]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - drivers/md/dm-snap.c
b7e179cdc5af906aed485c7f32b0a65e68533032
[~shefty/rdma-dev.git] / drivers / md / dm-snap.c
1 /*
2  * dm-snapshot.c
3  *
4  * Copyright (C) 2001-2002 Sistina Software (UK) Limited.
5  *
6  * This file is released under the GPL.
7  */
8
9 #include <linux/blkdev.h>
10 #include <linux/device-mapper.h>
11 #include <linux/delay.h>
12 #include <linux/fs.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
23 #include "dm-exception-store.h"
24
25 #define DM_MSG_PREFIX "snapshots"
26
27 static const char dm_snapshot_merge_target_name[] = "snapshot-merge";
28
29 #define dm_target_is_snapshot_merge(ti) \
30         ((ti)->type->name == dm_snapshot_merge_target_name)
31
32 /*
33  * The size of the mempool used to track chunks in use.
34  */
35 #define MIN_IOS 256
36
37 #define DM_TRACKED_CHUNK_HASH_SIZE      16
38 #define DM_TRACKED_CHUNK_HASH(x)        ((unsigned long)(x) & \
39                                          (DM_TRACKED_CHUNK_HASH_SIZE - 1))
40
41 struct dm_exception_table {
42         uint32_t hash_mask;
43         unsigned hash_shift;
44         struct list_head *table;
45 };
46
47 struct dm_snapshot {
48         struct rw_semaphore lock;
49
50         struct dm_dev *origin;
51         struct dm_dev *cow;
52
53         struct dm_target *ti;
54
55         /* List of snapshots per Origin */
56         struct list_head list;
57
58         /*
59          * You can't use a snapshot if this is 0 (e.g. if full).
60          * A snapshot-merge target never clears this.
61          */
62         int valid;
63
64         /* Origin writes don't trigger exceptions until this is set */
65         int active;
66
67         atomic_t pending_exceptions_count;
68
69         mempool_t *pending_pool;
70
71         struct dm_exception_table pending;
72         struct dm_exception_table complete;
73
74         /*
75          * pe_lock protects all pending_exception operations and access
76          * as well as the snapshot_bios list.
77          */
78         spinlock_t pe_lock;
79
80         /* Chunks with outstanding reads */
81         spinlock_t tracked_chunk_lock;
82         struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
83
84         /* The on disk metadata handler */
85         struct dm_exception_store *store;
86
87         struct dm_kcopyd_client *kcopyd_client;
88
89         /* Wait for events based on state_bits */
90         unsigned long state_bits;
91
92         /* Range of chunks currently being merged. */
93         chunk_t first_merging_chunk;
94         int num_merging_chunks;
95
96         /*
97          * The merge operation failed if this flag is set.
98          * Failure modes are handled as follows:
99          * - I/O error reading the header
100          *      => don't load the target; abort.
101          * - Header does not have "valid" flag set
102          *      => use the origin; forget about the snapshot.
103          * - I/O error when reading exceptions
104          *      => don't load the target; abort.
105          *         (We can't use the intermediate origin state.)
106          * - I/O error while merging
107          *      => stop merging; set merge_failed; process I/O normally.
108          */
109         int merge_failed;
110
111         /*
112          * Incoming bios that overlap with chunks being merged must wait
113          * for them to be committed.
114          */
115         struct bio_list bios_queued_during_merge;
116 };
117
118 /*
119  * state_bits:
120  *   RUNNING_MERGE  - Merge operation is in progress.
121  *   SHUTDOWN_MERGE - Set to signal that merge needs to be stopped;
122  *                    cleared afterwards.
123  */
124 #define RUNNING_MERGE          0
125 #define SHUTDOWN_MERGE         1
126
127 struct dm_dev *dm_snap_origin(struct dm_snapshot *s)
128 {
129         return s->origin;
130 }
131 EXPORT_SYMBOL(dm_snap_origin);
132
133 struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
134 {
135         return s->cow;
136 }
137 EXPORT_SYMBOL(dm_snap_cow);
138
139 static sector_t chunk_to_sector(struct dm_exception_store *store,
140                                 chunk_t chunk)
141 {
142         return chunk << store->chunk_shift;
143 }
144
145 static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
146 {
147         /*
148          * There is only ever one instance of a particular block
149          * device so we can compare pointers safely.
150          */
151         return lhs == rhs;
152 }
153
154 struct dm_snap_pending_exception {
155         struct dm_exception e;
156
157         /*
158          * Origin buffers waiting for this to complete are held
159          * in a bio list
160          */
161         struct bio_list origin_bios;
162         struct bio_list snapshot_bios;
163
164         /* Pointer back to snapshot context */
165         struct dm_snapshot *snap;
166
167         /*
168          * 1 indicates the exception has already been sent to
169          * kcopyd.
170          */
171         int started;
172
173         /*
174          * For writing a complete chunk, bypassing the copy.
175          */
176         struct bio *full_bio;
177         bio_end_io_t *full_bio_end_io;
178         void *full_bio_private;
179 };
180
181 /*
182  * Hash table mapping origin volumes to lists of snapshots and
183  * a lock to protect it
184  */
185 static struct kmem_cache *exception_cache;
186 static struct kmem_cache *pending_cache;
187
188 struct dm_snap_tracked_chunk {
189         struct hlist_node node;
190         chunk_t chunk;
191 };
192
193 static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s,
194                                                  struct bio *bio,
195                                                  chunk_t chunk)
196 {
197         struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
198
199         c->chunk = chunk;
200
201         spin_lock_irq(&s->tracked_chunk_lock);
202         hlist_add_head(&c->node,
203                        &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
204         spin_unlock_irq(&s->tracked_chunk_lock);
205
206         return c;
207 }
208
209 static void stop_tracking_chunk(struct dm_snapshot *s,
210                                 struct dm_snap_tracked_chunk *c)
211 {
212         unsigned long flags;
213
214         spin_lock_irqsave(&s->tracked_chunk_lock, flags);
215         hlist_del(&c->node);
216         spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
217 }
218
219 static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
220 {
221         struct dm_snap_tracked_chunk *c;
222         struct hlist_node *hn;
223         int found = 0;
224
225         spin_lock_irq(&s->tracked_chunk_lock);
226
227         hlist_for_each_entry(c, hn,
228             &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
229                 if (c->chunk == chunk) {
230                         found = 1;
231                         break;
232                 }
233         }
234
235         spin_unlock_irq(&s->tracked_chunk_lock);
236
237         return found;
238 }
239
240 /*
241  * This conflicting I/O is extremely improbable in the caller,
242  * so msleep(1) is sufficient and there is no need for a wait queue.
243  */
244 static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
245 {
246         while (__chunk_is_tracked(s, chunk))
247                 msleep(1);
248 }
249
250 /*
251  * One of these per registered origin, held in the snapshot_origins hash
252  */
253 struct origin {
254         /* The origin device */
255         struct block_device *bdev;
256
257         struct list_head hash_list;
258
259         /* List of snapshots for this origin */
260         struct list_head snapshots;
261 };
262
263 /*
264  * Size of the hash table for origin volumes. If we make this
265  * the size of the minors list then it should be nearly perfect
266  */
267 #define ORIGIN_HASH_SIZE 256
268 #define ORIGIN_MASK      0xFF
269 static struct list_head *_origins;
270 static struct rw_semaphore _origins_lock;
271
272 static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
273 static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
274 static uint64_t _pending_exceptions_done_count;
275
276 static int init_origin_hash(void)
277 {
278         int i;
279
280         _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
281                            GFP_KERNEL);
282         if (!_origins) {
283                 DMERR("unable to allocate memory");
284                 return -ENOMEM;
285         }
286
287         for (i = 0; i < ORIGIN_HASH_SIZE; i++)
288                 INIT_LIST_HEAD(_origins + i);
289         init_rwsem(&_origins_lock);
290
291         return 0;
292 }
293
294 static void exit_origin_hash(void)
295 {
296         kfree(_origins);
297 }
298
299 static unsigned origin_hash(struct block_device *bdev)
300 {
301         return bdev->bd_dev & ORIGIN_MASK;
302 }
303
304 static struct origin *__lookup_origin(struct block_device *origin)
305 {
306         struct list_head *ol;
307         struct origin *o;
308
309         ol = &_origins[origin_hash(origin)];
310         list_for_each_entry (o, ol, hash_list)
311                 if (bdev_equal(o->bdev, origin))
312                         return o;
313
314         return NULL;
315 }
316
317 static void __insert_origin(struct origin *o)
318 {
319         struct list_head *sl = &_origins[origin_hash(o->bdev)];
320         list_add_tail(&o->hash_list, sl);
321 }
322
323 /*
324  * _origins_lock must be held when calling this function.
325  * Returns number of snapshots registered using the supplied cow device, plus:
326  * snap_src - a snapshot suitable for use as a source of exception handover
327  * snap_dest - a snapshot capable of receiving exception handover.
328  * snap_merge - an existing snapshot-merge target linked to the same origin.
329  *   There can be at most one snapshot-merge target. The parameter is optional.
330  *
331  * Possible return values and states of snap_src and snap_dest.
332  *   0: NULL, NULL  - first new snapshot
333  *   1: snap_src, NULL - normal snapshot
334  *   2: snap_src, snap_dest  - waiting for handover
335  *   2: snap_src, NULL - handed over, waiting for old to be deleted
336  *   1: NULL, snap_dest - source got destroyed without handover
337  */
338 static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
339                                         struct dm_snapshot **snap_src,
340                                         struct dm_snapshot **snap_dest,
341                                         struct dm_snapshot **snap_merge)
342 {
343         struct dm_snapshot *s;
344         struct origin *o;
345         int count = 0;
346         int active;
347
348         o = __lookup_origin(snap->origin->bdev);
349         if (!o)
350                 goto out;
351
352         list_for_each_entry(s, &o->snapshots, list) {
353                 if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
354                         *snap_merge = s;
355                 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
356                         continue;
357
358                 down_read(&s->lock);
359                 active = s->active;
360                 up_read(&s->lock);
361
362                 if (active) {
363                         if (snap_src)
364                                 *snap_src = s;
365                 } else if (snap_dest)
366                         *snap_dest = s;
367
368                 count++;
369         }
370
371 out:
372         return count;
373 }
374
375 /*
376  * On success, returns 1 if this snapshot is a handover destination,
377  * otherwise returns 0.
378  */
379 static int __validate_exception_handover(struct dm_snapshot *snap)
380 {
381         struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
382         struct dm_snapshot *snap_merge = NULL;
383
384         /* Does snapshot need exceptions handed over to it? */
385         if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
386                                           &snap_merge) == 2) ||
387             snap_dest) {
388                 snap->ti->error = "Snapshot cow pairing for exception "
389                                   "table handover failed";
390                 return -EINVAL;
391         }
392
393         /*
394          * If no snap_src was found, snap cannot become a handover
395          * destination.
396          */
397         if (!snap_src)
398                 return 0;
399
400         /*
401          * Non-snapshot-merge handover?
402          */
403         if (!dm_target_is_snapshot_merge(snap->ti))
404                 return 1;
405
406         /*
407          * Do not allow more than one merging snapshot.
408          */
409         if (snap_merge) {
410                 snap->ti->error = "A snapshot is already merging.";
411                 return -EINVAL;
412         }
413
414         if (!snap_src->store->type->prepare_merge ||
415             !snap_src->store->type->commit_merge) {
416                 snap->ti->error = "Snapshot exception store does not "
417                                   "support snapshot-merge.";
418                 return -EINVAL;
419         }
420
421         return 1;
422 }
423
424 static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
425 {
426         struct dm_snapshot *l;
427
428         /* Sort the list according to chunk size, largest-first smallest-last */
429         list_for_each_entry(l, &o->snapshots, list)
430                 if (l->store->chunk_size < s->store->chunk_size)
431                         break;
432         list_add_tail(&s->list, &l->list);
433 }
434
435 /*
436  * Make a note of the snapshot and its origin so we can look it
437  * up when the origin has a write on it.
438  *
439  * Also validate snapshot exception store handovers.
440  * On success, returns 1 if this registration is a handover destination,
441  * otherwise returns 0.
442  */
443 static int register_snapshot(struct dm_snapshot *snap)
444 {
445         struct origin *o, *new_o = NULL;
446         struct block_device *bdev = snap->origin->bdev;
447         int r = 0;
448
449         new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
450         if (!new_o)
451                 return -ENOMEM;
452
453         down_write(&_origins_lock);
454
455         r = __validate_exception_handover(snap);
456         if (r < 0) {
457                 kfree(new_o);
458                 goto out;
459         }
460
461         o = __lookup_origin(bdev);
462         if (o)
463                 kfree(new_o);
464         else {
465                 /* New origin */
466                 o = new_o;
467
468                 /* Initialise the struct */
469                 INIT_LIST_HEAD(&o->snapshots);
470                 o->bdev = bdev;
471
472                 __insert_origin(o);
473         }
474
475         __insert_snapshot(o, snap);
476
477 out:
478         up_write(&_origins_lock);
479
480         return r;
481 }
482
483 /*
484  * Move snapshot to correct place in list according to chunk size.
485  */
486 static void reregister_snapshot(struct dm_snapshot *s)
487 {
488         struct block_device *bdev = s->origin->bdev;
489
490         down_write(&_origins_lock);
491
492         list_del(&s->list);
493         __insert_snapshot(__lookup_origin(bdev), s);
494
495         up_write(&_origins_lock);
496 }
497
498 static void unregister_snapshot(struct dm_snapshot *s)
499 {
500         struct origin *o;
501
502         down_write(&_origins_lock);
503         o = __lookup_origin(s->origin->bdev);
504
505         list_del(&s->list);
506         if (o && list_empty(&o->snapshots)) {
507                 list_del(&o->hash_list);
508                 kfree(o);
509         }
510
511         up_write(&_origins_lock);
512 }
513
514 /*
515  * Implementation of the exception hash tables.
516  * The lowest hash_shift bits of the chunk number are ignored, allowing
517  * some consecutive chunks to be grouped together.
518  */
519 static int dm_exception_table_init(struct dm_exception_table *et,
520                                    uint32_t size, unsigned hash_shift)
521 {
522         unsigned int i;
523
524         et->hash_shift = hash_shift;
525         et->hash_mask = size - 1;
526         et->table = dm_vcalloc(size, sizeof(struct list_head));
527         if (!et->table)
528                 return -ENOMEM;
529
530         for (i = 0; i < size; i++)
531                 INIT_LIST_HEAD(et->table + i);
532
533         return 0;
534 }
535
536 static void dm_exception_table_exit(struct dm_exception_table *et,
537                                     struct kmem_cache *mem)
538 {
539         struct list_head *slot;
540         struct dm_exception *ex, *next;
541         int i, size;
542
543         size = et->hash_mask + 1;
544         for (i = 0; i < size; i++) {
545                 slot = et->table + i;
546
547                 list_for_each_entry_safe (ex, next, slot, hash_list)
548                         kmem_cache_free(mem, ex);
549         }
550
551         vfree(et->table);
552 }
553
554 static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
555 {
556         return (chunk >> et->hash_shift) & et->hash_mask;
557 }
558
559 static void dm_remove_exception(struct dm_exception *e)
560 {
561         list_del(&e->hash_list);
562 }
563
564 /*
565  * Return the exception data for a sector, or NULL if not
566  * remapped.
567  */
568 static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
569                                                 chunk_t chunk)
570 {
571         struct list_head *slot;
572         struct dm_exception *e;
573
574         slot = &et->table[exception_hash(et, chunk)];
575         list_for_each_entry (e, slot, hash_list)
576                 if (chunk >= e->old_chunk &&
577                     chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
578                         return e;
579
580         return NULL;
581 }
582
583 static struct dm_exception *alloc_completed_exception(void)
584 {
585         struct dm_exception *e;
586
587         e = kmem_cache_alloc(exception_cache, GFP_NOIO);
588         if (!e)
589                 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
590
591         return e;
592 }
593
594 static void free_completed_exception(struct dm_exception *e)
595 {
596         kmem_cache_free(exception_cache, e);
597 }
598
599 static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
600 {
601         struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
602                                                              GFP_NOIO);
603
604         atomic_inc(&s->pending_exceptions_count);
605         pe->snap = s;
606
607         return pe;
608 }
609
610 static void free_pending_exception(struct dm_snap_pending_exception *pe)
611 {
612         struct dm_snapshot *s = pe->snap;
613
614         mempool_free(pe, s->pending_pool);
615         smp_mb__before_atomic_dec();
616         atomic_dec(&s->pending_exceptions_count);
617 }
618
619 static void dm_insert_exception(struct dm_exception_table *eh,
620                                 struct dm_exception *new_e)
621 {
622         struct list_head *l;
623         struct dm_exception *e = NULL;
624
625         l = &eh->table[exception_hash(eh, new_e->old_chunk)];
626
627         /* Add immediately if this table doesn't support consecutive chunks */
628         if (!eh->hash_shift)
629                 goto out;
630
631         /* List is ordered by old_chunk */
632         list_for_each_entry_reverse(e, l, hash_list) {
633                 /* Insert after an existing chunk? */
634                 if (new_e->old_chunk == (e->old_chunk +
635                                          dm_consecutive_chunk_count(e) + 1) &&
636                     new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
637                                          dm_consecutive_chunk_count(e) + 1)) {
638                         dm_consecutive_chunk_count_inc(e);
639                         free_completed_exception(new_e);
640                         return;
641                 }
642
643                 /* Insert before an existing chunk? */
644                 if (new_e->old_chunk == (e->old_chunk - 1) &&
645                     new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
646                         dm_consecutive_chunk_count_inc(e);
647                         e->old_chunk--;
648                         e->new_chunk--;
649                         free_completed_exception(new_e);
650                         return;
651                 }
652
653                 if (new_e->old_chunk > e->old_chunk)
654                         break;
655         }
656
657 out:
658         list_add(&new_e->hash_list, e ? &e->hash_list : l);
659 }
660
661 /*
662  * Callback used by the exception stores to load exceptions when
663  * initialising.
664  */
665 static int dm_add_exception(void *context, chunk_t old, chunk_t new)
666 {
667         struct dm_snapshot *s = context;
668         struct dm_exception *e;
669
670         e = alloc_completed_exception();
671         if (!e)
672                 return -ENOMEM;
673
674         e->old_chunk = old;
675
676         /* Consecutive_count is implicitly initialised to zero */
677         e->new_chunk = new;
678
679         dm_insert_exception(&s->complete, e);
680
681         return 0;
682 }
683
684 /*
685  * Return a minimum chunk size of all snapshots that have the specified origin.
686  * Return zero if the origin has no snapshots.
687  */
688 static uint32_t __minimum_chunk_size(struct origin *o)
689 {
690         struct dm_snapshot *snap;
691         unsigned chunk_size = 0;
692
693         if (o)
694                 list_for_each_entry(snap, &o->snapshots, list)
695                         chunk_size = min_not_zero(chunk_size,
696                                                   snap->store->chunk_size);
697
698         return (uint32_t) chunk_size;
699 }
700
701 /*
702  * Hard coded magic.
703  */
704 static int calc_max_buckets(void)
705 {
706         /* use a fixed size of 2MB */
707         unsigned long mem = 2 * 1024 * 1024;
708         mem /= sizeof(struct list_head);
709
710         return mem;
711 }
712
713 /*
714  * Allocate room for a suitable hash table.
715  */
716 static int init_hash_tables(struct dm_snapshot *s)
717 {
718         sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
719
720         /*
721          * Calculate based on the size of the original volume or
722          * the COW volume...
723          */
724         cow_dev_size = get_dev_size(s->cow->bdev);
725         origin_dev_size = get_dev_size(s->origin->bdev);
726         max_buckets = calc_max_buckets();
727
728         hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift;
729         hash_size = min(hash_size, max_buckets);
730
731         if (hash_size < 64)
732                 hash_size = 64;
733         hash_size = rounddown_pow_of_two(hash_size);
734         if (dm_exception_table_init(&s->complete, hash_size,
735                                     DM_CHUNK_CONSECUTIVE_BITS))
736                 return -ENOMEM;
737
738         /*
739          * Allocate hash table for in-flight exceptions
740          * Make this smaller than the real hash table
741          */
742         hash_size >>= 3;
743         if (hash_size < 64)
744                 hash_size = 64;
745
746         if (dm_exception_table_init(&s->pending, hash_size, 0)) {
747                 dm_exception_table_exit(&s->complete, exception_cache);
748                 return -ENOMEM;
749         }
750
751         return 0;
752 }
753
754 static void merge_shutdown(struct dm_snapshot *s)
755 {
756         clear_bit_unlock(RUNNING_MERGE, &s->state_bits);
757         smp_mb__after_clear_bit();
758         wake_up_bit(&s->state_bits, RUNNING_MERGE);
759 }
760
761 static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s)
762 {
763         s->first_merging_chunk = 0;
764         s->num_merging_chunks = 0;
765
766         return bio_list_get(&s->bios_queued_during_merge);
767 }
768
769 /*
770  * Remove one chunk from the index of completed exceptions.
771  */
772 static int __remove_single_exception_chunk(struct dm_snapshot *s,
773                                            chunk_t old_chunk)
774 {
775         struct dm_exception *e;
776
777         e = dm_lookup_exception(&s->complete, old_chunk);
778         if (!e) {
779                 DMERR("Corruption detected: exception for block %llu is "
780                       "on disk but not in memory",
781                       (unsigned long long)old_chunk);
782                 return -EINVAL;
783         }
784
785         /*
786          * If this is the only chunk using this exception, remove exception.
787          */
788         if (!dm_consecutive_chunk_count(e)) {
789                 dm_remove_exception(e);
790                 free_completed_exception(e);
791                 return 0;
792         }
793
794         /*
795          * The chunk may be either at the beginning or the end of a
796          * group of consecutive chunks - never in the middle.  We are
797          * removing chunks in the opposite order to that in which they
798          * were added, so this should always be true.
799          * Decrement the consecutive chunk counter and adjust the
800          * starting point if necessary.
801          */
802         if (old_chunk == e->old_chunk) {
803                 e->old_chunk++;
804                 e->new_chunk++;
805         } else if (old_chunk != e->old_chunk +
806                    dm_consecutive_chunk_count(e)) {
807                 DMERR("Attempt to merge block %llu from the "
808                       "middle of a chunk range [%llu - %llu]",
809                       (unsigned long long)old_chunk,
810                       (unsigned long long)e->old_chunk,
811                       (unsigned long long)
812                       e->old_chunk + dm_consecutive_chunk_count(e));
813                 return -EINVAL;
814         }
815
816         dm_consecutive_chunk_count_dec(e);
817
818         return 0;
819 }
820
821 static void flush_bios(struct bio *bio);
822
823 static int remove_single_exception_chunk(struct dm_snapshot *s)
824 {
825         struct bio *b = NULL;
826         int r;
827         chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1;
828
829         down_write(&s->lock);
830
831         /*
832          * Process chunks (and associated exceptions) in reverse order
833          * so that dm_consecutive_chunk_count_dec() accounting works.
834          */
835         do {
836                 r = __remove_single_exception_chunk(s, old_chunk);
837                 if (r)
838                         goto out;
839         } while (old_chunk-- > s->first_merging_chunk);
840
841         b = __release_queued_bios_after_merge(s);
842
843 out:
844         up_write(&s->lock);
845         if (b)
846                 flush_bios(b);
847
848         return r;
849 }
850
851 static int origin_write_extent(struct dm_snapshot *merging_snap,
852                                sector_t sector, unsigned chunk_size);
853
854 static void merge_callback(int read_err, unsigned long write_err,
855                            void *context);
856
857 static uint64_t read_pending_exceptions_done_count(void)
858 {
859         uint64_t pending_exceptions_done;
860
861         spin_lock(&_pending_exceptions_done_spinlock);
862         pending_exceptions_done = _pending_exceptions_done_count;
863         spin_unlock(&_pending_exceptions_done_spinlock);
864
865         return pending_exceptions_done;
866 }
867
868 static void increment_pending_exceptions_done_count(void)
869 {
870         spin_lock(&_pending_exceptions_done_spinlock);
871         _pending_exceptions_done_count++;
872         spin_unlock(&_pending_exceptions_done_spinlock);
873
874         wake_up_all(&_pending_exceptions_done);
875 }
876
877 static void snapshot_merge_next_chunks(struct dm_snapshot *s)
878 {
879         int i, linear_chunks;
880         chunk_t old_chunk, new_chunk;
881         struct dm_io_region src, dest;
882         sector_t io_size;
883         uint64_t previous_count;
884
885         BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
886         if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
887                 goto shut;
888
889         /*
890          * valid flag never changes during merge, so no lock required.
891          */
892         if (!s->valid) {
893                 DMERR("Snapshot is invalid: can't merge");
894                 goto shut;
895         }
896
897         linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
898                                                       &new_chunk);
899         if (linear_chunks <= 0) {
900                 if (linear_chunks < 0) {
901                         DMERR("Read error in exception store: "
902                               "shutting down merge");
903                         down_write(&s->lock);
904                         s->merge_failed = 1;
905                         up_write(&s->lock);
906                 }
907                 goto shut;
908         }
909
910         /* Adjust old_chunk and new_chunk to reflect start of linear region */
911         old_chunk = old_chunk + 1 - linear_chunks;
912         new_chunk = new_chunk + 1 - linear_chunks;
913
914         /*
915          * Use one (potentially large) I/O to copy all 'linear_chunks'
916          * from the exception store to the origin
917          */
918         io_size = linear_chunks * s->store->chunk_size;
919
920         dest.bdev = s->origin->bdev;
921         dest.sector = chunk_to_sector(s->store, old_chunk);
922         dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector);
923
924         src.bdev = s->cow->bdev;
925         src.sector = chunk_to_sector(s->store, new_chunk);
926         src.count = dest.count;
927
928         /*
929          * Reallocate any exceptions needed in other snapshots then
930          * wait for the pending exceptions to complete.
931          * Each time any pending exception (globally on the system)
932          * completes we are woken and repeat the process to find out
933          * if we can proceed.  While this may not seem a particularly
934          * efficient algorithm, it is not expected to have any
935          * significant impact on performance.
936          */
937         previous_count = read_pending_exceptions_done_count();
938         while (origin_write_extent(s, dest.sector, io_size)) {
939                 wait_event(_pending_exceptions_done,
940                            (read_pending_exceptions_done_count() !=
941                             previous_count));
942                 /* Retry after the wait, until all exceptions are done. */
943                 previous_count = read_pending_exceptions_done_count();
944         }
945
946         down_write(&s->lock);
947         s->first_merging_chunk = old_chunk;
948         s->num_merging_chunks = linear_chunks;
949         up_write(&s->lock);
950
951         /* Wait until writes to all 'linear_chunks' drain */
952         for (i = 0; i < linear_chunks; i++)
953                 __check_for_conflicting_io(s, old_chunk + i);
954
955         dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s);
956         return;
957
958 shut:
959         merge_shutdown(s);
960 }
961
962 static void error_bios(struct bio *bio);
963
964 static void merge_callback(int read_err, unsigned long write_err, void *context)
965 {
966         struct dm_snapshot *s = context;
967         struct bio *b = NULL;
968
969         if (read_err || write_err) {
970                 if (read_err)
971                         DMERR("Read error: shutting down merge.");
972                 else
973                         DMERR("Write error: shutting down merge.");
974                 goto shut;
975         }
976
977         if (s->store->type->commit_merge(s->store,
978                                          s->num_merging_chunks) < 0) {
979                 DMERR("Write error in exception store: shutting down merge");
980                 goto shut;
981         }
982
983         if (remove_single_exception_chunk(s) < 0)
984                 goto shut;
985
986         snapshot_merge_next_chunks(s);
987
988         return;
989
990 shut:
991         down_write(&s->lock);
992         s->merge_failed = 1;
993         b = __release_queued_bios_after_merge(s);
994         up_write(&s->lock);
995         error_bios(b);
996
997         merge_shutdown(s);
998 }
999
1000 static void start_merge(struct dm_snapshot *s)
1001 {
1002         if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits))
1003                 snapshot_merge_next_chunks(s);
1004 }
1005
1006 static int wait_schedule(void *ptr)
1007 {
1008         schedule();
1009
1010         return 0;
1011 }
1012
1013 /*
1014  * Stop the merging process and wait until it finishes.
1015  */
1016 static void stop_merge(struct dm_snapshot *s)
1017 {
1018         set_bit(SHUTDOWN_MERGE, &s->state_bits);
1019         wait_on_bit(&s->state_bits, RUNNING_MERGE, wait_schedule,
1020                     TASK_UNINTERRUPTIBLE);
1021         clear_bit(SHUTDOWN_MERGE, &s->state_bits);
1022 }
1023
1024 /*
1025  * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
1026  */
1027 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1028 {
1029         struct dm_snapshot *s;
1030         int i;
1031         int r = -EINVAL;
1032         char *origin_path, *cow_path;
1033         unsigned args_used, num_flush_requests = 1;
1034         fmode_t origin_mode = FMODE_READ;
1035
1036         if (argc != 4) {
1037                 ti->error = "requires exactly 4 arguments";
1038                 r = -EINVAL;
1039                 goto bad;
1040         }
1041
1042         if (dm_target_is_snapshot_merge(ti)) {
1043                 num_flush_requests = 2;
1044                 origin_mode = FMODE_WRITE;
1045         }
1046
1047         s = kmalloc(sizeof(*s), GFP_KERNEL);
1048         if (!s) {
1049                 ti->error = "Cannot allocate private snapshot structure";
1050                 r = -ENOMEM;
1051                 goto bad;
1052         }
1053
1054         origin_path = argv[0];
1055         argv++;
1056         argc--;
1057
1058         r = dm_get_device(ti, origin_path, origin_mode, &s->origin);
1059         if (r) {
1060                 ti->error = "Cannot get origin device";
1061                 goto bad_origin;
1062         }
1063
1064         cow_path = argv[0];
1065         argv++;
1066         argc--;
1067
1068         r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow);
1069         if (r) {
1070                 ti->error = "Cannot get COW device";
1071                 goto bad_cow;
1072         }
1073
1074         r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
1075         if (r) {
1076                 ti->error = "Couldn't create exception store";
1077                 r = -EINVAL;
1078                 goto bad_store;
1079         }
1080
1081         argv += args_used;
1082         argc -= args_used;
1083
1084         s->ti = ti;
1085         s->valid = 1;
1086         s->active = 0;
1087         atomic_set(&s->pending_exceptions_count, 0);
1088         init_rwsem(&s->lock);
1089         INIT_LIST_HEAD(&s->list);
1090         spin_lock_init(&s->pe_lock);
1091         s->state_bits = 0;
1092         s->merge_failed = 0;
1093         s->first_merging_chunk = 0;
1094         s->num_merging_chunks = 0;
1095         bio_list_init(&s->bios_queued_during_merge);
1096
1097         /* Allocate hash table for COW data */
1098         if (init_hash_tables(s)) {
1099                 ti->error = "Unable to allocate hash table space";
1100                 r = -ENOMEM;
1101                 goto bad_hash_tables;
1102         }
1103
1104         s->kcopyd_client = dm_kcopyd_client_create();
1105         if (IS_ERR(s->kcopyd_client)) {
1106                 r = PTR_ERR(s->kcopyd_client);
1107                 ti->error = "Could not create kcopyd client";
1108                 goto bad_kcopyd;
1109         }
1110
1111         s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
1112         if (!s->pending_pool) {
1113                 ti->error = "Could not allocate mempool for pending exceptions";
1114                 goto bad_pending_pool;
1115         }
1116
1117         for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1118                 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
1119
1120         spin_lock_init(&s->tracked_chunk_lock);
1121
1122         ti->private = s;
1123         ti->num_flush_requests = num_flush_requests;
1124         ti->per_bio_data_size = sizeof(struct dm_snap_tracked_chunk);
1125
1126         /* Add snapshot to the list of snapshots for this origin */
1127         /* Exceptions aren't triggered till snapshot_resume() is called */
1128         r = register_snapshot(s);
1129         if (r == -ENOMEM) {
1130                 ti->error = "Snapshot origin struct allocation failed";
1131                 goto bad_load_and_register;
1132         } else if (r < 0) {
1133                 /* invalid handover, register_snapshot has set ti->error */
1134                 goto bad_load_and_register;
1135         }
1136
1137         /*
1138          * Metadata must only be loaded into one table at once, so skip this
1139          * if metadata will be handed over during resume.
1140          * Chunk size will be set during the handover - set it to zero to
1141          * ensure it's ignored.
1142          */
1143         if (r > 0) {
1144                 s->store->chunk_size = 0;
1145                 return 0;
1146         }
1147
1148         r = s->store->type->read_metadata(s->store, dm_add_exception,
1149                                           (void *)s);
1150         if (r < 0) {
1151                 ti->error = "Failed to read snapshot metadata";
1152                 goto bad_read_metadata;
1153         } else if (r > 0) {
1154                 s->valid = 0;
1155                 DMWARN("Snapshot is marked invalid.");
1156         }
1157
1158         if (!s->store->chunk_size) {
1159                 ti->error = "Chunk size not set";
1160                 goto bad_read_metadata;
1161         }
1162
1163         r = dm_set_target_max_io_len(ti, s->store->chunk_size);
1164         if (r)
1165                 goto bad_read_metadata;
1166
1167         return 0;
1168
1169 bad_read_metadata:
1170         unregister_snapshot(s);
1171
1172 bad_load_and_register:
1173         mempool_destroy(s->pending_pool);
1174
1175 bad_pending_pool:
1176         dm_kcopyd_client_destroy(s->kcopyd_client);
1177
1178 bad_kcopyd:
1179         dm_exception_table_exit(&s->pending, pending_cache);
1180         dm_exception_table_exit(&s->complete, exception_cache);
1181
1182 bad_hash_tables:
1183         dm_exception_store_destroy(s->store);
1184
1185 bad_store:
1186         dm_put_device(ti, s->cow);
1187
1188 bad_cow:
1189         dm_put_device(ti, s->origin);
1190
1191 bad_origin:
1192         kfree(s);
1193
1194 bad:
1195         return r;
1196 }
1197
1198 static void __free_exceptions(struct dm_snapshot *s)
1199 {
1200         dm_kcopyd_client_destroy(s->kcopyd_client);
1201         s->kcopyd_client = NULL;
1202
1203         dm_exception_table_exit(&s->pending, pending_cache);
1204         dm_exception_table_exit(&s->complete, exception_cache);
1205 }
1206
1207 static void __handover_exceptions(struct dm_snapshot *snap_src,
1208                                   struct dm_snapshot *snap_dest)
1209 {
1210         union {
1211                 struct dm_exception_table table_swap;
1212                 struct dm_exception_store *store_swap;
1213         } u;
1214
1215         /*
1216          * Swap all snapshot context information between the two instances.
1217          */
1218         u.table_swap = snap_dest->complete;
1219         snap_dest->complete = snap_src->complete;
1220         snap_src->complete = u.table_swap;
1221
1222         u.store_swap = snap_dest->store;
1223         snap_dest->store = snap_src->store;
1224         snap_src->store = u.store_swap;
1225
1226         snap_dest->store->snap = snap_dest;
1227         snap_src->store->snap = snap_src;
1228
1229         snap_dest->ti->max_io_len = snap_dest->store->chunk_size;
1230         snap_dest->valid = snap_src->valid;
1231
1232         /*
1233          * Set source invalid to ensure it receives no further I/O.
1234          */
1235         snap_src->valid = 0;
1236 }
1237
1238 static void snapshot_dtr(struct dm_target *ti)
1239 {
1240 #ifdef CONFIG_DM_DEBUG
1241         int i;
1242 #endif
1243         struct dm_snapshot *s = ti->private;
1244         struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1245
1246         down_read(&_origins_lock);
1247         /* Check whether exception handover must be cancelled */
1248         (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1249         if (snap_src && snap_dest && (s == snap_src)) {
1250                 down_write(&snap_dest->lock);
1251                 snap_dest->valid = 0;
1252                 up_write(&snap_dest->lock);
1253                 DMERR("Cancelling snapshot handover.");
1254         }
1255         up_read(&_origins_lock);
1256
1257         if (dm_target_is_snapshot_merge(ti))
1258                 stop_merge(s);
1259
1260         /* Prevent further origin writes from using this snapshot. */
1261         /* After this returns there can be no new kcopyd jobs. */
1262         unregister_snapshot(s);
1263
1264         while (atomic_read(&s->pending_exceptions_count))
1265                 msleep(1);
1266         /*
1267          * Ensure instructions in mempool_destroy aren't reordered
1268          * before atomic_read.
1269          */
1270         smp_mb();
1271
1272 #ifdef CONFIG_DM_DEBUG
1273         for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1274                 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
1275 #endif
1276
1277         __free_exceptions(s);
1278
1279         mempool_destroy(s->pending_pool);
1280
1281         dm_exception_store_destroy(s->store);
1282
1283         dm_put_device(ti, s->cow);
1284
1285         dm_put_device(ti, s->origin);
1286
1287         kfree(s);
1288 }
1289
1290 /*
1291  * Flush a list of buffers.
1292  */
1293 static void flush_bios(struct bio *bio)
1294 {
1295         struct bio *n;
1296
1297         while (bio) {
1298                 n = bio->bi_next;
1299                 bio->bi_next = NULL;
1300                 generic_make_request(bio);
1301                 bio = n;
1302         }
1303 }
1304
1305 static int do_origin(struct dm_dev *origin, struct bio *bio);
1306
1307 /*
1308  * Flush a list of buffers.
1309  */
1310 static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
1311 {
1312         struct bio *n;
1313         int r;
1314
1315         while (bio) {
1316                 n = bio->bi_next;
1317                 bio->bi_next = NULL;
1318                 r = do_origin(s->origin, bio);
1319                 if (r == DM_MAPIO_REMAPPED)
1320                         generic_make_request(bio);
1321                 bio = n;
1322         }
1323 }
1324
1325 /*
1326  * Error a list of buffers.
1327  */
1328 static void error_bios(struct bio *bio)
1329 {
1330         struct bio *n;
1331
1332         while (bio) {
1333                 n = bio->bi_next;
1334                 bio->bi_next = NULL;
1335                 bio_io_error(bio);
1336                 bio = n;
1337         }
1338 }
1339
1340 static void __invalidate_snapshot(struct dm_snapshot *s, int err)
1341 {
1342         if (!s->valid)
1343                 return;
1344
1345         if (err == -EIO)
1346                 DMERR("Invalidating snapshot: Error reading/writing.");
1347         else if (err == -ENOMEM)
1348                 DMERR("Invalidating snapshot: Unable to allocate exception.");
1349
1350         if (s->store->type->drop_snapshot)
1351                 s->store->type->drop_snapshot(s->store);
1352
1353         s->valid = 0;
1354
1355         dm_table_event(s->ti->table);
1356 }
1357
1358 static void pending_complete(struct dm_snap_pending_exception *pe, int success)
1359 {
1360         struct dm_exception *e;
1361         struct dm_snapshot *s = pe->snap;
1362         struct bio *origin_bios = NULL;
1363         struct bio *snapshot_bios = NULL;
1364         struct bio *full_bio = NULL;
1365         int error = 0;
1366
1367         if (!success) {
1368                 /* Read/write error - snapshot is unusable */
1369                 down_write(&s->lock);
1370                 __invalidate_snapshot(s, -EIO);
1371                 error = 1;
1372                 goto out;
1373         }
1374
1375         e = alloc_completed_exception();
1376         if (!e) {
1377                 down_write(&s->lock);
1378                 __invalidate_snapshot(s, -ENOMEM);
1379                 error = 1;
1380                 goto out;
1381         }
1382         *e = pe->e;
1383
1384         down_write(&s->lock);
1385         if (!s->valid) {
1386                 free_completed_exception(e);
1387                 error = 1;
1388                 goto out;
1389         }
1390
1391         /* Check for conflicting reads */
1392         __check_for_conflicting_io(s, pe->e.old_chunk);
1393
1394         /*
1395          * Add a proper exception, and remove the
1396          * in-flight exception from the list.
1397          */
1398         dm_insert_exception(&s->complete, e);
1399
1400 out:
1401         dm_remove_exception(&pe->e);
1402         snapshot_bios = bio_list_get(&pe->snapshot_bios);
1403         origin_bios = bio_list_get(&pe->origin_bios);
1404         full_bio = pe->full_bio;
1405         if (full_bio) {
1406                 full_bio->bi_end_io = pe->full_bio_end_io;
1407                 full_bio->bi_private = pe->full_bio_private;
1408         }
1409         free_pending_exception(pe);
1410
1411         increment_pending_exceptions_done_count();
1412
1413         up_write(&s->lock);
1414
1415         /* Submit any pending write bios */
1416         if (error) {
1417                 if (full_bio)
1418                         bio_io_error(full_bio);
1419                 error_bios(snapshot_bios);
1420         } else {
1421                 if (full_bio)
1422                         bio_endio(full_bio, 0);
1423                 flush_bios(snapshot_bios);
1424         }
1425
1426         retry_origin_bios(s, origin_bios);
1427 }
1428
1429 static void commit_callback(void *context, int success)
1430 {
1431         struct dm_snap_pending_exception *pe = context;
1432
1433         pending_complete(pe, success);
1434 }
1435
1436 /*
1437  * Called when the copy I/O has finished.  kcopyd actually runs
1438  * this code so don't block.
1439  */
1440 static void copy_callback(int read_err, unsigned long write_err, void *context)
1441 {
1442         struct dm_snap_pending_exception *pe = context;
1443         struct dm_snapshot *s = pe->snap;
1444
1445         if (read_err || write_err)
1446                 pending_complete(pe, 0);
1447
1448         else
1449                 /* Update the metadata if we are persistent */
1450                 s->store->type->commit_exception(s->store, &pe->e,
1451                                                  commit_callback, pe);
1452 }
1453
1454 /*
1455  * Dispatches the copy operation to kcopyd.
1456  */
1457 static void start_copy(struct dm_snap_pending_exception *pe)
1458 {
1459         struct dm_snapshot *s = pe->snap;
1460         struct dm_io_region src, dest;
1461         struct block_device *bdev = s->origin->bdev;
1462         sector_t dev_size;
1463
1464         dev_size = get_dev_size(bdev);
1465
1466         src.bdev = bdev;
1467         src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
1468         src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
1469
1470         dest.bdev = s->cow->bdev;
1471         dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
1472         dest.count = src.count;
1473
1474         /* Hand over to kcopyd */
1475         dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe);
1476 }
1477
1478 static void full_bio_end_io(struct bio *bio, int error)
1479 {
1480         void *callback_data = bio->bi_private;
1481
1482         dm_kcopyd_do_callback(callback_data, 0, error ? 1 : 0);
1483 }
1484
1485 static void start_full_bio(struct dm_snap_pending_exception *pe,
1486                            struct bio *bio)
1487 {
1488         struct dm_snapshot *s = pe->snap;
1489         void *callback_data;
1490
1491         pe->full_bio = bio;
1492         pe->full_bio_end_io = bio->bi_end_io;
1493         pe->full_bio_private = bio->bi_private;
1494
1495         callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client,
1496                                                    copy_callback, pe);
1497
1498         bio->bi_end_io = full_bio_end_io;
1499         bio->bi_private = callback_data;
1500
1501         generic_make_request(bio);
1502 }
1503
1504 static struct dm_snap_pending_exception *
1505 __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1506 {
1507         struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
1508
1509         if (!e)
1510                 return NULL;
1511
1512         return container_of(e, struct dm_snap_pending_exception, e);
1513 }
1514
1515 /*
1516  * Looks to see if this snapshot already has a pending exception
1517  * for this chunk, otherwise it allocates a new one and inserts
1518  * it into the pending table.
1519  *
1520  * NOTE: a write lock must be held on snap->lock before calling
1521  * this.
1522  */
1523 static struct dm_snap_pending_exception *
1524 __find_pending_exception(struct dm_snapshot *s,
1525                          struct dm_snap_pending_exception *pe, chunk_t chunk)
1526 {
1527         struct dm_snap_pending_exception *pe2;
1528
1529         pe2 = __lookup_pending_exception(s, chunk);
1530         if (pe2) {
1531                 free_pending_exception(pe);
1532                 return pe2;
1533         }
1534
1535         pe->e.old_chunk = chunk;
1536         bio_list_init(&pe->origin_bios);
1537         bio_list_init(&pe->snapshot_bios);
1538         pe->started = 0;
1539         pe->full_bio = NULL;
1540
1541         if (s->store->type->prepare_exception(s->store, &pe->e)) {
1542                 free_pending_exception(pe);
1543                 return NULL;
1544         }
1545
1546         dm_insert_exception(&s->pending, &pe->e);
1547
1548         return pe;
1549 }
1550
1551 static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
1552                             struct bio *bio, chunk_t chunk)
1553 {
1554         bio->bi_bdev = s->cow->bdev;
1555         bio->bi_sector = chunk_to_sector(s->store,
1556                                          dm_chunk_number(e->new_chunk) +
1557                                          (chunk - e->old_chunk)) +
1558                                          (bio->bi_sector &
1559                                           s->store->chunk_mask);
1560 }
1561
1562 static int snapshot_map(struct dm_target *ti, struct bio *bio,
1563                         union map_info *map_context)
1564 {
1565         struct dm_exception *e;
1566         struct dm_snapshot *s = ti->private;
1567         int r = DM_MAPIO_REMAPPED;
1568         chunk_t chunk;
1569         struct dm_snap_pending_exception *pe = NULL;
1570
1571         if (bio->bi_rw & REQ_FLUSH) {
1572                 bio->bi_bdev = s->cow->bdev;
1573                 return DM_MAPIO_REMAPPED;
1574         }
1575
1576         chunk = sector_to_chunk(s->store, bio->bi_sector);
1577
1578         /* Full snapshots are not usable */
1579         /* To get here the table must be live so s->active is always set. */
1580         if (!s->valid)
1581                 return -EIO;
1582
1583         /* FIXME: should only take write lock if we need
1584          * to copy an exception */
1585         down_write(&s->lock);
1586
1587         if (!s->valid) {
1588                 r = -EIO;
1589                 goto out_unlock;
1590         }
1591
1592         /* If the block is already remapped - use that, else remap it */
1593         e = dm_lookup_exception(&s->complete, chunk);
1594         if (e) {
1595                 remap_exception(s, e, bio, chunk);
1596                 goto out_unlock;
1597         }
1598
1599         /*
1600          * Write to snapshot - higher level takes care of RW/RO
1601          * flags so we should only get this if we are
1602          * writeable.
1603          */
1604         if (bio_rw(bio) == WRITE) {
1605                 pe = __lookup_pending_exception(s, chunk);
1606                 if (!pe) {
1607                         up_write(&s->lock);
1608                         pe = alloc_pending_exception(s);
1609                         down_write(&s->lock);
1610
1611                         if (!s->valid) {
1612                                 free_pending_exception(pe);
1613                                 r = -EIO;
1614                                 goto out_unlock;
1615                         }
1616
1617                         e = dm_lookup_exception(&s->complete, chunk);
1618                         if (e) {
1619                                 free_pending_exception(pe);
1620                                 remap_exception(s, e, bio, chunk);
1621                                 goto out_unlock;
1622                         }
1623
1624                         pe = __find_pending_exception(s, pe, chunk);
1625                         if (!pe) {
1626                                 __invalidate_snapshot(s, -ENOMEM);
1627                                 r = -EIO;
1628                                 goto out_unlock;
1629                         }
1630                 }
1631
1632                 remap_exception(s, &pe->e, bio, chunk);
1633
1634                 r = DM_MAPIO_SUBMITTED;
1635
1636                 if (!pe->started &&
1637                     bio->bi_size == (s->store->chunk_size << SECTOR_SHIFT)) {
1638                         pe->started = 1;
1639                         up_write(&s->lock);
1640                         start_full_bio(pe, bio);
1641                         goto out;
1642                 }
1643
1644                 bio_list_add(&pe->snapshot_bios, bio);
1645
1646                 if (!pe->started) {
1647                         /* this is protected by snap->lock */
1648                         pe->started = 1;
1649                         up_write(&s->lock);
1650                         start_copy(pe);
1651                         goto out;
1652                 }
1653         } else {
1654                 bio->bi_bdev = s->origin->bdev;
1655                 map_context->ptr = track_chunk(s, bio, chunk);
1656         }
1657
1658 out_unlock:
1659         up_write(&s->lock);
1660 out:
1661         return r;
1662 }
1663
1664 /*
1665  * A snapshot-merge target behaves like a combination of a snapshot
1666  * target and a snapshot-origin target.  It only generates new
1667  * exceptions in other snapshots and not in the one that is being
1668  * merged.
1669  *
1670  * For each chunk, if there is an existing exception, it is used to
1671  * redirect I/O to the cow device.  Otherwise I/O is sent to the origin,
1672  * which in turn might generate exceptions in other snapshots.
1673  * If merging is currently taking place on the chunk in question, the
1674  * I/O is deferred by adding it to s->bios_queued_during_merge.
1675  */
1676 static int snapshot_merge_map(struct dm_target *ti, struct bio *bio,
1677                               union map_info *map_context)
1678 {
1679         struct dm_exception *e;
1680         struct dm_snapshot *s = ti->private;
1681         int r = DM_MAPIO_REMAPPED;
1682         chunk_t chunk;
1683
1684         if (bio->bi_rw & REQ_FLUSH) {
1685                 if (!dm_bio_get_target_request_nr(bio))
1686                         bio->bi_bdev = s->origin->bdev;
1687                 else
1688                         bio->bi_bdev = s->cow->bdev;
1689                 map_context->ptr = NULL;
1690                 return DM_MAPIO_REMAPPED;
1691         }
1692
1693         chunk = sector_to_chunk(s->store, bio->bi_sector);
1694
1695         down_write(&s->lock);
1696
1697         /* Full merging snapshots are redirected to the origin */
1698         if (!s->valid)
1699                 goto redirect_to_origin;
1700
1701         /* If the block is already remapped - use that */
1702         e = dm_lookup_exception(&s->complete, chunk);
1703         if (e) {
1704                 /* Queue writes overlapping with chunks being merged */
1705                 if (bio_rw(bio) == WRITE &&
1706                     chunk >= s->first_merging_chunk &&
1707                     chunk < (s->first_merging_chunk +
1708                              s->num_merging_chunks)) {
1709                         bio->bi_bdev = s->origin->bdev;
1710                         bio_list_add(&s->bios_queued_during_merge, bio);
1711                         r = DM_MAPIO_SUBMITTED;
1712                         goto out_unlock;
1713                 }
1714
1715                 remap_exception(s, e, bio, chunk);
1716
1717                 if (bio_rw(bio) == WRITE)
1718                         map_context->ptr = track_chunk(s, bio, chunk);
1719                 goto out_unlock;
1720         }
1721
1722 redirect_to_origin:
1723         bio->bi_bdev = s->origin->bdev;
1724
1725         if (bio_rw(bio) == WRITE) {
1726                 up_write(&s->lock);
1727                 return do_origin(s->origin, bio);
1728         }
1729
1730 out_unlock:
1731         up_write(&s->lock);
1732
1733         return r;
1734 }
1735
1736 static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1737                            int error, union map_info *map_context)
1738 {
1739         struct dm_snapshot *s = ti->private;
1740         struct dm_snap_tracked_chunk *c = map_context->ptr;
1741
1742         if (c)
1743                 stop_tracking_chunk(s, c);
1744
1745         return 0;
1746 }
1747
1748 static void snapshot_merge_presuspend(struct dm_target *ti)
1749 {
1750         struct dm_snapshot *s = ti->private;
1751
1752         stop_merge(s);
1753 }
1754
1755 static int snapshot_preresume(struct dm_target *ti)
1756 {
1757         int r = 0;
1758         struct dm_snapshot *s = ti->private;
1759         struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1760
1761         down_read(&_origins_lock);
1762         (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1763         if (snap_src && snap_dest) {
1764                 down_read(&snap_src->lock);
1765                 if (s == snap_src) {
1766                         DMERR("Unable to resume snapshot source until "
1767                               "handover completes.");
1768                         r = -EINVAL;
1769                 } else if (!dm_suspended(snap_src->ti)) {
1770                         DMERR("Unable to perform snapshot handover until "
1771                               "source is suspended.");
1772                         r = -EINVAL;
1773                 }
1774                 up_read(&snap_src->lock);
1775         }
1776         up_read(&_origins_lock);
1777
1778         return r;
1779 }
1780
1781 static void snapshot_resume(struct dm_target *ti)
1782 {
1783         struct dm_snapshot *s = ti->private;
1784         struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1785
1786         down_read(&_origins_lock);
1787         (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1788         if (snap_src && snap_dest) {
1789                 down_write(&snap_src->lock);
1790                 down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1791                 __handover_exceptions(snap_src, snap_dest);
1792                 up_write(&snap_dest->lock);
1793                 up_write(&snap_src->lock);
1794         }
1795         up_read(&_origins_lock);
1796
1797         /* Now we have correct chunk size, reregister */
1798         reregister_snapshot(s);
1799
1800         down_write(&s->lock);
1801         s->active = 1;
1802         up_write(&s->lock);
1803 }
1804
1805 static uint32_t get_origin_minimum_chunksize(struct block_device *bdev)
1806 {
1807         uint32_t min_chunksize;
1808
1809         down_read(&_origins_lock);
1810         min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
1811         up_read(&_origins_lock);
1812
1813         return min_chunksize;
1814 }
1815
1816 static void snapshot_merge_resume(struct dm_target *ti)
1817 {
1818         struct dm_snapshot *s = ti->private;
1819
1820         /*
1821          * Handover exceptions from existing snapshot.
1822          */
1823         snapshot_resume(ti);
1824
1825         /*
1826          * snapshot-merge acts as an origin, so set ti->max_io_len
1827          */
1828         ti->max_io_len = get_origin_minimum_chunksize(s->origin->bdev);
1829
1830         start_merge(s);
1831 }
1832
1833 static int snapshot_status(struct dm_target *ti, status_type_t type,
1834                            unsigned status_flags, char *result, unsigned maxlen)
1835 {
1836         unsigned sz = 0;
1837         struct dm_snapshot *snap = ti->private;
1838
1839         switch (type) {
1840         case STATUSTYPE_INFO:
1841
1842                 down_write(&snap->lock);
1843
1844                 if (!snap->valid)
1845                         DMEMIT("Invalid");
1846                 else if (snap->merge_failed)
1847                         DMEMIT("Merge failed");
1848                 else {
1849                         if (snap->store->type->usage) {
1850                                 sector_t total_sectors, sectors_allocated,
1851                                          metadata_sectors;
1852                                 snap->store->type->usage(snap->store,
1853                                                          &total_sectors,
1854                                                          &sectors_allocated,
1855                                                          &metadata_sectors);
1856                                 DMEMIT("%llu/%llu %llu",
1857                                        (unsigned long long)sectors_allocated,
1858                                        (unsigned long long)total_sectors,
1859                                        (unsigned long long)metadata_sectors);
1860                         }
1861                         else
1862                                 DMEMIT("Unknown");
1863                 }
1864
1865                 up_write(&snap->lock);
1866
1867                 break;
1868
1869         case STATUSTYPE_TABLE:
1870                 /*
1871                  * kdevname returns a static pointer so we need
1872                  * to make private copies if the output is to
1873                  * make sense.
1874                  */
1875                 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
1876                 snap->store->type->status(snap->store, type, result + sz,
1877                                           maxlen - sz);
1878                 break;
1879         }
1880
1881         return 0;
1882 }
1883
1884 static int snapshot_iterate_devices(struct dm_target *ti,
1885                                     iterate_devices_callout_fn fn, void *data)
1886 {
1887         struct dm_snapshot *snap = ti->private;
1888         int r;
1889
1890         r = fn(ti, snap->origin, 0, ti->len, data);
1891
1892         if (!r)
1893                 r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data);
1894
1895         return r;
1896 }
1897
1898
1899 /*-----------------------------------------------------------------
1900  * Origin methods
1901  *---------------------------------------------------------------*/
1902
1903 /*
1904  * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
1905  * supplied bio was ignored.  The caller may submit it immediately.
1906  * (No remapping actually occurs as the origin is always a direct linear
1907  * map.)
1908  *
1909  * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
1910  * and any supplied bio is added to a list to be submitted once all
1911  * the necessary exceptions exist.
1912  */
1913 static int __origin_write(struct list_head *snapshots, sector_t sector,
1914                           struct bio *bio)
1915 {
1916         int r = DM_MAPIO_REMAPPED;
1917         struct dm_snapshot *snap;
1918         struct dm_exception *e;
1919         struct dm_snap_pending_exception *pe;
1920         struct dm_snap_pending_exception *pe_to_start_now = NULL;
1921         struct dm_snap_pending_exception *pe_to_start_last = NULL;
1922         chunk_t chunk;
1923
1924         /* Do all the snapshots on this origin */
1925         list_for_each_entry (snap, snapshots, list) {
1926                 /*
1927                  * Don't make new exceptions in a merging snapshot
1928                  * because it has effectively been deleted
1929                  */
1930                 if (dm_target_is_snapshot_merge(snap->ti))
1931                         continue;
1932
1933                 down_write(&snap->lock);
1934
1935                 /* Only deal with valid and active snapshots */
1936                 if (!snap->valid || !snap->active)
1937                         goto next_snapshot;
1938
1939                 /* Nothing to do if writing beyond end of snapshot */
1940                 if (sector >= dm_table_get_size(snap->ti->table))
1941                         goto next_snapshot;
1942
1943                 /*
1944                  * Remember, different snapshots can have
1945                  * different chunk sizes.
1946                  */
1947                 chunk = sector_to_chunk(snap->store, sector);
1948
1949                 /*
1950                  * Check exception table to see if block
1951                  * is already remapped in this snapshot
1952                  * and trigger an exception if not.
1953                  */
1954                 e = dm_lookup_exception(&snap->complete, chunk);
1955                 if (e)
1956                         goto next_snapshot;
1957
1958                 pe = __lookup_pending_exception(snap, chunk);
1959                 if (!pe) {
1960                         up_write(&snap->lock);
1961                         pe = alloc_pending_exception(snap);
1962                         down_write(&snap->lock);
1963
1964                         if (!snap->valid) {
1965                                 free_pending_exception(pe);
1966                                 goto next_snapshot;
1967                         }
1968
1969                         e = dm_lookup_exception(&snap->complete, chunk);
1970                         if (e) {
1971                                 free_pending_exception(pe);
1972                                 goto next_snapshot;
1973                         }
1974
1975                         pe = __find_pending_exception(snap, pe, chunk);
1976                         if (!pe) {
1977                                 __invalidate_snapshot(snap, -ENOMEM);
1978                                 goto next_snapshot;
1979                         }
1980                 }
1981
1982                 r = DM_MAPIO_SUBMITTED;
1983
1984                 /*
1985                  * If an origin bio was supplied, queue it to wait for the
1986                  * completion of this exception, and start this one last,
1987                  * at the end of the function.
1988                  */
1989                 if (bio) {
1990                         bio_list_add(&pe->origin_bios, bio);
1991                         bio = NULL;
1992
1993                         if (!pe->started) {
1994                                 pe->started = 1;
1995                                 pe_to_start_last = pe;
1996                         }
1997                 }
1998
1999                 if (!pe->started) {
2000                         pe->started = 1;
2001                         pe_to_start_now = pe;
2002                 }
2003
2004 next_snapshot:
2005                 up_write(&snap->lock);
2006
2007                 if (pe_to_start_now) {
2008                         start_copy(pe_to_start_now);
2009                         pe_to_start_now = NULL;
2010                 }
2011         }
2012
2013         /*
2014          * Submit the exception against which the bio is queued last,
2015          * to give the other exceptions a head start.
2016          */
2017         if (pe_to_start_last)
2018                 start_copy(pe_to_start_last);
2019
2020         return r;
2021 }
2022
2023 /*
2024  * Called on a write from the origin driver.
2025  */
2026 static int do_origin(struct dm_dev *origin, struct bio *bio)
2027 {
2028         struct origin *o;
2029         int r = DM_MAPIO_REMAPPED;
2030
2031         down_read(&_origins_lock);
2032         o = __lookup_origin(origin->bdev);
2033         if (o)
2034                 r = __origin_write(&o->snapshots, bio->bi_sector, bio);
2035         up_read(&_origins_lock);
2036
2037         return r;
2038 }
2039
2040 /*
2041  * Trigger exceptions in all non-merging snapshots.
2042  *
2043  * The chunk size of the merging snapshot may be larger than the chunk
2044  * size of some other snapshot so we may need to reallocate multiple
2045  * chunks in other snapshots.
2046  *
2047  * We scan all the overlapping exceptions in the other snapshots.
2048  * Returns 1 if anything was reallocated and must be waited for,
2049  * otherwise returns 0.
2050  *
2051  * size must be a multiple of merging_snap's chunk_size.
2052  */
2053 static int origin_write_extent(struct dm_snapshot *merging_snap,
2054                                sector_t sector, unsigned size)
2055 {
2056         int must_wait = 0;
2057         sector_t n;
2058         struct origin *o;
2059
2060         /*
2061          * The origin's __minimum_chunk_size() got stored in max_io_len
2062          * by snapshot_merge_resume().
2063          */
2064         down_read(&_origins_lock);
2065         o = __lookup_origin(merging_snap->origin->bdev);
2066         for (n = 0; n < size; n += merging_snap->ti->max_io_len)
2067                 if (__origin_write(&o->snapshots, sector + n, NULL) ==
2068                     DM_MAPIO_SUBMITTED)
2069                         must_wait = 1;
2070         up_read(&_origins_lock);
2071
2072         return must_wait;
2073 }
2074
2075 /*
2076  * Origin: maps a linear range of a device, with hooks for snapshotting.
2077  */
2078
2079 /*
2080  * Construct an origin mapping: <dev_path>
2081  * The context for an origin is merely a 'struct dm_dev *'
2082  * pointing to the real device.
2083  */
2084 static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2085 {
2086         int r;
2087         struct dm_dev *dev;
2088
2089         if (argc != 1) {
2090                 ti->error = "origin: incorrect number of arguments";
2091                 return -EINVAL;
2092         }
2093
2094         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev);
2095         if (r) {
2096                 ti->error = "Cannot get target device";
2097                 return r;
2098         }
2099
2100         ti->private = dev;
2101         ti->num_flush_requests = 1;
2102
2103         return 0;
2104 }
2105
2106 static void origin_dtr(struct dm_target *ti)
2107 {
2108         struct dm_dev *dev = ti->private;
2109         dm_put_device(ti, dev);
2110 }
2111
2112 static int origin_map(struct dm_target *ti, struct bio *bio,
2113                       union map_info *map_context)
2114 {
2115         struct dm_dev *dev = ti->private;
2116         bio->bi_bdev = dev->bdev;
2117
2118         if (bio->bi_rw & REQ_FLUSH)
2119                 return DM_MAPIO_REMAPPED;
2120
2121         /* Only tell snapshots if this is a write */
2122         return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
2123 }
2124
2125 /*
2126  * Set the target "max_io_len" field to the minimum of all the snapshots'
2127  * chunk sizes.
2128  */
2129 static void origin_resume(struct dm_target *ti)
2130 {
2131         struct dm_dev *dev = ti->private;
2132
2133         ti->max_io_len = get_origin_minimum_chunksize(dev->bdev);
2134 }
2135
2136 static int origin_status(struct dm_target *ti, status_type_t type,
2137                          unsigned status_flags, char *result, unsigned maxlen)
2138 {
2139         struct dm_dev *dev = ti->private;
2140
2141         switch (type) {
2142         case STATUSTYPE_INFO:
2143                 result[0] = '\0';
2144                 break;
2145
2146         case STATUSTYPE_TABLE:
2147                 snprintf(result, maxlen, "%s", dev->name);
2148                 break;
2149         }
2150
2151         return 0;
2152 }
2153
2154 static int origin_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
2155                         struct bio_vec *biovec, int max_size)
2156 {
2157         struct dm_dev *dev = ti->private;
2158         struct request_queue *q = bdev_get_queue(dev->bdev);
2159
2160         if (!q->merge_bvec_fn)
2161                 return max_size;
2162
2163         bvm->bi_bdev = dev->bdev;
2164
2165         return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
2166 }
2167
2168 static int origin_iterate_devices(struct dm_target *ti,
2169                                   iterate_devices_callout_fn fn, void *data)
2170 {
2171         struct dm_dev *dev = ti->private;
2172
2173         return fn(ti, dev, 0, ti->len, data);
2174 }
2175
2176 static struct target_type origin_target = {
2177         .name    = "snapshot-origin",
2178         .version = {1, 8, 0},
2179         .module  = THIS_MODULE,
2180         .ctr     = origin_ctr,
2181         .dtr     = origin_dtr,
2182         .map     = origin_map,
2183         .resume  = origin_resume,
2184         .status  = origin_status,
2185         .merge   = origin_merge,
2186         .iterate_devices = origin_iterate_devices,
2187 };
2188
2189 static struct target_type snapshot_target = {
2190         .name    = "snapshot",
2191         .version = {1, 11, 0},
2192         .module  = THIS_MODULE,
2193         .ctr     = snapshot_ctr,
2194         .dtr     = snapshot_dtr,
2195         .map     = snapshot_map,
2196         .end_io  = snapshot_end_io,
2197         .preresume  = snapshot_preresume,
2198         .resume  = snapshot_resume,
2199         .status  = snapshot_status,
2200         .iterate_devices = snapshot_iterate_devices,
2201 };
2202
2203 static struct target_type merge_target = {
2204         .name    = dm_snapshot_merge_target_name,
2205         .version = {1, 2, 0},
2206         .module  = THIS_MODULE,
2207         .ctr     = snapshot_ctr,
2208         .dtr     = snapshot_dtr,
2209         .map     = snapshot_merge_map,
2210         .end_io  = snapshot_end_io,
2211         .presuspend = snapshot_merge_presuspend,
2212         .preresume  = snapshot_preresume,
2213         .resume  = snapshot_merge_resume,
2214         .status  = snapshot_status,
2215         .iterate_devices = snapshot_iterate_devices,
2216 };
2217
2218 static int __init dm_snapshot_init(void)
2219 {
2220         int r;
2221
2222         r = dm_exception_store_init();
2223         if (r) {
2224                 DMERR("Failed to initialize exception stores");
2225                 return r;
2226         }
2227
2228         r = dm_register_target(&snapshot_target);
2229         if (r < 0) {
2230                 DMERR("snapshot target register failed %d", r);
2231                 goto bad_register_snapshot_target;
2232         }
2233
2234         r = dm_register_target(&origin_target);
2235         if (r < 0) {
2236                 DMERR("Origin target register failed %d", r);
2237                 goto bad_register_origin_target;
2238         }
2239
2240         r = dm_register_target(&merge_target);
2241         if (r < 0) {
2242                 DMERR("Merge target register failed %d", r);
2243                 goto bad_register_merge_target;
2244         }
2245
2246         r = init_origin_hash();
2247         if (r) {
2248                 DMERR("init_origin_hash failed.");
2249                 goto bad_origin_hash;
2250         }
2251
2252         exception_cache = KMEM_CACHE(dm_exception, 0);
2253         if (!exception_cache) {
2254                 DMERR("Couldn't create exception cache.");
2255                 r = -ENOMEM;
2256                 goto bad_exception_cache;
2257         }
2258
2259         pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
2260         if (!pending_cache) {
2261                 DMERR("Couldn't create pending cache.");
2262                 r = -ENOMEM;
2263                 goto bad_pending_cache;
2264         }
2265
2266         return 0;
2267
2268 bad_pending_cache:
2269         kmem_cache_destroy(exception_cache);
2270 bad_exception_cache:
2271         exit_origin_hash();
2272 bad_origin_hash:
2273         dm_unregister_target(&merge_target);
2274 bad_register_merge_target:
2275         dm_unregister_target(&origin_target);
2276 bad_register_origin_target:
2277         dm_unregister_target(&snapshot_target);
2278 bad_register_snapshot_target:
2279         dm_exception_store_exit();
2280
2281         return r;
2282 }
2283
2284 static void __exit dm_snapshot_exit(void)
2285 {
2286         dm_unregister_target(&snapshot_target);
2287         dm_unregister_target(&origin_target);
2288         dm_unregister_target(&merge_target);
2289
2290         exit_origin_hash();
2291         kmem_cache_destroy(pending_cache);
2292         kmem_cache_destroy(exception_cache);
2293
2294         dm_exception_store_exit();
2295 }
2296
2297 /* Module hooks */
2298 module_init(dm_snapshot_init);
2299 module_exit(dm_snapshot_exit);
2300
2301 MODULE_DESCRIPTION(DM_NAME " snapshot target");
2302 MODULE_AUTHOR("Joe Thornber");
2303 MODULE_LICENSE("GPL");