap->origin->bdev;
417 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
421 down_write(&_origins_lock);
423 r = __validate_exception_handover(snap);
429 o = __lookup_origin(bdev);
436 /* Initialise the struct */
437 INIT_LIST_HEAD(&o->snapshots);
443 __insert_snapshot(o, snap);
446 up_write(&_origins_lock);
452 * Move snapshot to correct place in list according to chunk size.
454 static void reregister_snapshot(struct dm_snapshot *s)
456 struct block_device *bdev = s->origin->bdev;
458 down_write(&_origins_lock);
461 __insert_snapshot(__lookup_origin(bdev), s);
463 up_write(&_origins_lock);
466 static void unregister_snapshot(struct dm_snapshot *s)
470 down_write(&_origins_lock);
471 o = __lookup_origin(s->origin->bdev);
474 if (o && list_empty(&o->snapshots)) {
475 list_del(&o->hash_list);
479 up_write(&_origins_lock);
483 * Implementation of the exception hash tables.
484 * The lowest hash_shift bits of the chunk number are ignored, allowing
485 * some consecutive chunks to be grouped together.
487 static int dm_exception_table_init(struct dm_exception_table *et,
488 uint32_t size, unsigned hash_shift)
492 et->hash_shift = hash_shift;
493 et->hash_mask = size - 1;
494 et->table = dm_vcalloc(size, sizeof(struct list_head));
498 for (i = 0; i < size; i++)
499 INIT_LIST_HEAD(et->table + i);
504 static void dm_exception_table_exit(struct dm_exception_table *et,
505 struct kmem_cache *mem)
507 struct list_head *slot;
508 struct dm_exception *ex, *next;
511 size = et->hash_mask + 1;
512 for (i = 0; i < size; i++) {
513 slot = et->table + i;
515 list_for_each_entry_safe (ex, next, slot, hash_list)
516 kmem_cache_free(mem, ex);
522 static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
524 return (chunk >> et->hash_shift) & et->hash_mask;
527 static void dm_remove_exception(struct dm_exception *e)
529 list_del(&e->hash_list);
533 * Return the exception data for a sector, or NULL if not
536 static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
539 struct list_head *slot;
540 struct dm_exception *e;
542 slot = &et->table[exception_hash(et, chunk)];
543 list_for_each_entry (e, slot, hash_list)
544 if (chunk >= e->old_chunk &&
545 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
551 static struct dm_exception *alloc_completed_exception(void)
553 struct dm_exception *e;
555 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
557 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
562 static void free_completed_exception(struct dm_exception *e)
564 kmem_cache_free(exception_cache, e);
567 static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
569 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
572 atomic_inc(&s->pending_exceptions_count);
578 static void free_pending_exception(struct dm_snap_pending_exception *pe)
580 struct dm_snapshot *s = pe->snap;
582 mempool_free(pe, s->pending_pool);
583 smp_mb__before_atomic_dec();
584 atomic_dec(&s->pending_exceptions_count);
587 static void dm_insert_exception(struct dm_exception_table *eh,
588 struct dm_exception *new_e)
591 struct dm_exception *e = NULL;
593 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
595 /* Add immediately if this table doesn't support consecutive chunks */
599 /* List is ordered by old_chunk */
600 list_for_each_entry_reverse(e, l, hash_list) {
601 /* Insert after an existing chunk? */
602 if (new_e->old_chunk == (e->old_chunk +
603 dm_consecutive_chunk_count(e) + 1) &&
604 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
605 dm_consecutive_chunk_count(e) + 1)) {
606 dm_consecutive_chunk_count_inc(e);
607 free_completed_exception(new_e);
611 /* Insert before an existing chunk? */
612 if (new_e->old_chunk == (e->old_chunk - 1) &&
613 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
614 dm_consecutive_chunk_count_inc(e);
617 free_completed_exception(new_e);
621 if (new_e->old_chunk > e->old_chunk)
626 list_add(&new_e->hash_list, e ? &e->hash_list : l);
630 * Callback used by the exception stores to load exceptions when
633 static int dm_add_exception(void *context, chunk_t old, chunk_t new)
635 struct dm_snapshot *s = context;
636 struct dm_exception *e;
638 e = alloc_completed_exception();
644 /* Consecutive_count is implicitly initialised to zero */
647 dm_insert_exception(&s->complete, e);
652 #define min_not_zero(l, r) (((l) == 0) ? (r) : (((r) == 0) ? (l) : min(l, r)))
655 * Return a minimum chunk size of all snapshots that have the specified origin.
656 * Return zero if the origin has no snapshots.
658 static sector_t __minimum_chunk_size(struct origin *o)
660 struct dm_snapshot *snap;
661 unsigned chunk_size = 0;
664 list_for_each_entry(snap, &o->snapshots, list)
665 chunk_size = min_not_zero(chunk_size,
666 snap->store->chunk_size);
674 static int calc_max_buckets(void)
676 /* use a fixed size of 2MB */
677 unsigned long mem = 2 * 1024 * 1024;
678 mem /= sizeof(struct list_head);
684 * Allocate room for a suitable hash table.
686 static int init_hash_tables(struct dm_snapshot *s)
688 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
691 * Calculate based on the size of the original volume or
694 cow_dev_size = get_dev_size(s->cow->bdev);
695 origin_dev_size = get_dev_size(s->origin->bdev);
696 max_buckets = calc_max_buckets();
698 hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift;
699 hash_size = min(hash_size, max_buckets);
703 hash_size = rounddown_pow_of_two(hash_size);
704 if (dm_exception_table_init(&s->complete, hash_size,
705 DM_CHUNK_CONSECUTIVE_BITS))
709 * Allocate hash table for in-flight exceptions
710 * Make this smaller than the real hash table
716 if (dm_exception_table_init(&s->pending, hash_size, 0)) {
717 dm_exception_table_exit(&s->complete, exception_cache);
725 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
727 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
729 struct dm_snapshot *s;
732 char *origin_path, *cow_path;
733 unsigned args_used, num_flush_requests = 1;
734 fmode_t origin_mode = FMODE_READ;
737 ti->error = "requires exactly 4 arguments";
742 if (dm_target_is_snapshot_merge(ti)) {
743 num_flush_requests = 2;
744 origin_mode = FMODE_WRITE;
747 origin_path = argv[0];
751 s = kmalloc(sizeof(*s), GFP_KERNEL);
753 ti->error = "Cannot allocate snapshot context private "
763 r = dm_get_device(ti, cow_path, 0, 0,
764 FMODE_READ | FMODE_WRITE, &s->cow);
766 ti->error = "Cannot get COW device";
770 r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
772 ti->error = "Couldn't create exception store";
780 r = dm_get_device(ti, origin_path, 0, ti->len, origin_mode, &s->origin);
782 ti->error = "Cannot get origin device";
790 atomic_set(&s->pending_exceptions_count, 0);
791 init_rwsem(&s->lock);
792 INIT_LIST_HEAD(&s->list);
793 spin_lock_init(&s->pe_lock);
795 /* Allocate hash table for COW data */
796 if (init_hash_tables(s)) {
797 ti->error = "Unable to allocate hash table space";
799 goto bad_hash_tables;
802 r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client);
804 ti->error = "Could not create kcopyd client";
808 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
809 if (!s->pending_pool) {
810 ti->error = "Could not allocate mempool for pending exceptions";
811 goto bad_pending_pool;
814 s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS,
815 tracked_chunk_cache);
816 if (!s->tracked_chunk_pool) {
817 ti->error = "Could not allocate tracked_chunk mempool for "
819 goto bad_tracked_chunk_pool;
822 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
823 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
825 spin_lock_init(&s->tracked_chunk_lock);
827 bio_list_init(&s->queued_bios);
828 INIT_WORK(&s->queued_bios_work, flush_queued_bios);
831 ti->num_flush_requests = num_flush_requests;
833 /* Add snapshot to the list of snapshots for this origin */
834 /* Exceptions aren't triggered till snapshot_resume() is called */
835 r = register_snapshot(s);
837 ti->error = "Snapshot origin struct allocation failed";
838 goto bad_load_and_register;
840 /* invalid handover, register_snapshot has set ti->error */
841 goto bad_load_and_register;
845 * Metadata must only be loaded into one table at once, so skip this
846 * if metadata will be handed over during resume.
847 * Chunk size will be set during the handover - set it to zero to
848 * ensure it's ignored.
851 s->store->chunk_size = 0;
855 r = s->store->type->read_metadata(s->store, dm_add_exception,
858 ti->error = "Failed to read snapshot metadata";
859 goto bad_read_metadata;
862 DMWARN("Snapshot is marked invalid.");
865 if (!s->store->chunk_size) {
866 ti->error = "Chunk size not set";
867 goto bad_read_metadata;
869 ti->split_io = s->store->chunk_size;
874 unregister_snapshot(s);
876 bad_load_and_register:
877 mempool_destroy(s->tracked_chunk_pool);
879 bad_tracked_chunk_pool:
880 mempool_destroy(s->pending_pool);
883 dm_kcopyd_client_destroy(s->kcopyd_client);
886 dm_exception_table_exit(&s->pending, pending_cache);
887 dm_exception_table_exit(&s->complete, exception_cache);
890 dm_put_device(ti, s->origin);
893 dm_exception_store_destroy(s->store);
896 dm_put_device(ti, s->cow);
905 static void __free_exceptions(struct dm_snapshot *s)
907 dm_kcopyd_client_destroy(s->kcopyd_client);
908 s->kcopyd_client = NULL;
910 dm_exception_table_exit(&s->pending, pending_cache);
911 dm_exception_table_exit(&s->complete, exception_cache);
914 static void __handover_exceptions(struct dm_snapshot *snap_src,
915 struct dm_snapshot *snap_dest)
918 struct dm_exception_table table_swap;
919 struct dm_exception_store *store_swap;
923 * Swap all snapshot context information between the two instances.
925 u.table_swap = snap_dest->complete;
926 snap_dest->complete = snap_src->complete;
927 snap_src->complete = u.table_swap;
929 u.store_swap = snap_dest->store;
930 snap_dest->store = snap_src->store;
931 snap_src->store = u.store_swap;
933 snap_dest->store->snap = snap_dest;
934 snap_src->store->snap = snap_src;
936 snap_dest->ti->split_io = snap_dest->store->chunk_size;
937 snap_dest->valid = snap_src->valid;
940 * Set source invalid to ensure it receives no further I/O.
945 static void snapshot_dtr(struct dm_target *ti)
947 #ifdef CONFIG_DM_DEBUG
950 struct dm_snapshot *s = ti->private;
951 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
953 flush_workqueue(ksnapd);
955 down_read(&_origins_lock);
956 /* Check whether exception handover must be cancelled */
957 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
958 if (snap_src && snap_dest && (s == snap_src)) {
959 down_write(&snap_dest->lock);
960 snap_dest->valid = 0;
961 up_write(&snap_dest->lock);
962 DMERR("Cancelling snapshot handover.");
964 up_read(&_origins_lock);
966 /* Prevent further origin writes from using this snapshot. */
967 /* After this returns there can be no new kcopyd jobs. */
968 unregister_snapshot(s);
970 while (atomic_read(&s->pending_exceptions_count))
973 * Ensure instructions in mempool_destroy aren't reordered
974 * before atomic_read.
978 #ifdef CONFIG_DM_DEBUG
979 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
980 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
983 mempool_destroy(s->tracked_chunk_pool);
985 __free_exceptions(s);
987 mempool_destroy(s->pending_pool);
989 dm_put_device(ti, s->origin);
991 dm_exception_store_destroy(s->store);
993 dm_put_device(ti, s->cow);
999 * Flush a list of buffers.
1001 static void flush_bios(struct bio *bio)
1007 bio->bi_next = NULL;
1008 generic_make_request(bio);
1013 static void flush_queued_bios(struct work_struct *work)
1015 struct dm_snapshot *s =
1016 container_of(work, struct dm_snapshot, queued_bios_work);
1017 struct bio *queued_bios;
1018 unsigned long flags;
1020 spin_lock_irqsave(&s->pe_lock, flags);
1021 queued_bios = bio_list_get(&s->queued_bios);
1022 spin_unlock_irqrestore(&s->pe_lock, flags);
1024 flush_bios(queued_bios);
1027 static int do_origin(struct dm_dev *origin, struct bio *bio);
1030 * Flush a list of buffers.
1032 static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
1039 bio->bi_next = NULL;
1040 r = do_origin(s->origin, bio);
1041 if (r == DM_MAPIO_REMAPPED)
1042 generic_make_request(bio);
1048 * Error a list of buffers.
1050 static void error_bios(struct bio *bio)
1056 bio->bi_next = NULL;
1062 static void __invalidate_snapshot(struct dm_snapshot *s, int err)
1068 DMERR("Invalidating snapshot: Error reading/writing.");
1069 else if (err == -ENOMEM)
1070 DMERR("Invalidating snapshot: Unable to allocate exception.");
1072 if (s->store->type->drop_snapshot)
1073 s->store->type->drop_snapshot(s->store);
1077 dm_table_event(s->ti->table);
1080 static void pending_complete(struct dm_snap_pending_exception *pe, int success)
1082 struct dm_exception *e;
1083 struct dm_snapshot *s = pe->snap;
1084 struct bio *origin_bios = NULL;
1085 struct bio *snapshot_bios = NULL;
1089 /* Read/write error - snapshot is unusable */
1090 down_write(&s->lock);
1091 __invalidate_snapshot(s, -EIO);
1096 e = alloc_completed_exception();
1098 down_write(&s->lock);
1099 __invalidate_snapshot(s, -ENOMEM);
1105 down_write(&s->lock);
1107 free_completed_exception(e);
1112 /* Check for conflicting reads */
1113 __check_for_conflicting_io(s, pe->e.old_chunk);
1116 * Add a proper exception, and remove the
1117 * in-flight exception from the list.
1119 dm_insert_exception(&s->complete, e);
1122 dm_remove_exception(&pe->e);
1123 snapshot_bios = bio_list_get(&pe->snapshot_bios);
1124 origin_bios = bio_list_get(&pe->origin_bios);
1125 free_pending_exception(pe);
1129 /* Submit any pending write bios */
1131 error_bios(snapshot_bios);
1133 flush_bios(snapshot_bios);
1135 retry_origin_bios(s, origin_bios);
1138 static void commit_callback(void *context, int success)
1140 struct dm_snap_pending_exception *pe = context;
1142 pending_complete(pe, success);
1146 * Called when the copy I/O has finished. kcopyd actually runs
1147 * this code so don't block.
1149 static void copy_callback(int read_err, unsigned long write_err, void *context)
1151 struct dm_snap_pending_exception *pe = context;
1152 struct dm_snapshot *s = pe->snap;
1154 if (read_err || write_err)
1155 pending_complete(pe, 0);
1158 /* Update the metadata if we are persistent */
1159 s->store->type->commit_exception(s->store, &pe->e,
1160 commit_callback, pe);
1164 * Dispatches the copy operation to kcopyd.
1166 static void start_copy(struct dm_snap_pending_exception *pe)
1168 struct dm_snapshot *s = pe->snap;
1169 struct dm_io_region src, dest;
1170 struct block_device *bdev = s->origin->bdev;
1173 dev_size = get_dev_size(bdev);
1176 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
1177 src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
1179 dest.bdev = s->cow->bdev;
1180 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
1181 dest.count = src.count;
1183 /* Hand over to kcopyd */
1184 dm_kcopyd_copy(s->kcopyd_client,
1185 &src, 1, &dest, 0, copy_callback, pe);
1188 static struct dm_snap_pending_exception *
1189 __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1191 struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
1196 return container_of(e, struct dm_snap_pending_exception, e);
1200 * Looks to see if this snapshot already has a pending exception
1201 * for this chunk, otherwise it allocates a new one and inserts
1202 * it into the pending table.
1204 * NOTE: a write lock must be held on snap->lock before calling
1207 static struct dm_snap_pending_exception *
1208 __find_pending_exception(struct dm_snapshot *s,
1209 struct dm_snap_pending_exception *pe, chunk_t chunk)
1211 struct dm_snap_pending_exception *pe2;
1213 pe2 = __lookup_pending_exception(s, chunk);
1215 free_pending_exception(pe);
1219 pe->e.old_chunk = chunk;
1220 bio_list_init(&pe->origin_bios);
1221 bio_list_init(&pe->snapshot_bios);
1224 if (s->store->type->prepare_exception(s->store, &pe->e)) {
1225 free_pending_exception(pe);
1229 dm_insert_exception(&s->pending, &pe->e);
1234 static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
1235 struct bio *bio, chunk_t chunk)
1237 bio->bi_bdev = s->cow->bdev;
1238 bio->bi_sector = chunk_to_sector(s->store,
1239 dm_chunk_number(e->new_chunk) +
1240 (chunk - e->old_chunk)) +
1242 s->store->chunk_mask);
1245 static int snapshot_map(struct dm_target *ti, struct bio *bio,
1246 union map_info *map_context)
1248 struct dm_exception *e;
1249 struct dm_snapshot *s = ti->private;
1250 int r = DM_MAPIO_REMAPPED;
1252 struct dm_snap_pending_exception *pe = NULL;
1254 if (unlikely(bio_empty_barrier(bio))) {
1255 bio->bi_bdev = s->cow->bdev;
1256 return DM_MAPIO_REMAPPED;
1259 chunk = sector_to_chunk(s->store, bio->bi_sector);
1261 /* Full snapshots are not usable */
1262 /* To get here the table must be live so s->active is always set. */
1266 /* FIXME: should only take write lock if we need
1267 * to copy an exception */
1268 down_write(&s->lock);
1275 /* If the block is already remapped - use that, else remap it */
1276 e = dm_lookup_exception(&s->complete, chunk);
1278 remap_exception(s, e, bio, chunk);
1283 * Write to snapshot - higher level takes care of RW/RO
1284 * flags so we should only get this if we are
1287 if (bio_rw(bio) == WRITE) {
1288 pe = __lookup_pending_exception(s, chunk);
1291 pe = alloc_pending_exception(s);
1292 down_write(&s->lock);
1295 free_pending_exception(pe);
1300 e = dm_lookup_exception(&s->complete, chunk);
1302 free_pending_exception(pe);
1303 remap_exception(s, e, bio, chunk);
1307 pe = __find_pending_exception(s, pe, chunk);
1309 __invalidate_snapshot(s, -ENOMEM);
1315 remap_exception(s, &pe->e, bio, chunk);
1316 bio_list_add(&pe->snapshot_bios, bio);
1318 r = DM_MAPIO_SUBMITTED;
1321 /* this is protected by snap->lock */
1328 bio->bi_bdev = s->origin->bdev;
1329 map_context->ptr = track_chunk(s, chunk);
1339 * A snapshot-merge target behaves like a combination of a snapshot
1340 * target and a snapshot-origin target. It only generates new
1341 * exceptions in other snapshots and not in the one that is being
1344 * For each chunk, if there is an existing exception, it is used to
1345 * redirect I/O to the cow device. Otherwise I/O is sent to the origin,
1346 * which in turn might generate exceptions in other snapshots.
1348 static int snapshot_merge_map(struct dm_target *ti, struct bio *bio,
1349 union map_info *map_context)
1351 struct dm_exception *e;
1352 struct dm_snapshot *s = ti->private;
1353 int r = DM_MAPIO_REMAPPED;
1356 if (unlikely(bio_empty_barrier(bio))) {
1357 if (!map_context->flush_request)
1358 bio->bi_bdev = s->origin->bdev;
1360 bio->bi_bdev = s->cow->bdev;
1361 map_context->ptr = NULL;
1362 return DM_MAPIO_REMAPPED;
1365 chunk = sector_to_chunk(s->store, bio->bi_sector);
1367 down_read(&s->lock);
1369 /* Full snapshots are not usable */
1375 /* If the block is already remapped - use that */
1376 e = dm_lookup_exception(&s->complete, chunk);
1378 remap_exception(s, e, bio, chunk);
1382 bio->bi_bdev = s->origin->bdev;
1384 if (bio_rw(bio) == WRITE) {
1386 return do_origin(s->origin, bio);
1395 static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1396 int error, union map_info *map_context)
1398 struct dm_snapshot *s = ti->private;
1399 struct dm_snap_tracked_chunk *c = map_context->ptr;
1402 stop_tracking_chunk(s, c);
1407 static void snapshot_postsuspend(struct dm_target *ti)
1409 struct dm_snapshot *s = ti->private;
1411 down_write(&s->lock);
1416 static int snapshot_preresume(struct dm_target *ti)
1419 struct dm_snapshot *s = ti->private;
1420 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1422 down_read(&_origins_lock);
1423 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1424 if (snap_src && snap_dest) {
1425 down_read(&snap_src->lock);
1426 if (s == snap_src) {
1427 DMERR("Unable to resume snapshot source until "
1428 "handover completes.");
1430 } else if (!snap_src->suspended) {
1431 DMERR("Unable to perform snapshot handover until "
1432 "source is suspended.");
1435 up_read(&snap_src->lock);
1437 up_read(&_origins_lock);
1442 static void snapshot_resume(struct dm_target *ti)
1444 struct dm_snapshot *s = ti->private;
1445 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1447 down_read(&_origins_lock);
1448 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1449 if (snap_src && snap_dest) {
1450 down_write(&snap_src->lock);
1451 down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1452 __handover_exceptions(snap_src, snap_dest);
1453 up_write(&snap_dest->lock);
1454 up_write(&snap_src->lock);
1456 up_read(&_origins_lock);
1458 /* Now we have correct chunk size, reregister */
1459 reregister_snapshot(s);
1461 down_write(&s->lock);
1467 static int snapshot_status(struct dm_target *ti, status_type_t type,
1468 char *result, unsigned int maxlen)
1471 struct dm_snapshot *snap = ti->private;
1474 case STATUSTYPE_INFO:
1476 down_write(&snap->lock);
1481 if (snap->store->type->usage) {
1482 sector_t total_sectors, sectors_allocated,
1484 snap->store->type->usage(snap->store,
1488 DMEMIT("%llu/%llu %llu",
1489 (unsigned long long)sectors_allocated,
1490 (unsigned long long)total_sectors,
1491 (unsigned long long)metadata_sectors);
1497 up_write(&snap->lock);
1501 case STATUSTYPE_TABLE:
1503 * kdevname returns a static pointer so we need
1504 * to make private copies if the output is to
1507 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
1508 snap->store->type->status(snap->store, type, result + sz,
1516 static int snapshot_iterate_devices(struct dm_target *ti,
1517 iterate_devices_callout_fn fn, void *data)
1519 struct dm_snapshot *snap = ti->private;
1521 return fn(ti, snap->origin, 0, ti->len, data);
1525 /*-----------------------------------------------------------------
1527 *---------------------------------------------------------------*/
1530 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
1531 * supplied bio was ignored. The caller may submit it immediately.
1532 * (No remapping actually occurs as the origin is always a direct linear
1535 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
1536 * and any supplied bio is added to a list to be submitted once all
1537 * the necessary exceptions exist.
1539 static int __origin_write(struct list_head *snapshots, sector_t sector,
1542 int r = DM_MAPIO_REMAPPED;
1543 struct dm_snapshot *snap;
1544 struct dm_exception *e;
1545 struct dm_snap_pending_exception *pe;
1546 struct dm_snap_pending_exception *pe_to_start_now = NULL;
1547 struct dm_snap_pending_exception *pe_to_start_last = NULL;
1550 /* Do all the snapshots on this origin */
1551 list_for_each_entry (snap, snapshots, list) {
1553 * Don't make new exceptions in a merging snapshot
1554 * because it has effectively been deleted
1556 if (dm_target_is_snapshot_merge(snap->ti))
1559 down_write(&snap->lock);
1561 /* Only deal with valid and active snapshots */
1562 if (!snap->valid || !snap->active)
1565 /* Nothing to do if writing beyond end of snapshot */
1566 if (sector >= dm_table_get_size(snap->ti->table))
1570 * Remember, different snapshots can have
1571 * different chunk sizes.
1573 chunk = sector_to_chunk(snap->store, sector);
1576 * Check exception table to see if block
1577 * is already remapped in this snapshot
1578 * and trigger an exception if not.
1580 e = dm_lookup_exception(&snap->complete, chunk);
1584 pe = __lookup_pending_exception(snap, chunk);
1586 up_write(&snap->lock);
1587 pe = alloc_pending_exception(snap);
1588 down_write(&snap->lock);
1591 free_pending_exception(pe);
1595 e = dm_lookup_exception(&snap->complete, chunk);
1597 free_pending_exception(pe);
1601 pe = __find_pending_exception(snap, pe, chunk);
1603 __invalidate_snapshot(snap, -ENOMEM);
1608 r = DM_MAPIO_SUBMITTED;
1611 * If an origin bio was supplied, queue it to wait for the
1612 * completion of this exception, and start this one last,
1613 * at the end of the function.
1616 bio_list_add(&pe->origin_bios, bio);
1621 pe_to_start_last = pe;
1627 pe_to_start_now = pe;
1631 up_write(&snap->lock);
1633 if (pe_to_start_now) {
1634 start_copy(pe_to_start_now);
1635 pe_to_start_now = NULL;
1640 * Submit the exception against which the bio is queued last,
1641 * to give the other exceptions a head start.
1643 if (pe_to_start_last)
1644 start_copy(pe_to_start_last);
1650 * Called on a write from the origin driver.
1652 static int do_origin(struct dm_dev *origin, struct bio *bio)
1655 int r = DM_MAPIO_REMAPPED;
1657 down_read(&_origins_lock);
1658 o = __lookup_origin(origin->bdev);
1660 r = __origin_write(&o->snapshots, bio->bi_sector, bio);
1661 up_read(&_origins_lock);
1667 * Origin: maps a linear range of a device, with hooks for snapshotting.
1671 * Construct an origin mapping: <dev_path>
1672 * The context for an origin is merely a 'struct dm_dev *'
1673 * pointing to the real device.
1675 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1681 ti->error = "origin: incorrect number of arguments";
1685 r = dm_get_device(ti, argv[0], 0, ti->len,
1686 dm_table_get_mode(ti->table), &dev);
1688 ti->error = "Cannot get target device";
1693 ti->num_flush_requests = 1;
1698 static void origin_dtr(struct dm_target *ti)
1700 struct dm_dev *dev = ti->private;
1701 dm_put_device(ti, dev);
1704 static int origin_map(struct dm_target *ti, struct bio *bio,
1705 union map_info *map_context)
1707 struct dm_dev *dev = ti->private;
1708 bio->bi_bdev = dev->bdev;
1710 if (unlikely(bio_empty_barrier(bio)))
1711 return DM_MAPIO_REMAPPED;
1713 /* Only tell snapshots if this is a write */
1714 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
1718 * Set the target "split_io" field to the minimum of all the snapshots'
1721 static void origin_resume(struct dm_target *ti)
1723 struct dm_dev *dev = ti->private;
1725 down_read(&_origins_lock);
1727 ti->split_io = __minimum_chunk_size(__lookup_origin(dev->bdev));
1729 up_read(&_origins_lock);
1732 static int origin_status(struct dm_target *ti, status_type_t type, char *result,
1733 unsigned int maxlen)
1735 struct dm_dev *dev = ti->private;
1738 case STATUSTYPE_INFO:
1742 case STATUSTYPE_TABLE:
1743 snprintf(result, maxlen, "%s", dev->name);
1750 static int origin_iterate_devices(struct dm_target *ti,
1751 iterate_devices_callout_fn fn, void *data)
1753 struct dm_dev *dev = ti->private;
1755 return fn(ti, dev, 0, ti->len, data);
1758 static struct target_type origin_target = {
1759 .name = "snapshot-origin",
1760 .version = {1, 7, 0},
1761 .module = THIS_MODULE,
1765 .resume = origin_resume,
1766 .status = origin_status,
1767 .iterate_devices = origin_iterate_devices,
1770 static struct target_type snapshot_target = {
1772 .version = {1, 9, 0},
1773 .module = THIS_MODULE,
1774 .ctr = snapshot_ctr,
1775 .dtr = snapshot_dtr,
1776 .map = snapshot_map,
1777 .end_io = snapshot_end_io,
1778 .postsuspend = snapshot_postsuspend,
1779 .preresume = snapshot_preresume,
1780 .resume = snapshot_resume,
1781 .status = snapshot_status,
1782 .iterate_devices = snapshot_iterate_devices,
1785 static struct target_type merge_target = {
1786 .name = dm_snapshot_merge_target_name,
1787 .version = {1, 0, 0},
1788 .module = THIS_MODULE,
1789 .ctr = snapshot_ctr,
1790 .dtr = snapshot_dtr,
1791 .map = snapshot_merge_map,
1792 .end_io = snapshot_end_io,
1793 .postsuspend = snapshot_postsuspend,
1794 .preresume = snapshot_preresume,
1795 .resume = snapshot_resume,
1796 .status = snapshot_status,
1797 .iterate_devices = snapshot_iterate_devices,
1800 static int __init dm_snapshot_init(void)
1804 r = dm_exception_store_init();
1806 DMERR("Failed to initialize exception stores");
1810 r = dm_register_target(&snapshot_target);
1812 DMERR("snapshot target register failed %d", r);
1813 goto bad_register_snapshot_target;
1816 r = dm_register_target(&origin_target);
1818 DMERR("Origin target register failed %d", r);
1819 goto bad_register_origin_target;
1822 r = dm_register_target(&merge_target);
1824 DMERR("Merge target register failed %d", r);
1825 goto bad_register_merge_target;
1828 r = init_origin_hash();
1830 DMERR("init_origin_hash failed.");
1831 goto bad_origin_hash;
1834 exception_cache = KMEM_CACHE(dm_exception, 0);
1835 if (!exception_cache) {
1836 DMERR("Couldn't create exception cache.");
1838 goto bad_exception_cache;
1841 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
1842 if (!pending_cache) {
1843 DMERR("Couldn't create pending cache.");
1845 goto bad_pending_cache;
1848 tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0);
1849 if (!tracked_chunk_cache) {
1850 DMERR("Couldn't create cache to track chunks in use.");
1852 goto bad_tracked_chunk_cache;
1855 ksnapd = create_singlethread_workqueue("ksnapd");
1857 DMERR("Failed to create ksnapd workqueue.");
1859 goto bad_pending_pool;
1865 kmem_cache_destroy(tracked_chunk_cache);
1866 bad_tracked_chunk_cache:
1867 kmem_cache_destroy(pending_cache);
1869 kmem_cache_destroy(exception_cache);
1870 bad_exception_cache:
1873 dm_unregister_target(&merge_target);
1874 bad_register_merge_target:
1875 dm_unregister_target(&origin_target);
1876 bad_register_origin_target:
1877 dm_unregister_target(&snapshot_target);
1878 bad_register_snapshot_target:
1879 dm_exception_store_exit();
1884 static void __exit dm_snapshot_exit(void)
1886 destroy_workqueue(ksnapd);
1888 dm_unregister_target(&snapshot_target);
1889 dm_unregister_target(&origin_target);
1890 dm_unregister_target(&merge_target);
1893 kmem_cache_destroy(pending_cache);
1894 kmem_cache_destroy(exception_cache);
1895 kmem_cache_destroy(tracked_chunk_cache);
1897 dm_exception_store_exit();
1901 module_init(dm_snapshot_init);
1902 module_exit(dm_snapshot_exit);
1904 MODULE_DESCRIPTION(DM_NAME " snapshot target");
1905 MODULE_AUTHOR("Joe Thornber");
1906 MODULE_LICENSE("GPL");