xref: /linux/drivers/md/dm-era-target.c (revision 4f3c8320c78cdd11c8fdd23c33787407f719322e)
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  * Writeset
27  *--------------------------------------------------------------*/
28 struct writeset_metadata {
29 	uint32_t nr_bits;
30 	dm_block_t root;
31 };
32 
33 struct writeset {
34 	struct writeset_metadata md;
35 
36 	/*
37 	 * An in core copy of the bits to save constantly doing look ups on
38 	 * disk.
39 	 */
40 	unsigned long *bits;
41 };
42 
43 /*
44  * This does not free off the on disk bitset as this will normally be done
45  * after digesting into the era array.
46  */
47 static void writeset_free(struct writeset *ws)
48 {
49 	vfree(ws->bits);
50 }
51 
52 static int setup_on_disk_bitset(struct dm_disk_bitset *info,
53 				unsigned nr_bits, dm_block_t *root)
54 {
55 	int r;
56 
57 	r = dm_bitset_empty(info, root);
58 	if (r)
59 		return r;
60 
61 	return dm_bitset_resize(info, *root, 0, nr_bits, false, root);
62 }
63 
64 static size_t bitset_size(unsigned nr_bits)
65 {
66 	return sizeof(unsigned long) * dm_div_up(nr_bits, BITS_PER_LONG);
67 }
68 
69 /*
70  * Allocates memory for the in core bitset.
71  */
72 static int writeset_alloc(struct writeset *ws, dm_block_t nr_blocks)
73 {
74 	ws->md.nr_bits = nr_blocks;
75 	ws->md.root = INVALID_WRITESET_ROOT;
76 	ws->bits = vzalloc(bitset_size(nr_blocks));
77 	if (!ws->bits) {
78 		DMERR("%s: couldn't allocate in memory bitset", __func__);
79 		return -ENOMEM;
80 	}
81 
82 	return 0;
83 }
84 
85 /*
86  * Wipes the in-core bitset, and creates a new on disk bitset.
87  */
88 static int writeset_init(struct dm_disk_bitset *info, struct writeset *ws)
89 {
90 	int r;
91 
92 	memset(ws->bits, 0, bitset_size(ws->md.nr_bits));
93 
94 	r = setup_on_disk_bitset(info, ws->md.nr_bits, &ws->md.root);
95 	if (r) {
96 		DMERR("%s: setup_on_disk_bitset failed", __func__);
97 		return r;
98 	}
99 
100 	return 0;
101 }
102 
103 static bool writeset_marked(struct writeset *ws, dm_block_t block)
104 {
105 	return test_bit(block, ws->bits);
106 }
107 
108 static int writeset_marked_on_disk(struct dm_disk_bitset *info,
109 				   struct writeset_metadata *m, dm_block_t block,
110 				   bool *result)
111 {
112 	dm_block_t old = m->root;
113 
114 	/*
115 	 * The bitset was flushed when it was archived, so we know there'll
116 	 * be no change to the root.
117 	 */
118 	int r = dm_bitset_test_bit(info, m->root, block, &m->root, result);
119 	if (r) {
120 		DMERR("%s: dm_bitset_test_bit failed", __func__);
121 		return r;
122 	}
123 
124 	BUG_ON(m->root != old);
125 
126 	return r;
127 }
128 
129 /*
130  * Returns < 0 on error, 0 if the bit wasn't previously set, 1 if it was.
131  */
132 static int writeset_test_and_set(struct dm_disk_bitset *info,
133 				 struct writeset *ws, uint32_t block)
134 {
135 	int r;
136 
137 	if (!test_and_set_bit(block, ws->bits)) {
138 		r = dm_bitset_set_bit(info, ws->md.root, block, &ws->md.root);
139 		if (r) {
140 			/* FIXME: fail mode */
141 			return r;
142 		}
143 
144 		return 0;
145 	}
146 
147 	return 1;
148 }
149 
150 /*----------------------------------------------------------------
151  * On disk metadata layout
152  *--------------------------------------------------------------*/
153 #define SPACE_MAP_ROOT_SIZE 128
154 #define UUID_LEN 16
155 
156 struct writeset_disk {
157 	__le32 nr_bits;
158 	__le64 root;
159 } __packed;
160 
161 struct superblock_disk {
162 	__le32 csum;
163 	__le32 flags;
164 	__le64 blocknr;
165 
166 	__u8 uuid[UUID_LEN];
167 	__le64 magic;
168 	__le32 version;
169 
170 	__u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
171 
172 	__le32 data_block_size;
173 	__le32 metadata_block_size;
174 	__le32 nr_blocks;
175 
176 	__le32 current_era;
177 	struct writeset_disk current_writeset;
178 
179 	/*
180 	 * Only these two fields are valid within the metadata snapshot.
181 	 */
182 	__le64 writeset_tree_root;
183 	__le64 era_array_root;
184 
185 	__le64 metadata_snap;
186 } __packed;
187 
188 /*----------------------------------------------------------------
189  * Superblock validation
190  *--------------------------------------------------------------*/
191 static void sb_prepare_for_write(struct dm_block_validator *v,
192 				 struct dm_block *b,
193 				 size_t sb_block_size)
194 {
195 	struct superblock_disk *disk = dm_block_data(b);
196 
197 	disk->blocknr = cpu_to_le64(dm_block_location(b));
198 	disk->csum = cpu_to_le32(dm_bm_checksum(&disk->flags,
199 						sb_block_size - sizeof(__le32),
200 						SUPERBLOCK_CSUM_XOR));
201 }
202 
203 static int check_metadata_version(struct superblock_disk *disk)
204 {
205 	uint32_t metadata_version = le32_to_cpu(disk->version);
206 	if (metadata_version < MIN_ERA_VERSION || metadata_version > MAX_ERA_VERSION) {
207 		DMERR("Era metadata version %u found, but only versions between %u and %u supported.",
208 		      metadata_version, MIN_ERA_VERSION, MAX_ERA_VERSION);
209 		return -EINVAL;
210 	}
211 
212 	return 0;
213 }
214 
215 static int sb_check(struct dm_block_validator *v,
216 		    struct dm_block *b,
217 		    size_t sb_block_size)
218 {
219 	struct superblock_disk *disk = dm_block_data(b);
220 	__le32 csum_le;
221 
222 	if (dm_block_location(b) != le64_to_cpu(disk->blocknr)) {
223 		DMERR("sb_check failed: blocknr %llu: wanted %llu",
224 		      le64_to_cpu(disk->blocknr),
225 		      (unsigned long long)dm_block_location(b));
226 		return -ENOTBLK;
227 	}
228 
229 	if (le64_to_cpu(disk->magic) != SUPERBLOCK_MAGIC) {
230 		DMERR("sb_check failed: magic %llu: wanted %llu",
231 		      le64_to_cpu(disk->magic),
232 		      (unsigned long long) SUPERBLOCK_MAGIC);
233 		return -EILSEQ;
234 	}
235 
236 	csum_le = cpu_to_le32(dm_bm_checksum(&disk->flags,
237 					     sb_block_size - sizeof(__le32),
238 					     SUPERBLOCK_CSUM_XOR));
239 	if (csum_le != disk->csum) {
240 		DMERR("sb_check failed: csum %u: wanted %u",
241 		      le32_to_cpu(csum_le), le32_to_cpu(disk->csum));
242 		return -EILSEQ;
243 	}
244 
245 	return check_metadata_version(disk);
246 }
247 
248 static struct dm_block_validator sb_validator = {
249 	.name = "superblock",
250 	.prepare_for_write = sb_prepare_for_write,
251 	.check = sb_check
252 };
253 
254 /*----------------------------------------------------------------
255  * Low level metadata handling
256  *--------------------------------------------------------------*/
257 #define DM_ERA_METADATA_BLOCK_SIZE 4096
258 #define ERA_MAX_CONCURRENT_LOCKS 5
259 
260 struct 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 
300 static 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 
307 static 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 
314 static 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 */
322 static 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 	dm_bm_unlock(b);
347 
348 	return 0;
349 }
350 
351 /*----------------------------------------------------------------*/
352 
353 static void ws_pack(const struct writeset_metadata *core, struct writeset_disk *disk)
354 {
355 	disk->nr_bits = cpu_to_le32(core->nr_bits);
356 	disk->root = cpu_to_le64(core->root);
357 }
358 
359 static void ws_unpack(const struct writeset_disk *disk, struct writeset_metadata *core)
360 {
361 	core->nr_bits = le32_to_cpu(disk->nr_bits);
362 	core->root = le64_to_cpu(disk->root);
363 }
364 
365 static void ws_inc(void *context, const void *value)
366 {
367 	struct era_metadata *md = context;
368 	struct writeset_disk ws_d;
369 	dm_block_t b;
370 
371 	memcpy(&ws_d, value, sizeof(ws_d));
372 	b = le64_to_cpu(ws_d.root);
373 
374 	dm_tm_inc(md->tm, b);
375 }
376 
377 static void ws_dec(void *context, const void *value)
378 {
379 	struct era_metadata *md = context;
380 	struct writeset_disk ws_d;
381 	dm_block_t b;
382 
383 	memcpy(&ws_d, value, sizeof(ws_d));
384 	b = le64_to_cpu(ws_d.root);
385 
386 	dm_bitset_del(&md->bitset_info, b);
387 }
388 
389 static int ws_eq(void *context, const void *value1, const void *value2)
390 {
391 	return !memcmp(value1, value2, sizeof(struct writeset_metadata));
392 }
393 
394 /*----------------------------------------------------------------*/
395 
396 static void setup_writeset_tree_info(struct era_metadata *md)
397 {
398 	struct dm_btree_value_type *vt = &md->writeset_tree_info.value_type;
399 	md->writeset_tree_info.tm = md->tm;
400 	md->writeset_tree_info.levels = 1;
401 	vt->context = md;
402 	vt->size = sizeof(struct writeset_disk);
403 	vt->inc = ws_inc;
404 	vt->dec = ws_dec;
405 	vt->equal = ws_eq;
406 }
407 
408 static void setup_era_array_info(struct era_metadata *md)
409 
410 {
411 	struct dm_btree_value_type vt;
412 	vt.context = NULL;
413 	vt.size = sizeof(__le32);
414 	vt.inc = NULL;
415 	vt.dec = NULL;
416 	vt.equal = NULL;
417 
418 	dm_array_info_init(&md->era_array_info, md->tm, &vt);
419 }
420 
421 static void setup_infos(struct era_metadata *md)
422 {
423 	dm_disk_bitset_init(md->tm, &md->bitset_info);
424 	setup_writeset_tree_info(md);
425 	setup_era_array_info(md);
426 }
427 
428 /*----------------------------------------------------------------*/
429 
430 static int create_fresh_metadata(struct era_metadata *md)
431 {
432 	int r;
433 
434 	r = dm_tm_create_with_sm(md->bm, SUPERBLOCK_LOCATION,
435 				 &md->tm, &md->sm);
436 	if (r < 0) {
437 		DMERR("dm_tm_create_with_sm failed");
438 		return r;
439 	}
440 
441 	setup_infos(md);
442 
443 	r = dm_btree_empty(&md->writeset_tree_info, &md->writeset_tree_root);
444 	if (r) {
445 		DMERR("couldn't create new writeset tree");
446 		goto bad;
447 	}
448 
449 	r = dm_array_empty(&md->era_array_info, &md->era_array_root);
450 	if (r) {
451 		DMERR("couldn't create era array");
452 		goto bad;
453 	}
454 
455 	return 0;
456 
457 bad:
458 	dm_sm_destroy(md->sm);
459 	dm_tm_destroy(md->tm);
460 
461 	return r;
462 }
463 
464 static int save_sm_root(struct era_metadata *md)
465 {
466 	int r;
467 	size_t metadata_len;
468 
469 	r = dm_sm_root_size(md->sm, &metadata_len);
470 	if (r < 0)
471 		return r;
472 
473 	return dm_sm_copy_root(md->sm, &md->metadata_space_map_root,
474 			       metadata_len);
475 }
476 
477 static void copy_sm_root(struct era_metadata *md, struct superblock_disk *disk)
478 {
479 	memcpy(&disk->metadata_space_map_root,
480 	       &md->metadata_space_map_root,
481 	       sizeof(md->metadata_space_map_root));
482 }
483 
484 /*
485  * Writes a superblock, including the static fields that don't get updated
486  * with every commit (possible optimisation here).  'md' should be fully
487  * constructed when this is called.
488  */
489 static void prepare_superblock(struct era_metadata *md, struct superblock_disk *disk)
490 {
491 	disk->magic = cpu_to_le64(SUPERBLOCK_MAGIC);
492 	disk->flags = cpu_to_le32(0ul);
493 
494 	/* FIXME: can't keep blanking the uuid (uuid is currently unused though) */
495 	memset(disk->uuid, 0, sizeof(disk->uuid));
496 	disk->version = cpu_to_le32(MAX_ERA_VERSION);
497 
498 	copy_sm_root(md, disk);
499 
500 	disk->data_block_size = cpu_to_le32(md->block_size);
501 	disk->metadata_block_size = cpu_to_le32(DM_ERA_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
502 	disk->nr_blocks = cpu_to_le32(md->nr_blocks);
503 	disk->current_era = cpu_to_le32(md->current_era);
504 
505 	ws_pack(&md->current_writeset->md, &disk->current_writeset);
506 	disk->writeset_tree_root = cpu_to_le64(md->writeset_tree_root);
507 	disk->era_array_root = cpu_to_le64(md->era_array_root);
508 	disk->metadata_snap = cpu_to_le64(md->metadata_snap);
509 }
510 
511 static int write_superblock(struct era_metadata *md)
512 {
513 	int r;
514 	struct dm_block *sblock;
515 	struct superblock_disk *disk;
516 
517 	r = save_sm_root(md);
518 	if (r) {
519 		DMERR("%s: save_sm_root failed", __func__);
520 		return r;
521 	}
522 
523 	r = superblock_lock_zero(md, &sblock);
524 	if (r)
525 		return r;
526 
527 	disk = dm_block_data(sblock);
528 	prepare_superblock(md, disk);
529 
530 	return dm_tm_commit(md->tm, sblock);
531 }
532 
533 /*
534  * Assumes block_size and the infos are set.
535  */
536 static int format_metadata(struct era_metadata *md)
537 {
538 	int r;
539 
540 	r = create_fresh_metadata(md);
541 	if (r)
542 		return r;
543 
544 	r = write_superblock(md);
545 	if (r) {
546 		dm_sm_destroy(md->sm);
547 		dm_tm_destroy(md->tm);
548 		return r;
549 	}
550 
551 	return 0;
552 }
553 
554 static int open_metadata(struct era_metadata *md)
555 {
556 	int r;
557 	struct dm_block *sblock;
558 	struct superblock_disk *disk;
559 
560 	r = superblock_read_lock(md, &sblock);
561 	if (r) {
562 		DMERR("couldn't read_lock superblock");
563 		return r;
564 	}
565 
566 	disk = dm_block_data(sblock);
567 	r = dm_tm_open_with_sm(md->bm, SUPERBLOCK_LOCATION,
568 			       disk->metadata_space_map_root,
569 			       sizeof(disk->metadata_space_map_root),
570 			       &md->tm, &md->sm);
571 	if (r) {
572 		DMERR("dm_tm_open_with_sm failed");
573 		goto bad;
574 	}
575 
576 	setup_infos(md);
577 
578 	md->block_size = le32_to_cpu(disk->data_block_size);
579 	md->nr_blocks = le32_to_cpu(disk->nr_blocks);
580 	md->current_era = le32_to_cpu(disk->current_era);
581 
582 	md->writeset_tree_root = le64_to_cpu(disk->writeset_tree_root);
583 	md->era_array_root = le64_to_cpu(disk->era_array_root);
584 	md->metadata_snap = le64_to_cpu(disk->metadata_snap);
585 	md->archived_writesets = true;
586 
587 	dm_bm_unlock(sblock);
588 
589 	return 0;
590 
591 bad:
592 	dm_bm_unlock(sblock);
593 	return r;
594 }
595 
596 static int open_or_format_metadata(struct era_metadata *md,
597 				   bool may_format)
598 {
599 	int r;
600 	bool unformatted = false;
601 
602 	r = superblock_all_zeroes(md->bm, &unformatted);
603 	if (r)
604 		return r;
605 
606 	if (unformatted)
607 		return may_format ? format_metadata(md) : -EPERM;
608 
609 	return open_metadata(md);
610 }
611 
612 static int create_persistent_data_objects(struct era_metadata *md,
613 					  bool may_format)
614 {
615 	int r;
616 
617 	md->bm = dm_block_manager_create(md->bdev, DM_ERA_METADATA_BLOCK_SIZE,
618 					 ERA_MAX_CONCURRENT_LOCKS);
619 	if (IS_ERR(md->bm)) {
620 		DMERR("could not create block manager");
621 		return PTR_ERR(md->bm);
622 	}
623 
624 	r = open_or_format_metadata(md, may_format);
625 	if (r)
626 		dm_block_manager_destroy(md->bm);
627 
628 	return r;
629 }
630 
631 static void destroy_persistent_data_objects(struct era_metadata *md)
632 {
633 	dm_sm_destroy(md->sm);
634 	dm_tm_destroy(md->tm);
635 	dm_block_manager_destroy(md->bm);
636 }
637 
638 /*
639  * This waits until all era_map threads have picked up the new filter.
640  */
641 static void swap_writeset(struct era_metadata *md, struct writeset *new_writeset)
642 {
643 	rcu_assign_pointer(md->current_writeset, new_writeset);
644 	synchronize_rcu();
645 }
646 
647 /*----------------------------------------------------------------
648  * Writesets get 'digested' into the main era array.
649  *
650  * We're using a coroutine here so the worker thread can do the digestion,
651  * thus avoiding synchronisation of the metadata.  Digesting a whole
652  * writeset in one go would cause too much latency.
653  *--------------------------------------------------------------*/
654 struct digest {
655 	uint32_t era;
656 	unsigned nr_bits, current_bit;
657 	struct writeset_metadata writeset;
658 	__le32 value;
659 	struct dm_disk_bitset info;
660 
661 	int (*step)(struct era_metadata *, struct digest *);
662 };
663 
664 static int metadata_digest_lookup_writeset(struct era_metadata *md,
665 					   struct digest *d);
666 
667 static int metadata_digest_remove_writeset(struct era_metadata *md,
668 					   struct digest *d)
669 {
670 	int r;
671 	uint64_t key = d->era;
672 
673 	r = dm_btree_remove(&md->writeset_tree_info, md->writeset_tree_root,
674 			    &key, &md->writeset_tree_root);
675 	if (r) {
676 		DMERR("%s: dm_btree_remove failed", __func__);
677 		return r;
678 	}
679 
680 	d->step = metadata_digest_lookup_writeset;
681 	return 0;
682 }
683 
684 #define INSERTS_PER_STEP 100
685 
686 static int metadata_digest_transcribe_writeset(struct era_metadata *md,
687 					       struct digest *d)
688 {
689 	int r;
690 	bool marked;
691 	unsigned b, e = min(d->current_bit + INSERTS_PER_STEP, d->nr_bits);
692 
693 	for (b = d->current_bit; b < e; b++) {
694 		r = writeset_marked_on_disk(&d->info, &d->writeset, b, &marked);
695 		if (r) {
696 			DMERR("%s: writeset_marked_on_disk failed", __func__);
697 			return r;
698 		}
699 
700 		if (!marked)
701 			continue;
702 
703 		__dm_bless_for_disk(&d->value);
704 		r = dm_array_set_value(&md->era_array_info, md->era_array_root,
705 				       b, &d->value, &md->era_array_root);
706 		if (r) {
707 			DMERR("%s: dm_array_set_value failed", __func__);
708 			return r;
709 		}
710 	}
711 
712 	if (b == d->nr_bits)
713 		d->step = metadata_digest_remove_writeset;
714 	else
715 		d->current_bit = b;
716 
717 	return 0;
718 }
719 
720 static int metadata_digest_lookup_writeset(struct era_metadata *md,
721 					   struct digest *d)
722 {
723 	int r;
724 	uint64_t key;
725 	struct writeset_disk disk;
726 
727 	r = dm_btree_find_lowest_key(&md->writeset_tree_info,
728 				     md->writeset_tree_root, &key);
729 	if (r < 0)
730 		return r;
731 
732 	d->era = key;
733 
734 	r = dm_btree_lookup(&md->writeset_tree_info,
735 			    md->writeset_tree_root, &key, &disk);
736 	if (r) {
737 		if (r == -ENODATA) {
738 			d->step = NULL;
739 			return 0;
740 		}
741 
742 		DMERR("%s: dm_btree_lookup failed", __func__);
743 		return r;
744 	}
745 
746 	ws_unpack(&disk, &d->writeset);
747 	d->value = cpu_to_le32(key);
748 
749 	d->nr_bits = min(d->writeset.nr_bits, md->nr_blocks);
750 	d->current_bit = 0;
751 	d->step = metadata_digest_transcribe_writeset;
752 
753 	return 0;
754 }
755 
756 static int metadata_digest_start(struct era_metadata *md, struct digest *d)
757 {
758 	if (d->step)
759 		return 0;
760 
761 	memset(d, 0, sizeof(*d));
762 
763 	/*
764 	 * We initialise another bitset info to avoid any caching side
765 	 * effects with the previous one.
766 	 */
767 	dm_disk_bitset_init(md->tm, &d->info);
768 	d->step = metadata_digest_lookup_writeset;
769 
770 	return 0;
771 }
772 
773 /*----------------------------------------------------------------
774  * High level metadata interface.  Target methods should use these, and not
775  * the lower level ones.
776  *--------------------------------------------------------------*/
777 static struct era_metadata *metadata_open(struct block_device *bdev,
778 					  sector_t block_size,
779 					  bool may_format)
780 {
781 	int r;
782 	struct era_metadata *md = kzalloc(sizeof(*md), GFP_KERNEL);
783 
784 	if (!md)
785 		return NULL;
786 
787 	md->bdev = bdev;
788 	md->block_size = block_size;
789 
790 	md->writesets[0].md.root = INVALID_WRITESET_ROOT;
791 	md->writesets[1].md.root = INVALID_WRITESET_ROOT;
792 	md->current_writeset = &md->writesets[0];
793 
794 	r = create_persistent_data_objects(md, may_format);
795 	if (r) {
796 		kfree(md);
797 		return ERR_PTR(r);
798 	}
799 
800 	return md;
801 }
802 
803 static void metadata_close(struct era_metadata *md)
804 {
805 	destroy_persistent_data_objects(md);
806 	kfree(md);
807 }
808 
809 static bool valid_nr_blocks(dm_block_t n)
810 {
811 	/*
812 	 * dm_bitset restricts us to 2^32.  test_bit & co. restrict us
813 	 * further to 2^31 - 1
814 	 */
815 	return n < (1ull << 31);
816 }
817 
818 static int metadata_resize(struct era_metadata *md, void *arg)
819 {
820 	int r;
821 	dm_block_t *new_size = arg;
822 	__le32 value;
823 
824 	if (!valid_nr_blocks(*new_size)) {
825 		DMERR("Invalid number of origin blocks %llu",
826 		      (unsigned long long) *new_size);
827 		return -EINVAL;
828 	}
829 
830 	writeset_free(&md->writesets[0]);
831 	writeset_free(&md->writesets[1]);
832 
833 	r = writeset_alloc(&md->writesets[0], *new_size);
834 	if (r) {
835 		DMERR("%s: writeset_alloc failed for writeset 0", __func__);
836 		return r;
837 	}
838 
839 	r = writeset_alloc(&md->writesets[1], *new_size);
840 	if (r) {
841 		DMERR("%s: writeset_alloc failed for writeset 1", __func__);
842 		return r;
843 	}
844 
845 	value = cpu_to_le32(0u);
846 	__dm_bless_for_disk(&value);
847 	r = dm_array_resize(&md->era_array_info, md->era_array_root,
848 			    md->nr_blocks, *new_size,
849 			    &value, &md->era_array_root);
850 	if (r) {
851 		DMERR("%s: dm_array_resize failed", __func__);
852 		return r;
853 	}
854 
855 	md->nr_blocks = *new_size;
856 	return 0;
857 }
858 
859 static int metadata_era_archive(struct era_metadata *md)
860 {
861 	int r;
862 	uint64_t keys[1];
863 	struct writeset_disk value;
864 
865 	r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root,
866 			    &md->current_writeset->md.root);
867 	if (r) {
868 		DMERR("%s: dm_bitset_flush failed", __func__);
869 		return r;
870 	}
871 
872 	ws_pack(&md->current_writeset->md, &value);
873 	md->current_writeset->md.root = INVALID_WRITESET_ROOT;
874 
875 	keys[0] = md->current_era;
876 	__dm_bless_for_disk(&value);
877 	r = dm_btree_insert(&md->writeset_tree_info, md->writeset_tree_root,
878 			    keys, &value, &md->writeset_tree_root);
879 	if (r) {
880 		DMERR("%s: couldn't insert writeset into btree", __func__);
881 		/* FIXME: fail mode */
882 		return r;
883 	}
884 
885 	md->archived_writesets = true;
886 
887 	return 0;
888 }
889 
890 static struct writeset *next_writeset(struct era_metadata *md)
891 {
892 	return (md->current_writeset == &md->writesets[0]) ?
893 		&md->writesets[1] : &md->writesets[0];
894 }
895 
896 static int metadata_new_era(struct era_metadata *md)
897 {
898 	int r;
899 	struct writeset *new_writeset = next_writeset(md);
900 
901 	r = writeset_init(&md->bitset_info, new_writeset);
902 	if (r) {
903 		DMERR("%s: writeset_init failed", __func__);
904 		return r;
905 	}
906 
907 	swap_writeset(md, new_writeset);
908 	md->current_era++;
909 
910 	return 0;
911 }
912 
913 static int metadata_era_rollover(struct era_metadata *md)
914 {
915 	int r;
916 
917 	if (md->current_writeset->md.root != INVALID_WRITESET_ROOT) {
918 		r = metadata_era_archive(md);
919 		if (r) {
920 			DMERR("%s: metadata_archive_era failed", __func__);
921 			/* FIXME: fail mode? */
922 			return r;
923 		}
924 	}
925 
926 	r = metadata_new_era(md);
927 	if (r) {
928 		DMERR("%s: new era failed", __func__);
929 		/* FIXME: fail mode */
930 		return r;
931 	}
932 
933 	return 0;
934 }
935 
936 static bool metadata_current_marked(struct era_metadata *md, dm_block_t block)
937 {
938 	bool r;
939 	struct writeset *ws;
940 
941 	rcu_read_lock();
942 	ws = rcu_dereference(md->current_writeset);
943 	r = writeset_marked(ws, block);
944 	rcu_read_unlock();
945 
946 	return r;
947 }
948 
949 static int metadata_commit(struct era_metadata *md)
950 {
951 	int r;
952 	struct dm_block *sblock;
953 
954 	if (md->current_writeset->md.root != SUPERBLOCK_LOCATION) {
955 		r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root,
956 				    &md->current_writeset->md.root);
957 		if (r) {
958 			DMERR("%s: bitset flush failed", __func__);
959 			return r;
960 		}
961 	}
962 
963 	r = dm_tm_pre_commit(md->tm);
964 	if (r) {
965 		DMERR("%s: pre commit failed", __func__);
966 		return r;
967 	}
968 
969 	r = save_sm_root(md);
970 	if (r) {
971 		DMERR("%s: save_sm_root failed", __func__);
972 		return r;
973 	}
974 
975 	r = superblock_lock(md, &sblock);
976 	if (r) {
977 		DMERR("%s: superblock lock failed", __func__);
978 		return r;
979 	}
980 
981 	prepare_superblock(md, dm_block_data(sblock));
982 
983 	return dm_tm_commit(md->tm, sblock);
984 }
985 
986 static int metadata_checkpoint(struct era_metadata *md)
987 {
988 	/*
989 	 * For now we just rollover, but later I want to put a check in to
990 	 * avoid this if the filter is still pretty fresh.
991 	 */
992 	return metadata_era_rollover(md);
993 }
994 
995 /*
996  * Metadata snapshots allow userland to access era data.
997  */
998 static int metadata_take_snap(struct era_metadata *md)
999 {
1000 	int r, inc;
1001 	struct dm_block *clone;
1002 
1003 	if (md->metadata_snap != SUPERBLOCK_LOCATION) {
1004 		DMERR("%s: metadata snapshot already exists", __func__);
1005 		return -EINVAL;
1006 	}
1007 
1008 	r = metadata_era_rollover(md);
1009 	if (r) {
1010 		DMERR("%s: era rollover failed", __func__);
1011 		return r;
1012 	}
1013 
1014 	r = metadata_commit(md);
1015 	if (r) {
1016 		DMERR("%s: pre commit failed", __func__);
1017 		return r;
1018 	}
1019 
1020 	r = dm_sm_inc_block(md->sm, SUPERBLOCK_LOCATION);
1021 	if (r) {
1022 		DMERR("%s: couldn't increment superblock", __func__);
1023 		return r;
1024 	}
1025 
1026 	r = dm_tm_shadow_block(md->tm, SUPERBLOCK_LOCATION,
1027 			       &sb_validator, &clone, &inc);
1028 	if (r) {
1029 		DMERR("%s: couldn't shadow superblock", __func__);
1030 		dm_sm_dec_block(md->sm, SUPERBLOCK_LOCATION);
1031 		return r;
1032 	}
1033 	BUG_ON(!inc);
1034 
1035 	r = dm_sm_inc_block(md->sm, md->writeset_tree_root);
1036 	if (r) {
1037 		DMERR("%s: couldn't inc writeset tree root", __func__);
1038 		dm_tm_unlock(md->tm, clone);
1039 		return r;
1040 	}
1041 
1042 	r = dm_sm_inc_block(md->sm, md->era_array_root);
1043 	if (r) {
1044 		DMERR("%s: couldn't inc era tree root", __func__);
1045 		dm_sm_dec_block(md->sm, md->writeset_tree_root);
1046 		dm_tm_unlock(md->tm, clone);
1047 		return r;
1048 	}
1049 
1050 	md->metadata_snap = dm_block_location(clone);
1051 
1052 	dm_tm_unlock(md->tm, clone);
1053 
1054 	return 0;
1055 }
1056 
1057 static int metadata_drop_snap(struct era_metadata *md)
1058 {
1059 	int r;
1060 	dm_block_t location;
1061 	struct dm_block *clone;
1062 	struct superblock_disk *disk;
1063 
1064 	if (md->metadata_snap == SUPERBLOCK_LOCATION) {
1065 		DMERR("%s: no snap to drop", __func__);
1066 		return -EINVAL;
1067 	}
1068 
1069 	r = dm_tm_read_lock(md->tm, md->metadata_snap, &sb_validator, &clone);
1070 	if (r) {
1071 		DMERR("%s: couldn't read lock superblock clone", __func__);
1072 		return r;
1073 	}
1074 
1075 	/*
1076 	 * Whatever happens now we'll commit with no record of the metadata
1077 	 * snap.
1078 	 */
1079 	md->metadata_snap = SUPERBLOCK_LOCATION;
1080 
1081 	disk = dm_block_data(clone);
1082 	r = dm_btree_del(&md->writeset_tree_info,
1083 			 le64_to_cpu(disk->writeset_tree_root));
1084 	if (r) {
1085 		DMERR("%s: error deleting writeset tree clone", __func__);
1086 		dm_tm_unlock(md->tm, clone);
1087 		return r;
1088 	}
1089 
1090 	r = dm_array_del(&md->era_array_info, le64_to_cpu(disk->era_array_root));
1091 	if (r) {
1092 		DMERR("%s: error deleting era array clone", __func__);
1093 		dm_tm_unlock(md->tm, clone);
1094 		return r;
1095 	}
1096 
1097 	location = dm_block_location(clone);
1098 	dm_tm_unlock(md->tm, clone);
1099 
1100 	return dm_sm_dec_block(md->sm, location);
1101 }
1102 
1103 struct metadata_stats {
1104 	dm_block_t used;
1105 	dm_block_t total;
1106 	dm_block_t snap;
1107 	uint32_t era;
1108 };
1109 
1110 static int metadata_get_stats(struct era_metadata *md, void *ptr)
1111 {
1112 	int r;
1113 	struct metadata_stats *s = ptr;
1114 	dm_block_t nr_free, nr_total;
1115 
1116 	r = dm_sm_get_nr_free(md->sm, &nr_free);
1117 	if (r) {
1118 		DMERR("dm_sm_get_nr_free returned %d", r);
1119 		return r;
1120 	}
1121 
1122 	r = dm_sm_get_nr_blocks(md->sm, &nr_total);
1123 	if (r) {
1124 		DMERR("dm_pool_get_metadata_dev_size returned %d", r);
1125 		return r;
1126 	}
1127 
1128 	s->used = nr_total - nr_free;
1129 	s->total = nr_total;
1130 	s->snap = md->metadata_snap;
1131 	s->era = md->current_era;
1132 
1133 	return 0;
1134 }
1135 
1136 /*----------------------------------------------------------------*/
1137 
1138 struct era {
1139 	struct dm_target *ti;
1140 
1141 	struct dm_dev *metadata_dev;
1142 	struct dm_dev *origin_dev;
1143 
1144 	dm_block_t nr_blocks;
1145 	uint32_t sectors_per_block;
1146 	int sectors_per_block_shift;
1147 	struct era_metadata *md;
1148 
1149 	struct workqueue_struct *wq;
1150 	struct work_struct worker;
1151 
1152 	spinlock_t deferred_lock;
1153 	struct bio_list deferred_bios;
1154 
1155 	spinlock_t rpc_lock;
1156 	struct list_head rpc_calls;
1157 
1158 	struct digest digest;
1159 	atomic_t suspended;
1160 };
1161 
1162 struct rpc {
1163 	struct list_head list;
1164 
1165 	int (*fn0)(struct era_metadata *);
1166 	int (*fn1)(struct era_metadata *, void *);
1167 	void *arg;
1168 	int result;
1169 
1170 	struct completion complete;
1171 };
1172 
1173 /*----------------------------------------------------------------
1174  * Remapping.
1175  *---------------------------------------------------------------*/
1176 static bool block_size_is_power_of_two(struct era *era)
1177 {
1178 	return era->sectors_per_block_shift >= 0;
1179 }
1180 
1181 static dm_block_t get_block(struct era *era, struct bio *bio)
1182 {
1183 	sector_t block_nr = bio->bi_iter.bi_sector;
1184 
1185 	if (!block_size_is_power_of_two(era))
1186 		(void) sector_div(block_nr, era->sectors_per_block);
1187 	else
1188 		block_nr >>= era->sectors_per_block_shift;
1189 
1190 	return block_nr;
1191 }
1192 
1193 static void remap_to_origin(struct era *era, struct bio *bio)
1194 {
1195 	bio_set_dev(bio, era->origin_dev->bdev);
1196 }
1197 
1198 /*----------------------------------------------------------------
1199  * Worker thread
1200  *--------------------------------------------------------------*/
1201 static void wake_worker(struct era *era)
1202 {
1203 	if (!atomic_read(&era->suspended))
1204 		queue_work(era->wq, &era->worker);
1205 }
1206 
1207 static void process_old_eras(struct era *era)
1208 {
1209 	int r;
1210 
1211 	if (!era->digest.step)
1212 		return;
1213 
1214 	r = era->digest.step(era->md, &era->digest);
1215 	if (r < 0) {
1216 		DMERR("%s: digest step failed, stopping digestion", __func__);
1217 		era->digest.step = NULL;
1218 
1219 	} else if (era->digest.step)
1220 		wake_worker(era);
1221 }
1222 
1223 static void process_deferred_bios(struct era *era)
1224 {
1225 	int r;
1226 	struct bio_list deferred_bios, marked_bios;
1227 	struct bio *bio;
1228 	bool commit_needed = false;
1229 	bool failed = false;
1230 
1231 	bio_list_init(&deferred_bios);
1232 	bio_list_init(&marked_bios);
1233 
1234 	spin_lock(&era->deferred_lock);
1235 	bio_list_merge(&deferred_bios, &era->deferred_bios);
1236 	bio_list_init(&era->deferred_bios);
1237 	spin_unlock(&era->deferred_lock);
1238 
1239 	while ((bio = bio_list_pop(&deferred_bios))) {
1240 		r = writeset_test_and_set(&era->md->bitset_info,
1241 					  era->md->current_writeset,
1242 					  get_block(era, bio));
1243 		if (r < 0) {
1244 			/*
1245 			 * This is bad news, we need to rollback.
1246 			 * FIXME: finish.
1247 			 */
1248 			failed = true;
1249 
1250 		} else if (r == 0)
1251 			commit_needed = true;
1252 
1253 		bio_list_add(&marked_bios, bio);
1254 	}
1255 
1256 	if (commit_needed) {
1257 		r = metadata_commit(era->md);
1258 		if (r)
1259 			failed = true;
1260 	}
1261 
1262 	if (failed)
1263 		while ((bio = bio_list_pop(&marked_bios)))
1264 			bio_io_error(bio);
1265 	else
1266 		while ((bio = bio_list_pop(&marked_bios)))
1267 			submit_bio_noacct(bio);
1268 }
1269 
1270 static void process_rpc_calls(struct era *era)
1271 {
1272 	int r;
1273 	bool need_commit = false;
1274 	struct list_head calls;
1275 	struct rpc *rpc, *tmp;
1276 
1277 	INIT_LIST_HEAD(&calls);
1278 	spin_lock(&era->rpc_lock);
1279 	list_splice_init(&era->rpc_calls, &calls);
1280 	spin_unlock(&era->rpc_lock);
1281 
1282 	list_for_each_entry_safe(rpc, tmp, &calls, list) {
1283 		rpc->result = rpc->fn0 ? rpc->fn0(era->md) : rpc->fn1(era->md, rpc->arg);
1284 		need_commit = true;
1285 	}
1286 
1287 	if (need_commit) {
1288 		r = metadata_commit(era->md);
1289 		if (r)
1290 			list_for_each_entry_safe(rpc, tmp, &calls, list)
1291 				rpc->result = r;
1292 	}
1293 
1294 	list_for_each_entry_safe(rpc, tmp, &calls, list)
1295 		complete(&rpc->complete);
1296 }
1297 
1298 static void kick_off_digest(struct era *era)
1299 {
1300 	if (era->md->archived_writesets) {
1301 		era->md->archived_writesets = false;
1302 		metadata_digest_start(era->md, &era->digest);
1303 	}
1304 }
1305 
1306 static void do_work(struct work_struct *ws)
1307 {
1308 	struct era *era = container_of(ws, struct era, worker);
1309 
1310 	kick_off_digest(era);
1311 	process_old_eras(era);
1312 	process_deferred_bios(era);
1313 	process_rpc_calls(era);
1314 }
1315 
1316 static void defer_bio(struct era *era, struct bio *bio)
1317 {
1318 	spin_lock(&era->deferred_lock);
1319 	bio_list_add(&era->deferred_bios, bio);
1320 	spin_unlock(&era->deferred_lock);
1321 
1322 	wake_worker(era);
1323 }
1324 
1325 /*
1326  * Make an rpc call to the worker to change the metadata.
1327  */
1328 static int perform_rpc(struct era *era, struct rpc *rpc)
1329 {
1330 	rpc->result = 0;
1331 	init_completion(&rpc->complete);
1332 
1333 	spin_lock(&era->rpc_lock);
1334 	list_add(&rpc->list, &era->rpc_calls);
1335 	spin_unlock(&era->rpc_lock);
1336 
1337 	wake_worker(era);
1338 	wait_for_completion(&rpc->complete);
1339 
1340 	return rpc->result;
1341 }
1342 
1343 static int in_worker0(struct era *era, int (*fn)(struct era_metadata *))
1344 {
1345 	struct rpc rpc;
1346 	rpc.fn0 = fn;
1347 	rpc.fn1 = NULL;
1348 
1349 	return perform_rpc(era, &rpc);
1350 }
1351 
1352 static int in_worker1(struct era *era,
1353 		      int (*fn)(struct era_metadata *, void *), void *arg)
1354 {
1355 	struct rpc rpc;
1356 	rpc.fn0 = NULL;
1357 	rpc.fn1 = fn;
1358 	rpc.arg = arg;
1359 
1360 	return perform_rpc(era, &rpc);
1361 }
1362 
1363 static void start_worker(struct era *era)
1364 {
1365 	atomic_set(&era->suspended, 0);
1366 }
1367 
1368 static void stop_worker(struct era *era)
1369 {
1370 	atomic_set(&era->suspended, 1);
1371 	flush_workqueue(era->wq);
1372 }
1373 
1374 /*----------------------------------------------------------------
1375  * Target methods
1376  *--------------------------------------------------------------*/
1377 static void era_destroy(struct era *era)
1378 {
1379 	if (era->md)
1380 		metadata_close(era->md);
1381 
1382 	if (era->wq)
1383 		destroy_workqueue(era->wq);
1384 
1385 	if (era->origin_dev)
1386 		dm_put_device(era->ti, era->origin_dev);
1387 
1388 	if (era->metadata_dev)
1389 		dm_put_device(era->ti, era->metadata_dev);
1390 
1391 	kfree(era);
1392 }
1393 
1394 static dm_block_t calc_nr_blocks(struct era *era)
1395 {
1396 	return dm_sector_div_up(era->ti->len, era->sectors_per_block);
1397 }
1398 
1399 static bool valid_block_size(dm_block_t block_size)
1400 {
1401 	bool greater_than_zero = block_size > 0;
1402 	bool multiple_of_min_block_size = (block_size & (MIN_BLOCK_SIZE - 1)) == 0;
1403 
1404 	return greater_than_zero && multiple_of_min_block_size;
1405 }
1406 
1407 /*
1408  * <metadata dev> <data dev> <data block size (sectors)>
1409  */
1410 static int era_ctr(struct dm_target *ti, unsigned argc, char **argv)
1411 {
1412 	int r;
1413 	char dummy;
1414 	struct era *era;
1415 	struct era_metadata *md;
1416 
1417 	if (argc != 3) {
1418 		ti->error = "Invalid argument count";
1419 		return -EINVAL;
1420 	}
1421 
1422 	era = kzalloc(sizeof(*era), GFP_KERNEL);
1423 	if (!era) {
1424 		ti->error = "Error allocating era structure";
1425 		return -ENOMEM;
1426 	}
1427 
1428 	era->ti = ti;
1429 
1430 	r = dm_get_device(ti, argv[0], FMODE_READ | FMODE_WRITE, &era->metadata_dev);
1431 	if (r) {
1432 		ti->error = "Error opening metadata device";
1433 		era_destroy(era);
1434 		return -EINVAL;
1435 	}
1436 
1437 	r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &era->origin_dev);
1438 	if (r) {
1439 		ti->error = "Error opening data device";
1440 		era_destroy(era);
1441 		return -EINVAL;
1442 	}
1443 
1444 	r = sscanf(argv[2], "%u%c", &era->sectors_per_block, &dummy);
1445 	if (r != 1) {
1446 		ti->error = "Error parsing block size";
1447 		era_destroy(era);
1448 		return -EINVAL;
1449 	}
1450 
1451 	r = dm_set_target_max_io_len(ti, era->sectors_per_block);
1452 	if (r) {
1453 		ti->error = "could not set max io len";
1454 		era_destroy(era);
1455 		return -EINVAL;
1456 	}
1457 
1458 	if (!valid_block_size(era->sectors_per_block)) {
1459 		ti->error = "Invalid block size";
1460 		era_destroy(era);
1461 		return -EINVAL;
1462 	}
1463 	if (era->sectors_per_block & (era->sectors_per_block - 1))
1464 		era->sectors_per_block_shift = -1;
1465 	else
1466 		era->sectors_per_block_shift = __ffs(era->sectors_per_block);
1467 
1468 	md = metadata_open(era->metadata_dev->bdev, era->sectors_per_block, true);
1469 	if (IS_ERR(md)) {
1470 		ti->error = "Error reading metadata";
1471 		era_destroy(era);
1472 		return PTR_ERR(md);
1473 	}
1474 	era->md = md;
1475 
1476 	era->nr_blocks = calc_nr_blocks(era);
1477 
1478 	r = metadata_resize(era->md, &era->nr_blocks);
1479 	if (r) {
1480 		ti->error = "couldn't resize metadata";
1481 		era_destroy(era);
1482 		return -ENOMEM;
1483 	}
1484 
1485 	era->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
1486 	if (!era->wq) {
1487 		ti->error = "could not create workqueue for metadata object";
1488 		era_destroy(era);
1489 		return -ENOMEM;
1490 	}
1491 	INIT_WORK(&era->worker, do_work);
1492 
1493 	spin_lock_init(&era->deferred_lock);
1494 	bio_list_init(&era->deferred_bios);
1495 
1496 	spin_lock_init(&era->rpc_lock);
1497 	INIT_LIST_HEAD(&era->rpc_calls);
1498 
1499 	ti->private = era;
1500 	ti->num_flush_bios = 1;
1501 	ti->flush_supported = true;
1502 
1503 	ti->num_discard_bios = 1;
1504 
1505 	return 0;
1506 }
1507 
1508 static void era_dtr(struct dm_target *ti)
1509 {
1510 	era_destroy(ti->private);
1511 }
1512 
1513 static int era_map(struct dm_target *ti, struct bio *bio)
1514 {
1515 	struct era *era = ti->private;
1516 	dm_block_t block = get_block(era, bio);
1517 
1518 	/*
1519 	 * All bios get remapped to the origin device.  We do this now, but
1520 	 * it may not get issued until later.  Depending on whether the
1521 	 * block is marked in this era.
1522 	 */
1523 	remap_to_origin(era, bio);
1524 
1525 	/*
1526 	 * REQ_PREFLUSH bios carry no data, so we're not interested in them.
1527 	 */
1528 	if (!(bio->bi_opf & REQ_PREFLUSH) &&
1529 	    (bio_data_dir(bio) == WRITE) &&
1530 	    !metadata_current_marked(era->md, block)) {
1531 		defer_bio(era, bio);
1532 		return DM_MAPIO_SUBMITTED;
1533 	}
1534 
1535 	return DM_MAPIO_REMAPPED;
1536 }
1537 
1538 static void era_postsuspend(struct dm_target *ti)
1539 {
1540 	int r;
1541 	struct era *era = ti->private;
1542 
1543 	r = in_worker0(era, metadata_era_archive);
1544 	if (r) {
1545 		DMERR("%s: couldn't archive current era", __func__);
1546 		/* FIXME: fail mode */
1547 	}
1548 
1549 	stop_worker(era);
1550 }
1551 
1552 static int era_preresume(struct dm_target *ti)
1553 {
1554 	int r;
1555 	struct era *era = ti->private;
1556 	dm_block_t new_size = calc_nr_blocks(era);
1557 
1558 	if (era->nr_blocks != new_size) {
1559 		r = in_worker1(era, metadata_resize, &new_size);
1560 		if (r)
1561 			return r;
1562 
1563 		era->nr_blocks = new_size;
1564 	}
1565 
1566 	start_worker(era);
1567 
1568 	r = in_worker0(era, metadata_new_era);
1569 	if (r) {
1570 		DMERR("%s: metadata_era_rollover failed", __func__);
1571 		return r;
1572 	}
1573 
1574 	return 0;
1575 }
1576 
1577 /*
1578  * Status format:
1579  *
1580  * <metadata block size> <#used metadata blocks>/<#total metadata blocks>
1581  * <current era> <held metadata root | '-'>
1582  */
1583 static void era_status(struct dm_target *ti, status_type_t type,
1584 		       unsigned status_flags, char *result, unsigned maxlen)
1585 {
1586 	int r;
1587 	struct era *era = ti->private;
1588 	ssize_t sz = 0;
1589 	struct metadata_stats stats;
1590 	char buf[BDEVNAME_SIZE];
1591 
1592 	switch (type) {
1593 	case STATUSTYPE_INFO:
1594 		r = in_worker1(era, metadata_get_stats, &stats);
1595 		if (r)
1596 			goto err;
1597 
1598 		DMEMIT("%u %llu/%llu %u",
1599 		       (unsigned) (DM_ERA_METADATA_BLOCK_SIZE >> SECTOR_SHIFT),
1600 		       (unsigned long long) stats.used,
1601 		       (unsigned long long) stats.total,
1602 		       (unsigned) stats.era);
1603 
1604 		if (stats.snap != SUPERBLOCK_LOCATION)
1605 			DMEMIT(" %llu", stats.snap);
1606 		else
1607 			DMEMIT(" -");
1608 		break;
1609 
1610 	case STATUSTYPE_TABLE:
1611 		format_dev_t(buf, era->metadata_dev->bdev->bd_dev);
1612 		DMEMIT("%s ", buf);
1613 		format_dev_t(buf, era->origin_dev->bdev->bd_dev);
1614 		DMEMIT("%s %u", buf, era->sectors_per_block);
1615 		break;
1616 	}
1617 
1618 	return;
1619 
1620 err:
1621 	DMEMIT("Error");
1622 }
1623 
1624 static int era_message(struct dm_target *ti, unsigned argc, char **argv,
1625 		       char *result, unsigned maxlen)
1626 {
1627 	struct era *era = ti->private;
1628 
1629 	if (argc != 1) {
1630 		DMERR("incorrect number of message arguments");
1631 		return -EINVAL;
1632 	}
1633 
1634 	if (!strcasecmp(argv[0], "checkpoint"))
1635 		return in_worker0(era, metadata_checkpoint);
1636 
1637 	if (!strcasecmp(argv[0], "take_metadata_snap"))
1638 		return in_worker0(era, metadata_take_snap);
1639 
1640 	if (!strcasecmp(argv[0], "drop_metadata_snap"))
1641 		return in_worker0(era, metadata_drop_snap);
1642 
1643 	DMERR("unsupported message '%s'", argv[0]);
1644 	return -EINVAL;
1645 }
1646 
1647 static sector_t get_dev_size(struct dm_dev *dev)
1648 {
1649 	return i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT;
1650 }
1651 
1652 static int era_iterate_devices(struct dm_target *ti,
1653 			       iterate_devices_callout_fn fn, void *data)
1654 {
1655 	struct era *era = ti->private;
1656 	return fn(ti, era->origin_dev, 0, get_dev_size(era->origin_dev), data);
1657 }
1658 
1659 static void era_io_hints(struct dm_target *ti, struct queue_limits *limits)
1660 {
1661 	struct era *era = ti->private;
1662 	uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
1663 
1664 	/*
1665 	 * If the system-determined stacked limits are compatible with the
1666 	 * era device's blocksize (io_opt is a factor) do not override them.
1667 	 */
1668 	if (io_opt_sectors < era->sectors_per_block ||
1669 	    do_div(io_opt_sectors, era->sectors_per_block)) {
1670 		blk_limits_io_min(limits, 0);
1671 		blk_limits_io_opt(limits, era->sectors_per_block << SECTOR_SHIFT);
1672 	}
1673 }
1674 
1675 /*----------------------------------------------------------------*/
1676 
1677 static struct target_type era_target = {
1678 	.name = "era",
1679 	.version = {1, 0, 0},
1680 	.module = THIS_MODULE,
1681 	.ctr = era_ctr,
1682 	.dtr = era_dtr,
1683 	.map = era_map,
1684 	.postsuspend = era_postsuspend,
1685 	.preresume = era_preresume,
1686 	.status = era_status,
1687 	.message = era_message,
1688 	.iterate_devices = era_iterate_devices,
1689 	.io_hints = era_io_hints
1690 };
1691 
1692 static int __init dm_era_init(void)
1693 {
1694 	int r;
1695 
1696 	r = dm_register_target(&era_target);
1697 	if (r) {
1698 		DMERR("era target registration failed: %d", r);
1699 		return r;
1700 	}
1701 
1702 	return 0;
1703 }
1704 
1705 static void __exit dm_era_exit(void)
1706 {
1707 	dm_unregister_target(&era_target);
1708 }
1709 
1710 module_init(dm_era_init);
1711 module_exit(dm_era_exit);
1712 
1713 MODULE_DESCRIPTION(DM_NAME " era target");
1714 MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>");
1715 MODULE_LICENSE("GPL");
1716