1
2
3
4
5
6
7#include "dm-thin-metadata.h"
8#include "dm-bio-prison-v1.h"
9#include "dm.h"
10
11#include <linux/device-mapper.h>
12#include <linux/dm-io.h>
13#include <linux/dm-kcopyd.h>
14#include <linux/jiffies.h>
15#include <linux/log2.h>
16#include <linux/list.h>
17#include <linux/rculist.h>
18#include <linux/init.h>
19#include <linux/module.h>
20#include <linux/slab.h>
21#include <linux/vmalloc.h>
22#include <linux/sort.h>
23#include <linux/rbtree.h>
24
25#define DM_MSG_PREFIX "thin"
26
27
28
29
30#define ENDIO_HOOK_POOL_SIZE 1024
31#define MAPPING_POOL_SIZE 1024
32#define COMMIT_PERIOD HZ
33#define NO_SPACE_TIMEOUT_SECS 60
34
35static unsigned no_space_timeout_secs = NO_SPACE_TIMEOUT_SECS;
36
37DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle,
38 "A percentage of time allocated for copy on write");
39
40
41
42
43
44#define DATA_DEV_BLOCK_SIZE_MIN_SECTORS (64 * 1024 >> SECTOR_SHIFT)
45#define DATA_DEV_BLOCK_SIZE_MAX_SECTORS (1024 * 1024 * 1024 >> SECTOR_SHIFT)
46
47
48
49
50#define MAX_DEV_ID ((1 << 24) - 1)
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115enum lock_space {
116 VIRTUAL,
117 PHYSICAL
118};
119
120static void build_key(struct dm_thin_device *td, enum lock_space ls,
121 dm_block_t b, dm_block_t e, struct dm_cell_key *key)
122{
123 key->virtual = (ls == VIRTUAL);
124 key->dev = dm_thin_dev_id(td);
125 key->block_begin = b;
126 key->block_end = e;
127}
128
129static void build_data_key(struct dm_thin_device *td, dm_block_t b,
130 struct dm_cell_key *key)
131{
132 build_key(td, PHYSICAL, b, b + 1llu, key);
133}
134
135static void build_virtual_key(struct dm_thin_device *td, dm_block_t b,
136 struct dm_cell_key *key)
137{
138 build_key(td, VIRTUAL, b, b + 1llu, key);
139}
140
141
142
143#define THROTTLE_THRESHOLD (1 * HZ)
144
145struct throttle {
146 struct rw_semaphore lock;
147 unsigned long threshold;
148 bool throttle_applied;
149};
150
151static void throttle_init(struct throttle *t)
152{
153 init_rwsem(&t->lock);
154 t->throttle_applied = false;
155}
156
157static void throttle_work_start(struct throttle *t)
158{
159 t->threshold = jiffies + THROTTLE_THRESHOLD;
160}
161
162static void throttle_work_update(struct throttle *t)
163{
164 if (!t->throttle_applied && jiffies > t->threshold) {
165 down_write(&t->lock);
166 t->throttle_applied = true;
167 }
168}
169
170static void throttle_work_complete(struct throttle *t)
171{
172 if (t->throttle_applied) {
173 t->throttle_applied = false;
174 up_write(&t->lock);
175 }
176}
177
178static void throttle_lock(struct throttle *t)
179{
180 down_read(&t->lock);
181}
182
183static void throttle_unlock(struct throttle *t)
184{
185 up_read(&t->lock);
186}
187
188
189
190
191
192
193
194
195struct dm_thin_new_mapping;
196
197
198
199
200enum pool_mode {
201 PM_WRITE,
202 PM_OUT_OF_DATA_SPACE,
203
204
205
206
207 PM_OUT_OF_METADATA_SPACE,
208 PM_READ_ONLY,
209
210 PM_FAIL,
211};
212
213struct pool_features {
214 enum pool_mode mode;
215
216 bool zero_new_blocks:1;
217 bool discard_enabled:1;
218 bool discard_passdown:1;
219 bool error_if_no_space:1;
220};
221
222struct thin_c;
223typedef void (*process_bio_fn)(struct thin_c *tc, struct bio *bio);
224typedef void (*process_cell_fn)(struct thin_c *tc, struct dm_bio_prison_cell *cell);
225typedef void (*process_mapping_fn)(struct dm_thin_new_mapping *m);
226
227#define CELL_SORT_ARRAY_SIZE 8192
228
229struct pool {
230 struct list_head list;
231 struct dm_target *ti;
232
233 struct mapped_device *pool_md;
234 struct block_device *md_dev;
235 struct dm_pool_metadata *pmd;
236
237 dm_block_t low_water_blocks;
238 uint32_t sectors_per_block;
239 int sectors_per_block_shift;
240
241 struct pool_features pf;
242 bool low_water_triggered:1;
243 bool suspended:1;
244 bool out_of_data_space:1;
245
246 struct dm_bio_prison *prison;
247 struct dm_kcopyd_client *copier;
248
249 struct work_struct worker;
250 struct workqueue_struct *wq;
251 struct throttle throttle;
252 struct delayed_work waker;
253 struct delayed_work no_space_timeout;
254
255 unsigned long last_commit_jiffies;
256 unsigned ref_count;
257
258 spinlock_t lock;
259 struct bio_list deferred_flush_bios;
260 struct list_head prepared_mappings;
261 struct list_head prepared_discards;
262 struct list_head prepared_discards_pt2;
263 struct list_head active_thins;
264
265 struct dm_deferred_set *shared_read_ds;
266 struct dm_deferred_set *all_io_ds;
267
268 struct dm_thin_new_mapping *next_mapping;
269
270 process_bio_fn process_bio;
271 process_bio_fn process_discard;
272
273 process_cell_fn process_cell;
274 process_cell_fn process_discard_cell;
275
276 process_mapping_fn process_prepared_mapping;
277 process_mapping_fn process_prepared_discard;
278 process_mapping_fn process_prepared_discard_pt2;
279
280 struct dm_bio_prison_cell **cell_sort_array;
281
282 mempool_t mapping_pool;
283};
284
285static void metadata_operation_failed(struct pool *pool, const char *op, int r);
286
287static enum pool_mode get_pool_mode(struct pool *pool)
288{
289 return pool->pf.mode;
290}
291
292static void notify_of_pool_mode_change(struct pool *pool)
293{
294 const char *descs[] = {
295 "write",
296 "out-of-data-space",
297 "read-only",
298 "read-only",
299 "fail"
300 };
301 const char *extra_desc = NULL;
302 enum pool_mode mode = get_pool_mode(pool);
303
304 if (mode == PM_OUT_OF_DATA_SPACE) {
305 if (!pool->pf.error_if_no_space)
306 extra_desc = " (queue IO)";
307 else
308 extra_desc = " (error IO)";
309 }
310
311 dm_table_event(pool->ti->table);
312 DMINFO("%s: switching pool to %s%s mode",
313 dm_device_name(pool->pool_md),
314 descs[(int)mode], extra_desc ? : "");
315}
316
317
318
319
320struct pool_c {
321 struct dm_target *ti;
322 struct pool *pool;
323 struct dm_dev *data_dev;
324 struct dm_dev *metadata_dev;
325 struct dm_target_callbacks callbacks;
326
327 dm_block_t low_water_blocks;
328 struct pool_features requested_pf;
329 struct pool_features adjusted_pf;
330};
331
332
333
334
335struct thin_c {
336 struct list_head list;
337 struct dm_dev *pool_dev;
338 struct dm_dev *origin_dev;
339 sector_t origin_size;
340 dm_thin_id dev_id;
341
342 struct pool *pool;
343 struct dm_thin_device *td;
344 struct mapped_device *thin_md;
345
346 bool requeue_mode:1;
347 spinlock_t lock;
348 struct list_head deferred_cells;
349 struct bio_list deferred_bio_list;
350 struct bio_list retry_on_resume_list;
351 struct rb_root sort_bio_list;
352
353
354
355
356
357 refcount_t refcount;
358 struct completion can_destroy;
359};
360
361
362
363static bool block_size_is_power_of_two(struct pool *pool)
364{
365 return pool->sectors_per_block_shift >= 0;
366}
367
368static sector_t block_to_sectors(struct pool *pool, dm_block_t b)
369{
370 return block_size_is_power_of_two(pool) ?
371 (b << pool->sectors_per_block_shift) :
372 (b * pool->sectors_per_block);
373}
374
375
376
377struct discard_op {
378 struct thin_c *tc;
379 struct blk_plug plug;
380 struct bio *parent_bio;
381 struct bio *bio;
382};
383
384static void begin_discard(struct discard_op *op, struct thin_c *tc, struct bio *parent)
385{
386 BUG_ON(!parent);
387
388 op->tc = tc;
389 blk_start_plug(&op->plug);
390 op->parent_bio = parent;
391 op->bio = NULL;
392}
393
394static int issue_discard(struct discard_op *op, dm_block_t data_b, dm_block_t data_e)
395{
396 struct thin_c *tc = op->tc;
397 sector_t s = block_to_sectors(tc->pool, data_b);
398 sector_t len = block_to_sectors(tc->pool, data_e - data_b);
399
400 return __blkdev_issue_discard(tc->pool_dev->bdev, s, len,
401 GFP_NOWAIT, 0, &op->bio);
402}
403
404static void end_discard(struct discard_op *op, int r)
405{
406 if (op->bio) {
407
408
409
410
411 bio_chain(op->bio, op->parent_bio);
412 bio_set_op_attrs(op->bio, REQ_OP_DISCARD, 0);
413 submit_bio(op->bio);
414 }
415
416 blk_finish_plug(&op->plug);
417
418
419
420
421
422 if (r && !op->parent_bio->bi_status)
423 op->parent_bio->bi_status = errno_to_blk_status(r);
424 bio_endio(op->parent_bio);
425}
426
427
428
429
430
431
432
433static void wake_worker(struct pool *pool)
434{
435 queue_work(pool->wq, &pool->worker);
436}
437
438
439
440static int bio_detain(struct pool *pool, struct dm_cell_key *key, struct bio *bio,
441 struct dm_bio_prison_cell **cell_result)
442{
443 int r;
444 struct dm_bio_prison_cell *cell_prealloc;
445
446
447
448
449
450 cell_prealloc = dm_bio_prison_alloc_cell(pool->prison, GFP_NOIO);
451
452 r = dm_bio_detain(pool->prison, key, bio, cell_prealloc, cell_result);
453 if (r)
454
455
456
457
458 dm_bio_prison_free_cell(pool->prison, cell_prealloc);
459
460 return r;
461}
462
463static void cell_release(struct pool *pool,
464 struct dm_bio_prison_cell *cell,
465 struct bio_list *bios)
466{
467 dm_cell_release(pool->prison, cell, bios);
468 dm_bio_prison_free_cell(pool->prison, cell);
469}
470
471static void cell_visit_release(struct pool *pool,
472 void (*fn)(void *, struct dm_bio_prison_cell *),
473 void *context,
474 struct dm_bio_prison_cell *cell)
475{
476 dm_cell_visit_release(pool->prison, fn, context, cell);
477 dm_bio_prison_free_cell(pool->prison, cell);
478}
479
480static void cell_release_no_holder(struct pool *pool,
481 struct dm_bio_prison_cell *cell,
482 struct bio_list *bios)
483{
484 dm_cell_release_no_holder(pool->prison, cell, bios);
485 dm_bio_prison_free_cell(pool->prison, cell);
486}
487
488static void cell_error_with_code(struct pool *pool,
489 struct dm_bio_prison_cell *cell, blk_status_t error_code)
490{
491 dm_cell_error(pool->prison, cell, error_code);
492 dm_bio_prison_free_cell(pool->prison, cell);
493}
494
495static blk_status_t get_pool_io_error_code(struct pool *pool)
496{
497 return pool->out_of_data_space ? BLK_STS_NOSPC : BLK_STS_IOERR;
498}
499
500static void cell_error(struct pool *pool, struct dm_bio_prison_cell *cell)
501{
502 cell_error_with_code(pool, cell, get_pool_io_error_code(pool));
503}
504
505static void cell_success(struct pool *pool, struct dm_bio_prison_cell *cell)
506{
507 cell_error_with_code(pool, cell, 0);
508}
509
510static void cell_requeue(struct pool *pool, struct dm_bio_prison_cell *cell)
511{
512 cell_error_with_code(pool, cell, BLK_STS_DM_REQUEUE);
513}
514
515
516
517
518
519
520static struct dm_thin_pool_table {
521 struct mutex mutex;
522 struct list_head pools;
523} dm_thin_pool_table;
524
525static void pool_table_init(void)
526{
527 mutex_init(&dm_thin_pool_table.mutex);
528 INIT_LIST_HEAD(&dm_thin_pool_table.pools);
529}
530
531static void pool_table_exit(void)
532{
533 mutex_destroy(&dm_thin_pool_table.mutex);
534}
535
536static void __pool_table_insert(struct pool *pool)
537{
538 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
539 list_add(&pool->list, &dm_thin_pool_table.pools);
540}
541
542static void __pool_table_remove(struct pool *pool)
543{
544 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
545 list_del(&pool->list);
546}
547
548static struct pool *__pool_table_lookup(struct mapped_device *md)
549{
550 struct pool *pool = NULL, *tmp;
551
552 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
553
554 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
555 if (tmp->pool_md == md) {
556 pool = tmp;
557 break;
558 }
559 }
560
561 return pool;
562}
563
564static struct pool *__pool_table_lookup_metadata_dev(struct block_device *md_dev)
565{
566 struct pool *pool = NULL, *tmp;
567
568 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
569
570 list_for_each_entry(tmp, &dm_thin_pool_table.pools, list) {
571 if (tmp->md_dev == md_dev) {
572 pool = tmp;
573 break;
574 }
575 }
576
577 return pool;
578}
579
580
581
582struct dm_thin_endio_hook {
583 struct thin_c *tc;
584 struct dm_deferred_entry *shared_read_entry;
585 struct dm_deferred_entry *all_io_entry;
586 struct dm_thin_new_mapping *overwrite_mapping;
587 struct rb_node rb_node;
588 struct dm_bio_prison_cell *cell;
589};
590
591static void __merge_bio_list(struct bio_list *bios, struct bio_list *master)
592{
593 bio_list_merge(bios, master);
594 bio_list_init(master);
595}
596
597static void error_bio_list(struct bio_list *bios, blk_status_t error)
598{
599 struct bio *bio;
600
601 while ((bio = bio_list_pop(bios))) {
602 bio->bi_status = error;
603 bio_endio(bio);
604 }
605}
606
607static void error_thin_bio_list(struct thin_c *tc, struct bio_list *master,
608 blk_status_t error)
609{
610 struct bio_list bios;
611 unsigned long flags;
612
613 bio_list_init(&bios);
614
615 spin_lock_irqsave(&tc->lock, flags);
616 __merge_bio_list(&bios, master);
617 spin_unlock_irqrestore(&tc->lock, flags);
618
619 error_bio_list(&bios, error);
620}
621
622static void requeue_deferred_cells(struct thin_c *tc)
623{
624 struct pool *pool = tc->pool;
625 unsigned long flags;
626 struct list_head cells;
627 struct dm_bio_prison_cell *cell, *tmp;
628
629 INIT_LIST_HEAD(&cells);
630
631 spin_lock_irqsave(&tc->lock, flags);
632 list_splice_init(&tc->deferred_cells, &cells);
633 spin_unlock_irqrestore(&tc->lock, flags);
634
635 list_for_each_entry_safe(cell, tmp, &cells, user_list)
636 cell_requeue(pool, cell);
637}
638
639static void requeue_io(struct thin_c *tc)
640{
641 struct bio_list bios;
642 unsigned long flags;
643
644 bio_list_init(&bios);
645
646 spin_lock_irqsave(&tc->lock, flags);
647 __merge_bio_list(&bios, &tc->deferred_bio_list);
648 __merge_bio_list(&bios, &tc->retry_on_resume_list);
649 spin_unlock_irqrestore(&tc->lock, flags);
650
651 error_bio_list(&bios, BLK_STS_DM_REQUEUE);
652 requeue_deferred_cells(tc);
653}
654
655static void error_retry_list_with_code(struct pool *pool, blk_status_t error)
656{
657 struct thin_c *tc;
658
659 rcu_read_lock();
660 list_for_each_entry_rcu(tc, &pool->active_thins, list)
661 error_thin_bio_list(tc, &tc->retry_on_resume_list, error);
662 rcu_read_unlock();
663}
664
665static void error_retry_list(struct pool *pool)
666{
667 error_retry_list_with_code(pool, get_pool_io_error_code(pool));
668}
669
670
671
672
673
674
675
676
677static dm_block_t get_bio_block(struct thin_c *tc, struct bio *bio)
678{
679 struct pool *pool = tc->pool;
680 sector_t block_nr = bio->bi_iter.bi_sector;
681
682 if (block_size_is_power_of_two(pool))
683 block_nr >>= pool->sectors_per_block_shift;
684 else
685 (void) sector_div(block_nr, pool->sectors_per_block);
686
687 return block_nr;
688}
689
690
691
692
693static void get_bio_block_range(struct thin_c *tc, struct bio *bio,
694 dm_block_t *begin, dm_block_t *end)
695{
696 struct pool *pool = tc->pool;
697 sector_t b = bio->bi_iter.bi_sector;
698 sector_t e = b + (bio->bi_iter.bi_size >> SECTOR_SHIFT);
699
700 b += pool->sectors_per_block - 1ull;
701
702 if (block_size_is_power_of_two(pool)) {
703 b >>= pool->sectors_per_block_shift;
704 e >>= pool->sectors_per_block_shift;
705 } else {
706 (void) sector_div(b, pool->sectors_per_block);
707 (void) sector_div(e, pool->sectors_per_block);
708 }
709
710 if (e < b)
711
712 e = b;
713
714 *begin = b;
715 *end = e;
716}
717
718static void remap(struct thin_c *tc, struct bio *bio, dm_block_t block)
719{
720 struct pool *pool = tc->pool;
721 sector_t bi_sector = bio->bi_iter.bi_sector;
722
723 bio_set_dev(bio, tc->pool_dev->bdev);
724 if (block_size_is_power_of_two(pool))
725 bio->bi_iter.bi_sector =
726 (block << pool->sectors_per_block_shift) |
727 (bi_sector & (pool->sectors_per_block - 1));
728 else
729 bio->bi_iter.bi_sector = (block * pool->sectors_per_block) +
730 sector_div(bi_sector, pool->sectors_per_block);
731}
732
733static void remap_to_origin(struct thin_c *tc, struct bio *bio)
734{
735 bio_set_dev(bio, tc->origin_dev->bdev);
736}
737
738static int bio_triggers_commit(struct thin_c *tc, struct bio *bio)
739{
740 return op_is_flush(bio->bi_opf) &&
741 dm_thin_changed_this_transaction(tc->td);
742}
743
744static void inc_all_io_entry(struct pool *pool, struct bio *bio)
745{
746 struct dm_thin_endio_hook *h;
747
748 if (bio_op(bio) == REQ_OP_DISCARD)
749 return;
750
751 h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
752 h->all_io_entry = dm_deferred_entry_inc(pool->all_io_ds);
753}
754
755static void issue(struct thin_c *tc, struct bio *bio)
756{
757 struct pool *pool = tc->pool;
758 unsigned long flags;
759
760 if (!bio_triggers_commit(tc, bio)) {
761 generic_make_request(bio);
762 return;
763 }
764
765
766
767
768
769
770 if (dm_thin_aborted_changes(tc->td)) {
771 bio_io_error(bio);
772 return;
773 }
774
775
776
777
778
779 spin_lock_irqsave(&pool->lock, flags);
780 bio_list_add(&pool->deferred_flush_bios, bio);
781 spin_unlock_irqrestore(&pool->lock, flags);
782}
783
784static void remap_to_origin_and_issue(struct thin_c *tc, struct bio *bio)
785{
786 remap_to_origin(tc, bio);
787 issue(tc, bio);
788}
789
790static void remap_and_issue(struct thin_c *tc, struct bio *bio,
791 dm_block_t block)
792{
793 remap(tc, bio, block);
794 issue(tc, bio);
795}
796
797
798
799
800
801
802struct dm_thin_new_mapping {
803 struct list_head list;
804
805 bool pass_discard:1;
806 bool maybe_shared:1;
807
808
809
810
811
812
813 atomic_t prepare_actions;
814
815 blk_status_t status;
816 struct thin_c *tc;
817 dm_block_t virt_begin, virt_end;
818 dm_block_t data_block;
819 struct dm_bio_prison_cell *cell;
820
821
822
823
824
825
826
827 struct bio *bio;
828 bio_end_io_t *saved_bi_end_io;
829};
830
831static void __complete_mapping_preparation(struct dm_thin_new_mapping *m)
832{
833 struct pool *pool = m->tc->pool;
834
835 if (atomic_dec_and_test(&m->prepare_actions)) {
836 list_add_tail(&m->list, &pool->prepared_mappings);
837 wake_worker(pool);
838 }
839}
840
841static void complete_mapping_preparation(struct dm_thin_new_mapping *m)
842{
843 unsigned long flags;
844 struct pool *pool = m->tc->pool;
845
846 spin_lock_irqsave(&pool->lock, flags);
847 __complete_mapping_preparation(m);
848 spin_unlock_irqrestore(&pool->lock, flags);
849}
850
851static void copy_complete(int read_err, unsigned long write_err, void *context)
852{
853 struct dm_thin_new_mapping *m = context;
854
855 m->status = read_err || write_err ? BLK_STS_IOERR : 0;
856 complete_mapping_preparation(m);
857}
858
859static void overwrite_endio(struct bio *bio)
860{
861 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
862 struct dm_thin_new_mapping *m = h->overwrite_mapping;
863
864 bio->bi_end_io = m->saved_bi_end_io;
865
866 m->status = bio->bi_status;
867 complete_mapping_preparation(m);
868}
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *cell)
885{
886 struct pool *pool = tc->pool;
887 unsigned long flags;
888
889 spin_lock_irqsave(&tc->lock, flags);
890 cell_release_no_holder(pool, cell, &tc->deferred_bio_list);
891 spin_unlock_irqrestore(&tc->lock, flags);
892
893 wake_worker(pool);
894}
895
896static void thin_defer_bio(struct thin_c *tc, struct bio *bio);
897
898struct remap_info {
899 struct thin_c *tc;
900 struct bio_list defer_bios;
901 struct bio_list issue_bios;
902};
903
904static void __inc_remap_and_issue_cell(void *context,
905 struct dm_bio_prison_cell *cell)
906{
907 struct remap_info *info = context;
908 struct bio *bio;
909
910 while ((bio = bio_list_pop(&cell->bios))) {
911 if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD)
912 bio_list_add(&info->defer_bios, bio);
913 else {
914 inc_all_io_entry(info->tc->pool, bio);
915
916
917
918
919
920
921 bio_list_add(&info->issue_bios, bio);
922 }
923 }
924}
925
926static void inc_remap_and_issue_cell(struct thin_c *tc,
927 struct dm_bio_prison_cell *cell,
928 dm_block_t block)
929{
930 struct bio *bio;
931 struct remap_info info;
932
933 info.tc = tc;
934 bio_list_init(&info.defer_bios);
935 bio_list_init(&info.issue_bios);
936
937
938
939
940
941
942 cell_visit_release(tc->pool, __inc_remap_and_issue_cell,
943 &info, cell);
944
945 while ((bio = bio_list_pop(&info.defer_bios)))
946 thin_defer_bio(tc, bio);
947
948 while ((bio = bio_list_pop(&info.issue_bios)))
949 remap_and_issue(info.tc, bio, block);
950}
951
952static void process_prepared_mapping_fail(struct dm_thin_new_mapping *m)
953{
954 cell_error(m->tc->pool, m->cell);
955 list_del(&m->list);
956 mempool_free(m, &m->tc->pool->mapping_pool);
957}
958
959static void process_prepared_mapping(struct dm_thin_new_mapping *m)
960{
961 struct thin_c *tc = m->tc;
962 struct pool *pool = tc->pool;
963 struct bio *bio = m->bio;
964 int r;
965
966 if (m->status) {
967 cell_error(pool, m->cell);
968 goto out;
969 }
970
971
972
973
974
975
976 r = dm_thin_insert_block(tc->td, m->virt_begin, m->data_block);
977 if (r) {
978 metadata_operation_failed(pool, "dm_thin_insert_block", r);
979 cell_error(pool, m->cell);
980 goto out;
981 }
982
983
984
985
986
987
988
989 if (bio) {
990 inc_remap_and_issue_cell(tc, m->cell, m->data_block);
991 bio_endio(bio);
992 } else {
993 inc_all_io_entry(tc->pool, m->cell->holder);
994 remap_and_issue(tc, m->cell->holder, m->data_block);
995 inc_remap_and_issue_cell(tc, m->cell, m->data_block);
996 }
997
998out:
999 list_del(&m->list);
1000 mempool_free(m, &pool->mapping_pool);
1001}
1002
1003
1004
1005static void free_discard_mapping(struct dm_thin_new_mapping *m)
1006{
1007 struct thin_c *tc = m->tc;
1008 if (m->cell)
1009 cell_defer_no_holder(tc, m->cell);
1010 mempool_free(m, &tc->pool->mapping_pool);
1011}
1012
1013static void process_prepared_discard_fail(struct dm_thin_new_mapping *m)
1014{
1015 bio_io_error(m->bio);
1016 free_discard_mapping(m);
1017}
1018
1019static void process_prepared_discard_success(struct dm_thin_new_mapping *m)
1020{
1021 bio_endio(m->bio);
1022 free_discard_mapping(m);
1023}
1024
1025static void process_prepared_discard_no_passdown(struct dm_thin_new_mapping *m)
1026{
1027 int r;
1028 struct thin_c *tc = m->tc;
1029
1030 r = dm_thin_remove_range(tc->td, m->cell->key.block_begin, m->cell->key.block_end);
1031 if (r) {
1032 metadata_operation_failed(tc->pool, "dm_thin_remove_range", r);
1033 bio_io_error(m->bio);
1034 } else
1035 bio_endio(m->bio);
1036
1037 cell_defer_no_holder(tc, m->cell);
1038 mempool_free(m, &tc->pool->mapping_pool);
1039}
1040
1041
1042
1043static void passdown_double_checking_shared_status(struct dm_thin_new_mapping *m,
1044 struct bio *discard_parent)
1045{
1046
1047
1048
1049
1050 int r = 0;
1051 bool used = true;
1052 struct thin_c *tc = m->tc;
1053 struct pool *pool = tc->pool;
1054 dm_block_t b = m->data_block, e, end = m->data_block + m->virt_end - m->virt_begin;
1055 struct discard_op op;
1056
1057 begin_discard(&op, tc, discard_parent);
1058 while (b != end) {
1059
1060 for (; b < end; b++) {
1061 r = dm_pool_block_is_used(pool->pmd, b, &used);
1062 if (r)
1063 goto out;
1064
1065 if (!used)
1066 break;
1067 }
1068
1069 if (b == end)
1070 break;
1071
1072
1073 for (e = b + 1; e != end; e++) {
1074 r = dm_pool_block_is_used(pool->pmd, e, &used);
1075 if (r)
1076 goto out;
1077
1078 if (used)
1079 break;
1080 }
1081
1082 r = issue_discard(&op, b, e);
1083 if (r)
1084 goto out;
1085
1086 b = e;
1087 }
1088out:
1089 end_discard(&op, r);
1090}
1091
1092static void queue_passdown_pt2(struct dm_thin_new_mapping *m)
1093{
1094 unsigned long flags;
1095 struct pool *pool = m->tc->pool;
1096
1097 spin_lock_irqsave(&pool->lock, flags);
1098 list_add_tail(&m->list, &pool->prepared_discards_pt2);
1099 spin_unlock_irqrestore(&pool->lock, flags);
1100 wake_worker(pool);
1101}
1102
1103static void passdown_endio(struct bio *bio)
1104{
1105
1106
1107
1108
1109 queue_passdown_pt2(bio->bi_private);
1110 bio_put(bio);
1111}
1112
1113static void process_prepared_discard_passdown_pt1(struct dm_thin_new_mapping *m)
1114{
1115 int r;
1116 struct thin_c *tc = m->tc;
1117 struct pool *pool = tc->pool;
1118 struct bio *discard_parent;
1119 dm_block_t data_end = m->data_block + (m->virt_end - m->virt_begin);
1120
1121
1122
1123
1124
1125
1126 r = dm_thin_remove_range(tc->td, m->virt_begin, m->virt_end);
1127 if (r) {
1128 metadata_operation_failed(pool, "dm_thin_remove_range", r);
1129 bio_io_error(m->bio);
1130 cell_defer_no_holder(tc, m->cell);
1131 mempool_free(m, &pool->mapping_pool);
1132 return;
1133 }
1134
1135
1136
1137
1138
1139 r = dm_pool_inc_data_range(pool->pmd, m->data_block, data_end);
1140 if (r) {
1141 metadata_operation_failed(pool, "dm_pool_inc_data_range", r);
1142 bio_io_error(m->bio);
1143 cell_defer_no_holder(tc, m->cell);
1144 mempool_free(m, &pool->mapping_pool);
1145 return;
1146 }
1147
1148 discard_parent = bio_alloc(GFP_NOIO, 1);
1149 if (!discard_parent) {
1150 DMWARN("%s: unable to allocate top level discard bio for passdown. Skipping passdown.",
1151 dm_device_name(tc->pool->pool_md));
1152 queue_passdown_pt2(m);
1153
1154 } else {
1155 discard_parent->bi_end_io = passdown_endio;
1156 discard_parent->bi_private = m;
1157
1158 if (m->maybe_shared)
1159 passdown_double_checking_shared_status(m, discard_parent);
1160 else {
1161 struct discard_op op;
1162
1163 begin_discard(&op, tc, discard_parent);
1164 r = issue_discard(&op, m->data_block, data_end);
1165 end_discard(&op, r);
1166 }
1167 }
1168}
1169
1170static void process_prepared_discard_passdown_pt2(struct dm_thin_new_mapping *m)
1171{
1172 int r;
1173 struct thin_c *tc = m->tc;
1174 struct pool *pool = tc->pool;
1175
1176
1177
1178
1179
1180 r = dm_pool_dec_data_range(pool->pmd, m->data_block,
1181 m->data_block + (m->virt_end - m->virt_begin));
1182 if (r) {
1183 metadata_operation_failed(pool, "dm_pool_dec_data_range", r);
1184 bio_io_error(m->bio);
1185 } else
1186 bio_endio(m->bio);
1187
1188 cell_defer_no_holder(tc, m->cell);
1189 mempool_free(m, &pool->mapping_pool);
1190}
1191
1192static void process_prepared(struct pool *pool, struct list_head *head,
1193 process_mapping_fn *fn)
1194{
1195 unsigned long flags;
1196 struct list_head maps;
1197 struct dm_thin_new_mapping *m, *tmp;
1198
1199 INIT_LIST_HEAD(&maps);
1200 spin_lock_irqsave(&pool->lock, flags);
1201 list_splice_init(head, &maps);
1202 spin_unlock_irqrestore(&pool->lock, flags);
1203
1204 list_for_each_entry_safe(m, tmp, &maps, list)
1205 (*fn)(m);
1206}
1207
1208
1209
1210
1211static int io_overlaps_block(struct pool *pool, struct bio *bio)
1212{
1213 return bio->bi_iter.bi_size ==
1214 (pool->sectors_per_block << SECTOR_SHIFT);
1215}
1216
1217static int io_overwrites_block(struct pool *pool, struct bio *bio)
1218{
1219 return (bio_data_dir(bio) == WRITE) &&
1220 io_overlaps_block(pool, bio);
1221}
1222
1223static void save_and_set_endio(struct bio *bio, bio_end_io_t **save,
1224 bio_end_io_t *fn)
1225{
1226 *save = bio->bi_end_io;
1227 bio->bi_end_io = fn;
1228}
1229
1230static int ensure_next_mapping(struct pool *pool)
1231{
1232 if (pool->next_mapping)
1233 return 0;
1234
1235 pool->next_mapping = mempool_alloc(&pool->mapping_pool, GFP_ATOMIC);
1236
1237 return pool->next_mapping ? 0 : -ENOMEM;
1238}
1239
1240static struct dm_thin_new_mapping *get_next_mapping(struct pool *pool)
1241{
1242 struct dm_thin_new_mapping *m = pool->next_mapping;
1243
1244 BUG_ON(!pool->next_mapping);
1245
1246 memset(m, 0, sizeof(struct dm_thin_new_mapping));
1247 INIT_LIST_HEAD(&m->list);
1248 m->bio = NULL;
1249
1250 pool->next_mapping = NULL;
1251
1252 return m;
1253}
1254
1255static void ll_zero(struct thin_c *tc, struct dm_thin_new_mapping *m,
1256 sector_t begin, sector_t end)
1257{
1258 struct dm_io_region to;
1259
1260 to.bdev = tc->pool_dev->bdev;
1261 to.sector = begin;
1262 to.count = end - begin;
1263
1264 dm_kcopyd_zero(tc->pool->copier, 1, &to, 0, copy_complete, m);
1265}
1266
1267static void remap_and_issue_overwrite(struct thin_c *tc, struct bio *bio,
1268 dm_block_t data_begin,
1269 struct dm_thin_new_mapping *m)
1270{
1271 struct pool *pool = tc->pool;
1272 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1273
1274 h->overwrite_mapping = m;
1275 m->bio = bio;
1276 save_and_set_endio(bio, &m->saved_bi_end_io, overwrite_endio);
1277 inc_all_io_entry(pool, bio);
1278 remap_and_issue(tc, bio, data_begin);
1279}
1280
1281
1282
1283
1284static void schedule_copy(struct thin_c *tc, dm_block_t virt_block,
1285 struct dm_dev *origin, dm_block_t data_origin,
1286 dm_block_t data_dest,
1287 struct dm_bio_prison_cell *cell, struct bio *bio,
1288 sector_t len)
1289{
1290 struct pool *pool = tc->pool;
1291 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1292
1293 m->tc = tc;
1294 m->virt_begin = virt_block;
1295 m->virt_end = virt_block + 1u;
1296 m->data_block = data_dest;
1297 m->cell = cell;
1298
1299
1300
1301
1302
1303
1304 atomic_set(&m->prepare_actions, 3);
1305
1306 if (!dm_deferred_set_add_work(pool->shared_read_ds, &m->list))
1307 complete_mapping_preparation(m);
1308
1309
1310
1311
1312
1313
1314
1315 if (io_overwrites_block(pool, bio))
1316 remap_and_issue_overwrite(tc, bio, data_dest, m);
1317 else {
1318 struct dm_io_region from, to;
1319
1320 from.bdev = origin->bdev;
1321 from.sector = data_origin * pool->sectors_per_block;
1322 from.count = len;
1323
1324 to.bdev = tc->pool_dev->bdev;
1325 to.sector = data_dest * pool->sectors_per_block;
1326 to.count = len;
1327
1328 dm_kcopyd_copy(pool->copier, &from, 1, &to,
1329 0, copy_complete, m);
1330
1331
1332
1333
1334 if (len < pool->sectors_per_block && pool->pf.zero_new_blocks) {
1335 atomic_inc(&m->prepare_actions);
1336 ll_zero(tc, m,
1337 data_dest * pool->sectors_per_block + len,
1338 (data_dest + 1) * pool->sectors_per_block);
1339 }
1340 }
1341
1342 complete_mapping_preparation(m);
1343}
1344
1345static void schedule_internal_copy(struct thin_c *tc, dm_block_t virt_block,
1346 dm_block_t data_origin, dm_block_t data_dest,
1347 struct dm_bio_prison_cell *cell, struct bio *bio)
1348{
1349 schedule_copy(tc, virt_block, tc->pool_dev,
1350 data_origin, data_dest, cell, bio,
1351 tc->pool->sectors_per_block);
1352}
1353
1354static void schedule_zero(struct thin_c *tc, dm_block_t virt_block,
1355 dm_block_t data_block, struct dm_bio_prison_cell *cell,
1356 struct bio *bio)
1357{
1358 struct pool *pool = tc->pool;
1359 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1360
1361 atomic_set(&m->prepare_actions, 1);
1362 m->tc = tc;
1363 m->virt_begin = virt_block;
1364 m->virt_end = virt_block + 1u;
1365 m->data_block = data_block;
1366 m->cell = cell;
1367
1368
1369
1370
1371
1372
1373 if (pool->pf.zero_new_blocks) {
1374 if (io_overwrites_block(pool, bio))
1375 remap_and_issue_overwrite(tc, bio, data_block, m);
1376 else
1377 ll_zero(tc, m, data_block * pool->sectors_per_block,
1378 (data_block + 1) * pool->sectors_per_block);
1379 } else
1380 process_prepared_mapping(m);
1381}
1382
1383static void schedule_external_copy(struct thin_c *tc, dm_block_t virt_block,
1384 dm_block_t data_dest,
1385 struct dm_bio_prison_cell *cell, struct bio *bio)
1386{
1387 struct pool *pool = tc->pool;
1388 sector_t virt_block_begin = virt_block * pool->sectors_per_block;
1389 sector_t virt_block_end = (virt_block + 1) * pool->sectors_per_block;
1390
1391 if (virt_block_end <= tc->origin_size)
1392 schedule_copy(tc, virt_block, tc->origin_dev,
1393 virt_block, data_dest, cell, bio,
1394 pool->sectors_per_block);
1395
1396 else if (virt_block_begin < tc->origin_size)
1397 schedule_copy(tc, virt_block, tc->origin_dev,
1398 virt_block, data_dest, cell, bio,
1399 tc->origin_size - virt_block_begin);
1400
1401 else
1402 schedule_zero(tc, virt_block, data_dest, cell, bio);
1403}
1404
1405static void set_pool_mode(struct pool *pool, enum pool_mode new_mode);
1406
1407static void requeue_bios(struct pool *pool);
1408
1409static bool is_read_only_pool_mode(enum pool_mode mode)
1410{
1411 return (mode == PM_OUT_OF_METADATA_SPACE || mode == PM_READ_ONLY);
1412}
1413
1414static bool is_read_only(struct pool *pool)
1415{
1416 return is_read_only_pool_mode(get_pool_mode(pool));
1417}
1418
1419static void check_for_metadata_space(struct pool *pool)
1420{
1421 int r;
1422 const char *ooms_reason = NULL;
1423 dm_block_t nr_free;
1424
1425 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free);
1426 if (r)
1427 ooms_reason = "Could not get free metadata blocks";
1428 else if (!nr_free)
1429 ooms_reason = "No free metadata blocks";
1430
1431 if (ooms_reason && !is_read_only(pool)) {
1432 DMERR("%s", ooms_reason);
1433 set_pool_mode(pool, PM_OUT_OF_METADATA_SPACE);
1434 }
1435}
1436
1437static void check_for_data_space(struct pool *pool)
1438{
1439 int r;
1440 dm_block_t nr_free;
1441
1442 if (get_pool_mode(pool) != PM_OUT_OF_DATA_SPACE)
1443 return;
1444
1445 r = dm_pool_get_free_block_count(pool->pmd, &nr_free);
1446 if (r)
1447 return;
1448
1449 if (nr_free) {
1450 set_pool_mode(pool, PM_WRITE);
1451 requeue_bios(pool);
1452 }
1453}
1454
1455
1456
1457
1458
1459static int commit(struct pool *pool)
1460{
1461 int r;
1462
1463 if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE)
1464 return -EINVAL;
1465
1466 r = dm_pool_commit_metadata(pool->pmd);
1467 if (r)
1468 metadata_operation_failed(pool, "dm_pool_commit_metadata", r);
1469 else {
1470 check_for_metadata_space(pool);
1471 check_for_data_space(pool);
1472 }
1473
1474 return r;
1475}
1476
1477static void check_low_water_mark(struct pool *pool, dm_block_t free_blocks)
1478{
1479 unsigned long flags;
1480
1481 if (free_blocks <= pool->low_water_blocks && !pool->low_water_triggered) {
1482 DMWARN("%s: reached low water mark for data device: sending event.",
1483 dm_device_name(pool->pool_md));
1484 spin_lock_irqsave(&pool->lock, flags);
1485 pool->low_water_triggered = true;
1486 spin_unlock_irqrestore(&pool->lock, flags);
1487 dm_table_event(pool->ti->table);
1488 }
1489}
1490
1491static int alloc_data_block(struct thin_c *tc, dm_block_t *result)
1492{
1493 int r;
1494 dm_block_t free_blocks;
1495 struct pool *pool = tc->pool;
1496
1497 if (WARN_ON(get_pool_mode(pool) != PM_WRITE))
1498 return -EINVAL;
1499
1500 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1501 if (r) {
1502 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1503 return r;
1504 }
1505
1506 check_low_water_mark(pool, free_blocks);
1507
1508 if (!free_blocks) {
1509
1510
1511
1512
1513 r = commit(pool);
1514 if (r)
1515 return r;
1516
1517 r = dm_pool_get_free_block_count(pool->pmd, &free_blocks);
1518 if (r) {
1519 metadata_operation_failed(pool, "dm_pool_get_free_block_count", r);
1520 return r;
1521 }
1522
1523 if (!free_blocks) {
1524 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1525 return -ENOSPC;
1526 }
1527 }
1528
1529 r = dm_pool_alloc_data_block(pool->pmd, result);
1530 if (r) {
1531 if (r == -ENOSPC)
1532 set_pool_mode(pool, PM_OUT_OF_DATA_SPACE);
1533 else
1534 metadata_operation_failed(pool, "dm_pool_alloc_data_block", r);
1535 return r;
1536 }
1537
1538 r = dm_pool_get_free_metadata_block_count(pool->pmd, &free_blocks);
1539 if (r) {
1540 metadata_operation_failed(pool, "dm_pool_get_free_metadata_block_count", r);
1541 return r;
1542 }
1543
1544 if (!free_blocks) {
1545
1546 r = commit(pool);
1547 if (r)
1548 return r;
1549 }
1550
1551 return 0;
1552}
1553
1554
1555
1556
1557
1558static void retry_on_resume(struct bio *bio)
1559{
1560 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1561 struct thin_c *tc = h->tc;
1562 unsigned long flags;
1563
1564 spin_lock_irqsave(&tc->lock, flags);
1565 bio_list_add(&tc->retry_on_resume_list, bio);
1566 spin_unlock_irqrestore(&tc->lock, flags);
1567}
1568
1569static blk_status_t should_error_unserviceable_bio(struct pool *pool)
1570{
1571 enum pool_mode m = get_pool_mode(pool);
1572
1573 switch (m) {
1574 case PM_WRITE:
1575
1576 DMERR_LIMIT("bio unserviceable, yet pool is in PM_WRITE mode");
1577 return BLK_STS_IOERR;
1578
1579 case PM_OUT_OF_DATA_SPACE:
1580 return pool->pf.error_if_no_space ? BLK_STS_NOSPC : 0;
1581
1582 case PM_OUT_OF_METADATA_SPACE:
1583 case PM_READ_ONLY:
1584 case PM_FAIL:
1585 return BLK_STS_IOERR;
1586 default:
1587
1588 DMERR_LIMIT("bio unserviceable, yet pool has an unknown mode");
1589 return BLK_STS_IOERR;
1590 }
1591}
1592
1593static void handle_unserviceable_bio(struct pool *pool, struct bio *bio)
1594{
1595 blk_status_t error = should_error_unserviceable_bio(pool);
1596
1597 if (error) {
1598 bio->bi_status = error;
1599 bio_endio(bio);
1600 } else
1601 retry_on_resume(bio);
1602}
1603
1604static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *cell)
1605{
1606 struct bio *bio;
1607 struct bio_list bios;
1608 blk_status_t error;
1609
1610 error = should_error_unserviceable_bio(pool);
1611 if (error) {
1612 cell_error_with_code(pool, cell, error);
1613 return;
1614 }
1615
1616 bio_list_init(&bios);
1617 cell_release(pool, cell, &bios);
1618
1619 while ((bio = bio_list_pop(&bios)))
1620 retry_on_resume(bio);
1621}
1622
1623static void process_discard_cell_no_passdown(struct thin_c *tc,
1624 struct dm_bio_prison_cell *virt_cell)
1625{
1626 struct pool *pool = tc->pool;
1627 struct dm_thin_new_mapping *m = get_next_mapping(pool);
1628
1629
1630
1631
1632
1633 m->tc = tc;
1634 m->virt_begin = virt_cell->key.block_begin;
1635 m->virt_end = virt_cell->key.block_end;
1636 m->cell = virt_cell;
1637 m->bio = virt_cell->holder;
1638
1639 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
1640 pool->process_prepared_discard(m);
1641}
1642
1643static void break_up_discard_bio(struct thin_c *tc, dm_block_t begin, dm_block_t end,
1644 struct bio *bio)
1645{
1646 struct pool *pool = tc->pool;
1647
1648 int r;
1649 bool maybe_shared;
1650 struct dm_cell_key data_key;
1651 struct dm_bio_prison_cell *data_cell;
1652 struct dm_thin_new_mapping *m;
1653 dm_block_t virt_begin, virt_end, data_begin;
1654
1655 while (begin != end) {
1656 r = ensure_next_mapping(pool);
1657 if (r)
1658
1659 return;
1660
1661 r = dm_thin_find_mapped_range(tc->td, begin, end, &virt_begin, &virt_end,
1662 &data_begin, &maybe_shared);
1663 if (r)
1664
1665
1666
1667
1668 break;
1669
1670 build_key(tc->td, PHYSICAL, data_begin, data_begin + (virt_end - virt_begin), &data_key);
1671 if (bio_detain(tc->pool, &data_key, NULL, &data_cell)) {
1672
1673 begin = virt_end;
1674 continue;
1675 }
1676
1677
1678
1679
1680
1681 m = get_next_mapping(pool);
1682 m->tc = tc;
1683 m->maybe_shared = maybe_shared;
1684 m->virt_begin = virt_begin;
1685 m->virt_end = virt_end;
1686 m->data_block = data_begin;
1687 m->cell = data_cell;
1688 m->bio = bio;
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698 bio_inc_remaining(bio);
1699 if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
1700 pool->process_prepared_discard(m);
1701
1702 begin = virt_end;
1703 }
1704}
1705
1706static void process_discard_cell_passdown(struct thin_c *tc, struct dm_bio_prison_cell *virt_cell)
1707{
1708 struct bio *bio = virt_cell->holder;
1709 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1710
1711
1712
1713
1714
1715
1716 h->cell = virt_cell;
1717 break_up_discard_bio(tc, virt_cell->key.block_begin, virt_cell->key.block_end, bio);
1718
1719
1720
1721
1722
1723
1724 bio_endio(bio);
1725}
1726
1727static void process_discard_bio(struct thin_c *tc, struct bio *bio)
1728{
1729 dm_block_t begin, end;
1730 struct dm_cell_key virt_key;
1731 struct dm_bio_prison_cell *virt_cell;
1732
1733 get_bio_block_range(tc, bio, &begin, &end);
1734 if (begin == end) {
1735
1736
1737
1738 bio_endio(bio);
1739 return;
1740 }
1741
1742 build_key(tc->td, VIRTUAL, begin, end, &virt_key);
1743 if (bio_detain(tc->pool, &virt_key, bio, &virt_cell))
1744
1745
1746
1747
1748
1749
1750
1751 return;
1752
1753 tc->pool->process_discard_cell(tc, virt_cell);
1754}
1755
1756static void break_sharing(struct thin_c *tc, struct bio *bio, dm_block_t block,
1757 struct dm_cell_key *key,
1758 struct dm_thin_lookup_result *lookup_result,
1759 struct dm_bio_prison_cell *cell)
1760{
1761 int r;
1762 dm_block_t data_block;
1763 struct pool *pool = tc->pool;
1764
1765 r = alloc_data_block(tc, &data_block);
1766 switch (r) {
1767 case 0:
1768 schedule_internal_copy(tc, block, lookup_result->block,
1769 data_block, cell, bio);
1770 break;
1771
1772 case -ENOSPC:
1773 retry_bios_on_resume(pool, cell);
1774 break;
1775
1776 default:
1777 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1778 __func__, r);
1779 cell_error(pool, cell);
1780 break;
1781 }
1782}
1783
1784static void __remap_and_issue_shared_cell(void *context,
1785 struct dm_bio_prison_cell *cell)
1786{
1787 struct remap_info *info = context;
1788 struct bio *bio;
1789
1790 while ((bio = bio_list_pop(&cell->bios))) {
1791 if (bio_data_dir(bio) == WRITE || op_is_flush(bio->bi_opf) ||
1792 bio_op(bio) == REQ_OP_DISCARD)
1793 bio_list_add(&info->defer_bios, bio);
1794 else {
1795 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1796
1797 h->shared_read_entry = dm_deferred_entry_inc(info->tc->pool->shared_read_ds);
1798 inc_all_io_entry(info->tc->pool, bio);
1799 bio_list_add(&info->issue_bios, bio);
1800 }
1801 }
1802}
1803
1804static void remap_and_issue_shared_cell(struct thin_c *tc,
1805 struct dm_bio_prison_cell *cell,
1806 dm_block_t block)
1807{
1808 struct bio *bio;
1809 struct remap_info info;
1810
1811 info.tc = tc;
1812 bio_list_init(&info.defer_bios);
1813 bio_list_init(&info.issue_bios);
1814
1815 cell_visit_release(tc->pool, __remap_and_issue_shared_cell,
1816 &info, cell);
1817
1818 while ((bio = bio_list_pop(&info.defer_bios)))
1819 thin_defer_bio(tc, bio);
1820
1821 while ((bio = bio_list_pop(&info.issue_bios)))
1822 remap_and_issue(tc, bio, block);
1823}
1824
1825static void process_shared_bio(struct thin_c *tc, struct bio *bio,
1826 dm_block_t block,
1827 struct dm_thin_lookup_result *lookup_result,
1828 struct dm_bio_prison_cell *virt_cell)
1829{
1830 struct dm_bio_prison_cell *data_cell;
1831 struct pool *pool = tc->pool;
1832 struct dm_cell_key key;
1833
1834
1835
1836
1837
1838 build_data_key(tc->td, lookup_result->block, &key);
1839 if (bio_detain(pool, &key, bio, &data_cell)) {
1840 cell_defer_no_holder(tc, virt_cell);
1841 return;
1842 }
1843
1844 if (bio_data_dir(bio) == WRITE && bio->bi_iter.bi_size) {
1845 break_sharing(tc, bio, block, &key, lookup_result, data_cell);
1846 cell_defer_no_holder(tc, virt_cell);
1847 } else {
1848 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
1849
1850 h->shared_read_entry = dm_deferred_entry_inc(pool->shared_read_ds);
1851 inc_all_io_entry(pool, bio);
1852 remap_and_issue(tc, bio, lookup_result->block);
1853
1854 remap_and_issue_shared_cell(tc, data_cell, lookup_result->block);
1855 remap_and_issue_shared_cell(tc, virt_cell, lookup_result->block);
1856 }
1857}
1858
1859static void provision_block(struct thin_c *tc, struct bio *bio, dm_block_t block,
1860 struct dm_bio_prison_cell *cell)
1861{
1862 int r;
1863 dm_block_t data_block;
1864 struct pool *pool = tc->pool;
1865
1866
1867
1868
1869 if (!bio->bi_iter.bi_size) {
1870 inc_all_io_entry(pool, bio);
1871 cell_defer_no_holder(tc, cell);
1872
1873 remap_and_issue(tc, bio, 0);
1874 return;
1875 }
1876
1877
1878
1879
1880 if (bio_data_dir(bio) == READ) {
1881 zero_fill_bio(bio);
1882 cell_defer_no_holder(tc, cell);
1883 bio_endio(bio);
1884 return;
1885 }
1886
1887 r = alloc_data_block(tc, &data_block);
1888 switch (r) {
1889 case 0:
1890 if (tc->origin_dev)
1891 schedule_external_copy(tc, block, data_block, cell, bio);
1892 else
1893 schedule_zero(tc, block, data_block, cell, bio);
1894 break;
1895
1896 case -ENOSPC:
1897 retry_bios_on_resume(pool, cell);
1898 break;
1899
1900 default:
1901 DMERR_LIMIT("%s: alloc_data_block() failed: error = %d",
1902 __func__, r);
1903 cell_error(pool, cell);
1904 break;
1905 }
1906}
1907
1908static void process_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
1909{
1910 int r;
1911 struct pool *pool = tc->pool;
1912 struct bio *bio = cell->holder;
1913 dm_block_t block = get_bio_block(tc, bio);
1914 struct dm_thin_lookup_result lookup_result;
1915
1916 if (tc->requeue_mode) {
1917 cell_requeue(pool, cell);
1918 return;
1919 }
1920
1921 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1922 switch (r) {
1923 case 0:
1924 if (lookup_result.shared)
1925 process_shared_bio(tc, bio, block, &lookup_result, cell);
1926 else {
1927 inc_all_io_entry(pool, bio);
1928 remap_and_issue(tc, bio, lookup_result.block);
1929 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
1930 }
1931 break;
1932
1933 case -ENODATA:
1934 if (bio_data_dir(bio) == READ && tc->origin_dev) {
1935 inc_all_io_entry(pool, bio);
1936 cell_defer_no_holder(tc, cell);
1937
1938 if (bio_end_sector(bio) <= tc->origin_size)
1939 remap_to_origin_and_issue(tc, bio);
1940
1941 else if (bio->bi_iter.bi_sector < tc->origin_size) {
1942 zero_fill_bio(bio);
1943 bio->bi_iter.bi_size = (tc->origin_size - bio->bi_iter.bi_sector) << SECTOR_SHIFT;
1944 remap_to_origin_and_issue(tc, bio);
1945
1946 } else {
1947 zero_fill_bio(bio);
1948 bio_endio(bio);
1949 }
1950 } else
1951 provision_block(tc, bio, block, cell);
1952 break;
1953
1954 default:
1955 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
1956 __func__, r);
1957 cell_defer_no_holder(tc, cell);
1958 bio_io_error(bio);
1959 break;
1960 }
1961}
1962
1963static void process_bio(struct thin_c *tc, struct bio *bio)
1964{
1965 struct pool *pool = tc->pool;
1966 dm_block_t block = get_bio_block(tc, bio);
1967 struct dm_bio_prison_cell *cell;
1968 struct dm_cell_key key;
1969
1970
1971
1972
1973
1974 build_virtual_key(tc->td, block, &key);
1975 if (bio_detain(pool, &key, bio, &cell))
1976 return;
1977
1978 process_cell(tc, cell);
1979}
1980
1981static void __process_bio_read_only(struct thin_c *tc, struct bio *bio,
1982 struct dm_bio_prison_cell *cell)
1983{
1984 int r;
1985 int rw = bio_data_dir(bio);
1986 dm_block_t block = get_bio_block(tc, bio);
1987 struct dm_thin_lookup_result lookup_result;
1988
1989 r = dm_thin_find_block(tc->td, block, 1, &lookup_result);
1990 switch (r) {
1991 case 0:
1992 if (lookup_result.shared && (rw == WRITE) && bio->bi_iter.bi_size) {
1993 handle_unserviceable_bio(tc->pool, bio);
1994 if (cell)
1995 cell_defer_no_holder(tc, cell);
1996 } else {
1997 inc_all_io_entry(tc->pool, bio);
1998 remap_and_issue(tc, bio, lookup_result.block);
1999 if (cell)
2000 inc_remap_and_issue_cell(tc, cell, lookup_result.block);
2001 }
2002 break;
2003
2004 case -ENODATA:
2005 if (cell)
2006 cell_defer_no_holder(tc, cell);
2007 if (rw != READ) {
2008 handle_unserviceable_bio(tc->pool, bio);
2009 break;
2010 }
2011
2012 if (tc->origin_dev) {
2013 inc_all_io_entry(tc->pool, bio);
2014 remap_to_origin_and_issue(tc, bio);
2015 break;
2016 }
2017
2018 zero_fill_bio(bio);
2019 bio_endio(bio);
2020 break;
2021
2022 default:
2023 DMERR_LIMIT("%s: dm_thin_find_block() failed: error = %d",
2024 __func__, r);
2025 if (cell)
2026 cell_defer_no_holder(tc, cell);
2027 bio_io_error(bio);
2028 break;
2029 }
2030}
2031
2032static void process_bio_read_only(struct thin_c *tc, struct bio *bio)
2033{
2034 __process_bio_read_only(tc, bio, NULL);
2035}
2036
2037static void process_cell_read_only(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2038{
2039 __process_bio_read_only(tc, cell->holder, cell);
2040}
2041
2042static void process_bio_success(struct thin_c *tc, struct bio *bio)
2043{
2044 bio_endio(bio);
2045}
2046
2047static void process_bio_fail(struct thin_c *tc, struct bio *bio)
2048{
2049 bio_io_error(bio);
2050}
2051
2052static void process_cell_success(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2053{
2054 cell_success(tc->pool, cell);
2055}
2056
2057static void process_cell_fail(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2058{
2059 cell_error(tc->pool, cell);
2060}
2061
2062
2063
2064
2065
2066static int need_commit_due_to_time(struct pool *pool)
2067{
2068 return !time_in_range(jiffies, pool->last_commit_jiffies,
2069 pool->last_commit_jiffies + COMMIT_PERIOD);
2070}
2071
2072#define thin_pbd(node) rb_entry((node), struct dm_thin_endio_hook, rb_node)
2073#define thin_bio(pbd) dm_bio_from_per_bio_data((pbd), sizeof(struct dm_thin_endio_hook))
2074
2075static void __thin_bio_rb_add(struct thin_c *tc, struct bio *bio)
2076{
2077 struct rb_node **rbp, *parent;
2078 struct dm_thin_endio_hook *pbd;
2079 sector_t bi_sector = bio->bi_iter.bi_sector;
2080
2081 rbp = &tc->sort_bio_list.rb_node;
2082 parent = NULL;
2083 while (*rbp) {
2084 parent = *rbp;
2085 pbd = thin_pbd(parent);
2086
2087 if (bi_sector < thin_bio(pbd)->bi_iter.bi_sector)
2088 rbp = &(*rbp)->rb_left;
2089 else
2090 rbp = &(*rbp)->rb_right;
2091 }
2092
2093 pbd = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2094 rb_link_node(&pbd->rb_node, parent, rbp);
2095 rb_insert_color(&pbd->rb_node, &tc->sort_bio_list);
2096}
2097
2098static void __extract_sorted_bios(struct thin_c *tc)
2099{
2100 struct rb_node *node;
2101 struct dm_thin_endio_hook *pbd;
2102 struct bio *bio;
2103
2104 for (node = rb_first(&tc->sort_bio_list); node; node = rb_next(node)) {
2105 pbd = thin_pbd(node);
2106 bio = thin_bio(pbd);
2107
2108 bio_list_add(&tc->deferred_bio_list, bio);
2109 rb_erase(&pbd->rb_node, &tc->sort_bio_list);
2110 }
2111
2112 WARN_ON(!RB_EMPTY_ROOT(&tc->sort_bio_list));
2113}
2114
2115static void __sort_thin_deferred_bios(struct thin_c *tc)
2116{
2117 struct bio *bio;
2118 struct bio_list bios;
2119
2120 bio_list_init(&bios);
2121 bio_list_merge(&bios, &tc->deferred_bio_list);
2122 bio_list_init(&tc->deferred_bio_list);
2123
2124
2125 while ((bio = bio_list_pop(&bios)))
2126 __thin_bio_rb_add(tc, bio);
2127
2128
2129
2130
2131
2132
2133 __extract_sorted_bios(tc);
2134}
2135
2136static void process_thin_deferred_bios(struct thin_c *tc)
2137{
2138 struct pool *pool = tc->pool;
2139 unsigned long flags;
2140 struct bio *bio;
2141 struct bio_list bios;
2142 struct blk_plug plug;
2143 unsigned count = 0;
2144
2145 if (tc->requeue_mode) {
2146 error_thin_bio_list(tc, &tc->deferred_bio_list,
2147 BLK_STS_DM_REQUEUE);
2148 return;
2149 }
2150
2151 bio_list_init(&bios);
2152
2153 spin_lock_irqsave(&tc->lock, flags);
2154
2155 if (bio_list_empty(&tc->deferred_bio_list)) {
2156 spin_unlock_irqrestore(&tc->lock, flags);
2157 return;
2158 }
2159
2160 __sort_thin_deferred_bios(tc);
2161
2162 bio_list_merge(&bios, &tc->deferred_bio_list);
2163 bio_list_init(&tc->deferred_bio_list);
2164
2165 spin_unlock_irqrestore(&tc->lock, flags);
2166
2167 blk_start_plug(&plug);
2168 while ((bio = bio_list_pop(&bios))) {
2169
2170
2171
2172
2173
2174 if (ensure_next_mapping(pool)) {
2175 spin_lock_irqsave(&tc->lock, flags);
2176 bio_list_add(&tc->deferred_bio_list, bio);
2177 bio_list_merge(&tc->deferred_bio_list, &bios);
2178 spin_unlock_irqrestore(&tc->lock, flags);
2179 break;
2180 }
2181
2182 if (bio_op(bio) == REQ_OP_DISCARD)
2183 pool->process_discard(tc, bio);
2184 else
2185 pool->process_bio(tc, bio);
2186
2187 if ((count++ & 127) == 0) {
2188 throttle_work_update(&pool->throttle);
2189 dm_pool_issue_prefetches(pool->pmd);
2190 }
2191 }
2192 blk_finish_plug(&plug);
2193}
2194
2195static int cmp_cells(const void *lhs, const void *rhs)
2196{
2197 struct dm_bio_prison_cell *lhs_cell = *((struct dm_bio_prison_cell **) lhs);
2198 struct dm_bio_prison_cell *rhs_cell = *((struct dm_bio_prison_cell **) rhs);
2199
2200 BUG_ON(!lhs_cell->holder);
2201 BUG_ON(!rhs_cell->holder);
2202
2203 if (lhs_cell->holder->bi_iter.bi_sector < rhs_cell->holder->bi_iter.bi_sector)
2204 return -1;
2205
2206 if (lhs_cell->holder->bi_iter.bi_sector > rhs_cell->holder->bi_iter.bi_sector)
2207 return 1;
2208
2209 return 0;
2210}
2211
2212static unsigned sort_cells(struct pool *pool, struct list_head *cells)
2213{
2214 unsigned count = 0;
2215 struct dm_bio_prison_cell *cell, *tmp;
2216
2217 list_for_each_entry_safe(cell, tmp, cells, user_list) {
2218 if (count >= CELL_SORT_ARRAY_SIZE)
2219 break;
2220
2221 pool->cell_sort_array[count++] = cell;
2222 list_del(&cell->user_list);
2223 }
2224
2225 sort(pool->cell_sort_array, count, sizeof(cell), cmp_cells, NULL);
2226
2227 return count;
2228}
2229
2230static void process_thin_deferred_cells(struct thin_c *tc)
2231{
2232 struct pool *pool = tc->pool;
2233 unsigned long flags;
2234 struct list_head cells;
2235 struct dm_bio_prison_cell *cell;
2236 unsigned i, j, count;
2237
2238 INIT_LIST_HEAD(&cells);
2239
2240 spin_lock_irqsave(&tc->lock, flags);
2241 list_splice_init(&tc->deferred_cells, &cells);
2242 spin_unlock_irqrestore(&tc->lock, flags);
2243
2244 if (list_empty(&cells))
2245 return;
2246
2247 do {
2248 count = sort_cells(tc->pool, &cells);
2249
2250 for (i = 0; i < count; i++) {
2251 cell = pool->cell_sort_array[i];
2252 BUG_ON(!cell->holder);
2253
2254
2255
2256
2257
2258
2259 if (ensure_next_mapping(pool)) {
2260 for (j = i; j < count; j++)
2261 list_add(&pool->cell_sort_array[j]->user_list, &cells);
2262
2263 spin_lock_irqsave(&tc->lock, flags);
2264 list_splice(&cells, &tc->deferred_cells);
2265 spin_unlock_irqrestore(&tc->lock, flags);
2266 return;
2267 }
2268
2269 if (bio_op(cell->holder) == REQ_OP_DISCARD)
2270 pool->process_discard_cell(tc, cell);
2271 else
2272 pool->process_cell(tc, cell);
2273 }
2274 } while (!list_empty(&cells));
2275}
2276
2277static void thin_get(struct thin_c *tc);
2278static void thin_put(struct thin_c *tc);
2279
2280
2281
2282
2283
2284
2285static struct thin_c *get_first_thin(struct pool *pool)
2286{
2287 struct thin_c *tc = NULL;
2288
2289 rcu_read_lock();
2290 if (!list_empty(&pool->active_thins)) {
2291 tc = list_entry_rcu(pool->active_thins.next, struct thin_c, list);
2292 thin_get(tc);
2293 }
2294 rcu_read_unlock();
2295
2296 return tc;
2297}
2298
2299static struct thin_c *get_next_thin(struct pool *pool, struct thin_c *tc)
2300{
2301 struct thin_c *old_tc = tc;
2302
2303 rcu_read_lock();
2304 list_for_each_entry_continue_rcu(tc, &pool->active_thins, list) {
2305 thin_get(tc);
2306 thin_put(old_tc);
2307 rcu_read_unlock();
2308 return tc;
2309 }
2310 thin_put(old_tc);
2311 rcu_read_unlock();
2312
2313 return NULL;
2314}
2315
2316static void process_deferred_bios(struct pool *pool)
2317{
2318 unsigned long flags;
2319 struct bio *bio;
2320 struct bio_list bios;
2321 struct thin_c *tc;
2322
2323 tc = get_first_thin(pool);
2324 while (tc) {
2325 process_thin_deferred_cells(tc);
2326 process_thin_deferred_bios(tc);
2327 tc = get_next_thin(pool, tc);
2328 }
2329
2330
2331
2332
2333
2334 bio_list_init(&bios);
2335 spin_lock_irqsave(&pool->lock, flags);
2336 bio_list_merge(&bios, &pool->deferred_flush_bios);
2337 bio_list_init(&pool->deferred_flush_bios);
2338 spin_unlock_irqrestore(&pool->lock, flags);
2339
2340 if (bio_list_empty(&bios) &&
2341 !(dm_pool_changed_this_transaction(pool->pmd) && need_commit_due_to_time(pool)))
2342 return;
2343
2344 if (commit(pool)) {
2345 while ((bio = bio_list_pop(&bios)))
2346 bio_io_error(bio);
2347 return;
2348 }
2349 pool->last_commit_jiffies = jiffies;
2350
2351 while ((bio = bio_list_pop(&bios)))
2352 generic_make_request(bio);
2353}
2354
2355static void do_worker(struct work_struct *ws)
2356{
2357 struct pool *pool = container_of(ws, struct pool, worker);
2358
2359 throttle_work_start(&pool->throttle);
2360 dm_pool_issue_prefetches(pool->pmd);
2361 throttle_work_update(&pool->throttle);
2362 process_prepared(pool, &pool->prepared_mappings, &pool->process_prepared_mapping);
2363 throttle_work_update(&pool->throttle);
2364 process_prepared(pool, &pool->prepared_discards, &pool->process_prepared_discard);
2365 throttle_work_update(&pool->throttle);
2366 process_prepared(pool, &pool->prepared_discards_pt2, &pool->process_prepared_discard_pt2);
2367 throttle_work_update(&pool->throttle);
2368 process_deferred_bios(pool);
2369 throttle_work_complete(&pool->throttle);
2370}
2371
2372
2373
2374
2375
2376static void do_waker(struct work_struct *ws)
2377{
2378 struct pool *pool = container_of(to_delayed_work(ws), struct pool, waker);
2379 wake_worker(pool);
2380 queue_delayed_work(pool->wq, &pool->waker, COMMIT_PERIOD);
2381}
2382
2383
2384
2385
2386
2387
2388static void do_no_space_timeout(struct work_struct *ws)
2389{
2390 struct pool *pool = container_of(to_delayed_work(ws), struct pool,
2391 no_space_timeout);
2392
2393 if (get_pool_mode(pool) == PM_OUT_OF_DATA_SPACE && !pool->pf.error_if_no_space) {
2394 pool->pf.error_if_no_space = true;
2395 notify_of_pool_mode_change(pool);
2396 error_retry_list_with_code(pool, BLK_STS_NOSPC);
2397 }
2398}
2399
2400
2401
2402struct pool_work {
2403 struct work_struct worker;
2404 struct completion complete;
2405};
2406
2407static struct pool_work *to_pool_work(struct work_struct *ws)
2408{
2409 return container_of(ws, struct pool_work, worker);
2410}
2411
2412static void pool_work_complete(struct pool_work *pw)
2413{
2414 complete(&pw->complete);
2415}
2416
2417static void pool_work_wait(struct pool_work *pw, struct pool *pool,
2418 void (*fn)(struct work_struct *))
2419{
2420 INIT_WORK_ONSTACK(&pw->worker, fn);
2421 init_completion(&pw->complete);
2422 queue_work(pool->wq, &pw->worker);
2423 wait_for_completion(&pw->complete);
2424}
2425
2426
2427
2428struct noflush_work {
2429 struct pool_work pw;
2430 struct thin_c *tc;
2431};
2432
2433static struct noflush_work *to_noflush(struct work_struct *ws)
2434{
2435 return container_of(to_pool_work(ws), struct noflush_work, pw);
2436}
2437
2438static void do_noflush_start(struct work_struct *ws)
2439{
2440 struct noflush_work *w = to_noflush(ws);
2441 w->tc->requeue_mode = true;
2442 requeue_io(w->tc);
2443 pool_work_complete(&w->pw);
2444}
2445
2446static void do_noflush_stop(struct work_struct *ws)
2447{
2448 struct noflush_work *w = to_noflush(ws);
2449 w->tc->requeue_mode = false;
2450 pool_work_complete(&w->pw);
2451}
2452
2453static void noflush_work(struct thin_c *tc, void (*fn)(struct work_struct *))
2454{
2455 struct noflush_work w;
2456
2457 w.tc = tc;
2458 pool_work_wait(&w.pw, tc->pool, fn);
2459}
2460
2461
2462
2463static bool passdown_enabled(struct pool_c *pt)
2464{
2465 return pt->adjusted_pf.discard_passdown;
2466}
2467
2468static void set_discard_callbacks(struct pool *pool)
2469{
2470 struct pool_c *pt = pool->ti->private;
2471
2472 if (passdown_enabled(pt)) {
2473 pool->process_discard_cell = process_discard_cell_passdown;
2474 pool->process_prepared_discard = process_prepared_discard_passdown_pt1;
2475 pool->process_prepared_discard_pt2 = process_prepared_discard_passdown_pt2;
2476 } else {
2477 pool->process_discard_cell = process_discard_cell_no_passdown;
2478 pool->process_prepared_discard = process_prepared_discard_no_passdown;
2479 }
2480}
2481
2482static void set_pool_mode(struct pool *pool, enum pool_mode new_mode)
2483{
2484 struct pool_c *pt = pool->ti->private;
2485 bool needs_check = dm_pool_metadata_needs_check(pool->pmd);
2486 enum pool_mode old_mode = get_pool_mode(pool);
2487 unsigned long no_space_timeout = READ_ONCE(no_space_timeout_secs) * HZ;
2488
2489
2490
2491
2492
2493 if (new_mode == PM_WRITE && needs_check) {
2494 DMERR("%s: unable to switch pool to write mode until repaired.",
2495 dm_device_name(pool->pool_md));
2496 if (old_mode != new_mode)
2497 new_mode = old_mode;
2498 else
2499 new_mode = PM_READ_ONLY;
2500 }
2501
2502
2503
2504
2505
2506 if (old_mode == PM_FAIL)
2507 new_mode = old_mode;
2508
2509 switch (new_mode) {
2510 case PM_FAIL:
2511 dm_pool_metadata_read_only(pool->pmd);
2512 pool->process_bio = process_bio_fail;
2513 pool->process_discard = process_bio_fail;
2514 pool->process_cell = process_cell_fail;
2515 pool->process_discard_cell = process_cell_fail;
2516 pool->process_prepared_mapping = process_prepared_mapping_fail;
2517 pool->process_prepared_discard = process_prepared_discard_fail;
2518
2519 error_retry_list(pool);
2520 break;
2521
2522 case PM_OUT_OF_METADATA_SPACE:
2523 case PM_READ_ONLY:
2524 dm_pool_metadata_read_only(pool->pmd);
2525 pool->process_bio = process_bio_read_only;
2526 pool->process_discard = process_bio_success;
2527 pool->process_cell = process_cell_read_only;
2528 pool->process_discard_cell = process_cell_success;
2529 pool->process_prepared_mapping = process_prepared_mapping_fail;
2530 pool->process_prepared_discard = process_prepared_discard_success;
2531
2532 error_retry_list(pool);
2533 break;
2534
2535 case PM_OUT_OF_DATA_SPACE:
2536
2537
2538
2539
2540
2541
2542
2543
2544 pool->out_of_data_space = true;
2545 pool->process_bio = process_bio_read_only;
2546 pool->process_discard = process_discard_bio;
2547 pool->process_cell = process_cell_read_only;
2548 pool->process_prepared_mapping = process_prepared_mapping;
2549 set_discard_callbacks(pool);
2550
2551 if (!pool->pf.error_if_no_space && no_space_timeout)
2552 queue_delayed_work(pool->wq, &pool->no_space_timeout, no_space_timeout);
2553 break;
2554
2555 case PM_WRITE:
2556 if (old_mode == PM_OUT_OF_DATA_SPACE)
2557 cancel_delayed_work_sync(&pool->no_space_timeout);
2558 pool->out_of_data_space = false;
2559 pool->pf.error_if_no_space = pt->requested_pf.error_if_no_space;
2560 dm_pool_metadata_read_write(pool->pmd);
2561 pool->process_bio = process_bio;
2562 pool->process_discard = process_discard_bio;
2563 pool->process_cell = process_cell;
2564 pool->process_prepared_mapping = process_prepared_mapping;
2565 set_discard_callbacks(pool);
2566 break;
2567 }
2568
2569 pool->pf.mode = new_mode;
2570
2571
2572
2573
2574 pt->adjusted_pf.mode = new_mode;
2575
2576 if (old_mode != new_mode)
2577 notify_of_pool_mode_change(pool);
2578}
2579
2580static void abort_transaction(struct pool *pool)
2581{
2582 const char *dev_name = dm_device_name(pool->pool_md);
2583
2584 DMERR_LIMIT("%s: aborting current metadata transaction", dev_name);
2585 if (dm_pool_abort_metadata(pool->pmd)) {
2586 DMERR("%s: failed to abort metadata transaction", dev_name);
2587 set_pool_mode(pool, PM_FAIL);
2588 }
2589
2590 if (dm_pool_metadata_set_needs_check(pool->pmd)) {
2591 DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name);
2592 set_pool_mode(pool, PM_FAIL);
2593 }
2594}
2595
2596static void metadata_operation_failed(struct pool *pool, const char *op, int r)
2597{
2598 DMERR_LIMIT("%s: metadata operation '%s' failed: error = %d",
2599 dm_device_name(pool->pool_md), op, r);
2600
2601 abort_transaction(pool);
2602 set_pool_mode(pool, PM_READ_ONLY);
2603}
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614static void thin_defer_bio(struct thin_c *tc, struct bio *bio)
2615{
2616 unsigned long flags;
2617 struct pool *pool = tc->pool;
2618
2619 spin_lock_irqsave(&tc->lock, flags);
2620 bio_list_add(&tc->deferred_bio_list, bio);
2621 spin_unlock_irqrestore(&tc->lock, flags);
2622
2623 wake_worker(pool);
2624}
2625
2626static void thin_defer_bio_with_throttle(struct thin_c *tc, struct bio *bio)
2627{
2628 struct pool *pool = tc->pool;
2629
2630 throttle_lock(&pool->throttle);
2631 thin_defer_bio(tc, bio);
2632 throttle_unlock(&pool->throttle);
2633}
2634
2635static void thin_defer_cell(struct thin_c *tc, struct dm_bio_prison_cell *cell)
2636{
2637 unsigned long flags;
2638 struct pool *pool = tc->pool;
2639
2640 throttle_lock(&pool->throttle);
2641 spin_lock_irqsave(&tc->lock, flags);
2642 list_add_tail(&cell->user_list, &tc->deferred_cells);
2643 spin_unlock_irqrestore(&tc->lock, flags);
2644 throttle_unlock(&pool->throttle);
2645
2646 wake_worker(pool);
2647}
2648
2649static void thin_hook_bio(struct thin_c *tc, struct bio *bio)
2650{
2651 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
2652
2653 h->tc = tc;
2654 h->shared_read_entry = NULL;
2655 h->all_io_entry = NULL;
2656 h->overwrite_mapping = NULL;
2657 h->cell = NULL;
2658}
2659
2660
2661
2662
2663static int thin_bio_map(struct dm_target *ti, struct bio *bio)
2664{
2665 int r;
2666 struct thin_c *tc = ti->private;
2667 dm_block_t block = get_bio_block(tc, bio);
2668 struct dm_thin_device *td = tc->td;
2669 struct dm_thin_lookup_result result;
2670 struct dm_bio_prison_cell *virt_cell, *data_cell;
2671 struct dm_cell_key key;
2672
2673 thin_hook_bio(tc, bio);
2674
2675 if (tc->requeue_mode) {
2676 bio->bi_status = BLK_STS_DM_REQUEUE;
2677 bio_endio(bio);
2678 return DM_MAPIO_SUBMITTED;
2679 }
2680
2681 if (get_pool_mode(tc->pool) == PM_FAIL) {
2682 bio_io_error(bio);
2683 return DM_MAPIO_SUBMITTED;
2684 }
2685
2686 if (op_is_flush(bio->bi_opf) || bio_op(bio) == REQ_OP_DISCARD) {
2687 thin_defer_bio_with_throttle(tc, bio);
2688 return DM_MAPIO_SUBMITTED;
2689 }
2690
2691
2692
2693
2694
2695 build_virtual_key(tc->td, block, &key);
2696 if (bio_detain(tc->pool, &key, bio, &virt_cell))
2697 return DM_MAPIO_SUBMITTED;
2698
2699 r = dm_thin_find_block(td, block, 0, &result);
2700
2701
2702
2703
2704 switch (r) {
2705 case 0:
2706 if (unlikely(result.shared)) {
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721 thin_defer_cell(tc, virt_cell);
2722 return DM_MAPIO_SUBMITTED;
2723 }
2724
2725 build_data_key(tc->td, result.block, &key);
2726 if (bio_detain(tc->pool, &key, bio, &data_cell)) {
2727 cell_defer_no_holder(tc, virt_cell);
2728 return DM_MAPIO_SUBMITTED;
2729 }
2730
2731 inc_all_io_entry(tc->pool, bio);
2732 cell_defer_no_holder(tc, data_cell);
2733 cell_defer_no_holder(tc, virt_cell);
2734
2735 remap(tc, bio, result.block);
2736 return DM_MAPIO_REMAPPED;
2737
2738 case -ENODATA:
2739 case -EWOULDBLOCK:
2740 thin_defer_cell(tc, virt_cell);
2741 return DM_MAPIO_SUBMITTED;
2742
2743 default:
2744
2745
2746
2747
2748
2749 bio_io_error(bio);
2750 cell_defer_no_holder(tc, virt_cell);
2751 return DM_MAPIO_SUBMITTED;
2752 }
2753}
2754
2755static int pool_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
2756{
2757 struct pool_c *pt = container_of(cb, struct pool_c, callbacks);
2758 struct request_queue *q;
2759
2760 if (get_pool_mode(pt->pool) == PM_OUT_OF_DATA_SPACE)
2761 return 1;
2762
2763 q = bdev_get_queue(pt->data_dev->bdev);
2764 return bdi_congested(q->backing_dev_info, bdi_bits);
2765}
2766
2767static void requeue_bios(struct pool *pool)
2768{
2769 unsigned long flags;
2770 struct thin_c *tc;
2771
2772 rcu_read_lock();
2773 list_for_each_entry_rcu(tc, &pool->active_thins, list) {
2774 spin_lock_irqsave(&tc->lock, flags);
2775 bio_list_merge(&tc->deferred_bio_list, &tc->retry_on_resume_list);
2776 bio_list_init(&tc->retry_on_resume_list);
2777 spin_unlock_irqrestore(&tc->lock, flags);
2778 }
2779 rcu_read_unlock();
2780}
2781
2782
2783
2784
2785static bool data_dev_supports_discard(struct pool_c *pt)
2786{
2787 struct request_queue *q = bdev_get_queue(pt->data_dev->bdev);
2788
2789 return q && blk_queue_discard(q);
2790}
2791
2792static bool is_factor(sector_t block_size, uint32_t n)
2793{
2794 return !sector_div(block_size, n);
2795}
2796
2797
2798
2799
2800
2801static void disable_passdown_if_not_supported(struct pool_c *pt)
2802{
2803 struct pool *pool = pt->pool;
2804 struct block_device *data_bdev = pt->data_dev->bdev;
2805 struct queue_limits *data_limits = &bdev_get_queue(data_bdev)->limits;
2806 const char *reason = NULL;
2807 char buf[BDEVNAME_SIZE];
2808
2809 if (!pt->adjusted_pf.discard_passdown)
2810 return;
2811
2812 if (!data_dev_supports_discard(pt))
2813 reason = "discard unsupported";
2814
2815 else if (data_limits->max_discard_sectors < pool->sectors_per_block)
2816 reason = "max discard sectors smaller than a block";
2817
2818 if (reason) {
2819 DMWARN("Data device (%s) %s: Disabling discard passdown.", bdevname(data_bdev, buf), reason);
2820 pt->adjusted_pf.discard_passdown = false;
2821 }
2822}
2823
2824static int bind_control_target(struct pool *pool, struct dm_target *ti)
2825{
2826 struct pool_c *pt = ti->private;
2827
2828
2829
2830
2831 enum pool_mode old_mode = get_pool_mode(pool);
2832 enum pool_mode new_mode = pt->adjusted_pf.mode;
2833
2834
2835
2836
2837
2838
2839 pt->adjusted_pf.mode = old_mode;
2840
2841 pool->ti = ti;
2842 pool->pf = pt->adjusted_pf;
2843 pool->low_water_blocks = pt->low_water_blocks;
2844
2845 set_pool_mode(pool, new_mode);
2846
2847 return 0;
2848}
2849
2850static void unbind_control_target(struct pool *pool, struct dm_target *ti)
2851{
2852 if (pool->ti == ti)
2853 pool->ti = NULL;
2854}
2855
2856
2857
2858
2859
2860static void pool_features_init(struct pool_features *pf)
2861{
2862 pf->mode = PM_WRITE;
2863 pf->zero_new_blocks = true;
2864 pf->discard_enabled = true;
2865 pf->discard_passdown = true;
2866 pf->error_if_no_space = false;
2867}
2868
2869static void __pool_destroy(struct pool *pool)
2870{
2871 __pool_table_remove(pool);
2872
2873 vfree(pool->cell_sort_array);
2874 if (dm_pool_metadata_close(pool->pmd) < 0)
2875 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
2876
2877 dm_bio_prison_destroy(pool->prison);
2878 dm_kcopyd_client_destroy(pool->copier);
2879
2880 if (pool->wq)
2881 destroy_workqueue(pool->wq);
2882
2883 if (pool->next_mapping)
2884 mempool_free(pool->next_mapping, &pool->mapping_pool);
2885 mempool_exit(&pool->mapping_pool);
2886 dm_deferred_set_destroy(pool->shared_read_ds);
2887 dm_deferred_set_destroy(pool->all_io_ds);
2888 kfree(pool);
2889}
2890
2891static struct kmem_cache *_new_mapping_cache;
2892
2893static struct pool *pool_create(struct mapped_device *pool_md,
2894 struct block_device *metadata_dev,
2895 unsigned long block_size,
2896 int read_only, char **error)
2897{
2898 int r;
2899 void *err_p;
2900 struct pool *pool;
2901 struct dm_pool_metadata *pmd;
2902 bool format_device = read_only ? false : true;
2903
2904 pmd = dm_pool_metadata_open(metadata_dev, block_size, format_device);
2905 if (IS_ERR(pmd)) {
2906 *error = "Error creating metadata object";
2907 return (struct pool *)pmd;
2908 }
2909
2910 pool = kzalloc(sizeof(*pool), GFP_KERNEL);
2911 if (!pool) {
2912 *error = "Error allocating memory for pool";
2913 err_p = ERR_PTR(-ENOMEM);
2914 goto bad_pool;
2915 }
2916
2917 pool->pmd = pmd;
2918 pool->sectors_per_block = block_size;
2919 if (block_size & (block_size - 1))
2920 pool->sectors_per_block_shift = -1;
2921 else
2922 pool->sectors_per_block_shift = __ffs(block_size);
2923 pool->low_water_blocks = 0;
2924 pool_features_init(&pool->pf);
2925 pool->prison = dm_bio_prison_create();
2926 if (!pool->prison) {
2927 *error = "Error creating pool's bio prison";
2928 err_p = ERR_PTR(-ENOMEM);
2929 goto bad_prison;
2930 }
2931
2932 pool->copier = dm_kcopyd_client_create(&dm_kcopyd_throttle);
2933 if (IS_ERR(pool->copier)) {
2934 r = PTR_ERR(pool->copier);
2935 *error = "Error creating pool's kcopyd client";
2936 err_p = ERR_PTR(r);
2937 goto bad_kcopyd_client;
2938 }
2939
2940
2941
2942
2943
2944 pool->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
2945 if (!pool->wq) {
2946 *error = "Error creating pool's workqueue";
2947 err_p = ERR_PTR(-ENOMEM);
2948 goto bad_wq;
2949 }
2950
2951 throttle_init(&pool->throttle);
2952 INIT_WORK(&pool->worker, do_worker);
2953 INIT_DELAYED_WORK(&pool->waker, do_waker);
2954 INIT_DELAYED_WORK(&pool->no_space_timeout, do_no_space_timeout);
2955 spin_lock_init(&pool->lock);
2956 bio_list_init(&pool->deferred_flush_bios);
2957 INIT_LIST_HEAD(&pool->prepared_mappings);
2958 INIT_LIST_HEAD(&pool->prepared_discards);
2959 INIT_LIST_HEAD(&pool->prepared_discards_pt2);
2960 INIT_LIST_HEAD(&pool->active_thins);
2961 pool->low_water_triggered = false;
2962 pool->suspended = true;
2963 pool->out_of_data_space = false;
2964
2965 pool->shared_read_ds = dm_deferred_set_create();
2966 if (!pool->shared_read_ds) {
2967 *error = "Error creating pool's shared read deferred set";
2968 err_p = ERR_PTR(-ENOMEM);
2969 goto bad_shared_read_ds;
2970 }
2971
2972 pool->all_io_ds = dm_deferred_set_create();
2973 if (!pool->all_io_ds) {
2974 *error = "Error creating pool's all io deferred set";
2975 err_p = ERR_PTR(-ENOMEM);
2976 goto bad_all_io_ds;
2977 }
2978
2979 pool->next_mapping = NULL;
2980 r = mempool_init_slab_pool(&pool->mapping_pool, MAPPING_POOL_SIZE,
2981 _new_mapping_cache);
2982 if (r) {
2983 *error = "Error creating pool's mapping mempool";
2984 err_p = ERR_PTR(r);
2985 goto bad_mapping_pool;
2986 }
2987
2988 pool->cell_sort_array =
2989 vmalloc(array_size(CELL_SORT_ARRAY_SIZE,
2990 sizeof(*pool->cell_sort_array)));
2991 if (!pool->cell_sort_array) {
2992 *error = "Error allocating cell sort array";
2993 err_p = ERR_PTR(-ENOMEM);
2994 goto bad_sort_array;
2995 }
2996
2997 pool->ref_count = 1;
2998 pool->last_commit_jiffies = jiffies;
2999 pool->pool_md = pool_md;
3000 pool->md_dev = metadata_dev;
3001 __pool_table_insert(pool);
3002
3003 return pool;
3004
3005bad_sort_array:
3006 mempool_exit(&pool->mapping_pool);
3007bad_mapping_pool:
3008 dm_deferred_set_destroy(pool->all_io_ds);
3009bad_all_io_ds:
3010 dm_deferred_set_destroy(pool->shared_read_ds);
3011bad_shared_read_ds:
3012 destroy_workqueue(pool->wq);
3013bad_wq:
3014 dm_kcopyd_client_destroy(pool->copier);
3015bad_kcopyd_client:
3016 dm_bio_prison_destroy(pool->prison);
3017bad_prison:
3018 kfree(pool);
3019bad_pool:
3020 if (dm_pool_metadata_close(pmd))
3021 DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
3022
3023 return err_p;
3024}
3025
3026static void __pool_inc(struct pool *pool)
3027{
3028 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
3029 pool->ref_count++;
3030}
3031
3032static void __pool_dec(struct pool *pool)
3033{
3034 BUG_ON(!mutex_is_locked(&dm_thin_pool_table.mutex));
3035 BUG_ON(!pool->ref_count);
3036 if (!--pool->ref_count)
3037 __pool_destroy(pool);
3038}
3039
3040static struct pool *__pool_find(struct mapped_device *pool_md,
3041 struct block_device *metadata_dev,
3042 unsigned long block_size, int read_only,
3043 char **error, int *created)
3044{
3045 struct pool *pool = __pool_table_lookup_metadata_dev(metadata_dev);
3046
3047 if (pool) {
3048 if (pool->pool_md != pool_md) {
3049 *error = "metadata device already in use by a pool";
3050 return ERR_PTR(-EBUSY);
3051 }
3052 __pool_inc(pool);
3053
3054 } else {
3055 pool = __pool_table_lookup(pool_md);
3056 if (pool) {
3057 if (pool->md_dev != metadata_dev) {
3058 *error = "different pool cannot replace a pool";
3059 return ERR_PTR(-EINVAL);
3060 }
3061 __pool_inc(pool);
3062
3063 } else {
3064 pool = pool_create(pool_md, metadata_dev, block_size, read_only, error);
3065 *created = 1;
3066 }
3067 }
3068
3069 return pool;
3070}
3071
3072
3073
3074
3075static void pool_dtr(struct dm_target *ti)
3076{
3077 struct pool_c *pt = ti->private;
3078
3079 mutex_lock(&dm_thin_pool_table.mutex);
3080
3081 unbind_control_target(pt->pool, ti);
3082 __pool_dec(pt->pool);
3083 dm_put_device(ti, pt->metadata_dev);
3084 dm_put_device(ti, pt->data_dev);
3085 kfree(pt);
3086
3087 mutex_unlock(&dm_thin_pool_table.mutex);
3088}
3089
3090static int parse_pool_features(struct dm_arg_set *as, struct pool_features *pf,
3091 struct dm_target *ti)
3092{
3093 int r;
3094 unsigned argc;
3095 const char *arg_name;
3096
3097 static const struct dm_arg _args[] = {
3098 {0, 4, "Invalid number of pool feature arguments"},
3099 };
3100
3101
3102
3103
3104 if (!as->argc)
3105 return 0;
3106
3107 r = dm_read_arg_group(_args, as, &argc, &ti->error);
3108 if (r)
3109 return -EINVAL;
3110
3111 while (argc && !r) {
3112 arg_name = dm_shift_arg(as);
3113 argc--;
3114
3115 if (!strcasecmp(arg_name, "skip_block_zeroing"))
3116 pf->zero_new_blocks = false;
3117
3118 else if (!strcasecmp(arg_name, "ignore_discard"))
3119 pf->discard_enabled = false;
3120
3121 else if (!strcasecmp(arg_name, "no_discard_passdown"))
3122 pf->discard_passdown = false;
3123
3124 else if (!strcasecmp(arg_name, "read_only"))
3125 pf->mode = PM_READ_ONLY;
3126
3127 else if (!strcasecmp(arg_name, "error_if_no_space"))
3128 pf->error_if_no_space = true;
3129
3130 else {
3131 ti->error = "Unrecognised pool feature requested";
3132 r = -EINVAL;
3133 break;
3134 }
3135 }
3136
3137 return r;
3138}
3139
3140static void metadata_low_callback(void *context)
3141{
3142 struct pool *pool = context;
3143
3144 DMWARN("%s: reached low water mark for metadata device: sending event.",
3145 dm_device_name(pool->pool_md));
3146
3147 dm_table_event(pool->ti->table);
3148}
3149
3150static sector_t get_dev_size(struct block_device *bdev)
3151{
3152 return i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
3153}
3154
3155static void warn_if_metadata_device_too_big(struct block_device *bdev)
3156{
3157 sector_t metadata_dev_size = get_dev_size(bdev);
3158 char buffer[BDEVNAME_SIZE];
3159
3160 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS_WARNING)
3161 DMWARN("Metadata device %s is larger than %u sectors: excess space will not be used.",
3162 bdevname(bdev, buffer), THIN_METADATA_MAX_SECTORS);
3163}
3164
3165static sector_t get_metadata_dev_size(struct block_device *bdev)
3166{
3167 sector_t metadata_dev_size = get_dev_size(bdev);
3168
3169 if (metadata_dev_size > THIN_METADATA_MAX_SECTORS)
3170 metadata_dev_size = THIN_METADATA_MAX_SECTORS;
3171
3172 return metadata_dev_size;
3173}
3174
3175static dm_block_t get_metadata_dev_size_in_blocks(struct block_device *bdev)
3176{
3177 sector_t metadata_dev_size = get_metadata_dev_size(bdev);
3178
3179 sector_div(metadata_dev_size, THIN_METADATA_BLOCK_SIZE);
3180
3181 return metadata_dev_size;
3182}
3183
3184
3185
3186
3187
3188
3189
3190static dm_block_t calc_metadata_threshold(struct pool_c *pt)
3191{
3192
3193
3194
3195
3196
3197 dm_block_t quarter = get_metadata_dev_size_in_blocks(pt->metadata_dev->bdev) / 4;
3198 return min((dm_block_t)1024ULL , quarter);
3199}
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214static int pool_ctr(struct dm_target *ti, unsigned argc, char **argv)
3215{
3216 int r, pool_created = 0;
3217 struct pool_c *pt;
3218 struct pool *pool;
3219 struct pool_features pf;
3220 struct dm_arg_set as;
3221 struct dm_dev *data_dev;
3222 unsigned long block_size;
3223 dm_block_t low_water_blocks;
3224 struct dm_dev *metadata_dev;
3225 fmode_t metadata_mode;
3226
3227
3228
3229
3230 mutex_lock(&dm_thin_pool_table.mutex);
3231
3232 if (argc < 4) {
3233 ti->error = "Invalid argument count";
3234 r = -EINVAL;
3235 goto out_unlock;
3236 }
3237
3238 as.argc = argc;
3239 as.argv = argv;
3240
3241
3242
3243
3244 pool_features_init(&pf);
3245
3246 dm_consume_args(&as, 4);
3247 r = parse_pool_features(&as, &pf, ti);
3248 if (r)
3249 goto out_unlock;
3250
3251 metadata_mode = FMODE_READ | ((pf.mode == PM_READ_ONLY) ? 0 : FMODE_WRITE);
3252 r = dm_get_device(ti, argv[0], metadata_mode, &metadata_dev);
3253 if (r) {
3254 ti->error = "Error opening metadata block device";
3255 goto out_unlock;
3256 }
3257 warn_if_metadata_device_too_big(metadata_dev->bdev);
3258
3259 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &data_dev);
3260 if (r) {
3261 ti->error = "Error getting data device";
3262 goto out_metadata;
3263 }
3264
3265 if (kstrtoul(argv[2], 10, &block_size) || !block_size ||
3266 block_size < DATA_DEV_BLOCK_SIZE_MIN_SECTORS ||
3267 block_size > DATA_DEV_BLOCK_SIZE_MAX_SECTORS ||
3268 block_size & (DATA_DEV_BLOCK_SIZE_MIN_SECTORS - 1)) {
3269 ti->error = "Invalid block size";
3270 r = -EINVAL;
3271 goto out;
3272 }
3273
3274 if (kstrtoull(argv[3], 10, (unsigned long long *)&low_water_blocks)) {
3275 ti->error = "Invalid low water mark";
3276 r = -EINVAL;
3277 goto out;
3278 }
3279
3280 pt = kzalloc(sizeof(*pt), GFP_KERNEL);
3281 if (!pt) {
3282 r = -ENOMEM;
3283 goto out;
3284 }
3285
3286 pool = __pool_find(dm_table_get_md(ti->table), metadata_dev->bdev,
3287 block_size, pf.mode == PM_READ_ONLY, &ti->error, &pool_created);
3288 if (IS_ERR(pool)) {
3289 r = PTR_ERR(pool);
3290 goto out_free_pt;
3291 }
3292
3293
3294
3295
3296
3297
3298
3299 if (!pool_created && pf.discard_enabled != pool->pf.discard_enabled) {
3300 ti->error = "Discard support cannot be disabled once enabled";
3301 r = -EINVAL;
3302 goto out_flags_changed;
3303 }
3304
3305 pt->pool = pool;
3306 pt->ti = ti;
3307 pt->metadata_dev = metadata_dev;
3308 pt->data_dev = data_dev;
3309 pt->low_water_blocks = low_water_blocks;
3310 pt->adjusted_pf = pt->requested_pf = pf;
3311 ti->num_flush_bios = 1;
3312
3313
3314
3315
3316
3317
3318 if (pf.discard_enabled && pf.discard_passdown) {
3319 ti->num_discard_bios = 1;
3320
3321
3322
3323
3324
3325
3326 ti->discards_supported = true;
3327 }
3328 ti->private = pt;
3329
3330 r = dm_pool_register_metadata_threshold(pt->pool->pmd,
3331 calc_metadata_threshold(pt),
3332 metadata_low_callback,
3333 pool);
3334 if (r)
3335 goto out_flags_changed;
3336
3337 pt->callbacks.congested_fn = pool_is_congested;
3338 dm_table_add_target_callbacks(ti->table, &pt->callbacks);
3339
3340 mutex_unlock(&dm_thin_pool_table.mutex);
3341
3342 return 0;
3343
3344out_flags_changed:
3345 __pool_dec(pool);
3346out_free_pt:
3347 kfree(pt);
3348out:
3349 dm_put_device(ti, data_dev);
3350out_metadata:
3351 dm_put_device(ti, metadata_dev);
3352out_unlock:
3353 mutex_unlock(&dm_thin_pool_table.mutex);
3354
3355 return r;
3356}
3357
3358static int pool_map(struct dm_target *ti, struct bio *bio)
3359{
3360 int r;
3361 struct pool_c *pt = ti->private;
3362 struct pool *pool = pt->pool;
3363 unsigned long flags;
3364
3365
3366
3367
3368 spin_lock_irqsave(&pool->lock, flags);
3369 bio_set_dev(bio, pt->data_dev->bdev);
3370 r = DM_MAPIO_REMAPPED;
3371 spin_unlock_irqrestore(&pool->lock, flags);
3372
3373 return r;
3374}
3375
3376static int maybe_resize_data_dev(struct dm_target *ti, bool *need_commit)
3377{
3378 int r;
3379 struct pool_c *pt = ti->private;
3380 struct pool *pool = pt->pool;
3381 sector_t data_size = ti->len;
3382 dm_block_t sb_data_size;
3383
3384 *need_commit = false;
3385
3386 (void) sector_div(data_size, pool->sectors_per_block);
3387
3388 r = dm_pool_get_data_dev_size(pool->pmd, &sb_data_size);
3389 if (r) {
3390 DMERR("%s: failed to retrieve data device size",
3391 dm_device_name(pool->pool_md));
3392 return r;
3393 }
3394
3395 if (data_size < sb_data_size) {
3396 DMERR("%s: pool target (%llu blocks) too small: expected %llu",
3397 dm_device_name(pool->pool_md),
3398 (unsigned long long)data_size, sb_data_size);
3399 return -EINVAL;
3400
3401 } else if (data_size > sb_data_size) {
3402 if (dm_pool_metadata_needs_check(pool->pmd)) {
3403 DMERR("%s: unable to grow the data device until repaired.",
3404 dm_device_name(pool->pool_md));
3405 return 0;
3406 }
3407
3408 if (sb_data_size)
3409 DMINFO("%s: growing the data device from %llu to %llu blocks",
3410 dm_device_name(pool->pool_md),
3411 sb_data_size, (unsigned long long)data_size);
3412 r = dm_pool_resize_data_dev(pool->pmd, data_size);
3413 if (r) {
3414 metadata_operation_failed(pool, "dm_pool_resize_data_dev", r);
3415 return r;
3416 }
3417
3418 *need_commit = true;
3419 }
3420
3421 return 0;
3422}
3423
3424static int maybe_resize_metadata_dev(struct dm_target *ti, bool *need_commit)
3425{
3426 int r;
3427 struct pool_c *pt = ti->private;
3428 struct pool *pool = pt->pool;
3429 dm_block_t metadata_dev_size, sb_metadata_dev_size;
3430
3431 *need_commit = false;
3432
3433 metadata_dev_size = get_metadata_dev_size_in_blocks(pool->md_dev);
3434
3435 r = dm_pool_get_metadata_dev_size(pool->pmd, &sb_metadata_dev_size);
3436 if (r) {
3437 DMERR("%s: failed to retrieve metadata device size",
3438 dm_device_name(pool->pool_md));
3439 return r;
3440 }
3441
3442 if (metadata_dev_size < sb_metadata_dev_size) {
3443 DMERR("%s: metadata device (%llu blocks) too small: expected %llu",
3444 dm_device_name(pool->pool_md),
3445 metadata_dev_size, sb_metadata_dev_size);
3446 return -EINVAL;
3447
3448 } else if (metadata_dev_size > sb_metadata_dev_size) {
3449 if (dm_pool_metadata_needs_check(pool->pmd)) {
3450 DMERR("%s: unable to grow the metadata device until repaired.",
3451 dm_device_name(pool->pool_md));
3452 return 0;
3453 }
3454
3455 warn_if_metadata_device_too_big(pool->md_dev);
3456 DMINFO("%s: growing the metadata device from %llu to %llu blocks",
3457 dm_device_name(pool->pool_md),
3458 sb_metadata_dev_size, metadata_dev_size);
3459
3460 if (get_pool_mode(pool) == PM_OUT_OF_METADATA_SPACE)
3461 set_pool_mode(pool, PM_WRITE);
3462
3463 r = dm_pool_resize_metadata_dev(pool->pmd, metadata_dev_size);
3464 if (r) {
3465 metadata_operation_failed(pool, "dm_pool_resize_metadata_dev", r);
3466 return r;
3467 }
3468
3469 *need_commit = true;
3470 }
3471
3472 return 0;
3473}
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486static int pool_preresume(struct dm_target *ti)
3487{
3488 int r;
3489 bool need_commit1, need_commit2;
3490 struct pool_c *pt = ti->private;
3491 struct pool *pool = pt->pool;
3492
3493
3494
3495
3496 r = bind_control_target(pool, ti);
3497 if (r)
3498 return r;
3499
3500 r = maybe_resize_data_dev(ti, &need_commit1);
3501 if (r)
3502 return r;
3503
3504 r = maybe_resize_metadata_dev(ti, &need_commit2);
3505 if (r)
3506 return r;
3507
3508 if (need_commit1 || need_commit2)
3509 (void) commit(pool);
3510
3511 return 0;
3512}
3513
3514static void pool_suspend_active_thins(struct pool *pool)
3515{
3516 struct thin_c *tc;
3517
3518
3519 tc = get_first_thin(pool);
3520 while (tc) {
3521 dm_internal_suspend_noflush(tc->thin_md);
3522 tc = get_next_thin(pool, tc);
3523 }
3524}
3525
3526static void pool_resume_active_thins(struct pool *pool)
3527{
3528 struct thin_c *tc;
3529
3530
3531 tc = get_first_thin(pool);
3532 while (tc) {
3533 dm_internal_resume(tc->thin_md);
3534 tc = get_next_thin(pool, tc);
3535 }
3536}
3537
3538static void pool_resume(struct dm_target *ti)
3539{
3540 struct pool_c *pt = ti->private;
3541 struct pool *pool = pt->pool;
3542 unsigned long flags;
3543
3544
3545
3546
3547
3548 requeue_bios(pool);
3549 pool_resume_active_thins(pool);
3550
3551 spin_lock_irqsave(&pool->lock, flags);
3552 pool->low_water_triggered = false;
3553 pool->suspended = false;
3554 spin_unlock_irqrestore(&pool->lock, flags);
3555
3556 do_waker(&pool->waker.work);
3557}
3558
3559static void pool_presuspend(struct dm_target *ti)
3560{
3561 struct pool_c *pt = ti->private;
3562 struct pool *pool = pt->pool;
3563 unsigned long flags;
3564
3565 spin_lock_irqsave(&pool->lock, flags);
3566 pool->suspended = true;
3567 spin_unlock_irqrestore(&pool->lock, flags);
3568
3569 pool_suspend_active_thins(pool);
3570}
3571
3572static void pool_presuspend_undo(struct dm_target *ti)
3573{
3574 struct pool_c *pt = ti->private;
3575 struct pool *pool = pt->pool;
3576 unsigned long flags;
3577
3578 pool_resume_active_thins(pool);
3579
3580 spin_lock_irqsave(&pool->lock, flags);
3581 pool->suspended = false;
3582 spin_unlock_irqrestore(&pool->lock, flags);
3583}
3584
3585static void pool_postsuspend(struct dm_target *ti)
3586{
3587 struct pool_c *pt = ti->private;
3588 struct pool *pool = pt->pool;
3589
3590 cancel_delayed_work_sync(&pool->waker);
3591 cancel_delayed_work_sync(&pool->no_space_timeout);
3592 flush_workqueue(pool->wq);
3593 (void) commit(pool);
3594}
3595
3596static int check_arg_count(unsigned argc, unsigned args_required)
3597{
3598 if (argc != args_required) {
3599 DMWARN("Message received with %u arguments instead of %u.",
3600 argc, args_required);
3601 return -EINVAL;
3602 }
3603
3604 return 0;
3605}
3606
3607static int read_dev_id(char *arg, dm_thin_id *dev_id, int warning)
3608{
3609 if (!kstrtoull(arg, 10, (unsigned long long *)dev_id) &&
3610 *dev_id <= MAX_DEV_ID)
3611 return 0;
3612
3613 if (warning)
3614 DMWARN("Message received with invalid device id: %s", arg);
3615
3616 return -EINVAL;
3617}
3618
3619static int process_create_thin_mesg(unsigned argc, char **argv, struct pool *pool)
3620{
3621 dm_thin_id dev_id;
3622 int r;
3623
3624 r = check_arg_count(argc, 2);
3625 if (r)
3626 return r;
3627
3628 r = read_dev_id(argv[1], &dev_id, 1);
3629 if (r)
3630 return r;
3631
3632 r = dm_pool_create_thin(pool->pmd, dev_id);
3633 if (r) {
3634 DMWARN("Creation of new thinly-provisioned device with id %s failed.",
3635 argv[1]);
3636 return r;
3637 }
3638
3639 return 0;
3640}
3641
3642static int process_create_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3643{
3644 dm_thin_id dev_id;
3645 dm_thin_id origin_dev_id;
3646 int r;
3647
3648 r = check_arg_count(argc, 3);
3649 if (r)
3650 return r;
3651
3652 r = read_dev_id(argv[1], &dev_id, 1);
3653 if (r)
3654 return r;
3655
3656 r = read_dev_id(argv[2], &origin_dev_id, 1);
3657 if (r)
3658 return r;
3659
3660 r = dm_pool_create_snap(pool->pmd, dev_id, origin_dev_id);
3661 if (r) {
3662 DMWARN("Creation of new snapshot %s of device %s failed.",
3663 argv[1], argv[2]);
3664 return r;
3665 }
3666
3667 return 0;
3668}
3669
3670static int process_delete_mesg(unsigned argc, char **argv, struct pool *pool)
3671{
3672 dm_thin_id dev_id;
3673 int r;
3674
3675 r = check_arg_count(argc, 2);
3676 if (r)
3677 return r;
3678
3679 r = read_dev_id(argv[1], &dev_id, 1);
3680 if (r)
3681 return r;
3682
3683 r = dm_pool_delete_thin_device(pool->pmd, dev_id);
3684 if (r)
3685 DMWARN("Deletion of thin device %s failed.", argv[1]);
3686
3687 return r;
3688}
3689
3690static int process_set_transaction_id_mesg(unsigned argc, char **argv, struct pool *pool)
3691{
3692 dm_thin_id old_id, new_id;
3693 int r;
3694
3695 r = check_arg_count(argc, 3);
3696 if (r)
3697 return r;
3698
3699 if (kstrtoull(argv[1], 10, (unsigned long long *)&old_id)) {
3700 DMWARN("set_transaction_id message: Unrecognised id %s.", argv[1]);
3701 return -EINVAL;
3702 }
3703
3704 if (kstrtoull(argv[2], 10, (unsigned long long *)&new_id)) {
3705 DMWARN("set_transaction_id message: Unrecognised new id %s.", argv[2]);
3706 return -EINVAL;
3707 }
3708
3709 r = dm_pool_set_metadata_transaction_id(pool->pmd, old_id, new_id);
3710 if (r) {
3711 DMWARN("Failed to change transaction id from %s to %s.",
3712 argv[1], argv[2]);
3713 return r;
3714 }
3715
3716 return 0;
3717}
3718
3719static int process_reserve_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3720{
3721 int r;
3722
3723 r = check_arg_count(argc, 1);
3724 if (r)
3725 return r;
3726
3727 (void) commit(pool);
3728
3729 r = dm_pool_reserve_metadata_snap(pool->pmd);
3730 if (r)
3731 DMWARN("reserve_metadata_snap message failed.");
3732
3733 return r;
3734}
3735
3736static int process_release_metadata_snap_mesg(unsigned argc, char **argv, struct pool *pool)
3737{
3738 int r;
3739
3740 r = check_arg_count(argc, 1);
3741 if (r)
3742 return r;
3743
3744 r = dm_pool_release_metadata_snap(pool->pmd);
3745 if (r)
3746 DMWARN("release_metadata_snap message failed.");
3747
3748 return r;
3749}
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760static int pool_message(struct dm_target *ti, unsigned argc, char **argv,
3761 char *result, unsigned maxlen)
3762{
3763 int r = -EINVAL;
3764 struct pool_c *pt = ti->private;
3765 struct pool *pool = pt->pool;
3766
3767 if (get_pool_mode(pool) >= PM_OUT_OF_METADATA_SPACE) {
3768 DMERR("%s: unable to service pool target messages in READ_ONLY or FAIL mode",
3769 dm_device_name(pool->pool_md));
3770 return -EOPNOTSUPP;
3771 }
3772
3773 if (!strcasecmp(argv[0], "create_thin"))
3774 r = process_create_thin_mesg(argc, argv, pool);
3775
3776 else if (!strcasecmp(argv[0], "create_snap"))
3777 r = process_create_snap_mesg(argc, argv, pool);
3778
3779 else if (!strcasecmp(argv[0], "delete"))
3780 r = process_delete_mesg(argc, argv, pool);
3781
3782 else if (!strcasecmp(argv[0], "set_transaction_id"))
3783 r = process_set_transaction_id_mesg(argc, argv, pool);
3784
3785 else if (!strcasecmp(argv[0], "reserve_metadata_snap"))
3786 r = process_reserve_metadata_snap_mesg(argc, argv, pool);
3787
3788 else if (!strcasecmp(argv[0], "release_metadata_snap"))
3789 r = process_release_metadata_snap_mesg(argc, argv, pool);
3790
3791 else
3792 DMWARN("Unrecognised thin pool target message received: %s", argv[0]);
3793
3794 if (!r)
3795 (void) commit(pool);
3796
3797 return r;
3798}
3799
3800static void emit_flags(struct pool_features *pf, char *result,
3801 unsigned sz, unsigned maxlen)
3802{
3803 unsigned count = !pf->zero_new_blocks + !pf->discard_enabled +
3804 !pf->discard_passdown + (pf->mode == PM_READ_ONLY) +
3805 pf->error_if_no_space;
3806 DMEMIT("%u ", count);
3807
3808 if (!pf->zero_new_blocks)
3809 DMEMIT("skip_block_zeroing ");
3810
3811 if (!pf->discard_enabled)
3812 DMEMIT("ignore_discard ");
3813
3814 if (!pf->discard_passdown)
3815 DMEMIT("no_discard_passdown ");
3816
3817 if (pf->mode == PM_READ_ONLY)
3818 DMEMIT("read_only ");
3819
3820 if (pf->error_if_no_space)
3821 DMEMIT("error_if_no_space ");
3822}
3823
3824
3825
3826
3827
3828
3829
3830static void pool_status(struct dm_target *ti, status_type_t type,
3831 unsigned status_flags, char *result, unsigned maxlen)
3832{
3833 int r;
3834 unsigned sz = 0;
3835 uint64_t transaction_id;
3836 dm_block_t nr_free_blocks_data;
3837 dm_block_t nr_free_blocks_metadata;
3838 dm_block_t nr_blocks_data;
3839 dm_block_t nr_blocks_metadata;
3840 dm_block_t held_root;
3841 enum pool_mode mode;
3842 char buf[BDEVNAME_SIZE];
3843 char buf2[BDEVNAME_SIZE];
3844 struct pool_c *pt = ti->private;
3845 struct pool *pool = pt->pool;
3846
3847 switch (type) {
3848 case STATUSTYPE_INFO:
3849 if (get_pool_mode(pool) == PM_FAIL) {
3850 DMEMIT("Fail");
3851 break;
3852 }
3853
3854
3855 if (!(status_flags & DM_STATUS_NOFLUSH_FLAG) && !dm_suspended(ti))
3856 (void) commit(pool);
3857
3858 r = dm_pool_get_metadata_transaction_id(pool->pmd, &transaction_id);
3859 if (r) {
3860 DMERR("%s: dm_pool_get_metadata_transaction_id returned %d",
3861 dm_device_name(pool->pool_md), r);
3862 goto err;
3863 }
3864
3865 r = dm_pool_get_free_metadata_block_count(pool->pmd, &nr_free_blocks_metadata);
3866 if (r) {
3867 DMERR("%s: dm_pool_get_free_metadata_block_count returned %d",
3868 dm_device_name(pool->pool_md), r);
3869 goto err;
3870 }
3871
3872 r = dm_pool_get_metadata_dev_size(pool->pmd, &nr_blocks_metadata);
3873 if (r) {
3874 DMERR("%s: dm_pool_get_metadata_dev_size returned %d",
3875 dm_device_name(pool->pool_md), r);
3876 goto err;
3877 }
3878
3879 r = dm_pool_get_free_block_count(pool->pmd, &nr_free_blocks_data);
3880 if (r) {
3881 DMERR("%s: dm_pool_get_free_block_count returned %d",
3882 dm_device_name(pool->pool_md), r);
3883 goto err;
3884 }
3885
3886 r = dm_pool_get_data_dev_size(pool->pmd, &nr_blocks_data);
3887 if (r) {
3888 DMERR("%s: dm_pool_get_data_dev_size returned %d",
3889 dm_device_name(pool->pool_md), r);
3890 goto err;
3891 }
3892
3893 r = dm_pool_get_metadata_snap(pool->pmd, &held_root);
3894 if (r) {
3895 DMERR("%s: dm_pool_get_metadata_snap returned %d",
3896 dm_device_name(pool->pool_md), r);
3897 goto err;
3898 }
3899
3900 DMEMIT("%llu %llu/%llu %llu/%llu ",
3901 (unsigned long long)transaction_id,
3902 (unsigned long long)(nr_blocks_metadata - nr_free_blocks_metadata),
3903 (unsigned long long)nr_blocks_metadata,
3904 (unsigned long long)(nr_blocks_data - nr_free_blocks_data),
3905 (unsigned long long)nr_blocks_data);
3906
3907 if (held_root)
3908 DMEMIT("%llu ", held_root);
3909 else
3910 DMEMIT("- ");
3911
3912 mode = get_pool_mode(pool);
3913 if (mode == PM_OUT_OF_DATA_SPACE)
3914 DMEMIT("out_of_data_space ");
3915 else if (is_read_only_pool_mode(mode))
3916 DMEMIT("ro ");
3917 else
3918 DMEMIT("rw ");
3919
3920 if (!pool->pf.discard_enabled)
3921 DMEMIT("ignore_discard ");
3922 else if (pool->pf.discard_passdown)
3923 DMEMIT("discard_passdown ");
3924 else
3925 DMEMIT("no_discard_passdown ");
3926
3927 if (pool->pf.error_if_no_space)
3928 DMEMIT("error_if_no_space ");
3929 else
3930 DMEMIT("queue_if_no_space ");
3931
3932 if (dm_pool_metadata_needs_check(pool->pmd))
3933 DMEMIT("needs_check ");
3934 else
3935 DMEMIT("- ");
3936
3937 DMEMIT("%llu ", (unsigned long long)calc_metadata_threshold(pt));
3938
3939 break;
3940
3941 case STATUSTYPE_TABLE:
3942 DMEMIT("%s %s %lu %llu ",
3943 format_dev_t(buf, pt->metadata_dev->bdev->bd_dev),
3944 format_dev_t(buf2, pt->data_dev->bdev->bd_dev),
3945 (unsigned long)pool->sectors_per_block,
3946 (unsigned long long)pt->low_water_blocks);
3947 emit_flags(&pt->requested_pf, result, sz, maxlen);
3948 break;
3949 }
3950 return;
3951
3952err:
3953 DMEMIT("Error");
3954}
3955
3956static int pool_iterate_devices(struct dm_target *ti,
3957 iterate_devices_callout_fn fn, void *data)
3958{
3959 struct pool_c *pt = ti->private;
3960
3961 return fn(ti, pt->data_dev, 0, ti->len, data);
3962}
3963
3964static void pool_io_hints(struct dm_target *ti, struct queue_limits *limits)
3965{
3966 struct pool_c *pt = ti->private;
3967 struct pool *pool = pt->pool;
3968 sector_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979 if (limits->max_sectors < pool->sectors_per_block) {
3980 while (!is_factor(pool->sectors_per_block, limits->max_sectors)) {
3981 if ((limits->max_sectors & (limits->max_sectors - 1)) == 0)
3982 limits->max_sectors--;
3983 limits->max_sectors = rounddown_pow_of_two(limits->max_sectors);
3984 }
3985 }
3986
3987
3988
3989
3990
3991 if (io_opt_sectors < pool->sectors_per_block ||
3992 !is_factor(io_opt_sectors, pool->sectors_per_block)) {
3993 if (is_factor(pool->sectors_per_block, limits->max_sectors))
3994 blk_limits_io_min(limits, limits->max_sectors << SECTOR_SHIFT);
3995 else
3996 blk_limits_io_min(limits, pool->sectors_per_block << SECTOR_SHIFT);
3997 blk_limits_io_opt(limits, pool->sectors_per_block << SECTOR_SHIFT);
3998 }
3999
4000
4001
4002
4003
4004
4005 if (!pt->adjusted_pf.discard_enabled) {
4006
4007
4008
4009
4010
4011
4012 limits->discard_granularity = 0;
4013 return;
4014 }
4015
4016 disable_passdown_if_not_supported(pt);
4017
4018
4019
4020
4021
4022}
4023
4024static struct target_type pool_target = {
4025 .name = "thin-pool",
4026 .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE |
4027 DM_TARGET_IMMUTABLE,
4028 .version = {1, 21, 0},
4029 .module = THIS_MODULE,
4030 .ctr = pool_ctr,
4031 .dtr = pool_dtr,
4032 .map = pool_map,
4033 .presuspend = pool_presuspend,
4034 .presuspend_undo = pool_presuspend_undo,
4035 .postsuspend = pool_postsuspend,
4036 .preresume = pool_preresume,
4037 .resume = pool_resume,
4038 .message = pool_message,
4039 .status = pool_status,
4040 .iterate_devices = pool_iterate_devices,
4041 .io_hints = pool_io_hints,
4042};
4043
4044
4045
4046
4047static void thin_get(struct thin_c *tc)
4048{
4049 refcount_inc(&tc->refcount);
4050}
4051
4052static void thin_put(struct thin_c *tc)
4053{
4054 if (refcount_dec_and_test(&tc->refcount))
4055 complete(&tc->can_destroy);
4056}
4057
4058static void thin_dtr(struct dm_target *ti)
4059{
4060 struct thin_c *tc = ti->private;
4061 unsigned long flags;
4062
4063 spin_lock_irqsave(&tc->pool->lock, flags);
4064 list_del_rcu(&tc->list);
4065 spin_unlock_irqrestore(&tc->pool->lock, flags);
4066 synchronize_rcu();
4067
4068 thin_put(tc);
4069 wait_for_completion(&tc->can_destroy);
4070
4071 mutex_lock(&dm_thin_pool_table.mutex);
4072
4073 __pool_dec(tc->pool);
4074 dm_pool_close_thin_device(tc->td);
4075 dm_put_device(ti, tc->pool_dev);
4076 if (tc->origin_dev)
4077 dm_put_device(ti, tc->origin_dev);
4078 kfree(tc);
4079
4080 mutex_unlock(&dm_thin_pool_table.mutex);
4081}
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095static int thin_ctr(struct dm_target *ti, unsigned argc, char **argv)
4096{
4097 int r;
4098 struct thin_c *tc;
4099 struct dm_dev *pool_dev, *origin_dev;
4100 struct mapped_device *pool_md;
4101 unsigned long flags;
4102
4103 mutex_lock(&dm_thin_pool_table.mutex);
4104
4105 if (argc != 2 && argc != 3) {
4106 ti->error = "Invalid argument count";
4107 r = -EINVAL;
4108 goto out_unlock;
4109 }
4110
4111 tc = ti->private = kzalloc(sizeof(*tc), GFP_KERNEL);
4112 if (!tc) {
4113 ti->error = "Out of memory";
4114 r = -ENOMEM;
4115 goto out_unlock;
4116 }
4117 tc->thin_md = dm_table_get_md(ti->table);
4118 spin_lock_init(&tc->lock);
4119 INIT_LIST_HEAD(&tc->deferred_cells);
4120 bio_list_init(&tc->deferred_bio_list);
4121 bio_list_init(&tc->retry_on_resume_list);
4122 tc->sort_bio_list = RB_ROOT;
4123
4124 if (argc == 3) {
4125 r = dm_get_device(ti, argv[2], FMODE_READ, &origin_dev);
4126 if (r) {
4127 ti->error = "Error opening origin device";
4128 goto bad_origin_dev;
4129 }
4130 tc->origin_dev = origin_dev;
4131 }
4132
4133 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &pool_dev);
4134 if (r) {
4135 ti->error = "Error opening pool device";
4136 goto bad_pool_dev;
4137 }
4138 tc->pool_dev = pool_dev;
4139
4140 if (read_dev_id(argv[1], (unsigned long long *)&tc->dev_id, 0)) {
4141 ti->error = "Invalid device id";
4142 r = -EINVAL;
4143 goto bad_common;
4144 }
4145
4146 pool_md = dm_get_md(tc->pool_dev->bdev->bd_dev);
4147 if (!pool_md) {
4148 ti->error = "Couldn't get pool mapped device";
4149 r = -EINVAL;
4150 goto bad_common;
4151 }
4152
4153 tc->pool = __pool_table_lookup(pool_md);
4154 if (!tc->pool) {
4155 ti->error = "Couldn't find pool object";
4156 r = -EINVAL;
4157 goto bad_pool_lookup;
4158 }
4159 __pool_inc(tc->pool);
4160
4161 if (get_pool_mode(tc->pool) == PM_FAIL) {
4162 ti->error = "Couldn't open thin device, Pool is in fail mode";
4163 r = -EINVAL;
4164 goto bad_pool;
4165 }
4166
4167 r = dm_pool_open_thin_device(tc->pool->pmd, tc->dev_id, &tc->td);
4168 if (r) {
4169 ti->error = "Couldn't open thin internal device";
4170 goto bad_pool;
4171 }
4172
4173 r = dm_set_target_max_io_len(ti, tc->pool->sectors_per_block);
4174 if (r)
4175 goto bad;
4176
4177 ti->num_flush_bios = 1;
4178 ti->flush_supported = true;
4179 ti->per_io_data_size = sizeof(struct dm_thin_endio_hook);
4180
4181
4182 if (tc->pool->pf.discard_enabled) {
4183 ti->discards_supported = true;
4184 ti->num_discard_bios = 1;
4185 ti->split_discard_bios = false;
4186 }
4187
4188 mutex_unlock(&dm_thin_pool_table.mutex);
4189
4190 spin_lock_irqsave(&tc->pool->lock, flags);
4191 if (tc->pool->suspended) {
4192 spin_unlock_irqrestore(&tc->pool->lock, flags);
4193 mutex_lock(&dm_thin_pool_table.mutex);
4194 ti->error = "Unable to activate thin device while pool is suspended";
4195 r = -EINVAL;
4196 goto bad;
4197 }
4198 refcount_set(&tc->refcount, 1);
4199 init_completion(&tc->can_destroy);
4200 list_add_tail_rcu(&tc->list, &tc->pool->active_thins);
4201 spin_unlock_irqrestore(&tc->pool->lock, flags);
4202
4203
4204
4205
4206
4207
4208 synchronize_rcu();
4209
4210 dm_put(pool_md);
4211
4212 return 0;
4213
4214bad:
4215 dm_pool_close_thin_device(tc->td);
4216bad_pool:
4217 __pool_dec(tc->pool);
4218bad_pool_lookup:
4219 dm_put(pool_md);
4220bad_common:
4221 dm_put_device(ti, tc->pool_dev);
4222bad_pool_dev:
4223 if (tc->origin_dev)
4224 dm_put_device(ti, tc->origin_dev);
4225bad_origin_dev:
4226 kfree(tc);
4227out_unlock:
4228 mutex_unlock(&dm_thin_pool_table.mutex);
4229
4230 return r;
4231}
4232
4233static int thin_map(struct dm_target *ti, struct bio *bio)
4234{
4235 bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
4236
4237 return thin_bio_map(ti, bio);
4238}
4239
4240static int thin_endio(struct dm_target *ti, struct bio *bio,
4241 blk_status_t *err)
4242{
4243 unsigned long flags;
4244 struct dm_thin_endio_hook *h = dm_per_bio_data(bio, sizeof(struct dm_thin_endio_hook));
4245 struct list_head work;
4246 struct dm_thin_new_mapping *m, *tmp;
4247 struct pool *pool = h->tc->pool;
4248
4249 if (h->shared_read_entry) {
4250 INIT_LIST_HEAD(&work);
4251 dm_deferred_entry_dec(h->shared_read_entry, &work);
4252
4253 spin_lock_irqsave(&pool->lock, flags);
4254 list_for_each_entry_safe(m, tmp, &work, list) {
4255 list_del(&m->list);
4256 __complete_mapping_preparation(m);
4257 }
4258 spin_unlock_irqrestore(&pool->lock, flags);
4259 }
4260
4261 if (h->all_io_entry) {
4262 INIT_LIST_HEAD(&work);
4263 dm_deferred_entry_dec(h->all_io_entry, &work);
4264 if (!list_empty(&work)) {
4265 spin_lock_irqsave(&pool->lock, flags);
4266 list_for_each_entry_safe(m, tmp, &work, list)
4267 list_add_tail(&m->list, &pool->prepared_discards);
4268 spin_unlock_irqrestore(&pool->lock, flags);
4269 wake_worker(pool);
4270 }
4271 }
4272
4273 if (h->cell)
4274 cell_defer_no_holder(h->tc, h->cell);
4275
4276 return DM_ENDIO_DONE;
4277}
4278
4279static void thin_presuspend(struct dm_target *ti)
4280{
4281 struct thin_c *tc = ti->private;
4282
4283 if (dm_noflush_suspending(ti))
4284 noflush_work(tc, do_noflush_start);
4285}
4286
4287static void thin_postsuspend(struct dm_target *ti)
4288{
4289 struct thin_c *tc = ti->private;
4290
4291
4292
4293
4294
4295 noflush_work(tc, do_noflush_stop);
4296}
4297
4298static int thin_preresume(struct dm_target *ti)
4299{
4300 struct thin_c *tc = ti->private;
4301
4302 if (tc->origin_dev)
4303 tc->origin_size = get_dev_size(tc->origin_dev->bdev);
4304
4305 return 0;
4306}
4307
4308
4309
4310
4311static void thin_status(struct dm_target *ti, status_type_t type,
4312 unsigned status_flags, char *result, unsigned maxlen)
4313{
4314 int r;
4315 ssize_t sz = 0;
4316 dm_block_t mapped, highest;
4317 char buf[BDEVNAME_SIZE];
4318 struct thin_c *tc = ti->private;
4319
4320 if (get_pool_mode(tc->pool) == PM_FAIL) {
4321 DMEMIT("Fail");
4322 return;
4323 }
4324
4325 if (!tc->td)
4326 DMEMIT("-");
4327 else {
4328 switch (type) {
4329 case STATUSTYPE_INFO:
4330 r = dm_thin_get_mapped_count(tc->td, &mapped);
4331 if (r) {
4332 DMERR("dm_thin_get_mapped_count returned %d", r);
4333 goto err;
4334 }
4335
4336 r = dm_thin_get_highest_mapped_block(tc->td, &highest);
4337 if (r < 0) {
4338 DMERR("dm_thin_get_highest_mapped_block returned %d", r);
4339 goto err;
4340 }
4341
4342 DMEMIT("%llu ", mapped * tc->pool->sectors_per_block);
4343 if (r)
4344 DMEMIT("%llu", ((highest + 1) *
4345 tc->pool->sectors_per_block) - 1);
4346 else
4347 DMEMIT("-");
4348 break;
4349
4350 case STATUSTYPE_TABLE:
4351 DMEMIT("%s %lu",
4352 format_dev_t(buf, tc->pool_dev->bdev->bd_dev),
4353 (unsigned long) tc->dev_id);
4354 if (tc->origin_dev)
4355 DMEMIT(" %s", format_dev_t(buf, tc->origin_dev->bdev->bd_dev));
4356 break;
4357 }
4358 }
4359
4360 return;
4361
4362err:
4363 DMEMIT("Error");
4364}
4365
4366static int thin_iterate_devices(struct dm_target *ti,
4367 iterate_devices_callout_fn fn, void *data)
4368{
4369 sector_t blocks;
4370 struct thin_c *tc = ti->private;
4371 struct pool *pool = tc->pool;
4372
4373
4374
4375
4376
4377 if (!pool->ti)
4378 return 0;
4379
4380 blocks = pool->ti->len;
4381 (void) sector_div(blocks, pool->sectors_per_block);
4382 if (blocks)
4383 return fn(ti, tc->pool_dev, 0, pool->sectors_per_block * blocks, data);
4384
4385 return 0;
4386}
4387
4388static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits)
4389{
4390 struct thin_c *tc = ti->private;
4391 struct pool *pool = tc->pool;
4392
4393 if (!pool->pf.discard_enabled)
4394 return;
4395
4396 limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT;
4397 limits->max_discard_sectors = 2048 * 1024 * 16;
4398}
4399
4400static struct target_type thin_target = {
4401 .name = "thin",
4402 .version = {1, 21, 0},
4403 .module = THIS_MODULE,
4404 .ctr = thin_ctr,
4405 .dtr = thin_dtr,
4406 .map = thin_map,
4407 .end_io = thin_endio,
4408 .preresume = thin_preresume,
4409 .presuspend = thin_presuspend,
4410 .postsuspend = thin_postsuspend,
4411 .status = thin_status,
4412 .iterate_devices = thin_iterate_devices,
4413 .io_hints = thin_io_hints,
4414};
4415
4416
4417
4418static int __init dm_thin_init(void)
4419{
4420 int r = -ENOMEM;
4421
4422 pool_table_init();
4423
4424 _new_mapping_cache = KMEM_CACHE(dm_thin_new_mapping, 0);
4425 if (!_new_mapping_cache)
4426 return r;
4427
4428 r = dm_register_target(&thin_target);
4429 if (r)
4430 goto bad_new_mapping_cache;
4431
4432 r = dm_register_target(&pool_target);
4433 if (r)
4434 goto bad_thin_target;
4435
4436 return 0;
4437
4438bad_thin_target:
4439 dm_unregister_target(&thin_target);
4440bad_new_mapping_cache:
4441 kmem_cache_destroy(_new_mapping_cache);
4442
4443 return r;
4444}
4445
4446static void dm_thin_exit(void)
4447{
4448 dm_unregister_target(&thin_target);
4449 dm_unregister_target(&pool_target);
4450
4451 kmem_cache_destroy(_new_mapping_cache);
4452
4453 pool_table_exit();
4454}
4455
4456module_init(dm_thin_init);
4457module_exit(dm_thin_exit);
4458
4459module_param_named(no_space_timeout, no_space_timeout_secs, uint, S_IRUGO | S_IWUSR);
4460MODULE_PARM_DESC(no_space_timeout, "Out of data space queue IO timeout in seconds");
4461
4462MODULE_DESCRIPTION(DM_NAME " thin provisioning target");
4463MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
4464MODULE_LICENSE("GPL");
4465