4 * Copyright (C) 2001-2002 Sistina Software (UK) Limited.
6 * This file is released under the GPL.
9 #include <linux/blkdev.h>
10 #include <linux/device-mapper.h>
11 #include <linux/delay.h>
13 #include <linux/init.h>
14 #include <linux/kdev_t.h>
15 #include <linux/list.h>
16 #include <linux/mempool.h>
17 #include <linux/module.h>
18 #include <linux/slab.h>
19 #include <linux/vmalloc.h>
20 #include <linux/log2.h>
21 #include <linux/dm-kcopyd.h>
22 #include <linux/workqueue.h>
24 #include "dm-exception-store.h"
25 #include "dm-bio-list.h"
27 #define DM_MSG_PREFIX "snapshots"
30 * The percentage increment we will wake up users at
32 #define WAKE_UP_PERCENT 5
35 * kcopyd priority of snapshot operations
37 #define SNAPSHOT_COPY_PRIORITY 2
40 * Reserve 1MB for each snapshot initially (with minimum of 1 page).
42 #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1)
45 * The size of the mempool used to track chunks in use.
49 #define DM_TRACKED_CHUNK_HASH_SIZE 16
50 #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
51 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
53 struct exception_table {
56 struct list_head *table;
60 struct rw_semaphore lock;
62 struct dm_dev *origin;
64 /* List of snapshots per Origin */
65 struct list_head list;
67 /* You can't use a snapshot if this is 0 (e.g. if full) */
70 /* Origin writes don't trigger exceptions until this is set */
73 /* Used for display of table */
76 mempool_t *pending_pool;
78 atomic_t pending_exceptions_count;
80 struct exception_table pending;
81 struct exception_table complete;
84 * pe_lock protects all pending_exception operations and access
85 * as well as the snapshot_bios list.
89 /* The on disk metadata handler */
90 struct dm_exception_store *store;
92 struct dm_kcopyd_client *kcopyd_client;
94 /* Queue of snapshot writes for ksnapd to flush */
95 struct bio_list queued_bios;
96 struct work_struct queued_bios_work;
98 /* Chunks with outstanding reads */
99 mempool_t *tracked_chunk_pool;
100 spinlock_t tracked_chunk_lock;
101 struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
104 static struct workqueue_struct *ksnapd;
105 static void flush_queued_bios(struct work_struct *work);
107 static sector_t chunk_to_sector(struct dm_exception_store *store,
110 return chunk << store->chunk_shift;
113 static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
116 * There is only ever one instance of a particular block
117 * device so we can compare pointers safely.
122 struct dm_snap_pending_exception {
123 struct dm_snap_exception e;
126 * Origin buffers waiting for this to complete are held
129 struct bio_list origin_bios;
130 struct bio_list snapshot_bios;
133 * Short-term queue of pending exceptions prior to submission.
135 struct list_head list;
138 * The primary pending_exception is the one that holds
139 * the ref_count and the list of origin_bios for a
140 * group of pending_exceptions. It is always last to get freed.
141 * These fields get set up when writing to the origin.
143 struct dm_snap_pending_exception *primary_pe;
146 * Number of pending_exceptions processing this chunk.
147 * When this drops to zero we must complete the origin bios.
148 * If incrementing or decrementing this, hold pe->snap->lock for
149 * the sibling concerned and not pe->primary_pe->snap->lock unless
154 /* Pointer back to snapshot context */
155 struct dm_snapshot *snap;
158 * 1 indicates the exception has already been sent to
165 * Hash table mapping origin volumes to lists of snapshots and
166 * a lock to protect it
168 static struct kmem_cache *exception_cache;
169 static struct kmem_cache *pending_cache;
171 struct dm_snap_tracked_chunk {
172 struct hlist_node node;
176 static struct kmem_cache *tracked_chunk_cache;
178 static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s,
181 struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool,
187 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
188 hlist_add_head(&c->node,
189 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
190 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
195 static void stop_tracking_chunk(struct dm_snapshot *s,
196 struct dm_snap_tracked_chunk *c)
200 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
202 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
204 mempool_free(c, s->tracked_chunk_pool);
207 static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
209 struct dm_snap_tracked_chunk *c;
210 struct hlist_node *hn;
213 spin_lock_irq(&s->tracked_chunk_lock);
215 hlist_for_each_entry(c, hn,
216 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
217 if (c->chunk == chunk) {
223 spin_unlock_irq(&s->tracked_chunk_lock);
229 * One of these per registered origin, held in the snapshot_origins hash
232 /* The origin device */
233 struct block_device *bdev;
235 struct list_head hash_list;
237 /* List of snapshots for this origin */
238 struct list_head snapshots;
242 * Size of the hash table for origin volumes. If we make this
243 * the size of the minors list then it should be nearly perfect
245 #define ORIGIN_HASH_SIZE 256
246 #define ORIGIN_MASK 0xFF
247 static struct list_head *_origins;
248 static struct rw_semaphore _origins_lock;
250 static int init_origin_hash(void)
254 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
257 DMERR("unable to allocate memory");
261 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
262 INIT_LIST_HEAD(_origins + i);
263 init_rwsem(&_origins_lock);
268 static void exit_origin_hash(void)
273 static unsigned origin_hash(struct block_device *bdev)
275 return bdev->bd_dev & ORIGIN_MASK;
278 static struct origin *__lookup_origin(struct block_device *origin)
280 struct list_head *ol;
283 ol = &_origins[origin_hash(origin)];
284 list_for_each_entry (o, ol, hash_list)
285 if (bdev_equal(o->bdev, origin))
291 static void __insert_origin(struct origin *o)
293 struct list_head *sl = &_origins[origin_hash(o->bdev)];
294 list_add_tail(&o->hash_list, sl);
298 * Make a note of the snapshot and its origin so we can look it
299 * up when the origin has a write on it.
301 static int register_snapshot(struct dm_snapshot *snap)
303 struct origin *o, *new_o;
304 struct block_device *bdev = snap->origin->bdev;
306 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
310 down_write(&_origins_lock);
311 o = __lookup_origin(bdev);
319 /* Initialise the struct */
320 INIT_LIST_HEAD(&o->snapshots);
326 list_add_tail(&snap->list, &o->snapshots);
328 up_write(&_origins_lock);
332 static void unregister_snapshot(struct dm_snapshot *s)
336 down_write(&_origins_lock);
337 o = __lookup_origin(s->origin->bdev);
340 if (list_empty(&o->snapshots)) {
341 list_del(&o->hash_list);
345 up_write(&_origins_lock);
349 * Implementation of the exception hash tables.
350 * The lowest hash_shift bits of the chunk number are ignored, allowing
351 * some consecutive chunks to be grouped together.
353 static int init_exception_table(struct exception_table *et, uint32_t size,
358 et->hash_shift = hash_shift;
359 et->hash_mask = size - 1;
360 et->table = dm_vcalloc(size, sizeof(struct list_head));
364 for (i = 0; i < size; i++)
365 INIT_LIST_HEAD(et->table + i);
370 static void exit_exception_table(struct exception_table *et, struct kmem_cache *mem)
372 struct list_head *slot;
373 struct dm_snap_exception *ex, *next;
376 size = et->hash_mask + 1;
377 for (i = 0; i < size; i++) {
378 slot = et->table + i;
380 list_for_each_entry_safe (ex, next, slot, hash_list)
381 kmem_cache_free(mem, ex);
387 static uint32_t exception_hash(struct exception_table *et, chunk_t chunk)
389 return (chunk >> et->hash_shift) & et->hash_mask;
392 static void insert_exception(struct exception_table *eh,
393 struct dm_snap_exception *e)
395 struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)];
396 list_add(&e->hash_list, l);
399 static void remove_exception(struct dm_snap_exception *e)
401 list_del(&e->hash_list);
405 * Return the exception data for a sector, or NULL if not
408 static struct dm_snap_exception *lookup_exception(struct exception_table *et,
411 struct list_head *slot;
412 struct dm_snap_exception *e;
414 slot = &et->table[exception_hash(et, chunk)];
415 list_for_each_entry (e, slot, hash_list)
416 if (chunk >= e->old_chunk &&
417 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
423 static struct dm_snap_exception *alloc_exception(void)
425 struct dm_snap_exception *e;
427 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
429 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
434 static void free_exception(struct dm_snap_exception *e)
436 kmem_cache_free(exception_cache, e);
439 static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
441 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
444 atomic_inc(&s->pending_exceptions_count);
450 static void free_pending_exception(struct dm_snap_pending_exception *pe)
452 struct dm_snapshot *s = pe->snap;
454 mempool_free(pe, s->pending_pool);
455 smp_mb__before_atomic_dec();
456 atomic_dec(&s->pending_exceptions_count);
459 static void insert_completed_exception(struct dm_snapshot *s,
460 struct dm_snap_exception *new_e)
462 struct exception_table *eh = &s->complete;
464 struct dm_snap_exception *e = NULL;
466 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
468 /* Add immediately if this table doesn't support consecutive chunks */
472 /* List is ordered by old_chunk */
473 list_for_each_entry_reverse(e, l, hash_list) {
474 /* Insert after an existing chunk? */
475 if (new_e->old_chunk == (e->old_chunk +
476 dm_consecutive_chunk_count(e) + 1) &&
477 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
478 dm_consecutive_chunk_count(e) + 1)) {
479 dm_consecutive_chunk_count_inc(e);
480 free_exception(new_e);
484 /* Insert before an existing chunk? */
485 if (new_e->old_chunk == (e->old_chunk - 1) &&
486 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
487 dm_consecutive_chunk_count_inc(e);
490 free_exception(new_e);
494 if (new_e->old_chunk > e->old_chunk)
499 list_add(&new_e->hash_list, e ? &e->hash_list : l);
503 * Callback used by the exception stores to load exceptions when
506 static int dm_add_exception(void *context, chunk_t old, chunk_t new)
508 struct dm_snapshot *s = context;
509 struct dm_snap_exception *e;
511 e = alloc_exception();
517 /* Consecutive_count is implicitly initialised to zero */
520 insert_completed_exception(s, e);
528 static int calc_max_buckets(void)
530 /* use a fixed size of 2MB */
531 unsigned long mem = 2 * 1024 * 1024;
532 mem /= sizeof(struct list_head);
538 * Allocate room for a suitable hash table.
540 static int init_hash_tables(struct dm_snapshot *s)
542 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
545 * Calculate based on the size of the original volume or
548 cow_dev_size = get_dev_size(s->store->cow->bdev);
549 origin_dev_size = get_dev_size(s->origin->bdev);
550 max_buckets = calc_max_buckets();
552 hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift;
553 hash_size = min(hash_size, max_buckets);
555 hash_size = rounddown_pow_of_two(hash_size);
556 if (init_exception_table(&s->complete, hash_size,
557 DM_CHUNK_CONSECUTIVE_BITS))
561 * Allocate hash table for in-flight exceptions
562 * Make this smaller than the real hash table
568 if (init_exception_table(&s->pending, hash_size, 0)) {
569 exit_exception_table(&s->complete, exception_cache);
577 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
579 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
581 struct dm_snapshot *s;
585 struct dm_exception_store *store;
589 ti->error = "requires exactly 4 arguments";
594 origin_path = argv[0];
598 r = dm_exception_store_create(ti, argc, argv, &args_used, &store);
600 ti->error = "Couldn't create exception store";
608 s = kmalloc(sizeof(*s), GFP_KERNEL);
610 ti->error = "Cannot allocate snapshot context private "
616 r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin);
618 ti->error = "Cannot get origin device";
625 atomic_set(&s->pending_exceptions_count, 0);
626 init_rwsem(&s->lock);
627 spin_lock_init(&s->pe_lock);
629 /* Allocate hash table for COW data */
630 if (init_hash_tables(s)) {
631 ti->error = "Unable to allocate hash table space";
633 goto bad_hash_tables;
636 r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client);
638 ti->error = "Could not create kcopyd client";
642 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
643 if (!s->pending_pool) {
644 ti->error = "Could not allocate mempool for pending exceptions";
645 goto bad_pending_pool;
648 s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS,
649 tracked_chunk_cache);
650 if (!s->tracked_chunk_pool) {
651 ti->error = "Could not allocate tracked_chunk mempool for "
653 goto bad_tracked_chunk_pool;
656 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
657 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
659 spin_lock_init(&s->tracked_chunk_lock);
661 /* Metadata must only be loaded into one table at once */
662 r = s->store->type->read_metadata(s->store, dm_add_exception,
665 ti->error = "Failed to read snapshot metadata";
666 goto bad_load_and_register;
669 DMWARN("Snapshot is marked invalid.");
672 bio_list_init(&s->queued_bios);
673 INIT_WORK(&s->queued_bios_work, flush_queued_bios);
675 /* Add snapshot to the list of snapshots for this origin */
676 /* Exceptions aren't triggered till snapshot_resume() is called */
677 if (register_snapshot(s)) {
679 ti->error = "Cannot register snapshot origin";
680 goto bad_load_and_register;
684 ti->split_io = s->store->chunk_size;
688 bad_load_and_register:
689 mempool_destroy(s->tracked_chunk_pool);
691 bad_tracked_chunk_pool:
692 mempool_destroy(s->pending_pool);
695 dm_kcopyd_client_destroy(s->kcopyd_client);
698 exit_exception_table(&s->pending, pending_cache);
699 exit_exception_table(&s->complete, exception_cache);
702 dm_put_device(ti, s->origin);
708 dm_exception_store_destroy(store);
714 static void __free_exceptions(struct dm_snapshot *s)
716 dm_kcopyd_client_destroy(s->kcopyd_client);
717 s->kcopyd_client = NULL;
719 exit_exception_table(&s->pending, pending_cache);
720 exit_exception_table(&s->complete, exception_cache);
723 static void snapshot_dtr(struct dm_target *ti)
725 #ifdef CONFIG_DM_DEBUG
728 struct dm_snapshot *s = ti->private;
730 flush_workqueue(ksnapd);
732 /* Prevent further origin writes from using this snapshot. */
733 /* After this returns there can be no new kcopyd jobs. */
734 unregister_snapshot(s);
736 while (atomic_read(&s->pending_exceptions_count))
739 * Ensure instructions in mempool_destroy aren't reordered
740 * before atomic_read.
744 #ifdef CONFIG_DM_DEBUG
745 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
746 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
749 mempool_destroy(s->tracked_chunk_pool);
751 __free_exceptions(s);
753 mempool_destroy(s->pending_pool);
755 dm_put_device(ti, s->origin);
757 dm_exception_store_destroy(s->store);
763 * Flush a list of buffers.
765 static void flush_bios(struct bio *bio)
772 generic_make_request(bio);
777 static void flush_queued_bios(struct work_struct *work)
779 struct dm_snapshot *s =
780 container_of(work, struct dm_snapshot, queued_bios_work);
781 struct bio *queued_bios;
784 spin_lock_irqsave(&s->pe_lock, flags);
785 queued_bios = bio_list_get(&s->queued_bios);
786 spin_unlock_irqrestore(&s->pe_lock, flags);
788 flush_bios(queued_bios);
792 * Error a list of buffers.
794 static void error_bios(struct bio *bio)
806 static void __invalidate_snapshot(struct dm_snapshot *s, int err)
812 DMERR("Invalidating snapshot: Error reading/writing.");
813 else if (err == -ENOMEM)
814 DMERR("Invalidating snapshot: Unable to allocate exception.");
816 if (s->store->type->drop_snapshot)
817 s->store->type->drop_snapshot(s->store);
821 dm_table_event(s->store->ti->table);
824 static void get_pending_exception(struct dm_snap_pending_exception *pe)
826 atomic_inc(&pe->ref_count);
829 static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe)
831 struct dm_snap_pending_exception *primary_pe;
832 struct bio *origin_bios = NULL;
834 primary_pe = pe->primary_pe;
837 * If this pe is involved in a write to the origin and
838 * it is the last sibling to complete then release
839 * the bios for the original write to the origin.
842 atomic_dec_and_test(&primary_pe->ref_count)) {
843 origin_bios = bio_list_get(&primary_pe->origin_bios);
844 free_pending_exception(primary_pe);
848 * Free the pe if it's not linked to an origin write or if
849 * it's not itself a primary pe.
851 if (!primary_pe || primary_pe != pe)
852 free_pending_exception(pe);
857 static void pending_complete(struct dm_snap_pending_exception *pe, int success)
859 struct dm_snap_exception *e;
860 struct dm_snapshot *s = pe->snap;
861 struct bio *origin_bios = NULL;
862 struct bio *snapshot_bios = NULL;
866 /* Read/write error - snapshot is unusable */
867 down_write(&s->lock);
868 __invalidate_snapshot(s, -EIO);
873 e = alloc_exception();
875 down_write(&s->lock);
876 __invalidate_snapshot(s, -ENOMEM);
882 down_write(&s->lock);
890 * Check for conflicting reads. This is extremely improbable,
891 * so msleep(1) is sufficient and there is no need for a wait queue.
893 while (__chunk_is_tracked(s, pe->e.old_chunk))
897 * Add a proper exception, and remove the
898 * in-flight exception from the list.
900 insert_completed_exception(s, e);
903 remove_exception(&pe->e);
904 snapshot_bios = bio_list_get(&pe->snapshot_bios);
905 origin_bios = put_pending_exception(pe);
909 /* Submit any pending write bios */
911 error_bios(snapshot_bios);
913 flush_bios(snapshot_bios);
915 flush_bios(origin_bios);
918 static void commit_callback(void *context, int success)
920 struct dm_snap_pending_exception *pe = context;
922 pending_complete(pe, success);
926 * Called when the copy I/O has finished. kcopyd actually runs
927 * this code so don't block.
929 static void copy_callback(int read_err, unsigned long write_err, void *context)
931 struct dm_snap_pending_exception *pe = context;
932 struct dm_snapshot *s = pe->snap;
934 if (read_err || write_err)
935 pending_complete(pe, 0);
938 /* Update the metadata if we are persistent */
939 s->store->type->commit_exception(s->store, &pe->e,
940 commit_callback, pe);
944 * Dispatches the copy operation to kcopyd.
946 static void start_copy(struct dm_snap_pending_exception *pe)
948 struct dm_snapshot *s = pe->snap;
949 struct dm_io_region src, dest;
950 struct block_device *bdev = s->origin->bdev;
953 dev_size = get_dev_size(bdev);
956 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
957 src.count = min(s->store->chunk_size, dev_size - src.sector);
959 dest.bdev = s->store->cow->bdev;
960 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
961 dest.count = src.count;
963 /* Hand over to kcopyd */
964 dm_kcopyd_copy(s->kcopyd_client,
965 &src, 1, &dest, 0, copy_callback, pe);
968 static struct dm_snap_pending_exception *
969 __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
971 struct dm_snap_exception *e = lookup_exception(&s->pending, chunk);
976 return container_of(e, struct dm_snap_pending_exception, e);
980 * Looks to see if this snapshot already has a pending exception
981 * for this chunk, otherwise it allocates a new one and inserts
982 * it into the pending table.
984 * NOTE: a write lock must be held on snap->lock before calling
987 static struct dm_snap_pending_exception *
988 __find_pending_exception(struct dm_snapshot *s,
989 struct dm_snap_pending_exception *pe, chunk_t chunk)
991 struct dm_snap_pending_exception *pe2;
993 pe2 = __lookup_pending_exception(s, chunk);
995 free_pending_exception(pe);
999 pe->e.old_chunk = chunk;
1000 bio_list_init(&pe->origin_bios);
1001 bio_list_init(&pe->snapshot_bios);
1002 pe->primary_pe = NULL;
1003 atomic_set(&pe->ref_count, 0);
1006 if (s->store->type->prepare_exception(s->store, &pe->e)) {
1007 free_pending_exception(pe);
1011 get_pending_exception(pe);
1012 insert_exception(&s->pending, &pe->e);
1017 static void remap_exception(struct dm_snapshot *s, struct dm_snap_exception *e,
1018 struct bio *bio, chunk_t chunk)
1020 bio->bi_bdev = s->store->cow->bdev;
1021 bio->bi_sector = chunk_to_sector(s->store,
1022 dm_chunk_number(e->new_chunk) +
1023 (chunk - e->old_chunk)) +
1025 s->store->chunk_mask);
1028 static int snapshot_map(struct dm_target *ti, struct bio *bio,
1029 union map_info *map_context)
1031 struct dm_snap_exception *e;
1032 struct dm_snapshot *s = ti->private;
1033 int r = DM_MAPIO_REMAPPED;
1035 struct dm_snap_pending_exception *pe = NULL;
1037 chunk = sector_to_chunk(s->store, bio->bi_sector);
1039 /* Full snapshots are not usable */
1040 /* To get here the table must be live so s->active is always set. */
1044 /* FIXME: should only take write lock if we need
1045 * to copy an exception */
1046 down_write(&s->lock);
1053 /* If the block is already remapped - use that, else remap it */
1054 e = lookup_exception(&s->complete, chunk);
1056 remap_exception(s, e, bio, chunk);
1061 * Write to snapshot - higher level takes care of RW/RO
1062 * flags so we should only get this if we are
1065 if (bio_rw(bio) == WRITE) {
1066 pe = __lookup_pending_exception(s, chunk);
1069 pe = alloc_pending_exception(s);
1070 down_write(&s->lock);
1073 free_pending_exception(pe);
1078 e = lookup_exception(&s->complete, chunk);
1080 free_pending_exception(pe);
1081 remap_exception(s, e, bio, chunk);
1085 pe = __find_pending_exception(s, pe, chunk);
1087 __invalidate_snapshot(s, -ENOMEM);
1093 remap_exception(s, &pe->e, bio, chunk);
1094 bio_list_add(&pe->snapshot_bios, bio);
1096 r = DM_MAPIO_SUBMITTED;
1099 /* this is protected by snap->lock */
1106 bio->bi_bdev = s->origin->bdev;
1107 map_context->ptr = track_chunk(s, chunk);
1116 static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1117 int error, union map_info *map_context)
1119 struct dm_snapshot *s = ti->private;
1120 struct dm_snap_tracked_chunk *c = map_context->ptr;
1123 stop_tracking_chunk(s, c);
1128 static void snapshot_resume(struct dm_target *ti)
1130 struct dm_snapshot *s = ti->private;
1132 down_write(&s->lock);
1137 static int snapshot_status(struct dm_target *ti, status_type_t type,
1138 char *result, unsigned int maxlen)
1141 struct dm_snapshot *snap = ti->private;
1144 case STATUSTYPE_INFO:
1148 if (snap->store->type->fraction_full) {
1149 sector_t numerator, denominator;
1150 snap->store->type->fraction_full(snap->store,
1154 (unsigned long long)numerator,
1155 (unsigned long long)denominator);
1162 case STATUSTYPE_TABLE:
1164 * kdevname returns a static pointer so we need
1165 * to make private copies if the output is to
1168 DMEMIT("%s", snap->origin->name);
1169 DMEMIT(" %s %s %llu", snap->store->cow->name,
1170 snap->store->type->name,
1171 (unsigned long long)snap->store->chunk_size);
1178 /*-----------------------------------------------------------------
1180 *---------------------------------------------------------------*/
1181 static int __origin_write(struct list_head *snapshots, struct bio *bio)
1183 int r = DM_MAPIO_REMAPPED, first = 0;
1184 struct dm_snapshot *snap;
1185 struct dm_snap_exception *e;
1186 struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL;
1188 LIST_HEAD(pe_queue);
1190 /* Do all the snapshots on this origin */
1191 list_for_each_entry (snap, snapshots, list) {
1193 down_write(&snap->lock);
1195 /* Only deal with valid and active snapshots */
1196 if (!snap->valid || !snap->active)
1199 /* Nothing to do if writing beyond end of snapshot */
1200 if (bio->bi_sector >= dm_table_get_size(snap->store->ti->table))
1204 * Remember, different snapshots can have
1205 * different chunk sizes.
1207 chunk = sector_to_chunk(snap->store, bio->bi_sector);
1210 * Check exception table to see if block
1211 * is already remapped in this snapshot
1212 * and trigger an exception if not.
1214 * ref_count is initialised to 1 so pending_complete()
1215 * won't destroy the primary_pe while we're inside this loop.
1217 e = lookup_exception(&snap->complete, chunk);
1221 pe = __lookup_pending_exception(snap, chunk);
1223 up_write(&snap->lock);
1224 pe = alloc_pending_exception(snap);
1225 down_write(&snap->lock);
1228 free_pending_exception(pe);
1232 e = lookup_exception(&snap->complete, chunk);
1234 free_pending_exception(pe);
1238 pe = __find_pending_exception(snap, pe, chunk);
1240 __invalidate_snapshot(snap, -ENOMEM);
1247 * Either every pe here has same
1248 * primary_pe or none has one yet.
1251 primary_pe = pe->primary_pe;
1257 bio_list_add(&primary_pe->origin_bios, bio);
1259 r = DM_MAPIO_SUBMITTED;
1262 if (!pe->primary_pe) {
1263 pe->primary_pe = primary_pe;
1264 get_pending_exception(primary_pe);
1269 list_add_tail(&pe->list, &pe_queue);
1273 up_write(&snap->lock);
1280 * If this is the first time we're processing this chunk and
1281 * ref_count is now 1 it means all the pending exceptions
1282 * got completed while we were in the loop above, so it falls to
1283 * us here to remove the primary_pe and submit any origin_bios.
1286 if (first && atomic_dec_and_test(&primary_pe->ref_count)) {
1287 flush_bios(bio_list_get(&primary_pe->origin_bios));
1288 free_pending_exception(primary_pe);
1289 /* If we got here, pe_queue is necessarily empty. */
1294 * Now that we have a complete pe list we can start the copying.
1296 list_for_each_entry_safe(pe, next_pe, &pe_queue, list)
1303 * Called on a write from the origin driver.
1305 static int do_origin(struct dm_dev *origin, struct bio *bio)
1308 int r = DM_MAPIO_REMAPPED;
1310 down_read(&_origins_lock);
1311 o = __lookup_origin(origin->bdev);
1313 r = __origin_write(&o->snapshots, bio);
1314 up_read(&_origins_lock);
1320 * Origin: maps a linear range of a device, with hooks for snapshotting.
1324 * Construct an origin mapping: <dev_path>
1325 * The context for an origin is merely a 'struct dm_dev *'
1326 * pointing to the real device.
1328 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1334 ti->error = "origin: incorrect number of arguments";
1338 r = dm_get_device(ti, argv[0], 0, ti->len,
1339 dm_table_get_mode(ti->table), &dev);
1341 ti->error = "Cannot get target device";
1349 static void origin_dtr(struct dm_target *ti)
1351 struct dm_dev *dev = ti->private;
1352 dm_put_device(ti, dev);
1355 static int origin_map(struct dm_target *ti, struct bio *bio,
1356 union map_info *map_context)
1358 struct dm_dev *dev = ti->private;
1359 bio->bi_bdev = dev->bdev;
1361 /* Only tell snapshots if this is a write */
1362 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
1365 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
1368 * Set the target "split_io" field to the minimum of all the snapshots'
1371 static void origin_resume(struct dm_target *ti)
1373 struct dm_dev *dev = ti->private;
1374 struct dm_snapshot *snap;
1376 chunk_t chunk_size = 0;
1378 down_read(&_origins_lock);
1379 o = __lookup_origin(dev->bdev);
1381 list_for_each_entry (snap, &o->snapshots, list)
1382 chunk_size = min_not_zero(chunk_size,
1383 snap->store->chunk_size);
1384 up_read(&_origins_lock);
1386 ti->split_io = chunk_size;
1389 static int origin_status(struct dm_target *ti, status_type_t type, char *result,
1390 unsigned int maxlen)
1392 struct dm_dev *dev = ti->private;
1395 case STATUSTYPE_INFO:
1399 case STATUSTYPE_TABLE:
1400 snprintf(result, maxlen, "%s", dev->name);
1407 static struct target_type origin_target = {
1408 .name = "snapshot-origin",
1409 .version = {1, 6, 0},
1410 .module = THIS_MODULE,
1414 .resume = origin_resume,
1415 .status = origin_status,
1418 static struct target_type snapshot_target = {
1420 .version = {1, 6, 0},
1421 .module = THIS_MODULE,
1422 .ctr = snapshot_ctr,
1423 .dtr = snapshot_dtr,
1424 .map = snapshot_map,
1425 .end_io = snapshot_end_io,
1426 .resume = snapshot_resume,
1427 .status = snapshot_status,
1430 static int __init dm_snapshot_init(void)
1434 r = dm_exception_store_init();
1436 DMERR("Failed to initialize exception stores");
1440 r = dm_register_target(&snapshot_target);
1442 DMERR("snapshot target register failed %d", r);
1446 r = dm_register_target(&origin_target);
1448 DMERR("Origin target register failed %d", r);
1452 r = init_origin_hash();
1454 DMERR("init_origin_hash failed.");
1458 exception_cache = KMEM_CACHE(dm_snap_exception, 0);
1459 if (!exception_cache) {
1460 DMERR("Couldn't create exception cache.");
1465 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
1466 if (!pending_cache) {
1467 DMERR("Couldn't create pending cache.");
1472 tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0);
1473 if (!tracked_chunk_cache) {
1474 DMERR("Couldn't create cache to track chunks in use.");
1479 ksnapd = create_singlethread_workqueue("ksnapd");
1481 DMERR("Failed to create ksnapd workqueue.");
1483 goto bad_pending_pool;
1489 kmem_cache_destroy(tracked_chunk_cache);
1491 kmem_cache_destroy(pending_cache);
1493 kmem_cache_destroy(exception_cache);
1497 dm_unregister_target(&origin_target);
1499 dm_unregister_target(&snapshot_target);
1503 static void __exit dm_snapshot_exit(void)
1505 destroy_workqueue(ksnapd);
1507 dm_unregister_target(&snapshot_target);
1508 dm_unregister_target(&origin_target);
1511 kmem_cache_destroy(pending_cache);
1512 kmem_cache_destroy(exception_cache);
1513 kmem_cache_destroy(tracked_chunk_cache);
1515 dm_exception_store_exit();
1519 module_init(dm_snapshot_init);
1520 module_exit(dm_snapshot_exit);
1522 MODULE_DESCRIPTION(DM_NAME " snapshot target");
1523 MODULE_AUTHOR("Joe Thornber");
1524 MODULE_LICENSE("GPL");