xref: /openbmc/linux/drivers/md/dm-era-target.c (revision dea54fba)
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  *--------------------------------------------------------------*/
27 struct writeset_metadata {
28 	uint32_t nr_bits;
29 	dm_block_t root;
30 };
31 
32 struct 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  */
46 static void writeset_free(struct writeset *ws)
47 {
48 	vfree(ws->bits);
49 }
50 
51 static 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 
63 static 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  */
71 static 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  */
87 static 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 
102 static bool writeset_marked(struct writeset *ws, dm_block_t block)
103 {
104 	return test_bit(block, ws->bits);
105 }
106 
107 static 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  */
131 static 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 
155 struct writeset_disk {
156 	__le32 nr_bits;
157 	__le64 root;
158 } __packed;
159 
160 struct 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  *--------------------------------------------------------------*/
190 static 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 
202 static 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 
214 static 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 
247 static 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 ERA_MAX_CONCURRENT_LOCKS 5
258 
259 struct era_metadata {
260 	struct block_device *bdev;
261 	struct dm_block_manager *bm;
262 	struct dm_space_map *sm;
263 	struct dm_transaction_manager *tm;
264 
265 	dm_block_t block_size;
266 	uint32_t nr_blocks;
267 
268 	uint32_t current_era;
269 
270 	/*
271 	 * We preallocate 2 writesets.  When an era rolls over we
272 	 * switch between them. This means the allocation is done at
273 	 * preresume time, rather than on the io path.
274 	 */
275 	struct writeset writesets[2];
276 	struct writeset *current_writeset;
277 
278 	dm_block_t writeset_tree_root;
279 	dm_block_t era_array_root;
280 
281 	struct dm_disk_bitset bitset_info;
282 	struct dm_btree_info writeset_tree_info;
283 	struct dm_array_info era_array_info;
284 
285 	dm_block_t metadata_snap;
286 
287 	/*
288 	 * A flag that is set whenever a writeset has been archived.
289 	 */
290 	bool archived_writesets;
291 
292 	/*
293 	 * Reading the space map root can fail, so we read it into this
294 	 * buffer before the superblock is locked and updated.
295 	 */
296 	__u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
297 };
298 
299 static int superblock_read_lock(struct era_metadata *md,
300 				struct dm_block **sblock)
301 {
302 	return dm_bm_read_lock(md->bm, SUPERBLOCK_LOCATION,
303 			       &sb_validator, sblock);
304 }
305 
306 static int superblock_lock_zero(struct era_metadata *md,
307 				struct dm_block **sblock)
308 {
309 	return dm_bm_write_lock_zero(md->bm, SUPERBLOCK_LOCATION,
310 				     &sb_validator, sblock);
311 }
312 
313 static int superblock_lock(struct era_metadata *md,
314 			   struct dm_block **sblock)
315 {
316 	return dm_bm_write_lock(md->bm, SUPERBLOCK_LOCATION,
317 				&sb_validator, sblock);
318 }
319 
320 /* FIXME: duplication with cache and thin */
321 static int superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
322 {
323 	int r;
324 	unsigned i;
325 	struct dm_block *b;
326 	__le64 *data_le, zero = cpu_to_le64(0);
327 	unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
328 
329 	/*
330 	 * We can't use a validator here - it may be all zeroes.
331 	 */
332 	r = dm_bm_read_lock(bm, SUPERBLOCK_LOCATION, NULL, &b);
333 	if (r)
334 		return r;
335 
336 	data_le = dm_block_data(b);
337 	*result = true;
338 	for (i = 0; i < sb_block_size; i++) {
339 		if (data_le[i] != zero) {
340 			*result = false;
341 			break;
342 		}
343 	}
344 
345 	dm_bm_unlock(b);
346 
347 	return 0;
348 }
349 
350 /*----------------------------------------------------------------*/
351 
352 static void ws_pack(const struct writeset_metadata *core, struct writeset_disk *disk)
353 {
354 	disk->nr_bits = cpu_to_le32(core->nr_bits);
355 	disk->root = cpu_to_le64(core->root);
356 }
357 
358 static void ws_unpack(const struct writeset_disk *disk, struct writeset_metadata *core)
359 {
360 	core->nr_bits = le32_to_cpu(disk->nr_bits);
361 	core->root = le64_to_cpu(disk->root);
362 }
363 
364 static void ws_inc(void *context, const void *value)
365 {
366 	struct era_metadata *md = context;
367 	struct writeset_disk ws_d;
368 	dm_block_t b;
369 
370 	memcpy(&ws_d, value, sizeof(ws_d));
371 	b = le64_to_cpu(ws_d.root);
372 
373 	dm_tm_inc(md->tm, b);
374 }
375 
376 static void ws_dec(void *context, const void *value)
377 {
378 	struct era_metadata *md = context;
379 	struct writeset_disk ws_d;
380 	dm_block_t b;
381 
382 	memcpy(&ws_d, value, sizeof(ws_d));
383 	b = le64_to_cpu(ws_d.root);
384 
385 	dm_bitset_del(&md->bitset_info, b);
386 }
387 
388 static int ws_eq(void *context, const void *value1, const void *value2)
389 {
390 	return !memcmp(value1, value2, sizeof(struct writeset_metadata));
391 }
392 
393 /*----------------------------------------------------------------*/
394 
395 static void setup_writeset_tree_info(struct era_metadata *md)
396 {
397 	struct dm_btree_value_type *vt = &md->writeset_tree_info.value_type;
398 	md->writeset_tree_info.tm = md->tm;
399 	md->writeset_tree_info.levels = 1;
400 	vt->context = md;
401 	vt->size = sizeof(struct writeset_disk);
402 	vt->inc = ws_inc;
403 	vt->dec = ws_dec;
404 	vt->equal = ws_eq;
405 }
406 
407 static void setup_era_array_info(struct era_metadata *md)
408 
409 {
410 	struct dm_btree_value_type vt;
411 	vt.context = NULL;
412 	vt.size = sizeof(__le32);
413 	vt.inc = NULL;
414 	vt.dec = NULL;
415 	vt.equal = NULL;
416 
417 	dm_array_info_init(&md->era_array_info, md->tm, &vt);
418 }
419 
420 static void setup_infos(struct era_metadata *md)
421 {
422 	dm_disk_bitset_init(md->tm, &md->bitset_info);
423 	setup_writeset_tree_info(md);
424 	setup_era_array_info(md);
425 }
426 
427 /*----------------------------------------------------------------*/
428 
429 static int create_fresh_metadata(struct era_metadata *md)
430 {
431 	int r;
432 
433 	r = dm_tm_create_with_sm(md->bm, SUPERBLOCK_LOCATION,
434 				 &md->tm, &md->sm);
435 	if (r < 0) {
436 		DMERR("dm_tm_create_with_sm failed");
437 		return r;
438 	}
439 
440 	setup_infos(md);
441 
442 	r = dm_btree_empty(&md->writeset_tree_info, &md->writeset_tree_root);
443 	if (r) {
444 		DMERR("couldn't create new writeset tree");
445 		goto bad;
446 	}
447 
448 	r = dm_array_empty(&md->era_array_info, &md->era_array_root);
449 	if (r) {
450 		DMERR("couldn't create era array");
451 		goto bad;
452 	}
453 
454 	return 0;
455 
456 bad:
457 	dm_sm_destroy(md->sm);
458 	dm_tm_destroy(md->tm);
459 
460 	return r;
461 }
462 
463 static int save_sm_root(struct era_metadata *md)
464 {
465 	int r;
466 	size_t metadata_len;
467 
468 	r = dm_sm_root_size(md->sm, &metadata_len);
469 	if (r < 0)
470 		return r;
471 
472 	return dm_sm_copy_root(md->sm, &md->metadata_space_map_root,
473 			       metadata_len);
474 }
475 
476 static void copy_sm_root(struct era_metadata *md, struct superblock_disk *disk)
477 {
478 	memcpy(&disk->metadata_space_map_root,
479 	       &md->metadata_space_map_root,
480 	       sizeof(md->metadata_space_map_root));
481 }
482 
483 /*
484  * Writes a superblock, including the static fields that don't get updated
485  * with every commit (possible optimisation here).  'md' should be fully
486  * constructed when this is called.
487  */
488 static void prepare_superblock(struct era_metadata *md, struct superblock_disk *disk)
489 {
490 	disk->magic = cpu_to_le64(SUPERBLOCK_MAGIC);
491 	disk->flags = cpu_to_le32(0ul);
492 
493 	/* FIXME: can't keep blanking the uuid (uuid is currently unused though) */
494 	memset(disk->uuid, 0, sizeof(disk->uuid));
495 	disk->version = cpu_to_le32(MAX_ERA_VERSION);
496 
497 	copy_sm_root(md, disk);
498 
499 	disk->data_block_size = cpu_to_le32(md->block_size);
500 	disk->metadata_block_size = cpu_to_le32(DM_ERA_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
501 	disk->nr_blocks = cpu_to_le32(md->nr_blocks);
502 	disk->current_era = cpu_to_le32(md->current_era);
503 
504 	ws_pack(&md->current_writeset->md, &disk->current_writeset);
505 	disk->writeset_tree_root = cpu_to_le64(md->writeset_tree_root);
506 	disk->era_array_root = cpu_to_le64(md->era_array_root);
507 	disk->metadata_snap = cpu_to_le64(md->metadata_snap);
508 }
509 
510 static int write_superblock(struct era_metadata *md)
511 {
512 	int r;
513 	struct dm_block *sblock;
514 	struct superblock_disk *disk;
515 
516 	r = save_sm_root(md);
517 	if (r) {
518 		DMERR("%s: save_sm_root failed", __func__);
519 		return r;
520 	}
521 
522 	r = superblock_lock_zero(md, &sblock);
523 	if (r)
524 		return r;
525 
526 	disk = dm_block_data(sblock);
527 	prepare_superblock(md, disk);
528 
529 	return dm_tm_commit(md->tm, sblock);
530 }
531 
532 /*
533  * Assumes block_size and the infos are set.
534  */
535 static int format_metadata(struct era_metadata *md)
536 {
537 	int r;
538 
539 	r = create_fresh_metadata(md);
540 	if (r)
541 		return r;
542 
543 	r = write_superblock(md);
544 	if (r) {
545 		dm_sm_destroy(md->sm);
546 		dm_tm_destroy(md->tm);
547 		return r;
548 	}
549 
550 	return 0;
551 }
552 
553 static int open_metadata(struct era_metadata *md)
554 {
555 	int r;
556 	struct dm_block *sblock;
557 	struct superblock_disk *disk;
558 
559 	r = superblock_read_lock(md, &sblock);
560 	if (r) {
561 		DMERR("couldn't read_lock superblock");
562 		return r;
563 	}
564 
565 	disk = dm_block_data(sblock);
566 	r = dm_tm_open_with_sm(md->bm, SUPERBLOCK_LOCATION,
567 			       disk->metadata_space_map_root,
568 			       sizeof(disk->metadata_space_map_root),
569 			       &md->tm, &md->sm);
570 	if (r) {
571 		DMERR("dm_tm_open_with_sm failed");
572 		goto bad;
573 	}
574 
575 	setup_infos(md);
576 
577 	md->block_size = le32_to_cpu(disk->data_block_size);
578 	md->nr_blocks = le32_to_cpu(disk->nr_blocks);
579 	md->current_era = le32_to_cpu(disk->current_era);
580 
581 	md->writeset_tree_root = le64_to_cpu(disk->writeset_tree_root);
582 	md->era_array_root = le64_to_cpu(disk->era_array_root);
583 	md->metadata_snap = le64_to_cpu(disk->metadata_snap);
584 	md->archived_writesets = true;
585 
586 	dm_bm_unlock(sblock);
587 
588 	return 0;
589 
590 bad:
591 	dm_bm_unlock(sblock);
592 	return r;
593 }
594 
595 static int open_or_format_metadata(struct era_metadata *md,
596 				   bool may_format)
597 {
598 	int r;
599 	bool unformatted = false;
600 
601 	r = superblock_all_zeroes(md->bm, &unformatted);
602 	if (r)
603 		return r;
604 
605 	if (unformatted)
606 		return may_format ? format_metadata(md) : -EPERM;
607 
608 	return open_metadata(md);
609 }
610 
611 static int create_persistent_data_objects(struct era_metadata *md,
612 					  bool may_format)
613 {
614 	int r;
615 
616 	md->bm = dm_block_manager_create(md->bdev, DM_ERA_METADATA_BLOCK_SIZE,
617 					 ERA_MAX_CONCURRENT_LOCKS);
618 	if (IS_ERR(md->bm)) {
619 		DMERR("could not create block manager");
620 		return PTR_ERR(md->bm);
621 	}
622 
623 	r = open_or_format_metadata(md, may_format);
624 	if (r)
625 		dm_block_manager_destroy(md->bm);
626 
627 	return r;
628 }
629 
630 static void destroy_persistent_data_objects(struct era_metadata *md)
631 {
632 	dm_sm_destroy(md->sm);
633 	dm_tm_destroy(md->tm);
634 	dm_block_manager_destroy(md->bm);
635 }
636 
637 /*
638  * This waits until all era_map threads have picked up the new filter.
639  */
640 static void swap_writeset(struct era_metadata *md, struct writeset *new_writeset)
641 {
642 	rcu_assign_pointer(md->current_writeset, new_writeset);
643 	synchronize_rcu();
644 }
645 
646 /*----------------------------------------------------------------
647  * Writesets get 'digested' into the main era array.
648  *
649  * We're using a coroutine here so the worker thread can do the digestion,
650  * thus avoiding synchronisation of the metadata.  Digesting a whole
651  * writeset in one go would cause too much latency.
652  *--------------------------------------------------------------*/
653 struct digest {
654 	uint32_t era;
655 	unsigned nr_bits, current_bit;
656 	struct writeset_metadata writeset;
657 	__le32 value;
658 	struct dm_disk_bitset info;
659 
660 	int (*step)(struct era_metadata *, struct digest *);
661 };
662 
663 static int metadata_digest_lookup_writeset(struct era_metadata *md,
664 					   struct digest *d);
665 
666 static int metadata_digest_remove_writeset(struct era_metadata *md,
667 					   struct digest *d)
668 {
669 	int r;
670 	uint64_t key = d->era;
671 
672 	r = dm_btree_remove(&md->writeset_tree_info, md->writeset_tree_root,
673 			    &key, &md->writeset_tree_root);
674 	if (r) {
675 		DMERR("%s: dm_btree_remove failed", __func__);
676 		return r;
677 	}
678 
679 	d->step = metadata_digest_lookup_writeset;
680 	return 0;
681 }
682 
683 #define INSERTS_PER_STEP 100
684 
685 static int metadata_digest_transcribe_writeset(struct era_metadata *md,
686 					       struct digest *d)
687 {
688 	int r;
689 	bool marked;
690 	unsigned b, e = min(d->current_bit + INSERTS_PER_STEP, d->nr_bits);
691 
692 	for (b = d->current_bit; b < e; b++) {
693 		r = writeset_marked_on_disk(&d->info, &d->writeset, b, &marked);
694 		if (r) {
695 			DMERR("%s: writeset_marked_on_disk failed", __func__);
696 			return r;
697 		}
698 
699 		if (!marked)
700 			continue;
701 
702 		__dm_bless_for_disk(&d->value);
703 		r = dm_array_set_value(&md->era_array_info, md->era_array_root,
704 				       b, &d->value, &md->era_array_root);
705 		if (r) {
706 			DMERR("%s: dm_array_set_value failed", __func__);
707 			return r;
708 		}
709 	}
710 
711 	if (b == d->nr_bits)
712 		d->step = metadata_digest_remove_writeset;
713 	else
714 		d->current_bit = b;
715 
716 	return 0;
717 }
718 
719 static int metadata_digest_lookup_writeset(struct era_metadata *md,
720 					   struct digest *d)
721 {
722 	int r;
723 	uint64_t key;
724 	struct writeset_disk disk;
725 
726 	r = dm_btree_find_lowest_key(&md->writeset_tree_info,
727 				     md->writeset_tree_root, &key);
728 	if (r < 0)
729 		return r;
730 
731 	d->era = key;
732 
733 	r = dm_btree_lookup(&md->writeset_tree_info,
734 			    md->writeset_tree_root, &key, &disk);
735 	if (r) {
736 		if (r == -ENODATA) {
737 			d->step = NULL;
738 			return 0;
739 		}
740 
741 		DMERR("%s: dm_btree_lookup failed", __func__);
742 		return r;
743 	}
744 
745 	ws_unpack(&disk, &d->writeset);
746 	d->value = cpu_to_le32(key);
747 
748 	d->nr_bits = min(d->writeset.nr_bits, md->nr_blocks);
749 	d->current_bit = 0;
750 	d->step = metadata_digest_transcribe_writeset;
751 
752 	return 0;
753 }
754 
755 static int metadata_digest_start(struct era_metadata *md, struct digest *d)
756 {
757 	if (d->step)
758 		return 0;
759 
760 	memset(d, 0, sizeof(*d));
761 
762 	/*
763 	 * We initialise another bitset info to avoid any caching side
764 	 * effects with the previous one.
765 	 */
766 	dm_disk_bitset_init(md->tm, &d->info);
767 	d->step = metadata_digest_lookup_writeset;
768 
769 	return 0;
770 }
771 
772 /*----------------------------------------------------------------
773  * High level metadata interface.  Target methods should use these, and not
774  * the lower level ones.
775  *--------------------------------------------------------------*/
776 static struct era_metadata *metadata_open(struct block_device *bdev,
777 					  sector_t block_size,
778 					  bool may_format)
779 {
780 	int r;
781 	struct era_metadata *md = kzalloc(sizeof(*md), GFP_KERNEL);
782 
783 	if (!md)
784 		return NULL;
785 
786 	md->bdev = bdev;
787 	md->block_size = block_size;
788 
789 	md->writesets[0].md.root = INVALID_WRITESET_ROOT;
790 	md->writesets[1].md.root = INVALID_WRITESET_ROOT;
791 	md->current_writeset = &md->writesets[0];
792 
793 	r = create_persistent_data_objects(md, may_format);
794 	if (r) {
795 		kfree(md);
796 		return ERR_PTR(r);
797 	}
798 
799 	return md;
800 }
801 
802 static void metadata_close(struct era_metadata *md)
803 {
804 	destroy_persistent_data_objects(md);
805 	kfree(md);
806 }
807 
808 static bool valid_nr_blocks(dm_block_t n)
809 {
810 	/*
811 	 * dm_bitset restricts us to 2^32.  test_bit & co. restrict us
812 	 * further to 2^31 - 1
813 	 */
814 	return n < (1ull << 31);
815 }
816 
817 static int metadata_resize(struct era_metadata *md, void *arg)
818 {
819 	int r;
820 	dm_block_t *new_size = arg;
821 	__le32 value;
822 
823 	if (!valid_nr_blocks(*new_size)) {
824 		DMERR("Invalid number of origin blocks %llu",
825 		      (unsigned long long) *new_size);
826 		return -EINVAL;
827 	}
828 
829 	writeset_free(&md->writesets[0]);
830 	writeset_free(&md->writesets[1]);
831 
832 	r = writeset_alloc(&md->writesets[0], *new_size);
833 	if (r) {
834 		DMERR("%s: writeset_alloc failed for writeset 0", __func__);
835 		return r;
836 	}
837 
838 	r = writeset_alloc(&md->writesets[1], *new_size);
839 	if (r) {
840 		DMERR("%s: writeset_alloc failed for writeset 1", __func__);
841 		return r;
842 	}
843 
844 	value = cpu_to_le32(0u);
845 	__dm_bless_for_disk(&value);
846 	r = dm_array_resize(&md->era_array_info, md->era_array_root,
847 			    md->nr_blocks, *new_size,
848 			    &value, &md->era_array_root);
849 	if (r) {
850 		DMERR("%s: dm_array_resize failed", __func__);
851 		return r;
852 	}
853 
854 	md->nr_blocks = *new_size;
855 	return 0;
856 }
857 
858 static int metadata_era_archive(struct era_metadata *md)
859 {
860 	int r;
861 	uint64_t keys[1];
862 	struct writeset_disk value;
863 
864 	r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root,
865 			    &md->current_writeset->md.root);
866 	if (r) {
867 		DMERR("%s: dm_bitset_flush failed", __func__);
868 		return r;
869 	}
870 
871 	ws_pack(&md->current_writeset->md, &value);
872 	md->current_writeset->md.root = INVALID_WRITESET_ROOT;
873 
874 	keys[0] = md->current_era;
875 	__dm_bless_for_disk(&value);
876 	r = dm_btree_insert(&md->writeset_tree_info, md->writeset_tree_root,
877 			    keys, &value, &md->writeset_tree_root);
878 	if (r) {
879 		DMERR("%s: couldn't insert writeset into btree", __func__);
880 		/* FIXME: fail mode */
881 		return r;
882 	}
883 
884 	md->archived_writesets = true;
885 
886 	return 0;
887 }
888 
889 static struct writeset *next_writeset(struct era_metadata *md)
890 {
891 	return (md->current_writeset == &md->writesets[0]) ?
892 		&md->writesets[1] : &md->writesets[0];
893 }
894 
895 static int metadata_new_era(struct era_metadata *md)
896 {
897 	int r;
898 	struct writeset *new_writeset = next_writeset(md);
899 
900 	r = writeset_init(&md->bitset_info, new_writeset);
901 	if (r) {
902 		DMERR("%s: writeset_init failed", __func__);
903 		return r;
904 	}
905 
906 	swap_writeset(md, new_writeset);
907 	md->current_era++;
908 
909 	return 0;
910 }
911 
912 static int metadata_era_rollover(struct era_metadata *md)
913 {
914 	int r;
915 
916 	if (md->current_writeset->md.root != INVALID_WRITESET_ROOT) {
917 		r = metadata_era_archive(md);
918 		if (r) {
919 			DMERR("%s: metadata_archive_era failed", __func__);
920 			/* FIXME: fail mode? */
921 			return r;
922 		}
923 	}
924 
925 	r = metadata_new_era(md);
926 	if (r) {
927 		DMERR("%s: new era failed", __func__);
928 		/* FIXME: fail mode */
929 		return r;
930 	}
931 
932 	return 0;
933 }
934 
935 static bool metadata_current_marked(struct era_metadata *md, dm_block_t block)
936 {
937 	bool r;
938 	struct writeset *ws;
939 
940 	rcu_read_lock();
941 	ws = rcu_dereference(md->current_writeset);
942 	r = writeset_marked(ws, block);
943 	rcu_read_unlock();
944 
945 	return r;
946 }
947 
948 static int metadata_commit(struct era_metadata *md)
949 {
950 	int r;
951 	struct dm_block *sblock;
952 
953 	if (md->current_writeset->md.root != SUPERBLOCK_LOCATION) {
954 		r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root,
955 				    &md->current_writeset->md.root);
956 		if (r) {
957 			DMERR("%s: bitset flush failed", __func__);
958 			return r;
959 		}
960 	}
961 
962 	r = dm_tm_pre_commit(md->tm);
963 	if (r) {
964 		DMERR("%s: pre commit failed", __func__);
965 		return r;
966 	}
967 
968 	r = save_sm_root(md);
969 	if (r) {
970 		DMERR("%s: save_sm_root failed", __func__);
971 		return r;
972 	}
973 
974 	r = superblock_lock(md, &sblock);
975 	if (r) {
976 		DMERR("%s: superblock lock failed", __func__);
977 		return r;
978 	}
979 
980 	prepare_superblock(md, dm_block_data(sblock));
981 
982 	return dm_tm_commit(md->tm, sblock);
983 }
984 
985 static int metadata_checkpoint(struct era_metadata *md)
986 {
987 	/*
988 	 * For now we just rollover, but later I want to put a check in to
989 	 * avoid this if the filter is still pretty fresh.
990 	 */
991 	return metadata_era_rollover(md);
992 }
993 
994 /*
995  * Metadata snapshots allow userland to access era data.
996  */
997 static int metadata_take_snap(struct era_metadata *md)
998 {
999 	int r, inc;
1000 	struct dm_block *clone;
1001 
1002 	if (md->metadata_snap != SUPERBLOCK_LOCATION) {
1003 		DMERR("%s: metadata snapshot already exists", __func__);
1004 		return -EINVAL;
1005 	}
1006 
1007 	r = metadata_era_rollover(md);
1008 	if (r) {
1009 		DMERR("%s: era rollover failed", __func__);
1010 		return r;
1011 	}
1012 
1013 	r = metadata_commit(md);
1014 	if (r) {
1015 		DMERR("%s: pre commit failed", __func__);
1016 		return r;
1017 	}
1018 
1019 	r = dm_sm_inc_block(md->sm, SUPERBLOCK_LOCATION);
1020 	if (r) {
1021 		DMERR("%s: couldn't increment superblock", __func__);
1022 		return r;
1023 	}
1024 
1025 	r = dm_tm_shadow_block(md->tm, SUPERBLOCK_LOCATION,
1026 			       &sb_validator, &clone, &inc);
1027 	if (r) {
1028 		DMERR("%s: couldn't shadow superblock", __func__);
1029 		dm_sm_dec_block(md->sm, SUPERBLOCK_LOCATION);
1030 		return r;
1031 	}
1032 	BUG_ON(!inc);
1033 
1034 	r = dm_sm_inc_block(md->sm, md->writeset_tree_root);
1035 	if (r) {
1036 		DMERR("%s: couldn't inc writeset tree root", __func__);
1037 		dm_tm_unlock(md->tm, clone);
1038 		return r;
1039 	}
1040 
1041 	r = dm_sm_inc_block(md->sm, md->era_array_root);
1042 	if (r) {
1043 		DMERR("%s: couldn't inc era tree root", __func__);
1044 		dm_sm_dec_block(md->sm, md->writeset_tree_root);
1045 		dm_tm_unlock(md->tm, clone);
1046 		return r;
1047 	}
1048 
1049 	md->metadata_snap = dm_block_location(clone);
1050 
1051 	dm_tm_unlock(md->tm, clone);
1052 
1053 	return 0;
1054 }
1055 
1056 static int metadata_drop_snap(struct era_metadata *md)
1057 {
1058 	int r;
1059 	dm_block_t location;
1060 	struct dm_block *clone;
1061 	struct superblock_disk *disk;
1062 
1063 	if (md->metadata_snap == SUPERBLOCK_LOCATION) {
1064 		DMERR("%s: no snap to drop", __func__);
1065 		return -EINVAL;
1066 	}
1067 
1068 	r = dm_tm_read_lock(md->tm, md->metadata_snap, &sb_validator, &clone);
1069 	if (r) {
1070 		DMERR("%s: couldn't read lock superblock clone", __func__);
1071 		return r;
1072 	}
1073 
1074 	/*
1075 	 * Whatever happens now we'll commit with no record of the metadata
1076 	 * snap.
1077 	 */
1078 	md->metadata_snap = SUPERBLOCK_LOCATION;
1079 
1080 	disk = dm_block_data(clone);
1081 	r = dm_btree_del(&md->writeset_tree_info,
1082 			 le64_to_cpu(disk->writeset_tree_root));
1083 	if (r) {
1084 		DMERR("%s: error deleting writeset tree clone", __func__);
1085 		dm_tm_unlock(md->tm, clone);
1086 		return r;
1087 	}
1088 
1089 	r = dm_array_del(&md->era_array_info, le64_to_cpu(disk->era_array_root));
1090 	if (r) {
1091 		DMERR("%s: error deleting era array clone", __func__);
1092 		dm_tm_unlock(md->tm, clone);
1093 		return r;
1094 	}
1095 
1096 	location = dm_block_location(clone);
1097 	dm_tm_unlock(md->tm, clone);
1098 
1099 	return dm_sm_dec_block(md->sm, location);
1100 }
1101 
1102 struct metadata_stats {
1103 	dm_block_t used;
1104 	dm_block_t total;
1105 	dm_block_t snap;
1106 	uint32_t era;
1107 };
1108 
1109 static int metadata_get_stats(struct era_metadata *md, void *ptr)
1110 {
1111 	int r;
1112 	struct metadata_stats *s = ptr;
1113 	dm_block_t nr_free, nr_total;
1114 
1115 	r = dm_sm_get_nr_free(md->sm, &nr_free);
1116 	if (r) {
1117 		DMERR("dm_sm_get_nr_free returned %d", r);
1118 		return r;
1119 	}
1120 
1121 	r = dm_sm_get_nr_blocks(md->sm, &nr_total);
1122 	if (r) {
1123 		DMERR("dm_pool_get_metadata_dev_size returned %d", r);
1124 		return r;
1125 	}
1126 
1127 	s->used = nr_total - nr_free;
1128 	s->total = nr_total;
1129 	s->snap = md->metadata_snap;
1130 	s->era = md->current_era;
1131 
1132 	return 0;
1133 }
1134 
1135 /*----------------------------------------------------------------*/
1136 
1137 struct era {
1138 	struct dm_target *ti;
1139 	struct dm_target_callbacks callbacks;
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->bi_bdev = 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 			generic_make_request(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 int dev_is_congested(struct dm_dev *dev, int bdi_bits)
1378 {
1379 	struct request_queue *q = bdev_get_queue(dev->bdev);
1380 	return bdi_congested(q->backing_dev_info, bdi_bits);
1381 }
1382 
1383 static int era_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
1384 {
1385 	struct era *era = container_of(cb, struct era, callbacks);
1386 	return dev_is_congested(era->origin_dev, bdi_bits);
1387 }
1388 
1389 static void era_destroy(struct era *era)
1390 {
1391 	if (era->md)
1392 		metadata_close(era->md);
1393 
1394 	if (era->wq)
1395 		destroy_workqueue(era->wq);
1396 
1397 	if (era->origin_dev)
1398 		dm_put_device(era->ti, era->origin_dev);
1399 
1400 	if (era->metadata_dev)
1401 		dm_put_device(era->ti, era->metadata_dev);
1402 
1403 	kfree(era);
1404 }
1405 
1406 static dm_block_t calc_nr_blocks(struct era *era)
1407 {
1408 	return dm_sector_div_up(era->ti->len, era->sectors_per_block);
1409 }
1410 
1411 static bool valid_block_size(dm_block_t block_size)
1412 {
1413 	bool greater_than_zero = block_size > 0;
1414 	bool multiple_of_min_block_size = (block_size & (MIN_BLOCK_SIZE - 1)) == 0;
1415 
1416 	return greater_than_zero && multiple_of_min_block_size;
1417 }
1418 
1419 /*
1420  * <metadata dev> <data dev> <data block size (sectors)>
1421  */
1422 static int era_ctr(struct dm_target *ti, unsigned argc, char **argv)
1423 {
1424 	int r;
1425 	char dummy;
1426 	struct era *era;
1427 	struct era_metadata *md;
1428 
1429 	if (argc != 3) {
1430 		ti->error = "Invalid argument count";
1431 		return -EINVAL;
1432 	}
1433 
1434 	era = kzalloc(sizeof(*era), GFP_KERNEL);
1435 	if (!era) {
1436 		ti->error = "Error allocating era structure";
1437 		return -ENOMEM;
1438 	}
1439 
1440 	era->ti = ti;
1441 
1442 	r = dm_get_device(ti, argv[0], FMODE_READ | FMODE_WRITE, &era->metadata_dev);
1443 	if (r) {
1444 		ti->error = "Error opening metadata device";
1445 		era_destroy(era);
1446 		return -EINVAL;
1447 	}
1448 
1449 	r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &era->origin_dev);
1450 	if (r) {
1451 		ti->error = "Error opening data device";
1452 		era_destroy(era);
1453 		return -EINVAL;
1454 	}
1455 
1456 	r = sscanf(argv[2], "%u%c", &era->sectors_per_block, &dummy);
1457 	if (r != 1) {
1458 		ti->error = "Error parsing block size";
1459 		era_destroy(era);
1460 		return -EINVAL;
1461 	}
1462 
1463 	r = dm_set_target_max_io_len(ti, era->sectors_per_block);
1464 	if (r) {
1465 		ti->error = "could not set max io len";
1466 		era_destroy(era);
1467 		return -EINVAL;
1468 	}
1469 
1470 	if (!valid_block_size(era->sectors_per_block)) {
1471 		ti->error = "Invalid block size";
1472 		era_destroy(era);
1473 		return -EINVAL;
1474 	}
1475 	if (era->sectors_per_block & (era->sectors_per_block - 1))
1476 		era->sectors_per_block_shift = -1;
1477 	else
1478 		era->sectors_per_block_shift = __ffs(era->sectors_per_block);
1479 
1480 	md = metadata_open(era->metadata_dev->bdev, era->sectors_per_block, true);
1481 	if (IS_ERR(md)) {
1482 		ti->error = "Error reading metadata";
1483 		era_destroy(era);
1484 		return PTR_ERR(md);
1485 	}
1486 	era->md = md;
1487 
1488 	era->nr_blocks = calc_nr_blocks(era);
1489 
1490 	r = metadata_resize(era->md, &era->nr_blocks);
1491 	if (r) {
1492 		ti->error = "couldn't resize metadata";
1493 		era_destroy(era);
1494 		return -ENOMEM;
1495 	}
1496 
1497 	era->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
1498 	if (!era->wq) {
1499 		ti->error = "could not create workqueue for metadata object";
1500 		era_destroy(era);
1501 		return -ENOMEM;
1502 	}
1503 	INIT_WORK(&era->worker, do_work);
1504 
1505 	spin_lock_init(&era->deferred_lock);
1506 	bio_list_init(&era->deferred_bios);
1507 
1508 	spin_lock_init(&era->rpc_lock);
1509 	INIT_LIST_HEAD(&era->rpc_calls);
1510 
1511 	ti->private = era;
1512 	ti->num_flush_bios = 1;
1513 	ti->flush_supported = true;
1514 
1515 	ti->num_discard_bios = 1;
1516 	ti->discards_supported = true;
1517 	era->callbacks.congested_fn = era_is_congested;
1518 	dm_table_add_target_callbacks(ti->table, &era->callbacks);
1519 
1520 	return 0;
1521 }
1522 
1523 static void era_dtr(struct dm_target *ti)
1524 {
1525 	era_destroy(ti->private);
1526 }
1527 
1528 static int era_map(struct dm_target *ti, struct bio *bio)
1529 {
1530 	struct era *era = ti->private;
1531 	dm_block_t block = get_block(era, bio);
1532 
1533 	/*
1534 	 * All bios get remapped to the origin device.  We do this now, but
1535 	 * it may not get issued until later.  Depending on whether the
1536 	 * block is marked in this era.
1537 	 */
1538 	remap_to_origin(era, bio);
1539 
1540 	/*
1541 	 * REQ_PREFLUSH bios carry no data, so we're not interested in them.
1542 	 */
1543 	if (!(bio->bi_opf & REQ_PREFLUSH) &&
1544 	    (bio_data_dir(bio) == WRITE) &&
1545 	    !metadata_current_marked(era->md, block)) {
1546 		defer_bio(era, bio);
1547 		return DM_MAPIO_SUBMITTED;
1548 	}
1549 
1550 	return DM_MAPIO_REMAPPED;
1551 }
1552 
1553 static void era_postsuspend(struct dm_target *ti)
1554 {
1555 	int r;
1556 	struct era *era = ti->private;
1557 
1558 	r = in_worker0(era, metadata_era_archive);
1559 	if (r) {
1560 		DMERR("%s: couldn't archive current era", __func__);
1561 		/* FIXME: fail mode */
1562 	}
1563 
1564 	stop_worker(era);
1565 }
1566 
1567 static int era_preresume(struct dm_target *ti)
1568 {
1569 	int r;
1570 	struct era *era = ti->private;
1571 	dm_block_t new_size = calc_nr_blocks(era);
1572 
1573 	if (era->nr_blocks != new_size) {
1574 		r = in_worker1(era, metadata_resize, &new_size);
1575 		if (r)
1576 			return r;
1577 
1578 		era->nr_blocks = new_size;
1579 	}
1580 
1581 	start_worker(era);
1582 
1583 	r = in_worker0(era, metadata_new_era);
1584 	if (r) {
1585 		DMERR("%s: metadata_era_rollover failed", __func__);
1586 		return r;
1587 	}
1588 
1589 	return 0;
1590 }
1591 
1592 /*
1593  * Status format:
1594  *
1595  * <metadata block size> <#used metadata blocks>/<#total metadata blocks>
1596  * <current era> <held metadata root | '-'>
1597  */
1598 static void era_status(struct dm_target *ti, status_type_t type,
1599 		       unsigned status_flags, char *result, unsigned maxlen)
1600 {
1601 	int r;
1602 	struct era *era = ti->private;
1603 	ssize_t sz = 0;
1604 	struct metadata_stats stats;
1605 	char buf[BDEVNAME_SIZE];
1606 
1607 	switch (type) {
1608 	case STATUSTYPE_INFO:
1609 		r = in_worker1(era, metadata_get_stats, &stats);
1610 		if (r)
1611 			goto err;
1612 
1613 		DMEMIT("%u %llu/%llu %u",
1614 		       (unsigned) (DM_ERA_METADATA_BLOCK_SIZE >> SECTOR_SHIFT),
1615 		       (unsigned long long) stats.used,
1616 		       (unsigned long long) stats.total,
1617 		       (unsigned) stats.era);
1618 
1619 		if (stats.snap != SUPERBLOCK_LOCATION)
1620 			DMEMIT(" %llu", stats.snap);
1621 		else
1622 			DMEMIT(" -");
1623 		break;
1624 
1625 	case STATUSTYPE_TABLE:
1626 		format_dev_t(buf, era->metadata_dev->bdev->bd_dev);
1627 		DMEMIT("%s ", buf);
1628 		format_dev_t(buf, era->origin_dev->bdev->bd_dev);
1629 		DMEMIT("%s %u", buf, era->sectors_per_block);
1630 		break;
1631 	}
1632 
1633 	return;
1634 
1635 err:
1636 	DMEMIT("Error");
1637 }
1638 
1639 static int era_message(struct dm_target *ti, unsigned argc, char **argv)
1640 {
1641 	struct era *era = ti->private;
1642 
1643 	if (argc != 1) {
1644 		DMERR("incorrect number of message arguments");
1645 		return -EINVAL;
1646 	}
1647 
1648 	if (!strcasecmp(argv[0], "checkpoint"))
1649 		return in_worker0(era, metadata_checkpoint);
1650 
1651 	if (!strcasecmp(argv[0], "take_metadata_snap"))
1652 		return in_worker0(era, metadata_take_snap);
1653 
1654 	if (!strcasecmp(argv[0], "drop_metadata_snap"))
1655 		return in_worker0(era, metadata_drop_snap);
1656 
1657 	DMERR("unsupported message '%s'", argv[0]);
1658 	return -EINVAL;
1659 }
1660 
1661 static sector_t get_dev_size(struct dm_dev *dev)
1662 {
1663 	return i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT;
1664 }
1665 
1666 static int era_iterate_devices(struct dm_target *ti,
1667 			       iterate_devices_callout_fn fn, void *data)
1668 {
1669 	struct era *era = ti->private;
1670 	return fn(ti, era->origin_dev, 0, get_dev_size(era->origin_dev), data);
1671 }
1672 
1673 static void era_io_hints(struct dm_target *ti, struct queue_limits *limits)
1674 {
1675 	struct era *era = ti->private;
1676 	uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
1677 
1678 	/*
1679 	 * If the system-determined stacked limits are compatible with the
1680 	 * era device's blocksize (io_opt is a factor) do not override them.
1681 	 */
1682 	if (io_opt_sectors < era->sectors_per_block ||
1683 	    do_div(io_opt_sectors, era->sectors_per_block)) {
1684 		blk_limits_io_min(limits, 0);
1685 		blk_limits_io_opt(limits, era->sectors_per_block << SECTOR_SHIFT);
1686 	}
1687 }
1688 
1689 /*----------------------------------------------------------------*/
1690 
1691 static struct target_type era_target = {
1692 	.name = "era",
1693 	.version = {1, 0, 0},
1694 	.module = THIS_MODULE,
1695 	.ctr = era_ctr,
1696 	.dtr = era_dtr,
1697 	.map = era_map,
1698 	.postsuspend = era_postsuspend,
1699 	.preresume = era_preresume,
1700 	.status = era_status,
1701 	.message = era_message,
1702 	.iterate_devices = era_iterate_devices,
1703 	.io_hints = era_io_hints
1704 };
1705 
1706 static int __init dm_era_init(void)
1707 {
1708 	int r;
1709 
1710 	r = dm_register_target(&era_target);
1711 	if (r) {
1712 		DMERR("era target registration failed: %d", r);
1713 		return r;
1714 	}
1715 
1716 	return 0;
1717 }
1718 
1719 static void __exit dm_era_exit(void)
1720 {
1721 	dm_unregister_target(&era_target);
1722 }
1723 
1724 module_init(dm_era_init);
1725 module_exit(dm_era_exit);
1726 
1727 MODULE_DESCRIPTION(DM_NAME " era target");
1728 MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>");
1729 MODULE_LICENSE("GPL");
1730