Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2#include "dm.h"
3#include "persistent-data/dm-transaction-manager.h"
4#include "persistent-data/dm-bitset.h"
5#include "persistent-data/dm-space-map.h"
6
7#include <linux/dm-io.h>
8#include <linux/dm-kcopyd.h>
9#include <linux/init.h>
10#include <linux/mempool.h>
11#include <linux/module.h>
12#include <linux/slab.h>
13#include <linux/vmalloc.h>
14
15#define DM_MSG_PREFIX "era"
16
17#define SUPERBLOCK_LOCATION 0
18#define SUPERBLOCK_MAGIC 2126579579
19#define SUPERBLOCK_CSUM_XOR 146538381
20#define MIN_ERA_VERSION 1
21#define MAX_ERA_VERSION 1
22#define INVALID_WRITESET_ROOT SUPERBLOCK_LOCATION
23#define MIN_BLOCK_SIZE 8
24
25/*
26 *--------------------------------------------------------------
27 * Writeset
28 *--------------------------------------------------------------
29 */
30struct writeset_metadata {
31 uint32_t nr_bits;
32 dm_block_t root;
33};
34
35struct writeset {
36 struct writeset_metadata md;
37
38 /*
39 * An in core copy of the bits to save constantly doing look ups on
40 * disk.
41 */
42 unsigned long *bits;
43};
44
45/*
46 * This does not free off the on disk bitset as this will normally be done
47 * after digesting into the era array.
48 */
49static void writeset_free(struct writeset *ws)
50{
51 vfree(ws->bits);
52 ws->bits = NULL;
53}
54
55static int setup_on_disk_bitset(struct dm_disk_bitset *info,
56 unsigned int nr_bits, dm_block_t *root)
57{
58 int r;
59
60 r = dm_bitset_empty(info, root);
61 if (r)
62 return r;
63
64 return dm_bitset_resize(info, *root, 0, nr_bits, false, root);
65}
66
67static size_t bitset_size(unsigned int nr_bits)
68{
69 return sizeof(unsigned long) * dm_div_up(nr_bits, BITS_PER_LONG);
70}
71
72/*
73 * Allocates memory for the in core bitset.
74 */
75static int writeset_alloc(struct writeset *ws, dm_block_t nr_blocks)
76{
77 ws->bits = vzalloc(bitset_size(nr_blocks));
78 if (!ws->bits) {
79 DMERR("%s: couldn't allocate in memory bitset", __func__);
80 return -ENOMEM;
81 }
82
83 return 0;
84}
85
86/*
87 * Wipes the in-core bitset, and creates a new on disk bitset.
88 */
89static int writeset_init(struct dm_disk_bitset *info, struct writeset *ws,
90 dm_block_t nr_blocks)
91{
92 int r;
93
94 memset(ws->bits, 0, bitset_size(nr_blocks));
95
96 ws->md.nr_bits = nr_blocks;
97 r = setup_on_disk_bitset(info, ws->md.nr_bits, &ws->md.root);
98 if (r) {
99 DMERR("%s: setup_on_disk_bitset failed", __func__);
100 return r;
101 }
102
103 return 0;
104}
105
106static bool writeset_marked(struct writeset *ws, dm_block_t block)
107{
108 return test_bit(block, ws->bits);
109}
110
111static int writeset_marked_on_disk(struct dm_disk_bitset *info,
112 struct writeset_metadata *m, dm_block_t block,
113 bool *result)
114{
115 int r;
116 dm_block_t old = m->root;
117
118 /*
119 * The bitset was flushed when it was archived, so we know there'll
120 * be no change to the root.
121 */
122 r = dm_bitset_test_bit(info, m->root, block, &m->root, result);
123 if (r) {
124 DMERR("%s: dm_bitset_test_bit failed", __func__);
125 return r;
126 }
127
128 BUG_ON(m->root != old);
129
130 return r;
131}
132
133/*
134 * Returns < 0 on error, 0 if the bit wasn't previously set, 1 if it was.
135 */
136static int writeset_test_and_set(struct dm_disk_bitset *info,
137 struct writeset *ws, uint32_t block)
138{
139 int r;
140
141 if (!test_bit(block, ws->bits)) {
142 r = dm_bitset_set_bit(info, ws->md.root, block, &ws->md.root);
143 if (r) {
144 /* FIXME: fail mode */
145 return r;
146 }
147
148 return 0;
149 }
150
151 return 1;
152}
153
154/*
155 *--------------------------------------------------------------
156 * On disk metadata layout
157 *--------------------------------------------------------------
158 */
159#define SPACE_MAP_ROOT_SIZE 128
160#define UUID_LEN 16
161
162struct writeset_disk {
163 __le32 nr_bits;
164 __le64 root;
165} __packed;
166
167struct superblock_disk {
168 __le32 csum;
169 __le32 flags;
170 __le64 blocknr;
171
172 __u8 uuid[UUID_LEN];
173 __le64 magic;
174 __le32 version;
175
176 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
177
178 __le32 data_block_size;
179 __le32 metadata_block_size;
180 __le32 nr_blocks;
181
182 __le32 current_era;
183 struct writeset_disk current_writeset;
184
185 /*
186 * Only these two fields are valid within the metadata snapshot.
187 */
188 __le64 writeset_tree_root;
189 __le64 era_array_root;
190
191 __le64 metadata_snap;
192} __packed;
193
194/*
195 *--------------------------------------------------------------
196 * Superblock validation
197 *--------------------------------------------------------------
198 */
199static void sb_prepare_for_write(struct dm_block_validator *v,
200 struct dm_block *b,
201 size_t sb_block_size)
202{
203 struct superblock_disk *disk = dm_block_data(b);
204
205 disk->blocknr = cpu_to_le64(dm_block_location(b));
206 disk->csum = cpu_to_le32(dm_bm_checksum(&disk->flags,
207 sb_block_size - sizeof(__le32),
208 SUPERBLOCK_CSUM_XOR));
209}
210
211static int check_metadata_version(struct superblock_disk *disk)
212{
213 uint32_t metadata_version = le32_to_cpu(disk->version);
214
215 if (metadata_version < MIN_ERA_VERSION || metadata_version > MAX_ERA_VERSION) {
216 DMERR("Era metadata version %u found, but only versions between %u and %u supported.",
217 metadata_version, MIN_ERA_VERSION, MAX_ERA_VERSION);
218 return -EINVAL;
219 }
220
221 return 0;
222}
223
224static int sb_check(struct dm_block_validator *v,
225 struct dm_block *b,
226 size_t sb_block_size)
227{
228 struct superblock_disk *disk = dm_block_data(b);
229 __le32 csum_le;
230
231 if (dm_block_location(b) != le64_to_cpu(disk->blocknr)) {
232 DMERR("%s failed: blocknr %llu: wanted %llu",
233 __func__, le64_to_cpu(disk->blocknr),
234 (unsigned long long)dm_block_location(b));
235 return -ENOTBLK;
236 }
237
238 if (le64_to_cpu(disk->magic) != SUPERBLOCK_MAGIC) {
239 DMERR("%s failed: magic %llu: wanted %llu",
240 __func__, le64_to_cpu(disk->magic),
241 (unsigned long long) SUPERBLOCK_MAGIC);
242 return -EILSEQ;
243 }
244
245 csum_le = cpu_to_le32(dm_bm_checksum(&disk->flags,
246 sb_block_size - sizeof(__le32),
247 SUPERBLOCK_CSUM_XOR));
248 if (csum_le != disk->csum) {
249 DMERR("%s failed: csum %u: wanted %u",
250 __func__, le32_to_cpu(csum_le), le32_to_cpu(disk->csum));
251 return -EILSEQ;
252 }
253
254 return check_metadata_version(disk);
255}
256
257static struct dm_block_validator sb_validator = {
258 .name = "superblock",
259 .prepare_for_write = sb_prepare_for_write,
260 .check = sb_check
261};
262
263/*
264 *--------------------------------------------------------------
265 * Low level metadata handling
266 *--------------------------------------------------------------
267 */
268#define DM_ERA_METADATA_BLOCK_SIZE 4096
269#define ERA_MAX_CONCURRENT_LOCKS 5
270
271struct era_metadata {
272 struct block_device *bdev;
273 struct dm_block_manager *bm;
274 struct dm_space_map *sm;
275 struct dm_transaction_manager *tm;
276
277 dm_block_t block_size;
278 uint32_t nr_blocks;
279
280 uint32_t current_era;
281
282 /*
283 * We preallocate 2 writesets. When an era rolls over we
284 * switch between them. This means the allocation is done at
285 * preresume time, rather than on the io path.
286 */
287 struct writeset writesets[2];
288 struct writeset *current_writeset;
289
290 dm_block_t writeset_tree_root;
291 dm_block_t era_array_root;
292
293 struct dm_disk_bitset bitset_info;
294 struct dm_btree_info writeset_tree_info;
295 struct dm_array_info era_array_info;
296
297 dm_block_t metadata_snap;
298
299 /*
300 * A flag that is set whenever a writeset has been archived.
301 */
302 bool archived_writesets;
303
304 /*
305 * Reading the space map root can fail, so we read it into this
306 * buffer before the superblock is locked and updated.
307 */
308 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
309};
310
311static int superblock_read_lock(struct era_metadata *md,
312 struct dm_block **sblock)
313{
314 return dm_bm_read_lock(md->bm, SUPERBLOCK_LOCATION,
315 &sb_validator, sblock);
316}
317
318static int superblock_lock_zero(struct era_metadata *md,
319 struct dm_block **sblock)
320{
321 return dm_bm_write_lock_zero(md->bm, SUPERBLOCK_LOCATION,
322 &sb_validator, sblock);
323}
324
325static int superblock_lock(struct era_metadata *md,
326 struct dm_block **sblock)
327{
328 return dm_bm_write_lock(md->bm, SUPERBLOCK_LOCATION,
329 &sb_validator, sblock);
330}
331
332/* FIXME: duplication with cache and thin */
333static int superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
334{
335 int r;
336 unsigned int i;
337 struct dm_block *b;
338 __le64 *data_le, zero = cpu_to_le64(0);
339 unsigned int sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
340
341 /*
342 * We can't use a validator here - it may be all zeroes.
343 */
344 r = dm_bm_read_lock(bm, SUPERBLOCK_LOCATION, NULL, &b);
345 if (r)
346 return r;
347
348 data_le = dm_block_data(b);
349 *result = true;
350 for (i = 0; i < sb_block_size; i++) {
351 if (data_le[i] != zero) {
352 *result = false;
353 break;
354 }
355 }
356
357 dm_bm_unlock(b);
358
359 return 0;
360}
361
362/*----------------------------------------------------------------*/
363
364static void ws_pack(const struct writeset_metadata *core, struct writeset_disk *disk)
365{
366 disk->nr_bits = cpu_to_le32(core->nr_bits);
367 disk->root = cpu_to_le64(core->root);
368}
369
370static void ws_unpack(const struct writeset_disk *disk, struct writeset_metadata *core)
371{
372 core->nr_bits = le32_to_cpu(disk->nr_bits);
373 core->root = le64_to_cpu(disk->root);
374}
375
376static void ws_inc(void *context, const void *value, unsigned int count)
377{
378 struct era_metadata *md = context;
379 struct writeset_disk ws_d;
380 dm_block_t b;
381 unsigned int i;
382
383 for (i = 0; i < count; i++) {
384 memcpy(&ws_d, value + (i * sizeof(ws_d)), sizeof(ws_d));
385 b = le64_to_cpu(ws_d.root);
386 dm_tm_inc(md->tm, b);
387 }
388}
389
390static void ws_dec(void *context, const void *value, unsigned int count)
391{
392 struct era_metadata *md = context;
393 struct writeset_disk ws_d;
394 dm_block_t b;
395 unsigned int i;
396
397 for (i = 0; i < count; i++) {
398 memcpy(&ws_d, value + (i * sizeof(ws_d)), sizeof(ws_d));
399 b = le64_to_cpu(ws_d.root);
400 dm_bitset_del(&md->bitset_info, b);
401 }
402}
403
404static int ws_eq(void *context, const void *value1, const void *value2)
405{
406 return !memcmp(value1, value2, sizeof(struct writeset_disk));
407}
408
409/*----------------------------------------------------------------*/
410
411static void setup_writeset_tree_info(struct era_metadata *md)
412{
413 struct dm_btree_value_type *vt = &md->writeset_tree_info.value_type;
414
415 md->writeset_tree_info.tm = md->tm;
416 md->writeset_tree_info.levels = 1;
417 vt->context = md;
418 vt->size = sizeof(struct writeset_disk);
419 vt->inc = ws_inc;
420 vt->dec = ws_dec;
421 vt->equal = ws_eq;
422}
423
424static void setup_era_array_info(struct era_metadata *md)
425{
426 struct dm_btree_value_type vt;
427
428 vt.context = NULL;
429 vt.size = sizeof(__le32);
430 vt.inc = NULL;
431 vt.dec = NULL;
432 vt.equal = NULL;
433
434 dm_array_info_init(&md->era_array_info, md->tm, &vt);
435}
436
437static void setup_infos(struct era_metadata *md)
438{
439 dm_disk_bitset_init(md->tm, &md->bitset_info);
440 setup_writeset_tree_info(md);
441 setup_era_array_info(md);
442}
443
444/*----------------------------------------------------------------*/
445
446static int create_fresh_metadata(struct era_metadata *md)
447{
448 int r;
449
450 r = dm_tm_create_with_sm(md->bm, SUPERBLOCK_LOCATION,
451 &md->tm, &md->sm);
452 if (r < 0) {
453 DMERR("dm_tm_create_with_sm failed");
454 return r;
455 }
456
457 setup_infos(md);
458
459 r = dm_btree_empty(&md->writeset_tree_info, &md->writeset_tree_root);
460 if (r) {
461 DMERR("couldn't create new writeset tree");
462 goto bad;
463 }
464
465 r = dm_array_empty(&md->era_array_info, &md->era_array_root);
466 if (r) {
467 DMERR("couldn't create era array");
468 goto bad;
469 }
470
471 return 0;
472
473bad:
474 dm_sm_destroy(md->sm);
475 dm_tm_destroy(md->tm);
476
477 return r;
478}
479
480static int save_sm_root(struct era_metadata *md)
481{
482 int r;
483 size_t metadata_len;
484
485 r = dm_sm_root_size(md->sm, &metadata_len);
486 if (r < 0)
487 return r;
488
489 return dm_sm_copy_root(md->sm, &md->metadata_space_map_root,
490 metadata_len);
491}
492
493static void copy_sm_root(struct era_metadata *md, struct superblock_disk *disk)
494{
495 memcpy(&disk->metadata_space_map_root,
496 &md->metadata_space_map_root,
497 sizeof(md->metadata_space_map_root));
498}
499
500/*
501 * Writes a superblock, including the static fields that don't get updated
502 * with every commit (possible optimisation here). 'md' should be fully
503 * constructed when this is called.
504 */
505static void prepare_superblock(struct era_metadata *md, struct superblock_disk *disk)
506{
507 disk->magic = cpu_to_le64(SUPERBLOCK_MAGIC);
508 disk->flags = cpu_to_le32(0ul);
509
510 /* FIXME: can't keep blanking the uuid (uuid is currently unused though) */
511 memset(disk->uuid, 0, sizeof(disk->uuid));
512 disk->version = cpu_to_le32(MAX_ERA_VERSION);
513
514 copy_sm_root(md, disk);
515
516 disk->data_block_size = cpu_to_le32(md->block_size);
517 disk->metadata_block_size = cpu_to_le32(DM_ERA_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
518 disk->nr_blocks = cpu_to_le32(md->nr_blocks);
519 disk->current_era = cpu_to_le32(md->current_era);
520
521 ws_pack(&md->current_writeset->md, &disk->current_writeset);
522 disk->writeset_tree_root = cpu_to_le64(md->writeset_tree_root);
523 disk->era_array_root = cpu_to_le64(md->era_array_root);
524 disk->metadata_snap = cpu_to_le64(md->metadata_snap);
525}
526
527static int write_superblock(struct era_metadata *md)
528{
529 int r;
530 struct dm_block *sblock;
531 struct superblock_disk *disk;
532
533 r = save_sm_root(md);
534 if (r) {
535 DMERR("%s: save_sm_root failed", __func__);
536 return r;
537 }
538
539 r = superblock_lock_zero(md, &sblock);
540 if (r)
541 return r;
542
543 disk = dm_block_data(sblock);
544 prepare_superblock(md, disk);
545
546 return dm_tm_commit(md->tm, sblock);
547}
548
549/*
550 * Assumes block_size and the infos are set.
551 */
552static int format_metadata(struct era_metadata *md)
553{
554 int r;
555
556 r = create_fresh_metadata(md);
557 if (r)
558 return r;
559
560 r = write_superblock(md);
561 if (r) {
562 dm_sm_destroy(md->sm);
563 dm_tm_destroy(md->tm);
564 return r;
565 }
566
567 return 0;
568}
569
570static int open_metadata(struct era_metadata *md)
571{
572 int r;
573 struct dm_block *sblock;
574 struct superblock_disk *disk;
575
576 r = superblock_read_lock(md, &sblock);
577 if (r) {
578 DMERR("couldn't read_lock superblock");
579 return r;
580 }
581
582 disk = dm_block_data(sblock);
583
584 /* Verify the data block size hasn't changed */
585 if (le32_to_cpu(disk->data_block_size) != md->block_size) {
586 DMERR("changing the data block size (from %u to %llu) is not supported",
587 le32_to_cpu(disk->data_block_size), md->block_size);
588 r = -EINVAL;
589 goto bad;
590 }
591
592 r = dm_tm_open_with_sm(md->bm, SUPERBLOCK_LOCATION,
593 disk->metadata_space_map_root,
594 sizeof(disk->metadata_space_map_root),
595 &md->tm, &md->sm);
596 if (r) {
597 DMERR("dm_tm_open_with_sm failed");
598 goto bad;
599 }
600
601 setup_infos(md);
602
603 md->nr_blocks = le32_to_cpu(disk->nr_blocks);
604 md->current_era = le32_to_cpu(disk->current_era);
605
606 ws_unpack(&disk->current_writeset, &md->current_writeset->md);
607 md->writeset_tree_root = le64_to_cpu(disk->writeset_tree_root);
608 md->era_array_root = le64_to_cpu(disk->era_array_root);
609 md->metadata_snap = le64_to_cpu(disk->metadata_snap);
610 md->archived_writesets = true;
611
612 dm_bm_unlock(sblock);
613
614 return 0;
615
616bad:
617 dm_bm_unlock(sblock);
618 return r;
619}
620
621static int open_or_format_metadata(struct era_metadata *md,
622 bool may_format)
623{
624 int r;
625 bool unformatted = false;
626
627 r = superblock_all_zeroes(md->bm, &unformatted);
628 if (r)
629 return r;
630
631 if (unformatted)
632 return may_format ? format_metadata(md) : -EPERM;
633
634 return open_metadata(md);
635}
636
637static int create_persistent_data_objects(struct era_metadata *md,
638 bool may_format)
639{
640 int r;
641
642 md->bm = dm_block_manager_create(md->bdev, DM_ERA_METADATA_BLOCK_SIZE,
643 ERA_MAX_CONCURRENT_LOCKS);
644 if (IS_ERR(md->bm)) {
645 DMERR("could not create block manager");
646 return PTR_ERR(md->bm);
647 }
648
649 r = open_or_format_metadata(md, may_format);
650 if (r)
651 dm_block_manager_destroy(md->bm);
652
653 return r;
654}
655
656static void destroy_persistent_data_objects(struct era_metadata *md)
657{
658 dm_sm_destroy(md->sm);
659 dm_tm_destroy(md->tm);
660 dm_block_manager_destroy(md->bm);
661}
662
663/*
664 * This waits until all era_map threads have picked up the new filter.
665 */
666static void swap_writeset(struct era_metadata *md, struct writeset *new_writeset)
667{
668 rcu_assign_pointer(md->current_writeset, new_writeset);
669 synchronize_rcu();
670}
671
672/*
673 *------------------------------------------------------------------------
674 * Writesets get 'digested' into the main era array.
675 *
676 * We're using a coroutine here so the worker thread can do the digestion,
677 * thus avoiding synchronisation of the metadata. Digesting a whole
678 * writeset in one go would cause too much latency.
679 *------------------------------------------------------------------------
680 */
681struct digest {
682 uint32_t era;
683 unsigned int nr_bits, current_bit;
684 struct writeset_metadata writeset;
685 __le32 value;
686 struct dm_disk_bitset info;
687
688 int (*step)(struct era_metadata *md, struct digest *d);
689};
690
691static int metadata_digest_lookup_writeset(struct era_metadata *md,
692 struct digest *d);
693
694static int metadata_digest_remove_writeset(struct era_metadata *md,
695 struct digest *d)
696{
697 int r;
698 uint64_t key = d->era;
699
700 r = dm_btree_remove(&md->writeset_tree_info, md->writeset_tree_root,
701 &key, &md->writeset_tree_root);
702 if (r) {
703 DMERR("%s: dm_btree_remove failed", __func__);
704 return r;
705 }
706
707 d->step = metadata_digest_lookup_writeset;
708 return 0;
709}
710
711#define INSERTS_PER_STEP 100
712
713static int metadata_digest_transcribe_writeset(struct era_metadata *md,
714 struct digest *d)
715{
716 int r;
717 bool marked;
718 unsigned int b, e = min(d->current_bit + INSERTS_PER_STEP, d->nr_bits);
719
720 for (b = d->current_bit; b < e; b++) {
721 r = writeset_marked_on_disk(&d->info, &d->writeset, b, &marked);
722 if (r) {
723 DMERR("%s: writeset_marked_on_disk failed", __func__);
724 return r;
725 }
726
727 if (!marked)
728 continue;
729
730 __dm_bless_for_disk(&d->value);
731 r = dm_array_set_value(&md->era_array_info, md->era_array_root,
732 b, &d->value, &md->era_array_root);
733 if (r) {
734 DMERR("%s: dm_array_set_value failed", __func__);
735 return r;
736 }
737 }
738
739 if (b == d->nr_bits)
740 d->step = metadata_digest_remove_writeset;
741 else
742 d->current_bit = b;
743
744 return 0;
745}
746
747static int metadata_digest_lookup_writeset(struct era_metadata *md,
748 struct digest *d)
749{
750 int r;
751 uint64_t key;
752 struct writeset_disk disk;
753
754 r = dm_btree_find_lowest_key(&md->writeset_tree_info,
755 md->writeset_tree_root, &key);
756 if (r < 0)
757 return r;
758
759 d->era = key;
760
761 r = dm_btree_lookup(&md->writeset_tree_info,
762 md->writeset_tree_root, &key, &disk);
763 if (r) {
764 if (r == -ENODATA) {
765 d->step = NULL;
766 return 0;
767 }
768
769 DMERR("%s: dm_btree_lookup failed", __func__);
770 return r;
771 }
772
773 ws_unpack(&disk, &d->writeset);
774 d->value = cpu_to_le32(key);
775
776 /*
777 * We initialise another bitset info to avoid any caching side effects
778 * with the previous one.
779 */
780 dm_disk_bitset_init(md->tm, &d->info);
781
782 d->nr_bits = min(d->writeset.nr_bits, md->nr_blocks);
783 d->current_bit = 0;
784 d->step = metadata_digest_transcribe_writeset;
785
786 return 0;
787}
788
789static int metadata_digest_start(struct era_metadata *md, struct digest *d)
790{
791 if (d->step)
792 return 0;
793
794 memset(d, 0, sizeof(*d));
795 d->step = metadata_digest_lookup_writeset;
796
797 return 0;
798}
799
800/*
801 *-----------------------------------------------------------------
802 * High level metadata interface. Target methods should use these,
803 * and not the lower level ones.
804 *-----------------------------------------------------------------
805 */
806static struct era_metadata *metadata_open(struct block_device *bdev,
807 sector_t block_size,
808 bool may_format)
809{
810 int r;
811 struct era_metadata *md = kzalloc(sizeof(*md), GFP_KERNEL);
812
813 if (!md)
814 return NULL;
815
816 md->bdev = bdev;
817 md->block_size = block_size;
818
819 md->writesets[0].md.root = INVALID_WRITESET_ROOT;
820 md->writesets[1].md.root = INVALID_WRITESET_ROOT;
821 md->current_writeset = &md->writesets[0];
822
823 r = create_persistent_data_objects(md, may_format);
824 if (r) {
825 kfree(md);
826 return ERR_PTR(r);
827 }
828
829 return md;
830}
831
832static void metadata_close(struct era_metadata *md)
833{
834 writeset_free(&md->writesets[0]);
835 writeset_free(&md->writesets[1]);
836 destroy_persistent_data_objects(md);
837 kfree(md);
838}
839
840static bool valid_nr_blocks(dm_block_t n)
841{
842 /*
843 * dm_bitset restricts us to 2^32. test_bit & co. restrict us
844 * further to 2^31 - 1
845 */
846 return n < (1ull << 31);
847}
848
849static int metadata_resize(struct era_metadata *md, void *arg)
850{
851 int r;
852 dm_block_t *new_size = arg;
853 __le32 value;
854
855 if (!valid_nr_blocks(*new_size)) {
856 DMERR("Invalid number of origin blocks %llu",
857 (unsigned long long) *new_size);
858 return -EINVAL;
859 }
860
861 writeset_free(&md->writesets[0]);
862 writeset_free(&md->writesets[1]);
863
864 r = writeset_alloc(&md->writesets[0], *new_size);
865 if (r) {
866 DMERR("%s: writeset_alloc failed for writeset 0", __func__);
867 return r;
868 }
869
870 r = writeset_alloc(&md->writesets[1], *new_size);
871 if (r) {
872 DMERR("%s: writeset_alloc failed for writeset 1", __func__);
873 writeset_free(&md->writesets[0]);
874 return r;
875 }
876
877 value = cpu_to_le32(0u);
878 __dm_bless_for_disk(&value);
879 r = dm_array_resize(&md->era_array_info, md->era_array_root,
880 md->nr_blocks, *new_size,
881 &value, &md->era_array_root);
882 if (r) {
883 DMERR("%s: dm_array_resize failed", __func__);
884 writeset_free(&md->writesets[0]);
885 writeset_free(&md->writesets[1]);
886 return r;
887 }
888
889 md->nr_blocks = *new_size;
890 return 0;
891}
892
893static int metadata_era_archive(struct era_metadata *md)
894{
895 int r;
896 uint64_t keys[1];
897 struct writeset_disk value;
898
899 r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root,
900 &md->current_writeset->md.root);
901 if (r) {
902 DMERR("%s: dm_bitset_flush failed", __func__);
903 return r;
904 }
905
906 ws_pack(&md->current_writeset->md, &value);
907
908 keys[0] = md->current_era;
909 __dm_bless_for_disk(&value);
910 r = dm_btree_insert(&md->writeset_tree_info, md->writeset_tree_root,
911 keys, &value, &md->writeset_tree_root);
912 if (r) {
913 DMERR("%s: couldn't insert writeset into btree", __func__);
914 /* FIXME: fail mode */
915 return r;
916 }
917
918 md->current_writeset->md.root = INVALID_WRITESET_ROOT;
919 md->archived_writesets = true;
920
921 return 0;
922}
923
924static struct writeset *next_writeset(struct era_metadata *md)
925{
926 return (md->current_writeset == &md->writesets[0]) ?
927 &md->writesets[1] : &md->writesets[0];
928}
929
930static int metadata_new_era(struct era_metadata *md)
931{
932 int r;
933 struct writeset *new_writeset = next_writeset(md);
934
935 r = writeset_init(&md->bitset_info, new_writeset, md->nr_blocks);
936 if (r) {
937 DMERR("%s: writeset_init failed", __func__);
938 return r;
939 }
940
941 swap_writeset(md, new_writeset);
942 md->current_era++;
943
944 return 0;
945}
946
947static int metadata_era_rollover(struct era_metadata *md)
948{
949 int r;
950
951 if (md->current_writeset->md.root != INVALID_WRITESET_ROOT) {
952 r = metadata_era_archive(md);
953 if (r) {
954 DMERR("%s: metadata_archive_era failed", __func__);
955 /* FIXME: fail mode? */
956 return r;
957 }
958 }
959
960 r = metadata_new_era(md);
961 if (r) {
962 DMERR("%s: new era failed", __func__);
963 /* FIXME: fail mode */
964 return r;
965 }
966
967 return 0;
968}
969
970static bool metadata_current_marked(struct era_metadata *md, dm_block_t block)
971{
972 bool r;
973 struct writeset *ws;
974
975 rcu_read_lock();
976 ws = rcu_dereference(md->current_writeset);
977 r = writeset_marked(ws, block);
978 rcu_read_unlock();
979
980 return r;
981}
982
983static int metadata_commit(struct era_metadata *md)
984{
985 int r;
986 struct dm_block *sblock;
987
988 if (md->current_writeset->md.root != INVALID_WRITESET_ROOT) {
989 r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root,
990 &md->current_writeset->md.root);
991 if (r) {
992 DMERR("%s: bitset flush failed", __func__);
993 return r;
994 }
995 }
996
997 r = dm_tm_pre_commit(md->tm);
998 if (r) {
999 DMERR("%s: pre commit failed", __func__);
1000 return r;
1001 }
1002
1003 r = save_sm_root(md);
1004 if (r) {
1005 DMERR("%s: save_sm_root failed", __func__);
1006 return r;
1007 }
1008
1009 r = superblock_lock(md, &sblock);
1010 if (r) {
1011 DMERR("%s: superblock lock failed", __func__);
1012 return r;
1013 }
1014
1015 prepare_superblock(md, dm_block_data(sblock));
1016
1017 return dm_tm_commit(md->tm, sblock);
1018}
1019
1020static int metadata_checkpoint(struct era_metadata *md)
1021{
1022 /*
1023 * For now we just rollover, but later I want to put a check in to
1024 * avoid this if the filter is still pretty fresh.
1025 */
1026 return metadata_era_rollover(md);
1027}
1028
1029/*
1030 * Metadata snapshots allow userland to access era data.
1031 */
1032static int metadata_take_snap(struct era_metadata *md)
1033{
1034 int r, inc;
1035 struct dm_block *clone;
1036
1037 if (md->metadata_snap != SUPERBLOCK_LOCATION) {
1038 DMERR("%s: metadata snapshot already exists", __func__);
1039 return -EINVAL;
1040 }
1041
1042 r = metadata_era_rollover(md);
1043 if (r) {
1044 DMERR("%s: era rollover failed", __func__);
1045 return r;
1046 }
1047
1048 r = metadata_commit(md);
1049 if (r) {
1050 DMERR("%s: pre commit failed", __func__);
1051 return r;
1052 }
1053
1054 r = dm_sm_inc_block(md->sm, SUPERBLOCK_LOCATION);
1055 if (r) {
1056 DMERR("%s: couldn't increment superblock", __func__);
1057 return r;
1058 }
1059
1060 r = dm_tm_shadow_block(md->tm, SUPERBLOCK_LOCATION,
1061 &sb_validator, &clone, &inc);
1062 if (r) {
1063 DMERR("%s: couldn't shadow superblock", __func__);
1064 dm_sm_dec_block(md->sm, SUPERBLOCK_LOCATION);
1065 return r;
1066 }
1067 BUG_ON(!inc);
1068
1069 r = dm_sm_inc_block(md->sm, md->writeset_tree_root);
1070 if (r) {
1071 DMERR("%s: couldn't inc writeset tree root", __func__);
1072 dm_tm_unlock(md->tm, clone);
1073 return r;
1074 }
1075
1076 r = dm_sm_inc_block(md->sm, md->era_array_root);
1077 if (r) {
1078 DMERR("%s: couldn't inc era tree root", __func__);
1079 dm_sm_dec_block(md->sm, md->writeset_tree_root);
1080 dm_tm_unlock(md->tm, clone);
1081 return r;
1082 }
1083
1084 md->metadata_snap = dm_block_location(clone);
1085
1086 dm_tm_unlock(md->tm, clone);
1087
1088 return 0;
1089}
1090
1091static int metadata_drop_snap(struct era_metadata *md)
1092{
1093 int r;
1094 dm_block_t location;
1095 struct dm_block *clone;
1096 struct superblock_disk *disk;
1097
1098 if (md->metadata_snap == SUPERBLOCK_LOCATION) {
1099 DMERR("%s: no snap to drop", __func__);
1100 return -EINVAL;
1101 }
1102
1103 r = dm_tm_read_lock(md->tm, md->metadata_snap, &sb_validator, &clone);
1104 if (r) {
1105 DMERR("%s: couldn't read lock superblock clone", __func__);
1106 return r;
1107 }
1108
1109 /*
1110 * Whatever happens now we'll commit with no record of the metadata
1111 * snap.
1112 */
1113 md->metadata_snap = SUPERBLOCK_LOCATION;
1114
1115 disk = dm_block_data(clone);
1116 r = dm_btree_del(&md->writeset_tree_info,
1117 le64_to_cpu(disk->writeset_tree_root));
1118 if (r) {
1119 DMERR("%s: error deleting writeset tree clone", __func__);
1120 dm_tm_unlock(md->tm, clone);
1121 return r;
1122 }
1123
1124 r = dm_array_del(&md->era_array_info, le64_to_cpu(disk->era_array_root));
1125 if (r) {
1126 DMERR("%s: error deleting era array clone", __func__);
1127 dm_tm_unlock(md->tm, clone);
1128 return r;
1129 }
1130
1131 location = dm_block_location(clone);
1132 dm_tm_unlock(md->tm, clone);
1133
1134 return dm_sm_dec_block(md->sm, location);
1135}
1136
1137struct metadata_stats {
1138 dm_block_t used;
1139 dm_block_t total;
1140 dm_block_t snap;
1141 uint32_t era;
1142};
1143
1144static int metadata_get_stats(struct era_metadata *md, void *ptr)
1145{
1146 int r;
1147 struct metadata_stats *s = ptr;
1148 dm_block_t nr_free, nr_total;
1149
1150 r = dm_sm_get_nr_free(md->sm, &nr_free);
1151 if (r) {
1152 DMERR("dm_sm_get_nr_free returned %d", r);
1153 return r;
1154 }
1155
1156 r = dm_sm_get_nr_blocks(md->sm, &nr_total);
1157 if (r) {
1158 DMERR("dm_pool_get_metadata_dev_size returned %d", r);
1159 return r;
1160 }
1161
1162 s->used = nr_total - nr_free;
1163 s->total = nr_total;
1164 s->snap = md->metadata_snap;
1165 s->era = md->current_era;
1166
1167 return 0;
1168}
1169
1170/*----------------------------------------------------------------*/
1171
1172struct era {
1173 struct dm_target *ti;
1174
1175 struct dm_dev *metadata_dev;
1176 struct dm_dev *origin_dev;
1177
1178 dm_block_t nr_blocks;
1179 uint32_t sectors_per_block;
1180 int sectors_per_block_shift;
1181 struct era_metadata *md;
1182
1183 struct workqueue_struct *wq;
1184 struct work_struct worker;
1185
1186 spinlock_t deferred_lock;
1187 struct bio_list deferred_bios;
1188
1189 spinlock_t rpc_lock;
1190 struct list_head rpc_calls;
1191
1192 struct digest digest;
1193 atomic_t suspended;
1194};
1195
1196struct rpc {
1197 struct list_head list;
1198
1199 int (*fn0)(struct era_metadata *md);
1200 int (*fn1)(struct era_metadata *md, void *ref);
1201 void *arg;
1202 int result;
1203
1204 struct completion complete;
1205};
1206
1207/*
1208 *---------------------------------------------------------------
1209 * Remapping.
1210 *---------------------------------------------------------------
1211 */
1212static bool block_size_is_power_of_two(struct era *era)
1213{
1214 return era->sectors_per_block_shift >= 0;
1215}
1216
1217static dm_block_t get_block(struct era *era, struct bio *bio)
1218{
1219 sector_t block_nr = bio->bi_iter.bi_sector;
1220
1221 if (!block_size_is_power_of_two(era))
1222 (void) sector_div(block_nr, era->sectors_per_block);
1223 else
1224 block_nr >>= era->sectors_per_block_shift;
1225
1226 return block_nr;
1227}
1228
1229static void remap_to_origin(struct era *era, struct bio *bio)
1230{
1231 bio_set_dev(bio, era->origin_dev->bdev);
1232}
1233
1234/*
1235 *--------------------------------------------------------------
1236 * Worker thread
1237 *--------------------------------------------------------------
1238 */
1239static void wake_worker(struct era *era)
1240{
1241 if (!atomic_read(&era->suspended))
1242 queue_work(era->wq, &era->worker);
1243}
1244
1245static void process_old_eras(struct era *era)
1246{
1247 int r;
1248
1249 if (!era->digest.step)
1250 return;
1251
1252 r = era->digest.step(era->md, &era->digest);
1253 if (r < 0) {
1254 DMERR("%s: digest step failed, stopping digestion", __func__);
1255 era->digest.step = NULL;
1256
1257 } else if (era->digest.step)
1258 wake_worker(era);
1259}
1260
1261static void process_deferred_bios(struct era *era)
1262{
1263 int r;
1264 struct bio_list deferred_bios, marked_bios;
1265 struct bio *bio;
1266 struct blk_plug plug;
1267 bool commit_needed = false;
1268 bool failed = false;
1269 struct writeset *ws = era->md->current_writeset;
1270
1271 bio_list_init(&deferred_bios);
1272 bio_list_init(&marked_bios);
1273
1274 spin_lock(&era->deferred_lock);
1275 bio_list_merge(&deferred_bios, &era->deferred_bios);
1276 bio_list_init(&era->deferred_bios);
1277 spin_unlock(&era->deferred_lock);
1278
1279 if (bio_list_empty(&deferred_bios))
1280 return;
1281
1282 while ((bio = bio_list_pop(&deferred_bios))) {
1283 r = writeset_test_and_set(&era->md->bitset_info, ws,
1284 get_block(era, bio));
1285 if (r < 0) {
1286 /*
1287 * This is bad news, we need to rollback.
1288 * FIXME: finish.
1289 */
1290 failed = true;
1291 } else if (r == 0)
1292 commit_needed = true;
1293
1294 bio_list_add(&marked_bios, bio);
1295 }
1296
1297 if (commit_needed) {
1298 r = metadata_commit(era->md);
1299 if (r)
1300 failed = true;
1301 }
1302
1303 if (failed)
1304 while ((bio = bio_list_pop(&marked_bios)))
1305 bio_io_error(bio);
1306 else {
1307 blk_start_plug(&plug);
1308 while ((bio = bio_list_pop(&marked_bios))) {
1309 /*
1310 * Only update the in-core writeset if the on-disk one
1311 * was updated too.
1312 */
1313 if (commit_needed)
1314 set_bit(get_block(era, bio), ws->bits);
1315 submit_bio_noacct(bio);
1316 }
1317 blk_finish_plug(&plug);
1318 }
1319}
1320
1321static void process_rpc_calls(struct era *era)
1322{
1323 int r;
1324 bool need_commit = false;
1325 struct list_head calls;
1326 struct rpc *rpc, *tmp;
1327
1328 INIT_LIST_HEAD(&calls);
1329 spin_lock(&era->rpc_lock);
1330 list_splice_init(&era->rpc_calls, &calls);
1331 spin_unlock(&era->rpc_lock);
1332
1333 list_for_each_entry_safe(rpc, tmp, &calls, list) {
1334 rpc->result = rpc->fn0 ? rpc->fn0(era->md) : rpc->fn1(era->md, rpc->arg);
1335 need_commit = true;
1336 }
1337
1338 if (need_commit) {
1339 r = metadata_commit(era->md);
1340 if (r)
1341 list_for_each_entry_safe(rpc, tmp, &calls, list)
1342 rpc->result = r;
1343 }
1344
1345 list_for_each_entry_safe(rpc, tmp, &calls, list)
1346 complete(&rpc->complete);
1347}
1348
1349static void kick_off_digest(struct era *era)
1350{
1351 if (era->md->archived_writesets) {
1352 era->md->archived_writesets = false;
1353 metadata_digest_start(era->md, &era->digest);
1354 }
1355}
1356
1357static void do_work(struct work_struct *ws)
1358{
1359 struct era *era = container_of(ws, struct era, worker);
1360
1361 kick_off_digest(era);
1362 process_old_eras(era);
1363 process_deferred_bios(era);
1364 process_rpc_calls(era);
1365}
1366
1367static void defer_bio(struct era *era, struct bio *bio)
1368{
1369 spin_lock(&era->deferred_lock);
1370 bio_list_add(&era->deferred_bios, bio);
1371 spin_unlock(&era->deferred_lock);
1372
1373 wake_worker(era);
1374}
1375
1376/*
1377 * Make an rpc call to the worker to change the metadata.
1378 */
1379static int perform_rpc(struct era *era, struct rpc *rpc)
1380{
1381 rpc->result = 0;
1382 init_completion(&rpc->complete);
1383
1384 spin_lock(&era->rpc_lock);
1385 list_add(&rpc->list, &era->rpc_calls);
1386 spin_unlock(&era->rpc_lock);
1387
1388 wake_worker(era);
1389 wait_for_completion(&rpc->complete);
1390
1391 return rpc->result;
1392}
1393
1394static int in_worker0(struct era *era, int (*fn)(struct era_metadata *md))
1395{
1396 struct rpc rpc;
1397
1398 rpc.fn0 = fn;
1399 rpc.fn1 = NULL;
1400
1401 return perform_rpc(era, &rpc);
1402}
1403
1404static int in_worker1(struct era *era,
1405 int (*fn)(struct era_metadata *md, void *ref), void *arg)
1406{
1407 struct rpc rpc;
1408
1409 rpc.fn0 = NULL;
1410 rpc.fn1 = fn;
1411 rpc.arg = arg;
1412
1413 return perform_rpc(era, &rpc);
1414}
1415
1416static void start_worker(struct era *era)
1417{
1418 atomic_set(&era->suspended, 0);
1419}
1420
1421static void stop_worker(struct era *era)
1422{
1423 atomic_set(&era->suspended, 1);
1424 drain_workqueue(era->wq);
1425}
1426
1427/*
1428 *--------------------------------------------------------------
1429 * Target methods
1430 *--------------------------------------------------------------
1431 */
1432static void era_destroy(struct era *era)
1433{
1434 if (era->md)
1435 metadata_close(era->md);
1436
1437 if (era->wq)
1438 destroy_workqueue(era->wq);
1439
1440 if (era->origin_dev)
1441 dm_put_device(era->ti, era->origin_dev);
1442
1443 if (era->metadata_dev)
1444 dm_put_device(era->ti, era->metadata_dev);
1445
1446 kfree(era);
1447}
1448
1449static dm_block_t calc_nr_blocks(struct era *era)
1450{
1451 return dm_sector_div_up(era->ti->len, era->sectors_per_block);
1452}
1453
1454static bool valid_block_size(dm_block_t block_size)
1455{
1456 bool greater_than_zero = block_size > 0;
1457 bool multiple_of_min_block_size = (block_size & (MIN_BLOCK_SIZE - 1)) == 0;
1458
1459 return greater_than_zero && multiple_of_min_block_size;
1460}
1461
1462/*
1463 * <metadata dev> <data dev> <data block size (sectors)>
1464 */
1465static int era_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1466{
1467 int r;
1468 char dummy;
1469 struct era *era;
1470 struct era_metadata *md;
1471
1472 if (argc != 3) {
1473 ti->error = "Invalid argument count";
1474 return -EINVAL;
1475 }
1476
1477 era = kzalloc(sizeof(*era), GFP_KERNEL);
1478 if (!era) {
1479 ti->error = "Error allocating era structure";
1480 return -ENOMEM;
1481 }
1482
1483 era->ti = ti;
1484
1485 r = dm_get_device(ti, argv[0], BLK_OPEN_READ | BLK_OPEN_WRITE,
1486 &era->metadata_dev);
1487 if (r) {
1488 ti->error = "Error opening metadata device";
1489 era_destroy(era);
1490 return -EINVAL;
1491 }
1492
1493 r = dm_get_device(ti, argv[1], BLK_OPEN_READ | BLK_OPEN_WRITE,
1494 &era->origin_dev);
1495 if (r) {
1496 ti->error = "Error opening data device";
1497 era_destroy(era);
1498 return -EINVAL;
1499 }
1500
1501 r = sscanf(argv[2], "%u%c", &era->sectors_per_block, &dummy);
1502 if (r != 1) {
1503 ti->error = "Error parsing block size";
1504 era_destroy(era);
1505 return -EINVAL;
1506 }
1507
1508 r = dm_set_target_max_io_len(ti, era->sectors_per_block);
1509 if (r) {
1510 ti->error = "could not set max io len";
1511 era_destroy(era);
1512 return -EINVAL;
1513 }
1514
1515 if (!valid_block_size(era->sectors_per_block)) {
1516 ti->error = "Invalid block size";
1517 era_destroy(era);
1518 return -EINVAL;
1519 }
1520 if (era->sectors_per_block & (era->sectors_per_block - 1))
1521 era->sectors_per_block_shift = -1;
1522 else
1523 era->sectors_per_block_shift = __ffs(era->sectors_per_block);
1524
1525 md = metadata_open(era->metadata_dev->bdev, era->sectors_per_block, true);
1526 if (IS_ERR(md)) {
1527 ti->error = "Error reading metadata";
1528 era_destroy(era);
1529 return PTR_ERR(md);
1530 }
1531 era->md = md;
1532
1533 era->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
1534 if (!era->wq) {
1535 ti->error = "could not create workqueue for metadata object";
1536 era_destroy(era);
1537 return -ENOMEM;
1538 }
1539 INIT_WORK(&era->worker, do_work);
1540
1541 spin_lock_init(&era->deferred_lock);
1542 bio_list_init(&era->deferred_bios);
1543
1544 spin_lock_init(&era->rpc_lock);
1545 INIT_LIST_HEAD(&era->rpc_calls);
1546
1547 ti->private = era;
1548 ti->num_flush_bios = 1;
1549 ti->flush_supported = true;
1550
1551 ti->num_discard_bios = 1;
1552
1553 return 0;
1554}
1555
1556static void era_dtr(struct dm_target *ti)
1557{
1558 era_destroy(ti->private);
1559}
1560
1561static int era_map(struct dm_target *ti, struct bio *bio)
1562{
1563 struct era *era = ti->private;
1564 dm_block_t block = get_block(era, bio);
1565
1566 /*
1567 * All bios get remapped to the origin device. We do this now, but
1568 * it may not get issued until later. Depending on whether the
1569 * block is marked in this era.
1570 */
1571 remap_to_origin(era, bio);
1572
1573 /*
1574 * REQ_PREFLUSH bios carry no data, so we're not interested in them.
1575 */
1576 if (!(bio->bi_opf & REQ_PREFLUSH) &&
1577 (bio_data_dir(bio) == WRITE) &&
1578 !metadata_current_marked(era->md, block)) {
1579 defer_bio(era, bio);
1580 return DM_MAPIO_SUBMITTED;
1581 }
1582
1583 return DM_MAPIO_REMAPPED;
1584}
1585
1586static void era_postsuspend(struct dm_target *ti)
1587{
1588 int r;
1589 struct era *era = ti->private;
1590
1591 r = in_worker0(era, metadata_era_archive);
1592 if (r) {
1593 DMERR("%s: couldn't archive current era", __func__);
1594 /* FIXME: fail mode */
1595 }
1596
1597 stop_worker(era);
1598
1599 r = metadata_commit(era->md);
1600 if (r) {
1601 DMERR("%s: metadata_commit failed", __func__);
1602 /* FIXME: fail mode */
1603 }
1604}
1605
1606static int era_preresume(struct dm_target *ti)
1607{
1608 int r;
1609 struct era *era = ti->private;
1610 dm_block_t new_size = calc_nr_blocks(era);
1611
1612 if (era->nr_blocks != new_size) {
1613 r = metadata_resize(era->md, &new_size);
1614 if (r) {
1615 DMERR("%s: metadata_resize failed", __func__);
1616 return r;
1617 }
1618
1619 r = metadata_commit(era->md);
1620 if (r) {
1621 DMERR("%s: metadata_commit failed", __func__);
1622 return r;
1623 }
1624
1625 era->nr_blocks = new_size;
1626 }
1627
1628 start_worker(era);
1629
1630 r = in_worker0(era, metadata_era_rollover);
1631 if (r) {
1632 DMERR("%s: metadata_era_rollover failed", __func__);
1633 return r;
1634 }
1635
1636 return 0;
1637}
1638
1639/*
1640 * Status format:
1641 *
1642 * <metadata block size> <#used metadata blocks>/<#total metadata blocks>
1643 * <current era> <held metadata root | '-'>
1644 */
1645static void era_status(struct dm_target *ti, status_type_t type,
1646 unsigned int status_flags, char *result, unsigned int maxlen)
1647{
1648 int r;
1649 struct era *era = ti->private;
1650 ssize_t sz = 0;
1651 struct metadata_stats stats;
1652 char buf[BDEVNAME_SIZE];
1653
1654 switch (type) {
1655 case STATUSTYPE_INFO:
1656 r = in_worker1(era, metadata_get_stats, &stats);
1657 if (r)
1658 goto err;
1659
1660 DMEMIT("%u %llu/%llu %u",
1661 (unsigned int) (DM_ERA_METADATA_BLOCK_SIZE >> SECTOR_SHIFT),
1662 (unsigned long long) stats.used,
1663 (unsigned long long) stats.total,
1664 (unsigned int) stats.era);
1665
1666 if (stats.snap != SUPERBLOCK_LOCATION)
1667 DMEMIT(" %llu", stats.snap);
1668 else
1669 DMEMIT(" -");
1670 break;
1671
1672 case STATUSTYPE_TABLE:
1673 format_dev_t(buf, era->metadata_dev->bdev->bd_dev);
1674 DMEMIT("%s ", buf);
1675 format_dev_t(buf, era->origin_dev->bdev->bd_dev);
1676 DMEMIT("%s %u", buf, era->sectors_per_block);
1677 break;
1678
1679 case STATUSTYPE_IMA:
1680 *result = '\0';
1681 break;
1682 }
1683
1684 return;
1685
1686err:
1687 DMEMIT("Error");
1688}
1689
1690static int era_message(struct dm_target *ti, unsigned int argc, char **argv,
1691 char *result, unsigned int maxlen)
1692{
1693 struct era *era = ti->private;
1694
1695 if (argc != 1) {
1696 DMERR("incorrect number of message arguments");
1697 return -EINVAL;
1698 }
1699
1700 if (!strcasecmp(argv[0], "checkpoint"))
1701 return in_worker0(era, metadata_checkpoint);
1702
1703 if (!strcasecmp(argv[0], "take_metadata_snap"))
1704 return in_worker0(era, metadata_take_snap);
1705
1706 if (!strcasecmp(argv[0], "drop_metadata_snap"))
1707 return in_worker0(era, metadata_drop_snap);
1708
1709 DMERR("unsupported message '%s'", argv[0]);
1710 return -EINVAL;
1711}
1712
1713static sector_t get_dev_size(struct dm_dev *dev)
1714{
1715 return bdev_nr_sectors(dev->bdev);
1716}
1717
1718static int era_iterate_devices(struct dm_target *ti,
1719 iterate_devices_callout_fn fn, void *data)
1720{
1721 struct era *era = ti->private;
1722
1723 return fn(ti, era->origin_dev, 0, get_dev_size(era->origin_dev), data);
1724}
1725
1726static void era_io_hints(struct dm_target *ti, struct queue_limits *limits)
1727{
1728 struct era *era = ti->private;
1729 uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
1730
1731 /*
1732 * If the system-determined stacked limits are compatible with the
1733 * era device's blocksize (io_opt is a factor) do not override them.
1734 */
1735 if (io_opt_sectors < era->sectors_per_block ||
1736 do_div(io_opt_sectors, era->sectors_per_block)) {
1737 blk_limits_io_min(limits, 0);
1738 blk_limits_io_opt(limits, era->sectors_per_block << SECTOR_SHIFT);
1739 }
1740}
1741
1742/*----------------------------------------------------------------*/
1743
1744static struct target_type era_target = {
1745 .name = "era",
1746 .version = {1, 0, 0},
1747 .module = THIS_MODULE,
1748 .ctr = era_ctr,
1749 .dtr = era_dtr,
1750 .map = era_map,
1751 .postsuspend = era_postsuspend,
1752 .preresume = era_preresume,
1753 .status = era_status,
1754 .message = era_message,
1755 .iterate_devices = era_iterate_devices,
1756 .io_hints = era_io_hints
1757};
1758module_dm(era);
1759
1760MODULE_DESCRIPTION(DM_NAME " era target");
1761MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>");
1762MODULE_LICENSE("GPL");
1#include "dm.h"
2#include "persistent-data/dm-transaction-manager.h"
3#include "persistent-data/dm-bitset.h"
4#include "persistent-data/dm-space-map.h"
5
6#include <linux/dm-io.h>
7#include <linux/dm-kcopyd.h>
8#include <linux/init.h>
9#include <linux/mempool.h>
10#include <linux/module.h>
11#include <linux/slab.h>
12#include <linux/vmalloc.h>
13
14#define DM_MSG_PREFIX "era"
15
16#define SUPERBLOCK_LOCATION 0
17#define SUPERBLOCK_MAGIC 2126579579
18#define SUPERBLOCK_CSUM_XOR 146538381
19#define MIN_ERA_VERSION 1
20#define MAX_ERA_VERSION 1
21#define INVALID_WRITESET_ROOT SUPERBLOCK_LOCATION
22#define MIN_BLOCK_SIZE 8
23
24/*----------------------------------------------------------------
25 * Writeset
26 *--------------------------------------------------------------*/
27struct writeset_metadata {
28 uint32_t nr_bits;
29 dm_block_t root;
30};
31
32struct writeset {
33 struct writeset_metadata md;
34
35 /*
36 * An in core copy of the bits to save constantly doing look ups on
37 * disk.
38 */
39 unsigned long *bits;
40};
41
42/*
43 * This does not free off the on disk bitset as this will normally be done
44 * after digesting into the era array.
45 */
46static void writeset_free(struct writeset *ws)
47{
48 vfree(ws->bits);
49}
50
51static int setup_on_disk_bitset(struct dm_disk_bitset *info,
52 unsigned nr_bits, dm_block_t *root)
53{
54 int r;
55
56 r = dm_bitset_empty(info, root);
57 if (r)
58 return r;
59
60 return dm_bitset_resize(info, *root, 0, nr_bits, false, root);
61}
62
63static size_t bitset_size(unsigned nr_bits)
64{
65 return sizeof(unsigned long) * dm_div_up(nr_bits, BITS_PER_LONG);
66}
67
68/*
69 * Allocates memory for the in core bitset.
70 */
71static int writeset_alloc(struct writeset *ws, dm_block_t nr_blocks)
72{
73 ws->md.nr_bits = nr_blocks;
74 ws->md.root = INVALID_WRITESET_ROOT;
75 ws->bits = vzalloc(bitset_size(nr_blocks));
76 if (!ws->bits) {
77 DMERR("%s: couldn't allocate in memory bitset", __func__);
78 return -ENOMEM;
79 }
80
81 return 0;
82}
83
84/*
85 * Wipes the in-core bitset, and creates a new on disk bitset.
86 */
87static int writeset_init(struct dm_disk_bitset *info, struct writeset *ws)
88{
89 int r;
90
91 memset(ws->bits, 0, bitset_size(ws->md.nr_bits));
92
93 r = setup_on_disk_bitset(info, ws->md.nr_bits, &ws->md.root);
94 if (r) {
95 DMERR("%s: setup_on_disk_bitset failed", __func__);
96 return r;
97 }
98
99 return 0;
100}
101
102static bool writeset_marked(struct writeset *ws, dm_block_t block)
103{
104 return test_bit(block, ws->bits);
105}
106
107static int writeset_marked_on_disk(struct dm_disk_bitset *info,
108 struct writeset_metadata *m, dm_block_t block,
109 bool *result)
110{
111 dm_block_t old = m->root;
112
113 /*
114 * The bitset was flushed when it was archived, so we know there'll
115 * be no change to the root.
116 */
117 int r = dm_bitset_test_bit(info, m->root, block, &m->root, result);
118 if (r) {
119 DMERR("%s: dm_bitset_test_bit failed", __func__);
120 return r;
121 }
122
123 BUG_ON(m->root != old);
124
125 return r;
126}
127
128/*
129 * Returns < 0 on error, 0 if the bit wasn't previously set, 1 if it was.
130 */
131static int writeset_test_and_set(struct dm_disk_bitset *info,
132 struct writeset *ws, uint32_t block)
133{
134 int r;
135
136 if (!test_and_set_bit(block, ws->bits)) {
137 r = dm_bitset_set_bit(info, ws->md.root, block, &ws->md.root);
138 if (r) {
139 /* FIXME: fail mode */
140 return r;
141 }
142
143 return 0;
144 }
145
146 return 1;
147}
148
149/*----------------------------------------------------------------
150 * On disk metadata layout
151 *--------------------------------------------------------------*/
152#define SPACE_MAP_ROOT_SIZE 128
153#define UUID_LEN 16
154
155struct writeset_disk {
156 __le32 nr_bits;
157 __le64 root;
158} __packed;
159
160struct superblock_disk {
161 __le32 csum;
162 __le32 flags;
163 __le64 blocknr;
164
165 __u8 uuid[UUID_LEN];
166 __le64 magic;
167 __le32 version;
168
169 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
170
171 __le32 data_block_size;
172 __le32 metadata_block_size;
173 __le32 nr_blocks;
174
175 __le32 current_era;
176 struct writeset_disk current_writeset;
177
178 /*
179 * Only these two fields are valid within the metadata snapshot.
180 */
181 __le64 writeset_tree_root;
182 __le64 era_array_root;
183
184 __le64 metadata_snap;
185} __packed;
186
187/*----------------------------------------------------------------
188 * Superblock validation
189 *--------------------------------------------------------------*/
190static void sb_prepare_for_write(struct dm_block_validator *v,
191 struct dm_block *b,
192 size_t sb_block_size)
193{
194 struct superblock_disk *disk = dm_block_data(b);
195
196 disk->blocknr = cpu_to_le64(dm_block_location(b));
197 disk->csum = cpu_to_le32(dm_bm_checksum(&disk->flags,
198 sb_block_size - sizeof(__le32),
199 SUPERBLOCK_CSUM_XOR));
200}
201
202static int check_metadata_version(struct superblock_disk *disk)
203{
204 uint32_t metadata_version = le32_to_cpu(disk->version);
205 if (metadata_version < MIN_ERA_VERSION || metadata_version > MAX_ERA_VERSION) {
206 DMERR("Era metadata version %u found, but only versions between %u and %u supported.",
207 metadata_version, MIN_ERA_VERSION, MAX_ERA_VERSION);
208 return -EINVAL;
209 }
210
211 return 0;
212}
213
214static int sb_check(struct dm_block_validator *v,
215 struct dm_block *b,
216 size_t sb_block_size)
217{
218 struct superblock_disk *disk = dm_block_data(b);
219 __le32 csum_le;
220
221 if (dm_block_location(b) != le64_to_cpu(disk->blocknr)) {
222 DMERR("sb_check failed: blocknr %llu: wanted %llu",
223 le64_to_cpu(disk->blocknr),
224 (unsigned long long)dm_block_location(b));
225 return -ENOTBLK;
226 }
227
228 if (le64_to_cpu(disk->magic) != SUPERBLOCK_MAGIC) {
229 DMERR("sb_check failed: magic %llu: wanted %llu",
230 le64_to_cpu(disk->magic),
231 (unsigned long long) SUPERBLOCK_MAGIC);
232 return -EILSEQ;
233 }
234
235 csum_le = cpu_to_le32(dm_bm_checksum(&disk->flags,
236 sb_block_size - sizeof(__le32),
237 SUPERBLOCK_CSUM_XOR));
238 if (csum_le != disk->csum) {
239 DMERR("sb_check failed: csum %u: wanted %u",
240 le32_to_cpu(csum_le), le32_to_cpu(disk->csum));
241 return -EILSEQ;
242 }
243
244 return check_metadata_version(disk);
245}
246
247static struct dm_block_validator sb_validator = {
248 .name = "superblock",
249 .prepare_for_write = sb_prepare_for_write,
250 .check = sb_check
251};
252
253/*----------------------------------------------------------------
254 * Low level metadata handling
255 *--------------------------------------------------------------*/
256#define DM_ERA_METADATA_BLOCK_SIZE 4096
257#define DM_ERA_METADATA_CACHE_SIZE 64
258#define ERA_MAX_CONCURRENT_LOCKS 5
259
260struct era_metadata {
261 struct block_device *bdev;
262 struct dm_block_manager *bm;
263 struct dm_space_map *sm;
264 struct dm_transaction_manager *tm;
265
266 dm_block_t block_size;
267 uint32_t nr_blocks;
268
269 uint32_t current_era;
270
271 /*
272 * We preallocate 2 writesets. When an era rolls over we
273 * switch between them. This means the allocation is done at
274 * preresume time, rather than on the io path.
275 */
276 struct writeset writesets[2];
277 struct writeset *current_writeset;
278
279 dm_block_t writeset_tree_root;
280 dm_block_t era_array_root;
281
282 struct dm_disk_bitset bitset_info;
283 struct dm_btree_info writeset_tree_info;
284 struct dm_array_info era_array_info;
285
286 dm_block_t metadata_snap;
287
288 /*
289 * A flag that is set whenever a writeset has been archived.
290 */
291 bool archived_writesets;
292
293 /*
294 * Reading the space map root can fail, so we read it into this
295 * buffer before the superblock is locked and updated.
296 */
297 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
298};
299
300static int superblock_read_lock(struct era_metadata *md,
301 struct dm_block **sblock)
302{
303 return dm_bm_read_lock(md->bm, SUPERBLOCK_LOCATION,
304 &sb_validator, sblock);
305}
306
307static int superblock_lock_zero(struct era_metadata *md,
308 struct dm_block **sblock)
309{
310 return dm_bm_write_lock_zero(md->bm, SUPERBLOCK_LOCATION,
311 &sb_validator, sblock);
312}
313
314static int superblock_lock(struct era_metadata *md,
315 struct dm_block **sblock)
316{
317 return dm_bm_write_lock(md->bm, SUPERBLOCK_LOCATION,
318 &sb_validator, sblock);
319}
320
321/* FIXME: duplication with cache and thin */
322static int superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
323{
324 int r;
325 unsigned i;
326 struct dm_block *b;
327 __le64 *data_le, zero = cpu_to_le64(0);
328 unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
329
330 /*
331 * We can't use a validator here - it may be all zeroes.
332 */
333 r = dm_bm_read_lock(bm, SUPERBLOCK_LOCATION, NULL, &b);
334 if (r)
335 return r;
336
337 data_le = dm_block_data(b);
338 *result = true;
339 for (i = 0; i < sb_block_size; i++) {
340 if (data_le[i] != zero) {
341 *result = false;
342 break;
343 }
344 }
345
346 return dm_bm_unlock(b);
347}
348
349/*----------------------------------------------------------------*/
350
351static void ws_pack(const struct writeset_metadata *core, struct writeset_disk *disk)
352{
353 disk->nr_bits = cpu_to_le32(core->nr_bits);
354 disk->root = cpu_to_le64(core->root);
355}
356
357static void ws_unpack(const struct writeset_disk *disk, struct writeset_metadata *core)
358{
359 core->nr_bits = le32_to_cpu(disk->nr_bits);
360 core->root = le64_to_cpu(disk->root);
361}
362
363static void ws_inc(void *context, const void *value)
364{
365 struct era_metadata *md = context;
366 struct writeset_disk ws_d;
367 dm_block_t b;
368
369 memcpy(&ws_d, value, sizeof(ws_d));
370 b = le64_to_cpu(ws_d.root);
371
372 dm_tm_inc(md->tm, b);
373}
374
375static void ws_dec(void *context, const void *value)
376{
377 struct era_metadata *md = context;
378 struct writeset_disk ws_d;
379 dm_block_t b;
380
381 memcpy(&ws_d, value, sizeof(ws_d));
382 b = le64_to_cpu(ws_d.root);
383
384 dm_bitset_del(&md->bitset_info, b);
385}
386
387static int ws_eq(void *context, const void *value1, const void *value2)
388{
389 return !memcmp(value1, value2, sizeof(struct writeset_metadata));
390}
391
392/*----------------------------------------------------------------*/
393
394static void setup_writeset_tree_info(struct era_metadata *md)
395{
396 struct dm_btree_value_type *vt = &md->writeset_tree_info.value_type;
397 md->writeset_tree_info.tm = md->tm;
398 md->writeset_tree_info.levels = 1;
399 vt->context = md;
400 vt->size = sizeof(struct writeset_disk);
401 vt->inc = ws_inc;
402 vt->dec = ws_dec;
403 vt->equal = ws_eq;
404}
405
406static void setup_era_array_info(struct era_metadata *md)
407
408{
409 struct dm_btree_value_type vt;
410 vt.context = NULL;
411 vt.size = sizeof(__le32);
412 vt.inc = NULL;
413 vt.dec = NULL;
414 vt.equal = NULL;
415
416 dm_array_info_init(&md->era_array_info, md->tm, &vt);
417}
418
419static void setup_infos(struct era_metadata *md)
420{
421 dm_disk_bitset_init(md->tm, &md->bitset_info);
422 setup_writeset_tree_info(md);
423 setup_era_array_info(md);
424}
425
426/*----------------------------------------------------------------*/
427
428static int create_fresh_metadata(struct era_metadata *md)
429{
430 int r;
431
432 r = dm_tm_create_with_sm(md->bm, SUPERBLOCK_LOCATION,
433 &md->tm, &md->sm);
434 if (r < 0) {
435 DMERR("dm_tm_create_with_sm failed");
436 return r;
437 }
438
439 setup_infos(md);
440
441 r = dm_btree_empty(&md->writeset_tree_info, &md->writeset_tree_root);
442 if (r) {
443 DMERR("couldn't create new writeset tree");
444 goto bad;
445 }
446
447 r = dm_array_empty(&md->era_array_info, &md->era_array_root);
448 if (r) {
449 DMERR("couldn't create era array");
450 goto bad;
451 }
452
453 return 0;
454
455bad:
456 dm_sm_destroy(md->sm);
457 dm_tm_destroy(md->tm);
458
459 return r;
460}
461
462static int save_sm_root(struct era_metadata *md)
463{
464 int r;
465 size_t metadata_len;
466
467 r = dm_sm_root_size(md->sm, &metadata_len);
468 if (r < 0)
469 return r;
470
471 return dm_sm_copy_root(md->sm, &md->metadata_space_map_root,
472 metadata_len);
473}
474
475static void copy_sm_root(struct era_metadata *md, struct superblock_disk *disk)
476{
477 memcpy(&disk->metadata_space_map_root,
478 &md->metadata_space_map_root,
479 sizeof(md->metadata_space_map_root));
480}
481
482/*
483 * Writes a superblock, including the static fields that don't get updated
484 * with every commit (possible optimisation here). 'md' should be fully
485 * constructed when this is called.
486 */
487static void prepare_superblock(struct era_metadata *md, struct superblock_disk *disk)
488{
489 disk->magic = cpu_to_le64(SUPERBLOCK_MAGIC);
490 disk->flags = cpu_to_le32(0ul);
491
492 /* FIXME: can't keep blanking the uuid (uuid is currently unused though) */
493 memset(disk->uuid, 0, sizeof(disk->uuid));
494 disk->version = cpu_to_le32(MAX_ERA_VERSION);
495
496 copy_sm_root(md, disk);
497
498 disk->data_block_size = cpu_to_le32(md->block_size);
499 disk->metadata_block_size = cpu_to_le32(DM_ERA_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
500 disk->nr_blocks = cpu_to_le32(md->nr_blocks);
501 disk->current_era = cpu_to_le32(md->current_era);
502
503 ws_pack(&md->current_writeset->md, &disk->current_writeset);
504 disk->writeset_tree_root = cpu_to_le64(md->writeset_tree_root);
505 disk->era_array_root = cpu_to_le64(md->era_array_root);
506 disk->metadata_snap = cpu_to_le64(md->metadata_snap);
507}
508
509static int write_superblock(struct era_metadata *md)
510{
511 int r;
512 struct dm_block *sblock;
513 struct superblock_disk *disk;
514
515 r = save_sm_root(md);
516 if (r) {
517 DMERR("%s: save_sm_root failed", __func__);
518 return r;
519 }
520
521 r = superblock_lock_zero(md, &sblock);
522 if (r)
523 return r;
524
525 disk = dm_block_data(sblock);
526 prepare_superblock(md, disk);
527
528 return dm_tm_commit(md->tm, sblock);
529}
530
531/*
532 * Assumes block_size and the infos are set.
533 */
534static int format_metadata(struct era_metadata *md)
535{
536 int r;
537
538 r = create_fresh_metadata(md);
539 if (r)
540 return r;
541
542 r = write_superblock(md);
543 if (r) {
544 dm_sm_destroy(md->sm);
545 dm_tm_destroy(md->tm);
546 return r;
547 }
548
549 return 0;
550}
551
552static int open_metadata(struct era_metadata *md)
553{
554 int r;
555 struct dm_block *sblock;
556 struct superblock_disk *disk;
557
558 r = superblock_read_lock(md, &sblock);
559 if (r) {
560 DMERR("couldn't read_lock superblock");
561 return r;
562 }
563
564 disk = dm_block_data(sblock);
565 r = dm_tm_open_with_sm(md->bm, SUPERBLOCK_LOCATION,
566 disk->metadata_space_map_root,
567 sizeof(disk->metadata_space_map_root),
568 &md->tm, &md->sm);
569 if (r) {
570 DMERR("dm_tm_open_with_sm failed");
571 goto bad;
572 }
573
574 setup_infos(md);
575
576 md->block_size = le32_to_cpu(disk->data_block_size);
577 md->nr_blocks = le32_to_cpu(disk->nr_blocks);
578 md->current_era = le32_to_cpu(disk->current_era);
579
580 md->writeset_tree_root = le64_to_cpu(disk->writeset_tree_root);
581 md->era_array_root = le64_to_cpu(disk->era_array_root);
582 md->metadata_snap = le64_to_cpu(disk->metadata_snap);
583 md->archived_writesets = true;
584
585 return dm_bm_unlock(sblock);
586
587bad:
588 dm_bm_unlock(sblock);
589 return r;
590}
591
592static int open_or_format_metadata(struct era_metadata *md,
593 bool may_format)
594{
595 int r;
596 bool unformatted = false;
597
598 r = superblock_all_zeroes(md->bm, &unformatted);
599 if (r)
600 return r;
601
602 if (unformatted)
603 return may_format ? format_metadata(md) : -EPERM;
604
605 return open_metadata(md);
606}
607
608static int create_persistent_data_objects(struct era_metadata *md,
609 bool may_format)
610{
611 int r;
612
613 md->bm = dm_block_manager_create(md->bdev, DM_ERA_METADATA_BLOCK_SIZE,
614 DM_ERA_METADATA_CACHE_SIZE,
615 ERA_MAX_CONCURRENT_LOCKS);
616 if (IS_ERR(md->bm)) {
617 DMERR("could not create block manager");
618 return PTR_ERR(md->bm);
619 }
620
621 r = open_or_format_metadata(md, may_format);
622 if (r)
623 dm_block_manager_destroy(md->bm);
624
625 return r;
626}
627
628static void destroy_persistent_data_objects(struct era_metadata *md)
629{
630 dm_sm_destroy(md->sm);
631 dm_tm_destroy(md->tm);
632 dm_block_manager_destroy(md->bm);
633}
634
635/*
636 * This waits until all era_map threads have picked up the new filter.
637 */
638static void swap_writeset(struct era_metadata *md, struct writeset *new_writeset)
639{
640 rcu_assign_pointer(md->current_writeset, new_writeset);
641 synchronize_rcu();
642}
643
644/*----------------------------------------------------------------
645 * Writesets get 'digested' into the main era array.
646 *
647 * We're using a coroutine here so the worker thread can do the digestion,
648 * thus avoiding synchronisation of the metadata. Digesting a whole
649 * writeset in one go would cause too much latency.
650 *--------------------------------------------------------------*/
651struct digest {
652 uint32_t era;
653 unsigned nr_bits, current_bit;
654 struct writeset_metadata writeset;
655 __le32 value;
656 struct dm_disk_bitset info;
657
658 int (*step)(struct era_metadata *, struct digest *);
659};
660
661static int metadata_digest_lookup_writeset(struct era_metadata *md,
662 struct digest *d);
663
664static int metadata_digest_remove_writeset(struct era_metadata *md,
665 struct digest *d)
666{
667 int r;
668 uint64_t key = d->era;
669
670 r = dm_btree_remove(&md->writeset_tree_info, md->writeset_tree_root,
671 &key, &md->writeset_tree_root);
672 if (r) {
673 DMERR("%s: dm_btree_remove failed", __func__);
674 return r;
675 }
676
677 d->step = metadata_digest_lookup_writeset;
678 return 0;
679}
680
681#define INSERTS_PER_STEP 100
682
683static int metadata_digest_transcribe_writeset(struct era_metadata *md,
684 struct digest *d)
685{
686 int r;
687 bool marked;
688 unsigned b, e = min(d->current_bit + INSERTS_PER_STEP, d->nr_bits);
689
690 for (b = d->current_bit; b < e; b++) {
691 r = writeset_marked_on_disk(&d->info, &d->writeset, b, &marked);
692 if (r) {
693 DMERR("%s: writeset_marked_on_disk failed", __func__);
694 return r;
695 }
696
697 if (!marked)
698 continue;
699
700 __dm_bless_for_disk(&d->value);
701 r = dm_array_set_value(&md->era_array_info, md->era_array_root,
702 b, &d->value, &md->era_array_root);
703 if (r) {
704 DMERR("%s: dm_array_set_value failed", __func__);
705 return r;
706 }
707 }
708
709 if (b == d->nr_bits)
710 d->step = metadata_digest_remove_writeset;
711 else
712 d->current_bit = b;
713
714 return 0;
715}
716
717static int metadata_digest_lookup_writeset(struct era_metadata *md,
718 struct digest *d)
719{
720 int r;
721 uint64_t key;
722 struct writeset_disk disk;
723
724 r = dm_btree_find_lowest_key(&md->writeset_tree_info,
725 md->writeset_tree_root, &key);
726 if (r < 0)
727 return r;
728
729 d->era = key;
730
731 r = dm_btree_lookup(&md->writeset_tree_info,
732 md->writeset_tree_root, &key, &disk);
733 if (r) {
734 if (r == -ENODATA) {
735 d->step = NULL;
736 return 0;
737 }
738
739 DMERR("%s: dm_btree_lookup failed", __func__);
740 return r;
741 }
742
743 ws_unpack(&disk, &d->writeset);
744 d->value = cpu_to_le32(key);
745
746 d->nr_bits = min(d->writeset.nr_bits, md->nr_blocks);
747 d->current_bit = 0;
748 d->step = metadata_digest_transcribe_writeset;
749
750 return 0;
751}
752
753static int metadata_digest_start(struct era_metadata *md, struct digest *d)
754{
755 if (d->step)
756 return 0;
757
758 memset(d, 0, sizeof(*d));
759
760 /*
761 * We initialise another bitset info to avoid any caching side
762 * effects with the previous one.
763 */
764 dm_disk_bitset_init(md->tm, &d->info);
765 d->step = metadata_digest_lookup_writeset;
766
767 return 0;
768}
769
770/*----------------------------------------------------------------
771 * High level metadata interface. Target methods should use these, and not
772 * the lower level ones.
773 *--------------------------------------------------------------*/
774static struct era_metadata *metadata_open(struct block_device *bdev,
775 sector_t block_size,
776 bool may_format)
777{
778 int r;
779 struct era_metadata *md = kzalloc(sizeof(*md), GFP_KERNEL);
780
781 if (!md)
782 return NULL;
783
784 md->bdev = bdev;
785 md->block_size = block_size;
786
787 md->writesets[0].md.root = INVALID_WRITESET_ROOT;
788 md->writesets[1].md.root = INVALID_WRITESET_ROOT;
789 md->current_writeset = &md->writesets[0];
790
791 r = create_persistent_data_objects(md, may_format);
792 if (r) {
793 kfree(md);
794 return ERR_PTR(r);
795 }
796
797 return md;
798}
799
800static void metadata_close(struct era_metadata *md)
801{
802 destroy_persistent_data_objects(md);
803 kfree(md);
804}
805
806static bool valid_nr_blocks(dm_block_t n)
807{
808 /*
809 * dm_bitset restricts us to 2^32. test_bit & co. restrict us
810 * further to 2^31 - 1
811 */
812 return n < (1ull << 31);
813}
814
815static int metadata_resize(struct era_metadata *md, void *arg)
816{
817 int r;
818 dm_block_t *new_size = arg;
819 __le32 value;
820
821 if (!valid_nr_blocks(*new_size)) {
822 DMERR("Invalid number of origin blocks %llu",
823 (unsigned long long) *new_size);
824 return -EINVAL;
825 }
826
827 writeset_free(&md->writesets[0]);
828 writeset_free(&md->writesets[1]);
829
830 r = writeset_alloc(&md->writesets[0], *new_size);
831 if (r) {
832 DMERR("%s: writeset_alloc failed for writeset 0", __func__);
833 return r;
834 }
835
836 r = writeset_alloc(&md->writesets[1], *new_size);
837 if (r) {
838 DMERR("%s: writeset_alloc failed for writeset 1", __func__);
839 return r;
840 }
841
842 value = cpu_to_le32(0u);
843 __dm_bless_for_disk(&value);
844 r = dm_array_resize(&md->era_array_info, md->era_array_root,
845 md->nr_blocks, *new_size,
846 &value, &md->era_array_root);
847 if (r) {
848 DMERR("%s: dm_array_resize failed", __func__);
849 return r;
850 }
851
852 md->nr_blocks = *new_size;
853 return 0;
854}
855
856static int metadata_era_archive(struct era_metadata *md)
857{
858 int r;
859 uint64_t keys[1];
860 struct writeset_disk value;
861
862 r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root,
863 &md->current_writeset->md.root);
864 if (r) {
865 DMERR("%s: dm_bitset_flush failed", __func__);
866 return r;
867 }
868
869 ws_pack(&md->current_writeset->md, &value);
870 md->current_writeset->md.root = INVALID_WRITESET_ROOT;
871
872 keys[0] = md->current_era;
873 __dm_bless_for_disk(&value);
874 r = dm_btree_insert(&md->writeset_tree_info, md->writeset_tree_root,
875 keys, &value, &md->writeset_tree_root);
876 if (r) {
877 DMERR("%s: couldn't insert writeset into btree", __func__);
878 /* FIXME: fail mode */
879 return r;
880 }
881
882 md->archived_writesets = true;
883
884 return 0;
885}
886
887static struct writeset *next_writeset(struct era_metadata *md)
888{
889 return (md->current_writeset == &md->writesets[0]) ?
890 &md->writesets[1] : &md->writesets[0];
891}
892
893static int metadata_new_era(struct era_metadata *md)
894{
895 int r;
896 struct writeset *new_writeset = next_writeset(md);
897
898 r = writeset_init(&md->bitset_info, new_writeset);
899 if (r) {
900 DMERR("%s: writeset_init failed", __func__);
901 return r;
902 }
903
904 swap_writeset(md, new_writeset);
905 md->current_era++;
906
907 return 0;
908}
909
910static int metadata_era_rollover(struct era_metadata *md)
911{
912 int r;
913
914 if (md->current_writeset->md.root != INVALID_WRITESET_ROOT) {
915 r = metadata_era_archive(md);
916 if (r) {
917 DMERR("%s: metadata_archive_era failed", __func__);
918 /* FIXME: fail mode? */
919 return r;
920 }
921 }
922
923 r = metadata_new_era(md);
924 if (r) {
925 DMERR("%s: new era failed", __func__);
926 /* FIXME: fail mode */
927 return r;
928 }
929
930 return 0;
931}
932
933static bool metadata_current_marked(struct era_metadata *md, dm_block_t block)
934{
935 bool r;
936 struct writeset *ws;
937
938 rcu_read_lock();
939 ws = rcu_dereference(md->current_writeset);
940 r = writeset_marked(ws, block);
941 rcu_read_unlock();
942
943 return r;
944}
945
946static int metadata_commit(struct era_metadata *md)
947{
948 int r;
949 struct dm_block *sblock;
950
951 if (md->current_writeset->md.root != SUPERBLOCK_LOCATION) {
952 r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root,
953 &md->current_writeset->md.root);
954 if (r) {
955 DMERR("%s: bitset flush failed", __func__);
956 return r;
957 }
958 }
959
960 r = save_sm_root(md);
961 if (r) {
962 DMERR("%s: save_sm_root failed", __func__);
963 return r;
964 }
965
966 r = dm_tm_pre_commit(md->tm);
967 if (r) {
968 DMERR("%s: pre commit failed", __func__);
969 return r;
970 }
971
972 r = superblock_lock(md, &sblock);
973 if (r) {
974 DMERR("%s: superblock lock failed", __func__);
975 return r;
976 }
977
978 prepare_superblock(md, dm_block_data(sblock));
979
980 return dm_tm_commit(md->tm, sblock);
981}
982
983static int metadata_checkpoint(struct era_metadata *md)
984{
985 /*
986 * For now we just rollover, but later I want to put a check in to
987 * avoid this if the filter is still pretty fresh.
988 */
989 return metadata_era_rollover(md);
990}
991
992/*
993 * Metadata snapshots allow userland to access era data.
994 */
995static int metadata_take_snap(struct era_metadata *md)
996{
997 int r, inc;
998 struct dm_block *clone;
999
1000 if (md->metadata_snap != SUPERBLOCK_LOCATION) {
1001 DMERR("%s: metadata snapshot already exists", __func__);
1002 return -EINVAL;
1003 }
1004
1005 r = metadata_era_rollover(md);
1006 if (r) {
1007 DMERR("%s: era rollover failed", __func__);
1008 return r;
1009 }
1010
1011 r = metadata_commit(md);
1012 if (r) {
1013 DMERR("%s: pre commit failed", __func__);
1014 return r;
1015 }
1016
1017 r = dm_sm_inc_block(md->sm, SUPERBLOCK_LOCATION);
1018 if (r) {
1019 DMERR("%s: couldn't increment superblock", __func__);
1020 return r;
1021 }
1022
1023 r = dm_tm_shadow_block(md->tm, SUPERBLOCK_LOCATION,
1024 &sb_validator, &clone, &inc);
1025 if (r) {
1026 DMERR("%s: couldn't shadow superblock", __func__);
1027 dm_sm_dec_block(md->sm, SUPERBLOCK_LOCATION);
1028 return r;
1029 }
1030 BUG_ON(!inc);
1031
1032 r = dm_sm_inc_block(md->sm, md->writeset_tree_root);
1033 if (r) {
1034 DMERR("%s: couldn't inc writeset tree root", __func__);
1035 dm_tm_unlock(md->tm, clone);
1036 return r;
1037 }
1038
1039 r = dm_sm_inc_block(md->sm, md->era_array_root);
1040 if (r) {
1041 DMERR("%s: couldn't inc era tree root", __func__);
1042 dm_sm_dec_block(md->sm, md->writeset_tree_root);
1043 dm_tm_unlock(md->tm, clone);
1044 return r;
1045 }
1046
1047 md->metadata_snap = dm_block_location(clone);
1048
1049 r = dm_tm_unlock(md->tm, clone);
1050 if (r) {
1051 DMERR("%s: couldn't unlock clone", __func__);
1052 md->metadata_snap = SUPERBLOCK_LOCATION;
1053 return r;
1054 }
1055
1056 return 0;
1057}
1058
1059static int metadata_drop_snap(struct era_metadata *md)
1060{
1061 int r;
1062 dm_block_t location;
1063 struct dm_block *clone;
1064 struct superblock_disk *disk;
1065
1066 if (md->metadata_snap == SUPERBLOCK_LOCATION) {
1067 DMERR("%s: no snap to drop", __func__);
1068 return -EINVAL;
1069 }
1070
1071 r = dm_tm_read_lock(md->tm, md->metadata_snap, &sb_validator, &clone);
1072 if (r) {
1073 DMERR("%s: couldn't read lock superblock clone", __func__);
1074 return r;
1075 }
1076
1077 /*
1078 * Whatever happens now we'll commit with no record of the metadata
1079 * snap.
1080 */
1081 md->metadata_snap = SUPERBLOCK_LOCATION;
1082
1083 disk = dm_block_data(clone);
1084 r = dm_btree_del(&md->writeset_tree_info,
1085 le64_to_cpu(disk->writeset_tree_root));
1086 if (r) {
1087 DMERR("%s: error deleting writeset tree clone", __func__);
1088 dm_tm_unlock(md->tm, clone);
1089 return r;
1090 }
1091
1092 r = dm_array_del(&md->era_array_info, le64_to_cpu(disk->era_array_root));
1093 if (r) {
1094 DMERR("%s: error deleting era array clone", __func__);
1095 dm_tm_unlock(md->tm, clone);
1096 return r;
1097 }
1098
1099 location = dm_block_location(clone);
1100 dm_tm_unlock(md->tm, clone);
1101
1102 return dm_sm_dec_block(md->sm, location);
1103}
1104
1105struct metadata_stats {
1106 dm_block_t used;
1107 dm_block_t total;
1108 dm_block_t snap;
1109 uint32_t era;
1110};
1111
1112static int metadata_get_stats(struct era_metadata *md, void *ptr)
1113{
1114 int r;
1115 struct metadata_stats *s = ptr;
1116 dm_block_t nr_free, nr_total;
1117
1118 r = dm_sm_get_nr_free(md->sm, &nr_free);
1119 if (r) {
1120 DMERR("dm_sm_get_nr_free returned %d", r);
1121 return r;
1122 }
1123
1124 r = dm_sm_get_nr_blocks(md->sm, &nr_total);
1125 if (r) {
1126 DMERR("dm_pool_get_metadata_dev_size returned %d", r);
1127 return r;
1128 }
1129
1130 s->used = nr_total - nr_free;
1131 s->total = nr_total;
1132 s->snap = md->metadata_snap;
1133 s->era = md->current_era;
1134
1135 return 0;
1136}
1137
1138/*----------------------------------------------------------------*/
1139
1140struct era {
1141 struct dm_target *ti;
1142 struct dm_target_callbacks callbacks;
1143
1144 struct dm_dev *metadata_dev;
1145 struct dm_dev *origin_dev;
1146
1147 dm_block_t nr_blocks;
1148 uint32_t sectors_per_block;
1149 int sectors_per_block_shift;
1150 struct era_metadata *md;
1151
1152 struct workqueue_struct *wq;
1153 struct work_struct worker;
1154
1155 spinlock_t deferred_lock;
1156 struct bio_list deferred_bios;
1157
1158 spinlock_t rpc_lock;
1159 struct list_head rpc_calls;
1160
1161 struct digest digest;
1162 atomic_t suspended;
1163};
1164
1165struct rpc {
1166 struct list_head list;
1167
1168 int (*fn0)(struct era_metadata *);
1169 int (*fn1)(struct era_metadata *, void *);
1170 void *arg;
1171 int result;
1172
1173 struct completion complete;
1174};
1175
1176/*----------------------------------------------------------------
1177 * Remapping.
1178 *---------------------------------------------------------------*/
1179static bool block_size_is_power_of_two(struct era *era)
1180{
1181 return era->sectors_per_block_shift >= 0;
1182}
1183
1184static dm_block_t get_block(struct era *era, struct bio *bio)
1185{
1186 sector_t block_nr = bio->bi_iter.bi_sector;
1187
1188 if (!block_size_is_power_of_two(era))
1189 (void) sector_div(block_nr, era->sectors_per_block);
1190 else
1191 block_nr >>= era->sectors_per_block_shift;
1192
1193 return block_nr;
1194}
1195
1196static void remap_to_origin(struct era *era, struct bio *bio)
1197{
1198 bio->bi_bdev = era->origin_dev->bdev;
1199}
1200
1201/*----------------------------------------------------------------
1202 * Worker thread
1203 *--------------------------------------------------------------*/
1204static void wake_worker(struct era *era)
1205{
1206 if (!atomic_read(&era->suspended))
1207 queue_work(era->wq, &era->worker);
1208}
1209
1210static void process_old_eras(struct era *era)
1211{
1212 int r;
1213
1214 if (!era->digest.step)
1215 return;
1216
1217 r = era->digest.step(era->md, &era->digest);
1218 if (r < 0) {
1219 DMERR("%s: digest step failed, stopping digestion", __func__);
1220 era->digest.step = NULL;
1221
1222 } else if (era->digest.step)
1223 wake_worker(era);
1224}
1225
1226static void process_deferred_bios(struct era *era)
1227{
1228 int r;
1229 struct bio_list deferred_bios, marked_bios;
1230 struct bio *bio;
1231 bool commit_needed = false;
1232 bool failed = false;
1233
1234 bio_list_init(&deferred_bios);
1235 bio_list_init(&marked_bios);
1236
1237 spin_lock(&era->deferred_lock);
1238 bio_list_merge(&deferred_bios, &era->deferred_bios);
1239 bio_list_init(&era->deferred_bios);
1240 spin_unlock(&era->deferred_lock);
1241
1242 while ((bio = bio_list_pop(&deferred_bios))) {
1243 r = writeset_test_and_set(&era->md->bitset_info,
1244 era->md->current_writeset,
1245 get_block(era, bio));
1246 if (r < 0) {
1247 /*
1248 * This is bad news, we need to rollback.
1249 * FIXME: finish.
1250 */
1251 failed = true;
1252
1253 } else if (r == 0)
1254 commit_needed = true;
1255
1256 bio_list_add(&marked_bios, bio);
1257 }
1258
1259 if (commit_needed) {
1260 r = metadata_commit(era->md);
1261 if (r)
1262 failed = true;
1263 }
1264
1265 if (failed)
1266 while ((bio = bio_list_pop(&marked_bios)))
1267 bio_io_error(bio);
1268 else
1269 while ((bio = bio_list_pop(&marked_bios)))
1270 generic_make_request(bio);
1271}
1272
1273static void process_rpc_calls(struct era *era)
1274{
1275 int r;
1276 bool need_commit = false;
1277 struct list_head calls;
1278 struct rpc *rpc, *tmp;
1279
1280 INIT_LIST_HEAD(&calls);
1281 spin_lock(&era->rpc_lock);
1282 list_splice_init(&era->rpc_calls, &calls);
1283 spin_unlock(&era->rpc_lock);
1284
1285 list_for_each_entry_safe(rpc, tmp, &calls, list) {
1286 rpc->result = rpc->fn0 ? rpc->fn0(era->md) : rpc->fn1(era->md, rpc->arg);
1287 need_commit = true;
1288 }
1289
1290 if (need_commit) {
1291 r = metadata_commit(era->md);
1292 if (r)
1293 list_for_each_entry_safe(rpc, tmp, &calls, list)
1294 rpc->result = r;
1295 }
1296
1297 list_for_each_entry_safe(rpc, tmp, &calls, list)
1298 complete(&rpc->complete);
1299}
1300
1301static void kick_off_digest(struct era *era)
1302{
1303 if (era->md->archived_writesets) {
1304 era->md->archived_writesets = false;
1305 metadata_digest_start(era->md, &era->digest);
1306 }
1307}
1308
1309static void do_work(struct work_struct *ws)
1310{
1311 struct era *era = container_of(ws, struct era, worker);
1312
1313 kick_off_digest(era);
1314 process_old_eras(era);
1315 process_deferred_bios(era);
1316 process_rpc_calls(era);
1317}
1318
1319static void defer_bio(struct era *era, struct bio *bio)
1320{
1321 spin_lock(&era->deferred_lock);
1322 bio_list_add(&era->deferred_bios, bio);
1323 spin_unlock(&era->deferred_lock);
1324
1325 wake_worker(era);
1326}
1327
1328/*
1329 * Make an rpc call to the worker to change the metadata.
1330 */
1331static int perform_rpc(struct era *era, struct rpc *rpc)
1332{
1333 rpc->result = 0;
1334 init_completion(&rpc->complete);
1335
1336 spin_lock(&era->rpc_lock);
1337 list_add(&rpc->list, &era->rpc_calls);
1338 spin_unlock(&era->rpc_lock);
1339
1340 wake_worker(era);
1341 wait_for_completion(&rpc->complete);
1342
1343 return rpc->result;
1344}
1345
1346static int in_worker0(struct era *era, int (*fn)(struct era_metadata *))
1347{
1348 struct rpc rpc;
1349 rpc.fn0 = fn;
1350 rpc.fn1 = NULL;
1351
1352 return perform_rpc(era, &rpc);
1353}
1354
1355static int in_worker1(struct era *era,
1356 int (*fn)(struct era_metadata *, void *), void *arg)
1357{
1358 struct rpc rpc;
1359 rpc.fn0 = NULL;
1360 rpc.fn1 = fn;
1361 rpc.arg = arg;
1362
1363 return perform_rpc(era, &rpc);
1364}
1365
1366static void start_worker(struct era *era)
1367{
1368 atomic_set(&era->suspended, 0);
1369}
1370
1371static void stop_worker(struct era *era)
1372{
1373 atomic_set(&era->suspended, 1);
1374 flush_workqueue(era->wq);
1375}
1376
1377/*----------------------------------------------------------------
1378 * Target methods
1379 *--------------------------------------------------------------*/
1380static int dev_is_congested(struct dm_dev *dev, int bdi_bits)
1381{
1382 struct request_queue *q = bdev_get_queue(dev->bdev);
1383 return bdi_congested(&q->backing_dev_info, bdi_bits);
1384}
1385
1386static int era_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
1387{
1388 struct era *era = container_of(cb, struct era, callbacks);
1389 return dev_is_congested(era->origin_dev, bdi_bits);
1390}
1391
1392static void era_destroy(struct era *era)
1393{
1394 metadata_close(era->md);
1395
1396 if (era->wq)
1397 destroy_workqueue(era->wq);
1398
1399 if (era->origin_dev)
1400 dm_put_device(era->ti, era->origin_dev);
1401
1402 if (era->metadata_dev)
1403 dm_put_device(era->ti, era->metadata_dev);
1404
1405 kfree(era);
1406}
1407
1408static dm_block_t calc_nr_blocks(struct era *era)
1409{
1410 return dm_sector_div_up(era->ti->len, era->sectors_per_block);
1411}
1412
1413static bool valid_block_size(dm_block_t block_size)
1414{
1415 bool greater_than_zero = block_size > 0;
1416 bool multiple_of_min_block_size = (block_size & (MIN_BLOCK_SIZE - 1)) == 0;
1417
1418 return greater_than_zero && multiple_of_min_block_size;
1419}
1420
1421/*
1422 * <metadata dev> <data dev> <data block size (sectors)>
1423 */
1424static int era_ctr(struct dm_target *ti, unsigned argc, char **argv)
1425{
1426 int r;
1427 char dummy;
1428 struct era *era;
1429 struct era_metadata *md;
1430
1431 if (argc != 3) {
1432 ti->error = "Invalid argument count";
1433 return -EINVAL;
1434 }
1435
1436 era = kzalloc(sizeof(*era), GFP_KERNEL);
1437 if (!era) {
1438 ti->error = "Error allocating era structure";
1439 return -ENOMEM;
1440 }
1441
1442 era->ti = ti;
1443
1444 r = dm_get_device(ti, argv[0], FMODE_READ | FMODE_WRITE, &era->metadata_dev);
1445 if (r) {
1446 ti->error = "Error opening metadata device";
1447 era_destroy(era);
1448 return -EINVAL;
1449 }
1450
1451 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &era->origin_dev);
1452 if (r) {
1453 ti->error = "Error opening data device";
1454 era_destroy(era);
1455 return -EINVAL;
1456 }
1457
1458 r = sscanf(argv[2], "%u%c", &era->sectors_per_block, &dummy);
1459 if (r != 1) {
1460 ti->error = "Error parsing block size";
1461 era_destroy(era);
1462 return -EINVAL;
1463 }
1464
1465 r = dm_set_target_max_io_len(ti, era->sectors_per_block);
1466 if (r) {
1467 ti->error = "could not set max io len";
1468 era_destroy(era);
1469 return -EINVAL;
1470 }
1471
1472 if (!valid_block_size(era->sectors_per_block)) {
1473 ti->error = "Invalid block size";
1474 era_destroy(era);
1475 return -EINVAL;
1476 }
1477 if (era->sectors_per_block & (era->sectors_per_block - 1))
1478 era->sectors_per_block_shift = -1;
1479 else
1480 era->sectors_per_block_shift = __ffs(era->sectors_per_block);
1481
1482 md = metadata_open(era->metadata_dev->bdev, era->sectors_per_block, true);
1483 if (IS_ERR(md)) {
1484 ti->error = "Error reading metadata";
1485 era_destroy(era);
1486 return PTR_ERR(md);
1487 }
1488 era->md = md;
1489
1490 era->nr_blocks = calc_nr_blocks(era);
1491
1492 r = metadata_resize(era->md, &era->nr_blocks);
1493 if (r) {
1494 ti->error = "couldn't resize metadata";
1495 era_destroy(era);
1496 return -ENOMEM;
1497 }
1498
1499 era->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
1500 if (!era->wq) {
1501 ti->error = "could not create workqueue for metadata object";
1502 era_destroy(era);
1503 return -ENOMEM;
1504 }
1505 INIT_WORK(&era->worker, do_work);
1506
1507 spin_lock_init(&era->deferred_lock);
1508 bio_list_init(&era->deferred_bios);
1509
1510 spin_lock_init(&era->rpc_lock);
1511 INIT_LIST_HEAD(&era->rpc_calls);
1512
1513 ti->private = era;
1514 ti->num_flush_bios = 1;
1515 ti->flush_supported = true;
1516
1517 ti->num_discard_bios = 1;
1518 ti->discards_supported = true;
1519 era->callbacks.congested_fn = era_is_congested;
1520 dm_table_add_target_callbacks(ti->table, &era->callbacks);
1521
1522 return 0;
1523}
1524
1525static void era_dtr(struct dm_target *ti)
1526{
1527 era_destroy(ti->private);
1528}
1529
1530static int era_map(struct dm_target *ti, struct bio *bio)
1531{
1532 struct era *era = ti->private;
1533 dm_block_t block = get_block(era, bio);
1534
1535 /*
1536 * All bios get remapped to the origin device. We do this now, but
1537 * it may not get issued until later. Depending on whether the
1538 * block is marked in this era.
1539 */
1540 remap_to_origin(era, bio);
1541
1542 /*
1543 * REQ_FLUSH bios carry no data, so we're not interested in them.
1544 */
1545 if (!(bio->bi_rw & REQ_FLUSH) &&
1546 (bio_data_dir(bio) == WRITE) &&
1547 !metadata_current_marked(era->md, block)) {
1548 defer_bio(era, bio);
1549 return DM_MAPIO_SUBMITTED;
1550 }
1551
1552 return DM_MAPIO_REMAPPED;
1553}
1554
1555static void era_postsuspend(struct dm_target *ti)
1556{
1557 int r;
1558 struct era *era = ti->private;
1559
1560 r = in_worker0(era, metadata_era_archive);
1561 if (r) {
1562 DMERR("%s: couldn't archive current era", __func__);
1563 /* FIXME: fail mode */
1564 }
1565
1566 stop_worker(era);
1567}
1568
1569static int era_preresume(struct dm_target *ti)
1570{
1571 int r;
1572 struct era *era = ti->private;
1573 dm_block_t new_size = calc_nr_blocks(era);
1574
1575 if (era->nr_blocks != new_size) {
1576 r = in_worker1(era, metadata_resize, &new_size);
1577 if (r)
1578 return r;
1579
1580 era->nr_blocks = new_size;
1581 }
1582
1583 start_worker(era);
1584
1585 r = in_worker0(era, metadata_new_era);
1586 if (r) {
1587 DMERR("%s: metadata_era_rollover failed", __func__);
1588 return r;
1589 }
1590
1591 return 0;
1592}
1593
1594/*
1595 * Status format:
1596 *
1597 * <metadata block size> <#used metadata blocks>/<#total metadata blocks>
1598 * <current era> <held metadata root | '-'>
1599 */
1600static void era_status(struct dm_target *ti, status_type_t type,
1601 unsigned status_flags, char *result, unsigned maxlen)
1602{
1603 int r;
1604 struct era *era = ti->private;
1605 ssize_t sz = 0;
1606 struct metadata_stats stats;
1607 char buf[BDEVNAME_SIZE];
1608
1609 switch (type) {
1610 case STATUSTYPE_INFO:
1611 r = in_worker1(era, metadata_get_stats, &stats);
1612 if (r)
1613 goto err;
1614
1615 DMEMIT("%u %llu/%llu %u",
1616 (unsigned) (DM_ERA_METADATA_BLOCK_SIZE >> SECTOR_SHIFT),
1617 (unsigned long long) stats.used,
1618 (unsigned long long) stats.total,
1619 (unsigned) stats.era);
1620
1621 if (stats.snap != SUPERBLOCK_LOCATION)
1622 DMEMIT(" %llu", stats.snap);
1623 else
1624 DMEMIT(" -");
1625 break;
1626
1627 case STATUSTYPE_TABLE:
1628 format_dev_t(buf, era->metadata_dev->bdev->bd_dev);
1629 DMEMIT("%s ", buf);
1630 format_dev_t(buf, era->origin_dev->bdev->bd_dev);
1631 DMEMIT("%s %u", buf, era->sectors_per_block);
1632 break;
1633 }
1634
1635 return;
1636
1637err:
1638 DMEMIT("Error");
1639}
1640
1641static int era_message(struct dm_target *ti, unsigned argc, char **argv)
1642{
1643 struct era *era = ti->private;
1644
1645 if (argc != 1) {
1646 DMERR("incorrect number of message arguments");
1647 return -EINVAL;
1648 }
1649
1650 if (!strcasecmp(argv[0], "checkpoint"))
1651 return in_worker0(era, metadata_checkpoint);
1652
1653 if (!strcasecmp(argv[0], "take_metadata_snap"))
1654 return in_worker0(era, metadata_take_snap);
1655
1656 if (!strcasecmp(argv[0], "drop_metadata_snap"))
1657 return in_worker0(era, metadata_drop_snap);
1658
1659 DMERR("unsupported message '%s'", argv[0]);
1660 return -EINVAL;
1661}
1662
1663static sector_t get_dev_size(struct dm_dev *dev)
1664{
1665 return i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT;
1666}
1667
1668static int era_iterate_devices(struct dm_target *ti,
1669 iterate_devices_callout_fn fn, void *data)
1670{
1671 struct era *era = ti->private;
1672 return fn(ti, era->origin_dev, 0, get_dev_size(era->origin_dev), data);
1673}
1674
1675static int era_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
1676 struct bio_vec *biovec, int max_size)
1677{
1678 struct era *era = ti->private;
1679 struct request_queue *q = bdev_get_queue(era->origin_dev->bdev);
1680
1681 if (!q->merge_bvec_fn)
1682 return max_size;
1683
1684 bvm->bi_bdev = era->origin_dev->bdev;
1685
1686 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
1687}
1688
1689static void era_io_hints(struct dm_target *ti, struct queue_limits *limits)
1690{
1691 struct era *era = ti->private;
1692 uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
1693
1694 /*
1695 * If the system-determined stacked limits are compatible with the
1696 * era device's blocksize (io_opt is a factor) do not override them.
1697 */
1698 if (io_opt_sectors < era->sectors_per_block ||
1699 do_div(io_opt_sectors, era->sectors_per_block)) {
1700 blk_limits_io_min(limits, 0);
1701 blk_limits_io_opt(limits, era->sectors_per_block << SECTOR_SHIFT);
1702 }
1703}
1704
1705/*----------------------------------------------------------------*/
1706
1707static struct target_type era_target = {
1708 .name = "era",
1709 .version = {1, 0, 0},
1710 .module = THIS_MODULE,
1711 .ctr = era_ctr,
1712 .dtr = era_dtr,
1713 .map = era_map,
1714 .postsuspend = era_postsuspend,
1715 .preresume = era_preresume,
1716 .status = era_status,
1717 .message = era_message,
1718 .iterate_devices = era_iterate_devices,
1719 .merge = era_merge,
1720 .io_hints = era_io_hints
1721};
1722
1723static int __init dm_era_init(void)
1724{
1725 int r;
1726
1727 r = dm_register_target(&era_target);
1728 if (r) {
1729 DMERR("era target registration failed: %d", r);
1730 return r;
1731 }
1732
1733 return 0;
1734}
1735
1736static void __exit dm_era_exit(void)
1737{
1738 dm_unregister_target(&era_target);
1739}
1740
1741module_init(dm_era_init);
1742module_exit(dm_era_exit);
1743
1744MODULE_DESCRIPTION(DM_NAME " era target");
1745MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>");
1746MODULE_LICENSE("GPL");