1
2
3
4
5
6
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.h"
24
25#include "dm-exception-store.h"
26
27#define DM_MSG_PREFIX "snapshots"
28
29static const char dm_snapshot_merge_target_name[] = "snapshot-merge";
30
31#define dm_target_is_snapshot_merge(ti) \
32 ((ti)->type->name == dm_snapshot_merge_target_name)
33
34
35
36
37#define MIN_IOS 256
38
39#define DM_TRACKED_CHUNK_HASH_SIZE 16
40#define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
41 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
42
43struct dm_exception_table {
44 uint32_t hash_mask;
45 unsigned hash_shift;
46 struct list_head *table;
47};
48
49struct dm_snapshot {
50 struct mutex lock;
51
52 struct dm_dev *origin;
53 struct dm_dev *cow;
54
55 struct dm_target *ti;
56
57
58 struct list_head list;
59
60
61
62
63
64 int valid;
65
66
67
68
69
70
71 int snapshot_overflowed;
72
73
74 int active;
75
76 atomic_t pending_exceptions_count;
77
78
79 sector_t exception_start_sequence;
80
81
82 sector_t exception_complete_sequence;
83
84
85
86
87
88 struct list_head out_of_order_list;
89
90 mempool_t *pending_pool;
91
92 struct dm_exception_table pending;
93 struct dm_exception_table complete;
94
95
96
97
98
99 spinlock_t pe_lock;
100
101
102 spinlock_t tracked_chunk_lock;
103 struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
104
105
106 struct dm_exception_store *store;
107
108 struct dm_kcopyd_client *kcopyd_client;
109
110
111 unsigned long state_bits;
112
113
114 chunk_t first_merging_chunk;
115 int num_merging_chunks;
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130 int merge_failed;
131
132
133
134
135
136 struct bio_list bios_queued_during_merge;
137};
138
139
140
141
142
143
144
145#define RUNNING_MERGE 0
146#define SHUTDOWN_MERGE 1
147
148DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
149 "A percentage of time allocated for copy on write");
150
151struct dm_dev *dm_snap_origin(struct dm_snapshot *s)
152{
153 return s->origin;
154}
155EXPORT_SYMBOL(dm_snap_origin);
156
157struct dm_dev *dm_snap_cow(struct dm_snapshot *s)
158{
159 return s->cow;
160}
161EXPORT_SYMBOL(dm_snap_cow);
162
163static sector_t chunk_to_sector(struct dm_exception_store *store,
164 chunk_t chunk)
165{
166 return chunk << store->chunk_shift;
167}
168
169static int bdev_equal(struct block_device *lhs, struct block_device *rhs)
170{
171
172
173
174
175 return lhs == rhs;
176}
177
178struct dm_snap_pending_exception {
179 struct dm_exception e;
180
181
182
183
184
185 struct bio_list origin_bios;
186 struct bio_list snapshot_bios;
187
188
189 struct dm_snapshot *snap;
190
191
192
193
194
195 int started;
196
197
198 int copy_error;
199
200
201 sector_t exception_sequence;
202
203 struct list_head out_of_order_entry;
204
205
206
207
208 struct bio *full_bio;
209 bio_end_io_t *full_bio_end_io;
210};
211
212
213
214
215
216static struct kmem_cache *exception_cache;
217static struct kmem_cache *pending_cache;
218
219struct dm_snap_tracked_chunk {
220 struct hlist_node node;
221 chunk_t chunk;
222};
223
224static void init_tracked_chunk(struct bio *bio)
225{
226 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
227 INIT_HLIST_NODE(&c->node);
228}
229
230static bool is_bio_tracked(struct bio *bio)
231{
232 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
233 return !hlist_unhashed(&c->node);
234}
235
236static void track_chunk(struct dm_snapshot *s, struct bio *bio, chunk_t chunk)
237{
238 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
239
240 c->chunk = chunk;
241
242 spin_lock_irq(&s->tracked_chunk_lock);
243 hlist_add_head(&c->node,
244 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]);
245 spin_unlock_irq(&s->tracked_chunk_lock);
246}
247
248static void stop_tracking_chunk(struct dm_snapshot *s, struct bio *bio)
249{
250 struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk));
251 unsigned long flags;
252
253 spin_lock_irqsave(&s->tracked_chunk_lock, flags);
254 hlist_del(&c->node);
255 spin_unlock_irqrestore(&s->tracked_chunk_lock, flags);
256}
257
258static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
259{
260 struct dm_snap_tracked_chunk *c;
261 int found = 0;
262
263 spin_lock_irq(&s->tracked_chunk_lock);
264
265 hlist_for_each_entry(c,
266 &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) {
267 if (c->chunk == chunk) {
268 found = 1;
269 break;
270 }
271 }
272
273 spin_unlock_irq(&s->tracked_chunk_lock);
274
275 return found;
276}
277
278
279
280
281
282static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
283{
284 while (__chunk_is_tracked(s, chunk))
285 msleep(1);
286}
287
288
289
290
291struct origin {
292
293 struct block_device *bdev;
294
295 struct list_head hash_list;
296
297
298 struct list_head snapshots;
299};
300
301
302
303
304struct dm_origin {
305 struct dm_dev *dev;
306 struct dm_target *ti;
307 unsigned split_boundary;
308 struct list_head hash_list;
309};
310
311
312
313
314
315#define ORIGIN_HASH_SIZE 256
316#define ORIGIN_MASK 0xFF
317static struct list_head *_origins;
318static struct list_head *_dm_origins;
319static struct rw_semaphore _origins_lock;
320
321static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
322static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
323static uint64_t _pending_exceptions_done_count;
324
325static int init_origin_hash(void)
326{
327 int i;
328
329 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
330 GFP_KERNEL);
331 if (!_origins) {
332 DMERR("unable to allocate memory for _origins");
333 return -ENOMEM;
334 }
335 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
336 INIT_LIST_HEAD(_origins + i);
337
338 _dm_origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
339 GFP_KERNEL);
340 if (!_dm_origins) {
341 DMERR("unable to allocate memory for _dm_origins");
342 kfree(_origins);
343 return -ENOMEM;
344 }
345 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
346 INIT_LIST_HEAD(_dm_origins + i);
347
348 init_rwsem(&_origins_lock);
349
350 return 0;
351}
352
353static void exit_origin_hash(void)
354{
355 kfree(_origins);
356 kfree(_dm_origins);
357}
358
359static unsigned origin_hash(struct block_device *bdev)
360{
361 return bdev->bd_dev & ORIGIN_MASK;
362}
363
364static struct origin *__lookup_origin(struct block_device *origin)
365{
366 struct list_head *ol;
367 struct origin *o;
368
369 ol = &_origins[origin_hash(origin)];
370 list_for_each_entry (o, ol, hash_list)
371 if (bdev_equal(o->bdev, origin))
372 return o;
373
374 return NULL;
375}
376
377static void __insert_origin(struct origin *o)
378{
379 struct list_head *sl = &_origins[origin_hash(o->bdev)];
380 list_add_tail(&o->hash_list, sl);
381}
382
383static struct dm_origin *__lookup_dm_origin(struct block_device *origin)
384{
385 struct list_head *ol;
386 struct dm_origin *o;
387
388 ol = &_dm_origins[origin_hash(origin)];
389 list_for_each_entry (o, ol, hash_list)
390 if (bdev_equal(o->dev->bdev, origin))
391 return o;
392
393 return NULL;
394}
395
396static void __insert_dm_origin(struct dm_origin *o)
397{
398 struct list_head *sl = &_dm_origins[origin_hash(o->dev->bdev)];
399 list_add_tail(&o->hash_list, sl);
400}
401
402static void __remove_dm_origin(struct dm_origin *o)
403{
404 list_del(&o->hash_list);
405}
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422static int __find_snapshots_sharing_cow(struct dm_snapshot *snap,
423 struct dm_snapshot **snap_src,
424 struct dm_snapshot **snap_dest,
425 struct dm_snapshot **snap_merge)
426{
427 struct dm_snapshot *s;
428 struct origin *o;
429 int count = 0;
430 int active;
431
432 o = __lookup_origin(snap->origin->bdev);
433 if (!o)
434 goto out;
435
436 list_for_each_entry(s, &o->snapshots, list) {
437 if (dm_target_is_snapshot_merge(s->ti) && snap_merge)
438 *snap_merge = s;
439 if (!bdev_equal(s->cow->bdev, snap->cow->bdev))
440 continue;
441
442 mutex_lock(&s->lock);
443 active = s->active;
444 mutex_unlock(&s->lock);
445
446 if (active) {
447 if (snap_src)
448 *snap_src = s;
449 } else if (snap_dest)
450 *snap_dest = s;
451
452 count++;
453 }
454
455out:
456 return count;
457}
458
459
460
461
462
463static int __validate_exception_handover(struct dm_snapshot *snap)
464{
465 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
466 struct dm_snapshot *snap_merge = NULL;
467
468
469 if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest,
470 &snap_merge) == 2) ||
471 snap_dest) {
472 snap->ti->error = "Snapshot cow pairing for exception "
473 "table handover failed";
474 return -EINVAL;
475 }
476
477
478
479
480
481 if (!snap_src)
482 return 0;
483
484
485
486
487 if (!dm_target_is_snapshot_merge(snap->ti))
488 return 1;
489
490
491
492
493 if (snap_merge) {
494 snap->ti->error = "A snapshot is already merging.";
495 return -EINVAL;
496 }
497
498 if (!snap_src->store->type->prepare_merge ||
499 !snap_src->store->type->commit_merge) {
500 snap->ti->error = "Snapshot exception store does not "
501 "support snapshot-merge.";
502 return -EINVAL;
503 }
504
505 return 1;
506}
507
508static void __insert_snapshot(struct origin *o, struct dm_snapshot *s)
509{
510 struct dm_snapshot *l;
511
512
513 list_for_each_entry(l, &o->snapshots, list)
514 if (l->store->chunk_size < s->store->chunk_size)
515 break;
516 list_add_tail(&s->list, &l->list);
517}
518
519
520
521
522
523
524
525
526
527static int register_snapshot(struct dm_snapshot *snap)
528{
529 struct origin *o, *new_o = NULL;
530 struct block_device *bdev = snap->origin->bdev;
531 int r = 0;
532
533 new_o = kmalloc(sizeof(*new_o), GFP_KERNEL);
534 if (!new_o)
535 return -ENOMEM;
536
537 down_write(&_origins_lock);
538
539 r = __validate_exception_handover(snap);
540 if (r < 0) {
541 kfree(new_o);
542 goto out;
543 }
544
545 o = __lookup_origin(bdev);
546 if (o)
547 kfree(new_o);
548 else {
549
550 o = new_o;
551
552
553 INIT_LIST_HEAD(&o->snapshots);
554 o->bdev = bdev;
555
556 __insert_origin(o);
557 }
558
559 __insert_snapshot(o, snap);
560
561out:
562 up_write(&_origins_lock);
563
564 return r;
565}
566
567
568
569
570static void reregister_snapshot(struct dm_snapshot *s)
571{
572 struct block_device *bdev = s->origin->bdev;
573
574 down_write(&_origins_lock);
575
576 list_del(&s->list);
577 __insert_snapshot(__lookup_origin(bdev), s);
578
579 up_write(&_origins_lock);
580}
581
582static void unregister_snapshot(struct dm_snapshot *s)
583{
584 struct origin *o;
585
586 down_write(&_origins_lock);
587 o = __lookup_origin(s->origin->bdev);
588
589 list_del(&s->list);
590 if (o && list_empty(&o->snapshots)) {
591 list_del(&o->hash_list);
592 kfree(o);
593 }
594
595 up_write(&_origins_lock);
596}
597
598
599
600
601
602
603static int dm_exception_table_init(struct dm_exception_table *et,
604 uint32_t size, unsigned hash_shift)
605{
606 unsigned int i;
607
608 et->hash_shift = hash_shift;
609 et->hash_mask = size - 1;
610 et->table = dm_vcalloc(size, sizeof(struct list_head));
611 if (!et->table)
612 return -ENOMEM;
613
614 for (i = 0; i < size; i++)
615 INIT_LIST_HEAD(et->table + i);
616
617 return 0;
618}
619
620static void dm_exception_table_exit(struct dm_exception_table *et,
621 struct kmem_cache *mem)
622{
623 struct list_head *slot;
624 struct dm_exception *ex, *next;
625 int i, size;
626
627 size = et->hash_mask + 1;
628 for (i = 0; i < size; i++) {
629 slot = et->table + i;
630
631 list_for_each_entry_safe (ex, next, slot, hash_list)
632 kmem_cache_free(mem, ex);
633 }
634
635 vfree(et->table);
636}
637
638static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk)
639{
640 return (chunk >> et->hash_shift) & et->hash_mask;
641}
642
643static void dm_remove_exception(struct dm_exception *e)
644{
645 list_del(&e->hash_list);
646}
647
648
649
650
651
652static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et,
653 chunk_t chunk)
654{
655 struct list_head *slot;
656 struct dm_exception *e;
657
658 slot = &et->table[exception_hash(et, chunk)];
659 list_for_each_entry (e, slot, hash_list)
660 if (chunk >= e->old_chunk &&
661 chunk <= e->old_chunk + dm_consecutive_chunk_count(e))
662 return e;
663
664 return NULL;
665}
666
667static struct dm_exception *alloc_completed_exception(gfp_t gfp)
668{
669 struct dm_exception *e;
670
671 e = kmem_cache_alloc(exception_cache, gfp);
672 if (!e && gfp == GFP_NOIO)
673 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
674
675 return e;
676}
677
678static void free_completed_exception(struct dm_exception *e)
679{
680 kmem_cache_free(exception_cache, e);
681}
682
683static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s)
684{
685 struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool,
686 GFP_NOIO);
687
688 atomic_inc(&s->pending_exceptions_count);
689 pe->snap = s;
690
691 return pe;
692}
693
694static void free_pending_exception(struct dm_snap_pending_exception *pe)
695{
696 struct dm_snapshot *s = pe->snap;
697
698 mempool_free(pe, s->pending_pool);
699 smp_mb__before_atomic();
700 atomic_dec(&s->pending_exceptions_count);
701}
702
703static void dm_insert_exception(struct dm_exception_table *eh,
704 struct dm_exception *new_e)
705{
706 struct list_head *l;
707 struct dm_exception *e = NULL;
708
709 l = &eh->table[exception_hash(eh, new_e->old_chunk)];
710
711
712 if (!eh->hash_shift)
713 goto out;
714
715
716 list_for_each_entry_reverse(e, l, hash_list) {
717
718 if (new_e->old_chunk == (e->old_chunk +
719 dm_consecutive_chunk_count(e) + 1) &&
720 new_e->new_chunk == (dm_chunk_number(e->new_chunk) +
721 dm_consecutive_chunk_count(e) + 1)) {
722 dm_consecutive_chunk_count_inc(e);
723 free_completed_exception(new_e);
724 return;
725 }
726
727
728 if (new_e->old_chunk == (e->old_chunk - 1) &&
729 new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) {
730 dm_consecutive_chunk_count_inc(e);
731 e->old_chunk--;
732 e->new_chunk--;
733 free_completed_exception(new_e);
734 return;
735 }
736
737 if (new_e->old_chunk > e->old_chunk)
738 break;
739 }
740
741out:
742 list_add(&new_e->hash_list, e ? &e->hash_list : l);
743}
744
745
746
747
748
749static int dm_add_exception(void *context, chunk_t old, chunk_t new)
750{
751 struct dm_snapshot *s = context;
752 struct dm_exception *e;
753
754 e = alloc_completed_exception(GFP_KERNEL);
755 if (!e)
756 return -ENOMEM;
757
758 e->old_chunk = old;
759
760
761 e->new_chunk = new;
762
763 dm_insert_exception(&s->complete, e);
764
765 return 0;
766}
767
768
769
770
771
772static uint32_t __minimum_chunk_size(struct origin *o)
773{
774 struct dm_snapshot *snap;
775 unsigned chunk_size = 0;
776
777 if (o)
778 list_for_each_entry(snap, &o->snapshots, list)
779 chunk_size = min_not_zero(chunk_size,
780 snap->store->chunk_size);
781
782 return (uint32_t) chunk_size;
783}
784
785
786
787
788static int calc_max_buckets(void)
789{
790
791 unsigned long mem = 2 * 1024 * 1024;
792 mem /= sizeof(struct list_head);
793
794 return mem;
795}
796
797
798
799
800static int init_hash_tables(struct dm_snapshot *s)
801{
802 sector_t hash_size, cow_dev_size, max_buckets;
803
804
805
806
807
808 cow_dev_size = get_dev_size(s->cow->bdev);
809 max_buckets = calc_max_buckets();
810
811 hash_size = cow_dev_size >> s->store->chunk_shift;
812 hash_size = min(hash_size, max_buckets);
813
814 if (hash_size < 64)
815 hash_size = 64;
816 hash_size = rounddown_pow_of_two(hash_size);
817 if (dm_exception_table_init(&s->complete, hash_size,
818 DM_CHUNK_CONSECUTIVE_BITS))
819 return -ENOMEM;
820
821
822
823
824
825 hash_size >>= 3;
826 if (hash_size < 64)
827 hash_size = 64;
828
829 if (dm_exception_table_init(&s->pending, hash_size, 0)) {
830 dm_exception_table_exit(&s->complete, exception_cache);
831 return -ENOMEM;
832 }
833
834 return 0;
835}
836
837static void merge_shutdown(struct dm_snapshot *s)
838{
839 clear_bit_unlock(RUNNING_MERGE, &s->state_bits);
840 smp_mb__after_atomic();
841 wake_up_bit(&s->state_bits, RUNNING_MERGE);
842}
843
844static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s)
845{
846 s->first_merging_chunk = 0;
847 s->num_merging_chunks = 0;
848
849 return bio_list_get(&s->bios_queued_during_merge);
850}
851
852
853
854
855static int __remove_single_exception_chunk(struct dm_snapshot *s,
856 chunk_t old_chunk)
857{
858 struct dm_exception *e;
859
860 e = dm_lookup_exception(&s->complete, old_chunk);
861 if (!e) {
862 DMERR("Corruption detected: exception for block %llu is "
863 "on disk but not in memory",
864 (unsigned long long)old_chunk);
865 return -EINVAL;
866 }
867
868
869
870
871 if (!dm_consecutive_chunk_count(e)) {
872 dm_remove_exception(e);
873 free_completed_exception(e);
874 return 0;
875 }
876
877
878
879
880
881
882
883
884
885 if (old_chunk == e->old_chunk) {
886 e->old_chunk++;
887 e->new_chunk++;
888 } else if (old_chunk != e->old_chunk +
889 dm_consecutive_chunk_count(e)) {
890 DMERR("Attempt to merge block %llu from the "
891 "middle of a chunk range [%llu - %llu]",
892 (unsigned long long)old_chunk,
893 (unsigned long long)e->old_chunk,
894 (unsigned long long)
895 e->old_chunk + dm_consecutive_chunk_count(e));
896 return -EINVAL;
897 }
898
899 dm_consecutive_chunk_count_dec(e);
900
901 return 0;
902}
903
904static void flush_bios(struct bio *bio);
905
906static int remove_single_exception_chunk(struct dm_snapshot *s)
907{
908 struct bio *b = NULL;
909 int r;
910 chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1;
911
912 mutex_lock(&s->lock);
913
914
915
916
917
918 do {
919 r = __remove_single_exception_chunk(s, old_chunk);
920 if (r)
921 goto out;
922 } while (old_chunk-- > s->first_merging_chunk);
923
924 b = __release_queued_bios_after_merge(s);
925
926out:
927 mutex_unlock(&s->lock);
928 if (b)
929 flush_bios(b);
930
931 return r;
932}
933
934static int origin_write_extent(struct dm_snapshot *merging_snap,
935 sector_t sector, unsigned chunk_size);
936
937static void merge_callback(int read_err, unsigned long write_err,
938 void *context);
939
940static uint64_t read_pending_exceptions_done_count(void)
941{
942 uint64_t pending_exceptions_done;
943
944 spin_lock(&_pending_exceptions_done_spinlock);
945 pending_exceptions_done = _pending_exceptions_done_count;
946 spin_unlock(&_pending_exceptions_done_spinlock);
947
948 return pending_exceptions_done;
949}
950
951static void increment_pending_exceptions_done_count(void)
952{
953 spin_lock(&_pending_exceptions_done_spinlock);
954 _pending_exceptions_done_count++;
955 spin_unlock(&_pending_exceptions_done_spinlock);
956
957 wake_up_all(&_pending_exceptions_done);
958}
959
960static void snapshot_merge_next_chunks(struct dm_snapshot *s)
961{
962 int i, linear_chunks;
963 chunk_t old_chunk, new_chunk;
964 struct dm_io_region src, dest;
965 sector_t io_size;
966 uint64_t previous_count;
967
968 BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
969 if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
970 goto shut;
971
972
973
974
975 if (!s->valid) {
976 DMERR("Snapshot is invalid: can't merge");
977 goto shut;
978 }
979
980 linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
981 &new_chunk);
982 if (linear_chunks <= 0) {
983 if (linear_chunks < 0) {
984 DMERR("Read error in exception store: "
985 "shutting down merge");
986 mutex_lock(&s->lock);
987 s->merge_failed = 1;
988 mutex_unlock(&s->lock);
989 }
990 goto shut;
991 }
992
993
994 old_chunk = old_chunk + 1 - linear_chunks;
995 new_chunk = new_chunk + 1 - linear_chunks;
996
997
998
999
1000
1001 io_size = linear_chunks * s->store->chunk_size;
1002
1003 dest.bdev = s->origin->bdev;
1004 dest.sector = chunk_to_sector(s->store, old_chunk);
1005 dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector);
1006
1007 src.bdev = s->cow->bdev;
1008 src.sector = chunk_to_sector(s->store, new_chunk);
1009 src.count = dest.count;
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020 previous_count = read_pending_exceptions_done_count();
1021 while (origin_write_extent(s, dest.sector, io_size)) {
1022 wait_event(_pending_exceptions_done,
1023 (read_pending_exceptions_done_count() !=
1024 previous_count));
1025
1026 previous_count = read_pending_exceptions_done_count();
1027 }
1028
1029 mutex_lock(&s->lock);
1030 s->first_merging_chunk = old_chunk;
1031 s->num_merging_chunks = linear_chunks;
1032 mutex_unlock(&s->lock);
1033
1034
1035 for (i = 0; i < linear_chunks; i++)
1036 __check_for_conflicting_io(s, old_chunk + i);
1037
1038 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s);
1039 return;
1040
1041shut:
1042 merge_shutdown(s);
1043}
1044
1045static void error_bios(struct bio *bio);
1046
1047static void merge_callback(int read_err, unsigned long write_err, void *context)
1048{
1049 struct dm_snapshot *s = context;
1050 struct bio *b = NULL;
1051
1052 if (read_err || write_err) {
1053 if (read_err)
1054 DMERR("Read error: shutting down merge.");
1055 else
1056 DMERR("Write error: shutting down merge.");
1057 goto shut;
1058 }
1059
1060 if (s->store->type->commit_merge(s->store,
1061 s->num_merging_chunks) < 0) {
1062 DMERR("Write error in exception store: shutting down merge");
1063 goto shut;
1064 }
1065
1066 if (remove_single_exception_chunk(s) < 0)
1067 goto shut;
1068
1069 snapshot_merge_next_chunks(s);
1070
1071 return;
1072
1073shut:
1074 mutex_lock(&s->lock);
1075 s->merge_failed = 1;
1076 b = __release_queued_bios_after_merge(s);
1077 mutex_unlock(&s->lock);
1078 error_bios(b);
1079
1080 merge_shutdown(s);
1081}
1082
1083static void start_merge(struct dm_snapshot *s)
1084{
1085 if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits))
1086 snapshot_merge_next_chunks(s);
1087}
1088
1089
1090
1091
1092static void stop_merge(struct dm_snapshot *s)
1093{
1094 set_bit(SHUTDOWN_MERGE, &s->state_bits);
1095 wait_on_bit(&s->state_bits, RUNNING_MERGE, TASK_UNINTERRUPTIBLE);
1096 clear_bit(SHUTDOWN_MERGE, &s->state_bits);
1097}
1098
1099
1100
1101
1102static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1103{
1104 struct dm_snapshot *s;
1105 int i;
1106 int r = -EINVAL;
1107 char *origin_path, *cow_path;
1108 dev_t origin_dev, cow_dev;
1109 unsigned args_used, num_flush_bios = 1;
1110 fmode_t origin_mode = FMODE_READ;
1111
1112 if (argc != 4) {
1113 ti->error = "requires exactly 4 arguments";
1114 r = -EINVAL;
1115 goto bad;
1116 }
1117
1118 if (dm_target_is_snapshot_merge(ti)) {
1119 num_flush_bios = 2;
1120 origin_mode = FMODE_WRITE;
1121 }
1122
1123 s = kmalloc(sizeof(*s), GFP_KERNEL);
1124 if (!s) {
1125 ti->error = "Cannot allocate private snapshot structure";
1126 r = -ENOMEM;
1127 goto bad;
1128 }
1129
1130 origin_path = argv[0];
1131 argv++;
1132 argc--;
1133
1134 r = dm_get_device(ti, origin_path, origin_mode, &s->origin);
1135 if (r) {
1136 ti->error = "Cannot get origin device";
1137 goto bad_origin;
1138 }
1139 origin_dev = s->origin->bdev->bd_dev;
1140
1141 cow_path = argv[0];
1142 argv++;
1143 argc--;
1144
1145 cow_dev = dm_get_dev_t(cow_path);
1146 if (cow_dev && cow_dev == origin_dev) {
1147 ti->error = "COW device cannot be the same as origin device";
1148 r = -EINVAL;
1149 goto bad_cow;
1150 }
1151
1152 r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow);
1153 if (r) {
1154 ti->error = "Cannot get COW device";
1155 goto bad_cow;
1156 }
1157
1158 r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store);
1159 if (r) {
1160 ti->error = "Couldn't create exception store";
1161 r = -EINVAL;
1162 goto bad_store;
1163 }
1164
1165 argv += args_used;
1166 argc -= args_used;
1167
1168 s->ti = ti;
1169 s->valid = 1;
1170 s->snapshot_overflowed = 0;
1171 s->active = 0;
1172 atomic_set(&s->pending_exceptions_count, 0);
1173 s->exception_start_sequence = 0;
1174 s->exception_complete_sequence = 0;
1175 INIT_LIST_HEAD(&s->out_of_order_list);
1176 mutex_init(&s->lock);
1177 INIT_LIST_HEAD(&s->list);
1178 spin_lock_init(&s->pe_lock);
1179 s->state_bits = 0;
1180 s->merge_failed = 0;
1181 s->first_merging_chunk = 0;
1182 s->num_merging_chunks = 0;
1183 bio_list_init(&s->bios_queued_during_merge);
1184
1185
1186 if (init_hash_tables(s)) {
1187 ti->error = "Unable to allocate hash table space";
1188 r = -ENOMEM;
1189 goto bad_hash_tables;
1190 }
1191
1192 s->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle);
1193 if (IS_ERR(s->kcopyd_client)) {
1194 r = PTR_ERR(s->kcopyd_client);
1195 ti->error = "Could not create kcopyd client";
1196 goto bad_kcopyd;
1197 }
1198
1199 s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache);
1200 if (!s->pending_pool) {
1201 ti->error = "Could not allocate mempool for pending exceptions";
1202 r = -ENOMEM;
1203 goto bad_pending_pool;
1204 }
1205
1206 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1207 INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]);
1208
1209 spin_lock_init(&s->tracked_chunk_lock);
1210
1211 ti->private = s;
1212 ti->num_flush_bios = num_flush_bios;
1213 ti->per_io_data_size = sizeof(struct dm_snap_tracked_chunk);
1214
1215
1216
1217 r = register_snapshot(s);
1218 if (r == -ENOMEM) {
1219 ti->error = "Snapshot origin struct allocation failed";
1220 goto bad_load_and_register;
1221 } else if (r < 0) {
1222
1223 goto bad_load_and_register;
1224 }
1225
1226
1227
1228
1229
1230
1231
1232 if (r > 0) {
1233 s->store->chunk_size = 0;
1234 return 0;
1235 }
1236
1237 r = s->store->type->read_metadata(s->store, dm_add_exception,
1238 (void *)s);
1239 if (r < 0) {
1240 ti->error = "Failed to read snapshot metadata";
1241 goto bad_read_metadata;
1242 } else if (r > 0) {
1243 s->valid = 0;
1244 DMWARN("Snapshot is marked invalid.");
1245 }
1246
1247 if (!s->store->chunk_size) {
1248 ti->error = "Chunk size not set";
1249 goto bad_read_metadata;
1250 }
1251
1252 r = dm_set_target_max_io_len(ti, s->store->chunk_size);
1253 if (r)
1254 goto bad_read_metadata;
1255
1256 return 0;
1257
1258bad_read_metadata:
1259 unregister_snapshot(s);
1260
1261bad_load_and_register:
1262 mempool_destroy(s->pending_pool);
1263
1264bad_pending_pool:
1265 dm_kcopyd_client_destroy(s->kcopyd_client);
1266
1267bad_kcopyd:
1268 dm_exception_table_exit(&s->pending, pending_cache);
1269 dm_exception_table_exit(&s->complete, exception_cache);
1270
1271bad_hash_tables:
1272 dm_exception_store_destroy(s->store);
1273
1274bad_store:
1275 dm_put_device(ti, s->cow);
1276
1277bad_cow:
1278 dm_put_device(ti, s->origin);
1279
1280bad_origin:
1281 kfree(s);
1282
1283bad:
1284 return r;
1285}
1286
1287static void __free_exceptions(struct dm_snapshot *s)
1288{
1289 dm_kcopyd_client_destroy(s->kcopyd_client);
1290 s->kcopyd_client = NULL;
1291
1292 dm_exception_table_exit(&s->pending, pending_cache);
1293 dm_exception_table_exit(&s->complete, exception_cache);
1294}
1295
1296static void __handover_exceptions(struct dm_snapshot *snap_src,
1297 struct dm_snapshot *snap_dest)
1298{
1299 union {
1300 struct dm_exception_table table_swap;
1301 struct dm_exception_store *store_swap;
1302 } u;
1303
1304
1305
1306
1307 u.table_swap = snap_dest->complete;
1308 snap_dest->complete = snap_src->complete;
1309 snap_src->complete = u.table_swap;
1310
1311 u.store_swap = snap_dest->store;
1312 snap_dest->store = snap_src->store;
1313 snap_dest->store->userspace_supports_overflow = u.store_swap->userspace_supports_overflow;
1314 snap_src->store = u.store_swap;
1315
1316 snap_dest->store->snap = snap_dest;
1317 snap_src->store->snap = snap_src;
1318
1319 snap_dest->ti->max_io_len = snap_dest->store->chunk_size;
1320 snap_dest->valid = snap_src->valid;
1321 snap_dest->snapshot_overflowed = snap_src->snapshot_overflowed;
1322
1323
1324
1325
1326 snap_src->valid = 0;
1327}
1328
1329static void snapshot_dtr(struct dm_target *ti)
1330{
1331#ifdef CONFIG_DM_DEBUG
1332 int i;
1333#endif
1334 struct dm_snapshot *s = ti->private;
1335 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1336
1337 down_read(&_origins_lock);
1338
1339 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1340 if (snap_src && snap_dest && (s == snap_src)) {
1341 mutex_lock(&snap_dest->lock);
1342 snap_dest->valid = 0;
1343 mutex_unlock(&snap_dest->lock);
1344 DMERR("Cancelling snapshot handover.");
1345 }
1346 up_read(&_origins_lock);
1347
1348 if (dm_target_is_snapshot_merge(ti))
1349 stop_merge(s);
1350
1351
1352
1353 unregister_snapshot(s);
1354
1355 while (atomic_read(&s->pending_exceptions_count))
1356 msleep(1);
1357
1358
1359
1360
1361 smp_mb();
1362
1363#ifdef CONFIG_DM_DEBUG
1364 for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++)
1365 BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i]));
1366#endif
1367
1368 __free_exceptions(s);
1369
1370 mempool_destroy(s->pending_pool);
1371
1372 dm_exception_store_destroy(s->store);
1373
1374 mutex_destroy(&s->lock);
1375
1376 dm_put_device(ti, s->cow);
1377
1378 dm_put_device(ti, s->origin);
1379
1380 kfree(s);
1381}
1382
1383
1384
1385
1386static void flush_bios(struct bio *bio)
1387{
1388 struct bio *n;
1389
1390 while (bio) {
1391 n = bio->bi_next;
1392 bio->bi_next = NULL;
1393 generic_make_request(bio);
1394 bio = n;
1395 }
1396}
1397
1398static int do_origin(struct dm_dev *origin, struct bio *bio);
1399
1400
1401
1402
1403static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio)
1404{
1405 struct bio *n;
1406 int r;
1407
1408 while (bio) {
1409 n = bio->bi_next;
1410 bio->bi_next = NULL;
1411 r = do_origin(s->origin, bio);
1412 if (r == DM_MAPIO_REMAPPED)
1413 generic_make_request(bio);
1414 bio = n;
1415 }
1416}
1417
1418
1419
1420
1421static void error_bios(struct bio *bio)
1422{
1423 struct bio *n;
1424
1425 while (bio) {
1426 n = bio->bi_next;
1427 bio->bi_next = NULL;
1428 bio_io_error(bio);
1429 bio = n;
1430 }
1431}
1432
1433static void __invalidate_snapshot(struct dm_snapshot *s, int err)
1434{
1435 if (!s->valid)
1436 return;
1437
1438 if (err == -EIO)
1439 DMERR("Invalidating snapshot: Error reading/writing.");
1440 else if (err == -ENOMEM)
1441 DMERR("Invalidating snapshot: Unable to allocate exception.");
1442
1443 if (s->store->type->drop_snapshot)
1444 s->store->type->drop_snapshot(s->store);
1445
1446 s->valid = 0;
1447
1448 dm_table_event(s->ti->table);
1449}
1450
1451static void pending_complete(void *context, int success)
1452{
1453 struct dm_snap_pending_exception *pe = context;
1454 struct dm_exception *e;
1455 struct dm_snapshot *s = pe->snap;
1456 struct bio *origin_bios = NULL;
1457 struct bio *snapshot_bios = NULL;
1458 struct bio *full_bio = NULL;
1459 int error = 0;
1460
1461 if (!success) {
1462
1463 mutex_lock(&s->lock);
1464 __invalidate_snapshot(s, -EIO);
1465 error = 1;
1466 goto out;
1467 }
1468
1469 e = alloc_completed_exception(GFP_NOIO);
1470 if (!e) {
1471 mutex_lock(&s->lock);
1472 __invalidate_snapshot(s, -ENOMEM);
1473 error = 1;
1474 goto out;
1475 }
1476 *e = pe->e;
1477
1478 mutex_lock(&s->lock);
1479 if (!s->valid) {
1480 free_completed_exception(e);
1481 error = 1;
1482 goto out;
1483 }
1484
1485
1486 __check_for_conflicting_io(s, pe->e.old_chunk);
1487
1488
1489
1490
1491
1492 dm_insert_exception(&s->complete, e);
1493
1494out:
1495 dm_remove_exception(&pe->e);
1496 snapshot_bios = bio_list_get(&pe->snapshot_bios);
1497 origin_bios = bio_list_get(&pe->origin_bios);
1498 full_bio = pe->full_bio;
1499 if (full_bio)
1500 full_bio->bi_end_io = pe->full_bio_end_io;
1501 increment_pending_exceptions_done_count();
1502
1503 mutex_unlock(&s->lock);
1504
1505
1506 if (error) {
1507 if (full_bio)
1508 bio_io_error(full_bio);
1509 error_bios(snapshot_bios);
1510 } else {
1511 if (full_bio)
1512 bio_endio(full_bio);
1513 flush_bios(snapshot_bios);
1514 }
1515
1516 retry_origin_bios(s, origin_bios);
1517
1518 free_pending_exception(pe);
1519}
1520
1521static void complete_exception(struct dm_snap_pending_exception *pe)
1522{
1523 struct dm_snapshot *s = pe->snap;
1524
1525
1526 s->store->type->commit_exception(s->store, &pe->e, !pe->copy_error,
1527 pending_complete, pe);
1528}
1529
1530
1531
1532
1533
1534static void copy_callback(int read_err, unsigned long write_err, void *context)
1535{
1536 struct dm_snap_pending_exception *pe = context;
1537 struct dm_snapshot *s = pe->snap;
1538
1539 pe->copy_error = read_err || write_err;
1540
1541 if (pe->exception_sequence == s->exception_complete_sequence) {
1542 s->exception_complete_sequence++;
1543 complete_exception(pe);
1544
1545 while (!list_empty(&s->out_of_order_list)) {
1546 pe = list_entry(s->out_of_order_list.next,
1547 struct dm_snap_pending_exception, out_of_order_entry);
1548 if (pe->exception_sequence != s->exception_complete_sequence)
1549 break;
1550 s->exception_complete_sequence++;
1551 list_del(&pe->out_of_order_entry);
1552 complete_exception(pe);
1553 }
1554 } else {
1555 struct list_head *lh;
1556 struct dm_snap_pending_exception *pe2;
1557
1558 list_for_each_prev(lh, &s->out_of_order_list) {
1559 pe2 = list_entry(lh, struct dm_snap_pending_exception, out_of_order_entry);
1560 if (pe2->exception_sequence < pe->exception_sequence)
1561 break;
1562 }
1563 list_add(&pe->out_of_order_entry, lh);
1564 }
1565}
1566
1567
1568
1569
1570static void start_copy(struct dm_snap_pending_exception *pe)
1571{
1572 struct dm_snapshot *s = pe->snap;
1573 struct dm_io_region src, dest;
1574 struct block_device *bdev = s->origin->bdev;
1575 sector_t dev_size;
1576
1577 dev_size = get_dev_size(bdev);
1578
1579 src.bdev = bdev;
1580 src.sector = chunk_to_sector(s->store, pe->e.old_chunk);
1581 src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector);
1582
1583 dest.bdev = s->cow->bdev;
1584 dest.sector = chunk_to_sector(s->store, pe->e.new_chunk);
1585 dest.count = src.count;
1586
1587
1588 dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe);
1589}
1590
1591static void full_bio_end_io(struct bio *bio)
1592{
1593 void *callback_data = bio->bi_private;
1594
1595 dm_kcopyd_do_callback(callback_data, 0, bio->bi_status ? 1 : 0);
1596}
1597
1598static void start_full_bio(struct dm_snap_pending_exception *pe,
1599 struct bio *bio)
1600{
1601 struct dm_snapshot *s = pe->snap;
1602 void *callback_data;
1603
1604 pe->full_bio = bio;
1605 pe->full_bio_end_io = bio->bi_end_io;
1606
1607 callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client,
1608 copy_callback, pe);
1609
1610 bio->bi_end_io = full_bio_end_io;
1611 bio->bi_private = callback_data;
1612
1613 generic_make_request(bio);
1614}
1615
1616static struct dm_snap_pending_exception *
1617__lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk)
1618{
1619 struct dm_exception *e = dm_lookup_exception(&s->pending, chunk);
1620
1621 if (!e)
1622 return NULL;
1623
1624 return container_of(e, struct dm_snap_pending_exception, e);
1625}
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635static struct dm_snap_pending_exception *
1636__find_pending_exception(struct dm_snapshot *s,
1637 struct dm_snap_pending_exception *pe, chunk_t chunk)
1638{
1639 struct dm_snap_pending_exception *pe2;
1640
1641 pe2 = __lookup_pending_exception(s, chunk);
1642 if (pe2) {
1643 free_pending_exception(pe);
1644 return pe2;
1645 }
1646
1647 pe->e.old_chunk = chunk;
1648 bio_list_init(&pe->origin_bios);
1649 bio_list_init(&pe->snapshot_bios);
1650 pe->started = 0;
1651 pe->full_bio = NULL;
1652
1653 if (s->store->type->prepare_exception(s->store, &pe->e)) {
1654 free_pending_exception(pe);
1655 return NULL;
1656 }
1657
1658 pe->exception_sequence = s->exception_start_sequence++;
1659
1660 dm_insert_exception(&s->pending, &pe->e);
1661
1662 return pe;
1663}
1664
1665static void remap_exception(struct dm_snapshot *s, struct dm_exception *e,
1666 struct bio *bio, chunk_t chunk)
1667{
1668 bio_set_dev(bio, s->cow->bdev);
1669 bio->bi_iter.bi_sector =
1670 chunk_to_sector(s->store, dm_chunk_number(e->new_chunk) +
1671 (chunk - e->old_chunk)) +
1672 (bio->bi_iter.bi_sector & s->store->chunk_mask);
1673}
1674
1675static int snapshot_map(struct dm_target *ti, struct bio *bio)
1676{
1677 struct dm_exception *e;
1678 struct dm_snapshot *s = ti->private;
1679 int r = DM_MAPIO_REMAPPED;
1680 chunk_t chunk;
1681 struct dm_snap_pending_exception *pe = NULL;
1682
1683 init_tracked_chunk(bio);
1684
1685 if (bio->bi_opf & REQ_PREFLUSH) {
1686 bio_set_dev(bio, s->cow->bdev);
1687 return DM_MAPIO_REMAPPED;
1688 }
1689
1690 chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector);
1691
1692
1693
1694 if (!s->valid)
1695 return DM_MAPIO_KILL;
1696
1697
1698
1699 mutex_lock(&s->lock);
1700
1701 if (!s->valid || (unlikely(s->snapshot_overflowed) &&
1702 bio_data_dir(bio) == WRITE)) {
1703 r = DM_MAPIO_KILL;
1704 goto out_unlock;
1705 }
1706
1707
1708 e = dm_lookup_exception(&s->complete, chunk);
1709 if (e) {
1710 remap_exception(s, e, bio, chunk);
1711 goto out_unlock;
1712 }
1713
1714
1715
1716
1717
1718
1719 if (bio_data_dir(bio) == WRITE) {
1720 pe = __lookup_pending_exception(s, chunk);
1721 if (!pe) {
1722 mutex_unlock(&s->lock);
1723 pe = alloc_pending_exception(s);
1724 mutex_lock(&s->lock);
1725
1726 if (!s->valid || s->snapshot_overflowed) {
1727 free_pending_exception(pe);
1728 r = DM_MAPIO_KILL;
1729 goto out_unlock;
1730 }
1731
1732 e = dm_lookup_exception(&s->complete, chunk);
1733 if (e) {
1734 free_pending_exception(pe);
1735 remap_exception(s, e, bio, chunk);
1736 goto out_unlock;
1737 }
1738
1739 pe = __find_pending_exception(s, pe, chunk);
1740 if (!pe) {
1741 if (s->store->userspace_supports_overflow) {
1742 s->snapshot_overflowed = 1;
1743 DMERR("Snapshot overflowed: Unable to allocate exception.");
1744 } else
1745 __invalidate_snapshot(s, -ENOMEM);
1746 r = DM_MAPIO_KILL;
1747 goto out_unlock;
1748 }
1749 }
1750
1751 remap_exception(s, &pe->e, bio, chunk);
1752
1753 r = DM_MAPIO_SUBMITTED;
1754
1755 if (!pe->started &&
1756 bio->bi_iter.bi_size ==
1757 (s->store->chunk_size << SECTOR_SHIFT)) {
1758 pe->started = 1;
1759 mutex_unlock(&s->lock);
1760 start_full_bio(pe, bio);
1761 goto out;
1762 }
1763
1764 bio_list_add(&pe->snapshot_bios, bio);
1765
1766 if (!pe->started) {
1767
1768 pe->started = 1;
1769 mutex_unlock(&s->lock);
1770 start_copy(pe);
1771 goto out;
1772 }
1773 } else {
1774 bio_set_dev(bio, s->origin->bdev);
1775 track_chunk(s, bio, chunk);
1776 }
1777
1778out_unlock:
1779 mutex_unlock(&s->lock);
1780out:
1781 return r;
1782}
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796static int snapshot_merge_map(struct dm_target *ti, struct bio *bio)
1797{
1798 struct dm_exception *e;
1799 struct dm_snapshot *s = ti->private;
1800 int r = DM_MAPIO_REMAPPED;
1801 chunk_t chunk;
1802
1803 init_tracked_chunk(bio);
1804
1805 if (bio->bi_opf & REQ_PREFLUSH) {
1806 if (!dm_bio_get_target_bio_nr(bio))
1807 bio_set_dev(bio, s->origin->bdev);
1808 else
1809 bio_set_dev(bio, s->cow->bdev);
1810 return DM_MAPIO_REMAPPED;
1811 }
1812
1813 chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector);
1814
1815 mutex_lock(&s->lock);
1816
1817
1818 if (!s->valid)
1819 goto redirect_to_origin;
1820
1821
1822 e = dm_lookup_exception(&s->complete, chunk);
1823 if (e) {
1824
1825 if (bio_data_dir(bio) == WRITE &&
1826 chunk >= s->first_merging_chunk &&
1827 chunk < (s->first_merging_chunk +
1828 s->num_merging_chunks)) {
1829 bio_set_dev(bio, s->origin->bdev);
1830 bio_list_add(&s->bios_queued_during_merge, bio);
1831 r = DM_MAPIO_SUBMITTED;
1832 goto out_unlock;
1833 }
1834
1835 remap_exception(s, e, bio, chunk);
1836
1837 if (bio_data_dir(bio) == WRITE)
1838 track_chunk(s, bio, chunk);
1839 goto out_unlock;
1840 }
1841
1842redirect_to_origin:
1843 bio_set_dev(bio, s->origin->bdev);
1844
1845 if (bio_data_dir(bio) == WRITE) {
1846 mutex_unlock(&s->lock);
1847 return do_origin(s->origin, bio);
1848 }
1849
1850out_unlock:
1851 mutex_unlock(&s->lock);
1852
1853 return r;
1854}
1855
1856static int snapshot_end_io(struct dm_target *ti, struct bio *bio,
1857 blk_status_t *error)
1858{
1859 struct dm_snapshot *s = ti->private;
1860
1861 if (is_bio_tracked(bio))
1862 stop_tracking_chunk(s, bio);
1863
1864 return DM_ENDIO_DONE;
1865}
1866
1867static void snapshot_merge_presuspend(struct dm_target *ti)
1868{
1869 struct dm_snapshot *s = ti->private;
1870
1871 stop_merge(s);
1872}
1873
1874static int snapshot_preresume(struct dm_target *ti)
1875{
1876 int r = 0;
1877 struct dm_snapshot *s = ti->private;
1878 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL;
1879
1880 down_read(&_origins_lock);
1881 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1882 if (snap_src && snap_dest) {
1883 mutex_lock(&snap_src->lock);
1884 if (s == snap_src) {
1885 DMERR("Unable to resume snapshot source until "
1886 "handover completes.");
1887 r = -EINVAL;
1888 } else if (!dm_suspended(snap_src->ti)) {
1889 DMERR("Unable to perform snapshot handover until "
1890 "source is suspended.");
1891 r = -EINVAL;
1892 }
1893 mutex_unlock(&snap_src->lock);
1894 }
1895 up_read(&_origins_lock);
1896
1897 return r;
1898}
1899
1900static void snapshot_resume(struct dm_target *ti)
1901{
1902 struct dm_snapshot *s = ti->private;
1903 struct dm_snapshot *snap_src = NULL, *snap_dest = NULL, *snap_merging = NULL;
1904 struct dm_origin *o;
1905 struct mapped_device *origin_md = NULL;
1906 bool must_restart_merging = false;
1907
1908 down_read(&_origins_lock);
1909
1910 o = __lookup_dm_origin(s->origin->bdev);
1911 if (o)
1912 origin_md = dm_table_get_md(o->ti->table);
1913 if (!origin_md) {
1914 (void) __find_snapshots_sharing_cow(s, NULL, NULL, &snap_merging);
1915 if (snap_merging)
1916 origin_md = dm_table_get_md(snap_merging->ti->table);
1917 }
1918 if (origin_md == dm_table_get_md(ti->table))
1919 origin_md = NULL;
1920 if (origin_md) {
1921 if (dm_hold(origin_md))
1922 origin_md = NULL;
1923 }
1924
1925 up_read(&_origins_lock);
1926
1927 if (origin_md) {
1928 dm_internal_suspend_fast(origin_md);
1929 if (snap_merging && test_bit(RUNNING_MERGE, &snap_merging->state_bits)) {
1930 must_restart_merging = true;
1931 stop_merge(snap_merging);
1932 }
1933 }
1934
1935 down_read(&_origins_lock);
1936
1937 (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL);
1938 if (snap_src && snap_dest) {
1939 mutex_lock(&snap_src->lock);
1940 mutex_lock_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING);
1941 __handover_exceptions(snap_src, snap_dest);
1942 mutex_unlock(&snap_dest->lock);
1943 mutex_unlock(&snap_src->lock);
1944 }
1945
1946 up_read(&_origins_lock);
1947
1948 if (origin_md) {
1949 if (must_restart_merging)
1950 start_merge(snap_merging);
1951 dm_internal_resume_fast(origin_md);
1952 dm_put(origin_md);
1953 }
1954
1955
1956 reregister_snapshot(s);
1957
1958 mutex_lock(&s->lock);
1959 s->active = 1;
1960 mutex_unlock(&s->lock);
1961}
1962
1963static uint32_t get_origin_minimum_chunksize(struct block_device *bdev)
1964{
1965 uint32_t min_chunksize;
1966
1967 down_read(&_origins_lock);
1968 min_chunksize = __minimum_chunk_size(__lookup_origin(bdev));
1969 up_read(&_origins_lock);
1970
1971 return min_chunksize;
1972}
1973
1974static void snapshot_merge_resume(struct dm_target *ti)
1975{
1976 struct dm_snapshot *s = ti->private;
1977
1978
1979
1980
1981 snapshot_resume(ti);
1982
1983
1984
1985
1986 ti->max_io_len = get_origin_minimum_chunksize(s->origin->bdev);
1987
1988 start_merge(s);
1989}
1990
1991static void snapshot_status(struct dm_target *ti, status_type_t type,
1992 unsigned status_flags, char *result, unsigned maxlen)
1993{
1994 unsigned sz = 0;
1995 struct dm_snapshot *snap = ti->private;
1996
1997 switch (type) {
1998 case STATUSTYPE_INFO:
1999
2000 mutex_lock(&snap->lock);
2001
2002 if (!snap->valid)
2003 DMEMIT("Invalid");
2004 else if (snap->merge_failed)
2005 DMEMIT("Merge failed");
2006 else if (snap->snapshot_overflowed)
2007 DMEMIT("Overflow");
2008 else {
2009 if (snap->store->type->usage) {
2010 sector_t total_sectors, sectors_allocated,
2011 metadata_sectors;
2012 snap->store->type->usage(snap->store,
2013 &total_sectors,
2014 §ors_allocated,
2015 &metadata_sectors);
2016 DMEMIT("%llu/%llu %llu",
2017 (unsigned long long)sectors_allocated,
2018 (unsigned long long)total_sectors,
2019 (unsigned long long)metadata_sectors);
2020 }
2021 else
2022 DMEMIT("Unknown");
2023 }
2024
2025 mutex_unlock(&snap->lock);
2026
2027 break;
2028
2029 case STATUSTYPE_TABLE:
2030
2031
2032
2033
2034
2035 DMEMIT("%s %s", snap->origin->name, snap->cow->name);
2036 snap->store->type->status(snap->store, type, result + sz,
2037 maxlen - sz);
2038 break;
2039 }
2040}
2041
2042static int snapshot_iterate_devices(struct dm_target *ti,
2043 iterate_devices_callout_fn fn, void *data)
2044{
2045 struct dm_snapshot *snap = ti->private;
2046 int r;
2047
2048 r = fn(ti, snap->origin, 0, ti->len, data);
2049
2050 if (!r)
2051 r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data);
2052
2053 return r;
2054}
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071static int __origin_write(struct list_head *snapshots, sector_t sector,
2072 struct bio *bio)
2073{
2074 int r = DM_MAPIO_REMAPPED;
2075 struct dm_snapshot *snap;
2076 struct dm_exception *e;
2077 struct dm_snap_pending_exception *pe;
2078 struct dm_snap_pending_exception *pe_to_start_now = NULL;
2079 struct dm_snap_pending_exception *pe_to_start_last = NULL;
2080 chunk_t chunk;
2081
2082
2083 list_for_each_entry (snap, snapshots, list) {
2084
2085
2086
2087
2088 if (dm_target_is_snapshot_merge(snap->ti))
2089 continue;
2090
2091 mutex_lock(&snap->lock);
2092
2093
2094 if (!snap->valid || !snap->active)
2095 goto next_snapshot;
2096
2097
2098 if (sector >= dm_table_get_size(snap->ti->table))
2099 goto next_snapshot;
2100
2101
2102
2103
2104
2105 chunk = sector_to_chunk(snap->store, sector);
2106
2107
2108
2109
2110
2111
2112 e = dm_lookup_exception(&snap->complete, chunk);
2113 if (e)
2114 goto next_snapshot;
2115
2116 pe = __lookup_pending_exception(snap, chunk);
2117 if (!pe) {
2118 mutex_unlock(&snap->lock);
2119 pe = alloc_pending_exception(snap);
2120 mutex_lock(&snap->lock);
2121
2122 if (!snap->valid) {
2123 free_pending_exception(pe);
2124 goto next_snapshot;
2125 }
2126
2127 e = dm_lookup_exception(&snap->complete, chunk);
2128 if (e) {
2129 free_pending_exception(pe);
2130 goto next_snapshot;
2131 }
2132
2133 pe = __find_pending_exception(snap, pe, chunk);
2134 if (!pe) {
2135 __invalidate_snapshot(snap, -ENOMEM);
2136 goto next_snapshot;
2137 }
2138 }
2139
2140 r = DM_MAPIO_SUBMITTED;
2141
2142
2143
2144
2145
2146
2147 if (bio) {
2148 bio_list_add(&pe->origin_bios, bio);
2149 bio = NULL;
2150
2151 if (!pe->started) {
2152 pe->started = 1;
2153 pe_to_start_last = pe;
2154 }
2155 }
2156
2157 if (!pe->started) {
2158 pe->started = 1;
2159 pe_to_start_now = pe;
2160 }
2161
2162next_snapshot:
2163 mutex_unlock(&snap->lock);
2164
2165 if (pe_to_start_now) {
2166 start_copy(pe_to_start_now);
2167 pe_to_start_now = NULL;
2168 }
2169 }
2170
2171
2172
2173
2174
2175 if (pe_to_start_last)
2176 start_copy(pe_to_start_last);
2177
2178 return r;
2179}
2180
2181
2182
2183
2184static int do_origin(struct dm_dev *origin, struct bio *bio)
2185{
2186 struct origin *o;
2187 int r = DM_MAPIO_REMAPPED;
2188
2189 down_read(&_origins_lock);
2190 o = __lookup_origin(origin->bdev);
2191 if (o)
2192 r = __origin_write(&o->snapshots, bio->bi_iter.bi_sector, bio);
2193 up_read(&_origins_lock);
2194
2195 return r;
2196}
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211static int origin_write_extent(struct dm_snapshot *merging_snap,
2212 sector_t sector, unsigned size)
2213{
2214 int must_wait = 0;
2215 sector_t n;
2216 struct origin *o;
2217
2218
2219
2220
2221
2222 down_read(&_origins_lock);
2223 o = __lookup_origin(merging_snap->origin->bdev);
2224 for (n = 0; n < size; n += merging_snap->ti->max_io_len)
2225 if (__origin_write(&o->snapshots, sector + n, NULL) ==
2226 DM_MAPIO_SUBMITTED)
2227 must_wait = 1;
2228 up_read(&_origins_lock);
2229
2230 return must_wait;
2231}
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2243{
2244 int r;
2245 struct dm_origin *o;
2246
2247 if (argc != 1) {
2248 ti->error = "origin: incorrect number of arguments";
2249 return -EINVAL;
2250 }
2251
2252 o = kmalloc(sizeof(struct dm_origin), GFP_KERNEL);
2253 if (!o) {
2254 ti->error = "Cannot allocate private origin structure";
2255 r = -ENOMEM;
2256 goto bad_alloc;
2257 }
2258
2259 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &o->dev);
2260 if (r) {
2261 ti->error = "Cannot get target device";
2262 goto bad_open;
2263 }
2264
2265 o->ti = ti;
2266 ti->private = o;
2267 ti->num_flush_bios = 1;
2268
2269 return 0;
2270
2271bad_open:
2272 kfree(o);
2273bad_alloc:
2274 return r;
2275}
2276
2277static void origin_dtr(struct dm_target *ti)
2278{
2279 struct dm_origin *o = ti->private;
2280
2281 dm_put_device(ti, o->dev);
2282 kfree(o);
2283}
2284
2285static int origin_map(struct dm_target *ti, struct bio *bio)
2286{
2287 struct dm_origin *o = ti->private;
2288 unsigned available_sectors;
2289
2290 bio_set_dev(bio, o->dev->bdev);
2291
2292 if (unlikely(bio->bi_opf & REQ_PREFLUSH))
2293 return DM_MAPIO_REMAPPED;
2294
2295 if (bio_data_dir(bio) != WRITE)
2296 return DM_MAPIO_REMAPPED;
2297
2298 available_sectors = o->split_boundary -
2299 ((unsigned)bio->bi_iter.bi_sector & (o->split_boundary - 1));
2300
2301 if (bio_sectors(bio) > available_sectors)
2302 dm_accept_partial_bio(bio, available_sectors);
2303
2304
2305 return do_origin(o->dev, bio);
2306}
2307
2308static long origin_dax_direct_access(struct dm_target *ti, pgoff_t pgoff,
2309 long nr_pages, void **kaddr, pfn_t *pfn)
2310{
2311 DMWARN("device does not support dax.");
2312 return -EIO;
2313}
2314
2315
2316
2317
2318
2319static void origin_resume(struct dm_target *ti)
2320{
2321 struct dm_origin *o = ti->private;
2322
2323 o->split_boundary = get_origin_minimum_chunksize(o->dev->bdev);
2324
2325 down_write(&_origins_lock);
2326 __insert_dm_origin(o);
2327 up_write(&_origins_lock);
2328}
2329
2330static void origin_postsuspend(struct dm_target *ti)
2331{
2332 struct dm_origin *o = ti->private;
2333
2334 down_write(&_origins_lock);
2335 __remove_dm_origin(o);
2336 up_write(&_origins_lock);
2337}
2338
2339static void origin_status(struct dm_target *ti, status_type_t type,
2340 unsigned status_flags, char *result, unsigned maxlen)
2341{
2342 struct dm_origin *o = ti->private;
2343
2344 switch (type) {
2345 case STATUSTYPE_INFO:
2346 result[0] = '\0';
2347 break;
2348
2349 case STATUSTYPE_TABLE:
2350 snprintf(result, maxlen, "%s", o->dev->name);
2351 break;
2352 }
2353}
2354
2355static int origin_iterate_devices(struct dm_target *ti,
2356 iterate_devices_callout_fn fn, void *data)
2357{
2358 struct dm_origin *o = ti->private;
2359
2360 return fn(ti, o->dev, 0, ti->len, data);
2361}
2362
2363static struct target_type origin_target = {
2364 .name = "snapshot-origin",
2365 .version = {1, 9, 0},
2366 .module = THIS_MODULE,
2367 .ctr = origin_ctr,
2368 .dtr = origin_dtr,
2369 .map = origin_map,
2370 .resume = origin_resume,
2371 .postsuspend = origin_postsuspend,
2372 .status = origin_status,
2373 .iterate_devices = origin_iterate_devices,
2374 .direct_access = origin_dax_direct_access,
2375};
2376
2377static struct target_type snapshot_target = {
2378 .name = "snapshot",
2379 .version = {1, 15, 0},
2380 .module = THIS_MODULE,
2381 .ctr = snapshot_ctr,
2382 .dtr = snapshot_dtr,
2383 .map = snapshot_map,
2384 .end_io = snapshot_end_io,
2385 .preresume = snapshot_preresume,
2386 .resume = snapshot_resume,
2387 .status = snapshot_status,
2388 .iterate_devices = snapshot_iterate_devices,
2389};
2390
2391static struct target_type merge_target = {
2392 .name = dm_snapshot_merge_target_name,
2393 .version = {1, 4, 0},
2394 .module = THIS_MODULE,
2395 .ctr = snapshot_ctr,
2396 .dtr = snapshot_dtr,
2397 .map = snapshot_merge_map,
2398 .end_io = snapshot_end_io,
2399 .presuspend = snapshot_merge_presuspend,
2400 .preresume = snapshot_preresume,
2401 .resume = snapshot_merge_resume,
2402 .status = snapshot_status,
2403 .iterate_devices = snapshot_iterate_devices,
2404};
2405
2406static int __init dm_snapshot_init(void)
2407{
2408 int r;
2409
2410 r = dm_exception_store_init();
2411 if (r) {
2412 DMERR("Failed to initialize exception stores");
2413 return r;
2414 }
2415
2416 r = init_origin_hash();
2417 if (r) {
2418 DMERR("init_origin_hash failed.");
2419 goto bad_origin_hash;
2420 }
2421
2422 exception_cache = KMEM_CACHE(dm_exception, 0);
2423 if (!exception_cache) {
2424 DMERR("Couldn't create exception cache.");
2425 r = -ENOMEM;
2426 goto bad_exception_cache;
2427 }
2428
2429 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
2430 if (!pending_cache) {
2431 DMERR("Couldn't create pending cache.");
2432 r = -ENOMEM;
2433 goto bad_pending_cache;
2434 }
2435
2436 r = dm_register_target(&snapshot_target);
2437 if (r < 0) {
2438 DMERR("snapshot target register failed %d", r);
2439 goto bad_register_snapshot_target;
2440 }
2441
2442 r = dm_register_target(&origin_target);
2443 if (r < 0) {
2444 DMERR("Origin target register failed %d", r);
2445 goto bad_register_origin_target;
2446 }
2447
2448 r = dm_register_target(&merge_target);
2449 if (r < 0) {
2450 DMERR("Merge target register failed %d", r);
2451 goto bad_register_merge_target;
2452 }
2453
2454 return 0;
2455
2456bad_register_merge_target:
2457 dm_unregister_target(&origin_target);
2458bad_register_origin_target:
2459 dm_unregister_target(&snapshot_target);
2460bad_register_snapshot_target:
2461 kmem_cache_destroy(pending_cache);
2462bad_pending_cache:
2463 kmem_cache_destroy(exception_cache);
2464bad_exception_cache:
2465 exit_origin_hash();
2466bad_origin_hash:
2467 dm_exception_store_exit();
2468
2469 return r;
2470}
2471
2472static void __exit dm_snapshot_exit(void)
2473{
2474 dm_unregister_target(&snapshot_target);
2475 dm_unregister_target(&origin_target);
2476 dm_unregister_target(&merge_target);
2477
2478 exit_origin_hash();
2479 kmem_cache_destroy(pending_cache);
2480 kmem_cache_destroy(exception_cache);
2481
2482 dm_exception_store_exit();
2483}
2484
2485
2486module_init(dm_snapshot_init);
2487module_exit(dm_snapshot_exit);
2488
2489MODULE_DESCRIPTION(DM_NAME " snapshot target");
2490MODULE_AUTHOR("Joe Thornber");
2491MODULE_LICENSE("GPL");
2492MODULE_ALIAS("dm-snapshot-origin");
2493MODULE_ALIAS("dm-snapshot-merge");
2494