xref: /openbmc/linux/drivers/md/dm-era-target.c (revision f71a261a)
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 	ws->bits = NULL;
51 }
52 
53 static int setup_on_disk_bitset(struct dm_disk_bitset *info,
54 				unsigned nr_bits, dm_block_t *root)
55 {
56 	int r;
57 
58 	r = dm_bitset_empty(info, root);
59 	if (r)
60 		return r;
61 
62 	return dm_bitset_resize(info, *root, 0, nr_bits, false, root);
63 }
64 
65 static size_t bitset_size(unsigned nr_bits)
66 {
67 	return sizeof(unsigned long) * dm_div_up(nr_bits, BITS_PER_LONG);
68 }
69 
70 /*
71  * Allocates memory for the in core bitset.
72  */
73 static int writeset_alloc(struct writeset *ws, dm_block_t nr_blocks)
74 {
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 			 dm_block_t nr_blocks)
89 {
90 	int r;
91 
92 	memset(ws->bits, 0, bitset_size(nr_blocks));
93 
94 	ws->md.nr_bits = nr_blocks;
95 	r = setup_on_disk_bitset(info, ws->md.nr_bits, &ws->md.root);
96 	if (r) {
97 		DMERR("%s: setup_on_disk_bitset failed", __func__);
98 		return r;
99 	}
100 
101 	return 0;
102 }
103 
104 static bool writeset_marked(struct writeset *ws, dm_block_t block)
105 {
106 	return test_bit(block, ws->bits);
107 }
108 
109 static int writeset_marked_on_disk(struct dm_disk_bitset *info,
110 				   struct writeset_metadata *m, dm_block_t block,
111 				   bool *result)
112 {
113 	dm_block_t old = m->root;
114 
115 	/*
116 	 * The bitset was flushed when it was archived, so we know there'll
117 	 * be no change to the root.
118 	 */
119 	int r = dm_bitset_test_bit(info, m->root, block, &m->root, result);
120 	if (r) {
121 		DMERR("%s: dm_bitset_test_bit failed", __func__);
122 		return r;
123 	}
124 
125 	BUG_ON(m->root != old);
126 
127 	return r;
128 }
129 
130 /*
131  * Returns < 0 on error, 0 if the bit wasn't previously set, 1 if it was.
132  */
133 static int writeset_test_and_set(struct dm_disk_bitset *info,
134 				 struct writeset *ws, uint32_t block)
135 {
136 	int r;
137 
138 	if (!test_bit(block, ws->bits)) {
139 		r = dm_bitset_set_bit(info, ws->md.root, block, &ws->md.root);
140 		if (r) {
141 			/* FIXME: fail mode */
142 			return r;
143 		}
144 
145 		return 0;
146 	}
147 
148 	return 1;
149 }
150 
151 /*----------------------------------------------------------------
152  * On disk metadata layout
153  *--------------------------------------------------------------*/
154 #define SPACE_MAP_ROOT_SIZE 128
155 #define UUID_LEN 16
156 
157 struct writeset_disk {
158 	__le32 nr_bits;
159 	__le64 root;
160 } __packed;
161 
162 struct superblock_disk {
163 	__le32 csum;
164 	__le32 flags;
165 	__le64 blocknr;
166 
167 	__u8 uuid[UUID_LEN];
168 	__le64 magic;
169 	__le32 version;
170 
171 	__u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
172 
173 	__le32 data_block_size;
174 	__le32 metadata_block_size;
175 	__le32 nr_blocks;
176 
177 	__le32 current_era;
178 	struct writeset_disk current_writeset;
179 
180 	/*
181 	 * Only these two fields are valid within the metadata snapshot.
182 	 */
183 	__le64 writeset_tree_root;
184 	__le64 era_array_root;
185 
186 	__le64 metadata_snap;
187 } __packed;
188 
189 /*----------------------------------------------------------------
190  * Superblock validation
191  *--------------------------------------------------------------*/
192 static void sb_prepare_for_write(struct dm_block_validator *v,
193 				 struct dm_block *b,
194 				 size_t sb_block_size)
195 {
196 	struct superblock_disk *disk = dm_block_data(b);
197 
198 	disk->blocknr = cpu_to_le64(dm_block_location(b));
199 	disk->csum = cpu_to_le32(dm_bm_checksum(&disk->flags,
200 						sb_block_size - sizeof(__le32),
201 						SUPERBLOCK_CSUM_XOR));
202 }
203 
204 static int check_metadata_version(struct superblock_disk *disk)
205 {
206 	uint32_t metadata_version = le32_to_cpu(disk->version);
207 	if (metadata_version < MIN_ERA_VERSION || metadata_version > MAX_ERA_VERSION) {
208 		DMERR("Era metadata version %u found, but only versions between %u and %u supported.",
209 		      metadata_version, MIN_ERA_VERSION, MAX_ERA_VERSION);
210 		return -EINVAL;
211 	}
212 
213 	return 0;
214 }
215 
216 static int sb_check(struct dm_block_validator *v,
217 		    struct dm_block *b,
218 		    size_t sb_block_size)
219 {
220 	struct superblock_disk *disk = dm_block_data(b);
221 	__le32 csum_le;
222 
223 	if (dm_block_location(b) != le64_to_cpu(disk->blocknr)) {
224 		DMERR("sb_check failed: blocknr %llu: wanted %llu",
225 		      le64_to_cpu(disk->blocknr),
226 		      (unsigned long long)dm_block_location(b));
227 		return -ENOTBLK;
228 	}
229 
230 	if (le64_to_cpu(disk->magic) != SUPERBLOCK_MAGIC) {
231 		DMERR("sb_check failed: magic %llu: wanted %llu",
232 		      le64_to_cpu(disk->magic),
233 		      (unsigned long long) SUPERBLOCK_MAGIC);
234 		return -EILSEQ;
235 	}
236 
237 	csum_le = cpu_to_le32(dm_bm_checksum(&disk->flags,
238 					     sb_block_size - sizeof(__le32),
239 					     SUPERBLOCK_CSUM_XOR));
240 	if (csum_le != disk->csum) {
241 		DMERR("sb_check failed: csum %u: wanted %u",
242 		      le32_to_cpu(csum_le), le32_to_cpu(disk->csum));
243 		return -EILSEQ;
244 	}
245 
246 	return check_metadata_version(disk);
247 }
248 
249 static struct dm_block_validator sb_validator = {
250 	.name = "superblock",
251 	.prepare_for_write = sb_prepare_for_write,
252 	.check = sb_check
253 };
254 
255 /*----------------------------------------------------------------
256  * Low level metadata handling
257  *--------------------------------------------------------------*/
258 #define DM_ERA_METADATA_BLOCK_SIZE 4096
259 #define ERA_MAX_CONCURRENT_LOCKS 5
260 
261 struct era_metadata {
262 	struct block_device *bdev;
263 	struct dm_block_manager *bm;
264 	struct dm_space_map *sm;
265 	struct dm_transaction_manager *tm;
266 
267 	dm_block_t block_size;
268 	uint32_t nr_blocks;
269 
270 	uint32_t current_era;
271 
272 	/*
273 	 * We preallocate 2 writesets.  When an era rolls over we
274 	 * switch between them. This means the allocation is done at
275 	 * preresume time, rather than on the io path.
276 	 */
277 	struct writeset writesets[2];
278 	struct writeset *current_writeset;
279 
280 	dm_block_t writeset_tree_root;
281 	dm_block_t era_array_root;
282 
283 	struct dm_disk_bitset bitset_info;
284 	struct dm_btree_info writeset_tree_info;
285 	struct dm_array_info era_array_info;
286 
287 	dm_block_t metadata_snap;
288 
289 	/*
290 	 * A flag that is set whenever a writeset has been archived.
291 	 */
292 	bool archived_writesets;
293 
294 	/*
295 	 * Reading the space map root can fail, so we read it into this
296 	 * buffer before the superblock is locked and updated.
297 	 */
298 	__u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
299 };
300 
301 static int superblock_read_lock(struct era_metadata *md,
302 				struct dm_block **sblock)
303 {
304 	return dm_bm_read_lock(md->bm, SUPERBLOCK_LOCATION,
305 			       &sb_validator, sblock);
306 }
307 
308 static int superblock_lock_zero(struct era_metadata *md,
309 				struct dm_block **sblock)
310 {
311 	return dm_bm_write_lock_zero(md->bm, SUPERBLOCK_LOCATION,
312 				     &sb_validator, sblock);
313 }
314 
315 static int superblock_lock(struct era_metadata *md,
316 			   struct dm_block **sblock)
317 {
318 	return dm_bm_write_lock(md->bm, SUPERBLOCK_LOCATION,
319 				&sb_validator, sblock);
320 }
321 
322 /* FIXME: duplication with cache and thin */
323 static int superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
324 {
325 	int r;
326 	unsigned i;
327 	struct dm_block *b;
328 	__le64 *data_le, zero = cpu_to_le64(0);
329 	unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
330 
331 	/*
332 	 * We can't use a validator here - it may be all zeroes.
333 	 */
334 	r = dm_bm_read_lock(bm, SUPERBLOCK_LOCATION, NULL, &b);
335 	if (r)
336 		return r;
337 
338 	data_le = dm_block_data(b);
339 	*result = true;
340 	for (i = 0; i < sb_block_size; i++) {
341 		if (data_le[i] != zero) {
342 			*result = false;
343 			break;
344 		}
345 	}
346 
347 	dm_bm_unlock(b);
348 
349 	return 0;
350 }
351 
352 /*----------------------------------------------------------------*/
353 
354 static void ws_pack(const struct writeset_metadata *core, struct writeset_disk *disk)
355 {
356 	disk->nr_bits = cpu_to_le32(core->nr_bits);
357 	disk->root = cpu_to_le64(core->root);
358 }
359 
360 static void ws_unpack(const struct writeset_disk *disk, struct writeset_metadata *core)
361 {
362 	core->nr_bits = le32_to_cpu(disk->nr_bits);
363 	core->root = le64_to_cpu(disk->root);
364 }
365 
366 static void ws_inc(void *context, const void *value, unsigned count)
367 {
368 	struct era_metadata *md = context;
369 	struct writeset_disk ws_d;
370 	dm_block_t b;
371 	unsigned i;
372 
373 	for (i = 0; i < count; i++) {
374 		memcpy(&ws_d, value + (i * sizeof(ws_d)), sizeof(ws_d));
375 		b = le64_to_cpu(ws_d.root);
376 		dm_tm_inc(md->tm, b);
377 	}
378 }
379 
380 static void ws_dec(void *context, const void *value, unsigned count)
381 {
382 	struct era_metadata *md = context;
383 	struct writeset_disk ws_d;
384 	dm_block_t b;
385 	unsigned i;
386 
387 	for (i = 0; i < count; i++) {
388 		memcpy(&ws_d, value + (i * sizeof(ws_d)), sizeof(ws_d));
389 		b = le64_to_cpu(ws_d.root);
390 		dm_bitset_del(&md->bitset_info, b);
391 	}
392 }
393 
394 static int ws_eq(void *context, const void *value1, const void *value2)
395 {
396 	return !memcmp(value1, value2, sizeof(struct writeset_disk));
397 }
398 
399 /*----------------------------------------------------------------*/
400 
401 static void setup_writeset_tree_info(struct era_metadata *md)
402 {
403 	struct dm_btree_value_type *vt = &md->writeset_tree_info.value_type;
404 	md->writeset_tree_info.tm = md->tm;
405 	md->writeset_tree_info.levels = 1;
406 	vt->context = md;
407 	vt->size = sizeof(struct writeset_disk);
408 	vt->inc = ws_inc;
409 	vt->dec = ws_dec;
410 	vt->equal = ws_eq;
411 }
412 
413 static void setup_era_array_info(struct era_metadata *md)
414 
415 {
416 	struct dm_btree_value_type vt;
417 	vt.context = NULL;
418 	vt.size = sizeof(__le32);
419 	vt.inc = NULL;
420 	vt.dec = NULL;
421 	vt.equal = NULL;
422 
423 	dm_array_info_init(&md->era_array_info, md->tm, &vt);
424 }
425 
426 static void setup_infos(struct era_metadata *md)
427 {
428 	dm_disk_bitset_init(md->tm, &md->bitset_info);
429 	setup_writeset_tree_info(md);
430 	setup_era_array_info(md);
431 }
432 
433 /*----------------------------------------------------------------*/
434 
435 static int create_fresh_metadata(struct era_metadata *md)
436 {
437 	int r;
438 
439 	r = dm_tm_create_with_sm(md->bm, SUPERBLOCK_LOCATION,
440 				 &md->tm, &md->sm);
441 	if (r < 0) {
442 		DMERR("dm_tm_create_with_sm failed");
443 		return r;
444 	}
445 
446 	setup_infos(md);
447 
448 	r = dm_btree_empty(&md->writeset_tree_info, &md->writeset_tree_root);
449 	if (r) {
450 		DMERR("couldn't create new writeset tree");
451 		goto bad;
452 	}
453 
454 	r = dm_array_empty(&md->era_array_info, &md->era_array_root);
455 	if (r) {
456 		DMERR("couldn't create era array");
457 		goto bad;
458 	}
459 
460 	return 0;
461 
462 bad:
463 	dm_sm_destroy(md->sm);
464 	dm_tm_destroy(md->tm);
465 
466 	return r;
467 }
468 
469 static int save_sm_root(struct era_metadata *md)
470 {
471 	int r;
472 	size_t metadata_len;
473 
474 	r = dm_sm_root_size(md->sm, &metadata_len);
475 	if (r < 0)
476 		return r;
477 
478 	return dm_sm_copy_root(md->sm, &md->metadata_space_map_root,
479 			       metadata_len);
480 }
481 
482 static void copy_sm_root(struct era_metadata *md, struct superblock_disk *disk)
483 {
484 	memcpy(&disk->metadata_space_map_root,
485 	       &md->metadata_space_map_root,
486 	       sizeof(md->metadata_space_map_root));
487 }
488 
489 /*
490  * Writes a superblock, including the static fields that don't get updated
491  * with every commit (possible optimisation here).  'md' should be fully
492  * constructed when this is called.
493  */
494 static void prepare_superblock(struct era_metadata *md, struct superblock_disk *disk)
495 {
496 	disk->magic = cpu_to_le64(SUPERBLOCK_MAGIC);
497 	disk->flags = cpu_to_le32(0ul);
498 
499 	/* FIXME: can't keep blanking the uuid (uuid is currently unused though) */
500 	memset(disk->uuid, 0, sizeof(disk->uuid));
501 	disk->version = cpu_to_le32(MAX_ERA_VERSION);
502 
503 	copy_sm_root(md, disk);
504 
505 	disk->data_block_size = cpu_to_le32(md->block_size);
506 	disk->metadata_block_size = cpu_to_le32(DM_ERA_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
507 	disk->nr_blocks = cpu_to_le32(md->nr_blocks);
508 	disk->current_era = cpu_to_le32(md->current_era);
509 
510 	ws_pack(&md->current_writeset->md, &disk->current_writeset);
511 	disk->writeset_tree_root = cpu_to_le64(md->writeset_tree_root);
512 	disk->era_array_root = cpu_to_le64(md->era_array_root);
513 	disk->metadata_snap = cpu_to_le64(md->metadata_snap);
514 }
515 
516 static int write_superblock(struct era_metadata *md)
517 {
518 	int r;
519 	struct dm_block *sblock;
520 	struct superblock_disk *disk;
521 
522 	r = save_sm_root(md);
523 	if (r) {
524 		DMERR("%s: save_sm_root failed", __func__);
525 		return r;
526 	}
527 
528 	r = superblock_lock_zero(md, &sblock);
529 	if (r)
530 		return r;
531 
532 	disk = dm_block_data(sblock);
533 	prepare_superblock(md, disk);
534 
535 	return dm_tm_commit(md->tm, sblock);
536 }
537 
538 /*
539  * Assumes block_size and the infos are set.
540  */
541 static int format_metadata(struct era_metadata *md)
542 {
543 	int r;
544 
545 	r = create_fresh_metadata(md);
546 	if (r)
547 		return r;
548 
549 	r = write_superblock(md);
550 	if (r) {
551 		dm_sm_destroy(md->sm);
552 		dm_tm_destroy(md->tm);
553 		return r;
554 	}
555 
556 	return 0;
557 }
558 
559 static int open_metadata(struct era_metadata *md)
560 {
561 	int r;
562 	struct dm_block *sblock;
563 	struct superblock_disk *disk;
564 
565 	r = superblock_read_lock(md, &sblock);
566 	if (r) {
567 		DMERR("couldn't read_lock superblock");
568 		return r;
569 	}
570 
571 	disk = dm_block_data(sblock);
572 
573 	/* Verify the data block size hasn't changed */
574 	if (le32_to_cpu(disk->data_block_size) != md->block_size) {
575 		DMERR("changing the data block size (from %u to %llu) is not supported",
576 		      le32_to_cpu(disk->data_block_size), md->block_size);
577 		r = -EINVAL;
578 		goto bad;
579 	}
580 
581 	r = dm_tm_open_with_sm(md->bm, SUPERBLOCK_LOCATION,
582 			       disk->metadata_space_map_root,
583 			       sizeof(disk->metadata_space_map_root),
584 			       &md->tm, &md->sm);
585 	if (r) {
586 		DMERR("dm_tm_open_with_sm failed");
587 		goto bad;
588 	}
589 
590 	setup_infos(md);
591 
592 	md->nr_blocks = le32_to_cpu(disk->nr_blocks);
593 	md->current_era = le32_to_cpu(disk->current_era);
594 
595 	ws_unpack(&disk->current_writeset, &md->current_writeset->md);
596 	md->writeset_tree_root = le64_to_cpu(disk->writeset_tree_root);
597 	md->era_array_root = le64_to_cpu(disk->era_array_root);
598 	md->metadata_snap = le64_to_cpu(disk->metadata_snap);
599 	md->archived_writesets = true;
600 
601 	dm_bm_unlock(sblock);
602 
603 	return 0;
604 
605 bad:
606 	dm_bm_unlock(sblock);
607 	return r;
608 }
609 
610 static int open_or_format_metadata(struct era_metadata *md,
611 				   bool may_format)
612 {
613 	int r;
614 	bool unformatted = false;
615 
616 	r = superblock_all_zeroes(md->bm, &unformatted);
617 	if (r)
618 		return r;
619 
620 	if (unformatted)
621 		return may_format ? format_metadata(md) : -EPERM;
622 
623 	return open_metadata(md);
624 }
625 
626 static int create_persistent_data_objects(struct era_metadata *md,
627 					  bool may_format)
628 {
629 	int r;
630 
631 	md->bm = dm_block_manager_create(md->bdev, DM_ERA_METADATA_BLOCK_SIZE,
632 					 ERA_MAX_CONCURRENT_LOCKS);
633 	if (IS_ERR(md->bm)) {
634 		DMERR("could not create block manager");
635 		return PTR_ERR(md->bm);
636 	}
637 
638 	r = open_or_format_metadata(md, may_format);
639 	if (r)
640 		dm_block_manager_destroy(md->bm);
641 
642 	return r;
643 }
644 
645 static void destroy_persistent_data_objects(struct era_metadata *md)
646 {
647 	dm_sm_destroy(md->sm);
648 	dm_tm_destroy(md->tm);
649 	dm_block_manager_destroy(md->bm);
650 }
651 
652 /*
653  * This waits until all era_map threads have picked up the new filter.
654  */
655 static void swap_writeset(struct era_metadata *md, struct writeset *new_writeset)
656 {
657 	rcu_assign_pointer(md->current_writeset, new_writeset);
658 	synchronize_rcu();
659 }
660 
661 /*----------------------------------------------------------------
662  * Writesets get 'digested' into the main era array.
663  *
664  * We're using a coroutine here so the worker thread can do the digestion,
665  * thus avoiding synchronisation of the metadata.  Digesting a whole
666  * writeset in one go would cause too much latency.
667  *--------------------------------------------------------------*/
668 struct digest {
669 	uint32_t era;
670 	unsigned nr_bits, current_bit;
671 	struct writeset_metadata writeset;
672 	__le32 value;
673 	struct dm_disk_bitset info;
674 
675 	int (*step)(struct era_metadata *, struct digest *);
676 };
677 
678 static int metadata_digest_lookup_writeset(struct era_metadata *md,
679 					   struct digest *d);
680 
681 static int metadata_digest_remove_writeset(struct era_metadata *md,
682 					   struct digest *d)
683 {
684 	int r;
685 	uint64_t key = d->era;
686 
687 	r = dm_btree_remove(&md->writeset_tree_info, md->writeset_tree_root,
688 			    &key, &md->writeset_tree_root);
689 	if (r) {
690 		DMERR("%s: dm_btree_remove failed", __func__);
691 		return r;
692 	}
693 
694 	d->step = metadata_digest_lookup_writeset;
695 	return 0;
696 }
697 
698 #define INSERTS_PER_STEP 100
699 
700 static int metadata_digest_transcribe_writeset(struct era_metadata *md,
701 					       struct digest *d)
702 {
703 	int r;
704 	bool marked;
705 	unsigned b, e = min(d->current_bit + INSERTS_PER_STEP, d->nr_bits);
706 
707 	for (b = d->current_bit; b < e; b++) {
708 		r = writeset_marked_on_disk(&d->info, &d->writeset, b, &marked);
709 		if (r) {
710 			DMERR("%s: writeset_marked_on_disk failed", __func__);
711 			return r;
712 		}
713 
714 		if (!marked)
715 			continue;
716 
717 		__dm_bless_for_disk(&d->value);
718 		r = dm_array_set_value(&md->era_array_info, md->era_array_root,
719 				       b, &d->value, &md->era_array_root);
720 		if (r) {
721 			DMERR("%s: dm_array_set_value failed", __func__);
722 			return r;
723 		}
724 	}
725 
726 	if (b == d->nr_bits)
727 		d->step = metadata_digest_remove_writeset;
728 	else
729 		d->current_bit = b;
730 
731 	return 0;
732 }
733 
734 static int metadata_digest_lookup_writeset(struct era_metadata *md,
735 					   struct digest *d)
736 {
737 	int r;
738 	uint64_t key;
739 	struct writeset_disk disk;
740 
741 	r = dm_btree_find_lowest_key(&md->writeset_tree_info,
742 				     md->writeset_tree_root, &key);
743 	if (r < 0)
744 		return r;
745 
746 	d->era = key;
747 
748 	r = dm_btree_lookup(&md->writeset_tree_info,
749 			    md->writeset_tree_root, &key, &disk);
750 	if (r) {
751 		if (r == -ENODATA) {
752 			d->step = NULL;
753 			return 0;
754 		}
755 
756 		DMERR("%s: dm_btree_lookup failed", __func__);
757 		return r;
758 	}
759 
760 	ws_unpack(&disk, &d->writeset);
761 	d->value = cpu_to_le32(key);
762 
763 	/*
764 	 * We initialise another bitset info to avoid any caching side effects
765 	 * with the previous one.
766 	 */
767 	dm_disk_bitset_init(md->tm, &d->info);
768 
769 	d->nr_bits = min(d->writeset.nr_bits, md->nr_blocks);
770 	d->current_bit = 0;
771 	d->step = metadata_digest_transcribe_writeset;
772 
773 	return 0;
774 }
775 
776 static int metadata_digest_start(struct era_metadata *md, struct digest *d)
777 {
778 	if (d->step)
779 		return 0;
780 
781 	memset(d, 0, sizeof(*d));
782 	d->step = metadata_digest_lookup_writeset;
783 
784 	return 0;
785 }
786 
787 /*----------------------------------------------------------------
788  * High level metadata interface.  Target methods should use these, and not
789  * the lower level ones.
790  *--------------------------------------------------------------*/
791 static struct era_metadata *metadata_open(struct block_device *bdev,
792 					  sector_t block_size,
793 					  bool may_format)
794 {
795 	int r;
796 	struct era_metadata *md = kzalloc(sizeof(*md), GFP_KERNEL);
797 
798 	if (!md)
799 		return NULL;
800 
801 	md->bdev = bdev;
802 	md->block_size = block_size;
803 
804 	md->writesets[0].md.root = INVALID_WRITESET_ROOT;
805 	md->writesets[1].md.root = INVALID_WRITESET_ROOT;
806 	md->current_writeset = &md->writesets[0];
807 
808 	r = create_persistent_data_objects(md, may_format);
809 	if (r) {
810 		kfree(md);
811 		return ERR_PTR(r);
812 	}
813 
814 	return md;
815 }
816 
817 static void metadata_close(struct era_metadata *md)
818 {
819 	writeset_free(&md->writesets[0]);
820 	writeset_free(&md->writesets[1]);
821 	destroy_persistent_data_objects(md);
822 	kfree(md);
823 }
824 
825 static bool valid_nr_blocks(dm_block_t n)
826 {
827 	/*
828 	 * dm_bitset restricts us to 2^32.  test_bit & co. restrict us
829 	 * further to 2^31 - 1
830 	 */
831 	return n < (1ull << 31);
832 }
833 
834 static int metadata_resize(struct era_metadata *md, void *arg)
835 {
836 	int r;
837 	dm_block_t *new_size = arg;
838 	__le32 value;
839 
840 	if (!valid_nr_blocks(*new_size)) {
841 		DMERR("Invalid number of origin blocks %llu",
842 		      (unsigned long long) *new_size);
843 		return -EINVAL;
844 	}
845 
846 	writeset_free(&md->writesets[0]);
847 	writeset_free(&md->writesets[1]);
848 
849 	r = writeset_alloc(&md->writesets[0], *new_size);
850 	if (r) {
851 		DMERR("%s: writeset_alloc failed for writeset 0", __func__);
852 		return r;
853 	}
854 
855 	r = writeset_alloc(&md->writesets[1], *new_size);
856 	if (r) {
857 		DMERR("%s: writeset_alloc failed for writeset 1", __func__);
858 		writeset_free(&md->writesets[0]);
859 		return r;
860 	}
861 
862 	value = cpu_to_le32(0u);
863 	__dm_bless_for_disk(&value);
864 	r = dm_array_resize(&md->era_array_info, md->era_array_root,
865 			    md->nr_blocks, *new_size,
866 			    &value, &md->era_array_root);
867 	if (r) {
868 		DMERR("%s: dm_array_resize failed", __func__);
869 		writeset_free(&md->writesets[0]);
870 		writeset_free(&md->writesets[1]);
871 		return r;
872 	}
873 
874 	md->nr_blocks = *new_size;
875 	return 0;
876 }
877 
878 static int metadata_era_archive(struct era_metadata *md)
879 {
880 	int r;
881 	uint64_t keys[1];
882 	struct writeset_disk value;
883 
884 	r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root,
885 			    &md->current_writeset->md.root);
886 	if (r) {
887 		DMERR("%s: dm_bitset_flush failed", __func__);
888 		return r;
889 	}
890 
891 	ws_pack(&md->current_writeset->md, &value);
892 
893 	keys[0] = md->current_era;
894 	__dm_bless_for_disk(&value);
895 	r = dm_btree_insert(&md->writeset_tree_info, md->writeset_tree_root,
896 			    keys, &value, &md->writeset_tree_root);
897 	if (r) {
898 		DMERR("%s: couldn't insert writeset into btree", __func__);
899 		/* FIXME: fail mode */
900 		return r;
901 	}
902 
903 	md->current_writeset->md.root = INVALID_WRITESET_ROOT;
904 	md->archived_writesets = true;
905 
906 	return 0;
907 }
908 
909 static struct writeset *next_writeset(struct era_metadata *md)
910 {
911 	return (md->current_writeset == &md->writesets[0]) ?
912 		&md->writesets[1] : &md->writesets[0];
913 }
914 
915 static int metadata_new_era(struct era_metadata *md)
916 {
917 	int r;
918 	struct writeset *new_writeset = next_writeset(md);
919 
920 	r = writeset_init(&md->bitset_info, new_writeset, md->nr_blocks);
921 	if (r) {
922 		DMERR("%s: writeset_init failed", __func__);
923 		return r;
924 	}
925 
926 	swap_writeset(md, new_writeset);
927 	md->current_era++;
928 
929 	return 0;
930 }
931 
932 static int metadata_era_rollover(struct era_metadata *md)
933 {
934 	int r;
935 
936 	if (md->current_writeset->md.root != INVALID_WRITESET_ROOT) {
937 		r = metadata_era_archive(md);
938 		if (r) {
939 			DMERR("%s: metadata_archive_era failed", __func__);
940 			/* FIXME: fail mode? */
941 			return r;
942 		}
943 	}
944 
945 	r = metadata_new_era(md);
946 	if (r) {
947 		DMERR("%s: new era failed", __func__);
948 		/* FIXME: fail mode */
949 		return r;
950 	}
951 
952 	return 0;
953 }
954 
955 static bool metadata_current_marked(struct era_metadata *md, dm_block_t block)
956 {
957 	bool r;
958 	struct writeset *ws;
959 
960 	rcu_read_lock();
961 	ws = rcu_dereference(md->current_writeset);
962 	r = writeset_marked(ws, block);
963 	rcu_read_unlock();
964 
965 	return r;
966 }
967 
968 static int metadata_commit(struct era_metadata *md)
969 {
970 	int r;
971 	struct dm_block *sblock;
972 
973 	if (md->current_writeset->md.root != INVALID_WRITESET_ROOT) {
974 		r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root,
975 				    &md->current_writeset->md.root);
976 		if (r) {
977 			DMERR("%s: bitset flush failed", __func__);
978 			return r;
979 		}
980 	}
981 
982 	r = dm_tm_pre_commit(md->tm);
983 	if (r) {
984 		DMERR("%s: pre commit failed", __func__);
985 		return r;
986 	}
987 
988 	r = save_sm_root(md);
989 	if (r) {
990 		DMERR("%s: save_sm_root failed", __func__);
991 		return r;
992 	}
993 
994 	r = superblock_lock(md, &sblock);
995 	if (r) {
996 		DMERR("%s: superblock lock failed", __func__);
997 		return r;
998 	}
999 
1000 	prepare_superblock(md, dm_block_data(sblock));
1001 
1002 	return dm_tm_commit(md->tm, sblock);
1003 }
1004 
1005 static int metadata_checkpoint(struct era_metadata *md)
1006 {
1007 	/*
1008 	 * For now we just rollover, but later I want to put a check in to
1009 	 * avoid this if the filter is still pretty fresh.
1010 	 */
1011 	return metadata_era_rollover(md);
1012 }
1013 
1014 /*
1015  * Metadata snapshots allow userland to access era data.
1016  */
1017 static int metadata_take_snap(struct era_metadata *md)
1018 {
1019 	int r, inc;
1020 	struct dm_block *clone;
1021 
1022 	if (md->metadata_snap != SUPERBLOCK_LOCATION) {
1023 		DMERR("%s: metadata snapshot already exists", __func__);
1024 		return -EINVAL;
1025 	}
1026 
1027 	r = metadata_era_rollover(md);
1028 	if (r) {
1029 		DMERR("%s: era rollover failed", __func__);
1030 		return r;
1031 	}
1032 
1033 	r = metadata_commit(md);
1034 	if (r) {
1035 		DMERR("%s: pre commit failed", __func__);
1036 		return r;
1037 	}
1038 
1039 	r = dm_sm_inc_block(md->sm, SUPERBLOCK_LOCATION);
1040 	if (r) {
1041 		DMERR("%s: couldn't increment superblock", __func__);
1042 		return r;
1043 	}
1044 
1045 	r = dm_tm_shadow_block(md->tm, SUPERBLOCK_LOCATION,
1046 			       &sb_validator, &clone, &inc);
1047 	if (r) {
1048 		DMERR("%s: couldn't shadow superblock", __func__);
1049 		dm_sm_dec_block(md->sm, SUPERBLOCK_LOCATION);
1050 		return r;
1051 	}
1052 	BUG_ON(!inc);
1053 
1054 	r = dm_sm_inc_block(md->sm, md->writeset_tree_root);
1055 	if (r) {
1056 		DMERR("%s: couldn't inc writeset tree root", __func__);
1057 		dm_tm_unlock(md->tm, clone);
1058 		return r;
1059 	}
1060 
1061 	r = dm_sm_inc_block(md->sm, md->era_array_root);
1062 	if (r) {
1063 		DMERR("%s: couldn't inc era tree root", __func__);
1064 		dm_sm_dec_block(md->sm, md->writeset_tree_root);
1065 		dm_tm_unlock(md->tm, clone);
1066 		return r;
1067 	}
1068 
1069 	md->metadata_snap = dm_block_location(clone);
1070 
1071 	dm_tm_unlock(md->tm, clone);
1072 
1073 	return 0;
1074 }
1075 
1076 static int metadata_drop_snap(struct era_metadata *md)
1077 {
1078 	int r;
1079 	dm_block_t location;
1080 	struct dm_block *clone;
1081 	struct superblock_disk *disk;
1082 
1083 	if (md->metadata_snap == SUPERBLOCK_LOCATION) {
1084 		DMERR("%s: no snap to drop", __func__);
1085 		return -EINVAL;
1086 	}
1087 
1088 	r = dm_tm_read_lock(md->tm, md->metadata_snap, &sb_validator, &clone);
1089 	if (r) {
1090 		DMERR("%s: couldn't read lock superblock clone", __func__);
1091 		return r;
1092 	}
1093 
1094 	/*
1095 	 * Whatever happens now we'll commit with no record of the metadata
1096 	 * snap.
1097 	 */
1098 	md->metadata_snap = SUPERBLOCK_LOCATION;
1099 
1100 	disk = dm_block_data(clone);
1101 	r = dm_btree_del(&md->writeset_tree_info,
1102 			 le64_to_cpu(disk->writeset_tree_root));
1103 	if (r) {
1104 		DMERR("%s: error deleting writeset tree clone", __func__);
1105 		dm_tm_unlock(md->tm, clone);
1106 		return r;
1107 	}
1108 
1109 	r = dm_array_del(&md->era_array_info, le64_to_cpu(disk->era_array_root));
1110 	if (r) {
1111 		DMERR("%s: error deleting era array clone", __func__);
1112 		dm_tm_unlock(md->tm, clone);
1113 		return r;
1114 	}
1115 
1116 	location = dm_block_location(clone);
1117 	dm_tm_unlock(md->tm, clone);
1118 
1119 	return dm_sm_dec_block(md->sm, location);
1120 }
1121 
1122 struct metadata_stats {
1123 	dm_block_t used;
1124 	dm_block_t total;
1125 	dm_block_t snap;
1126 	uint32_t era;
1127 };
1128 
1129 static int metadata_get_stats(struct era_metadata *md, void *ptr)
1130 {
1131 	int r;
1132 	struct metadata_stats *s = ptr;
1133 	dm_block_t nr_free, nr_total;
1134 
1135 	r = dm_sm_get_nr_free(md->sm, &nr_free);
1136 	if (r) {
1137 		DMERR("dm_sm_get_nr_free returned %d", r);
1138 		return r;
1139 	}
1140 
1141 	r = dm_sm_get_nr_blocks(md->sm, &nr_total);
1142 	if (r) {
1143 		DMERR("dm_pool_get_metadata_dev_size returned %d", r);
1144 		return r;
1145 	}
1146 
1147 	s->used = nr_total - nr_free;
1148 	s->total = nr_total;
1149 	s->snap = md->metadata_snap;
1150 	s->era = md->current_era;
1151 
1152 	return 0;
1153 }
1154 
1155 /*----------------------------------------------------------------*/
1156 
1157 struct era {
1158 	struct dm_target *ti;
1159 
1160 	struct dm_dev *metadata_dev;
1161 	struct dm_dev *origin_dev;
1162 
1163 	dm_block_t nr_blocks;
1164 	uint32_t sectors_per_block;
1165 	int sectors_per_block_shift;
1166 	struct era_metadata *md;
1167 
1168 	struct workqueue_struct *wq;
1169 	struct work_struct worker;
1170 
1171 	spinlock_t deferred_lock;
1172 	struct bio_list deferred_bios;
1173 
1174 	spinlock_t rpc_lock;
1175 	struct list_head rpc_calls;
1176 
1177 	struct digest digest;
1178 	atomic_t suspended;
1179 };
1180 
1181 struct rpc {
1182 	struct list_head list;
1183 
1184 	int (*fn0)(struct era_metadata *);
1185 	int (*fn1)(struct era_metadata *, void *);
1186 	void *arg;
1187 	int result;
1188 
1189 	struct completion complete;
1190 };
1191 
1192 /*----------------------------------------------------------------
1193  * Remapping.
1194  *---------------------------------------------------------------*/
1195 static bool block_size_is_power_of_two(struct era *era)
1196 {
1197 	return era->sectors_per_block_shift >= 0;
1198 }
1199 
1200 static dm_block_t get_block(struct era *era, struct bio *bio)
1201 {
1202 	sector_t block_nr = bio->bi_iter.bi_sector;
1203 
1204 	if (!block_size_is_power_of_two(era))
1205 		(void) sector_div(block_nr, era->sectors_per_block);
1206 	else
1207 		block_nr >>= era->sectors_per_block_shift;
1208 
1209 	return block_nr;
1210 }
1211 
1212 static void remap_to_origin(struct era *era, struct bio *bio)
1213 {
1214 	bio_set_dev(bio, era->origin_dev->bdev);
1215 }
1216 
1217 /*----------------------------------------------------------------
1218  * Worker thread
1219  *--------------------------------------------------------------*/
1220 static void wake_worker(struct era *era)
1221 {
1222 	if (!atomic_read(&era->suspended))
1223 		queue_work(era->wq, &era->worker);
1224 }
1225 
1226 static void process_old_eras(struct era *era)
1227 {
1228 	int r;
1229 
1230 	if (!era->digest.step)
1231 		return;
1232 
1233 	r = era->digest.step(era->md, &era->digest);
1234 	if (r < 0) {
1235 		DMERR("%s: digest step failed, stopping digestion", __func__);
1236 		era->digest.step = NULL;
1237 
1238 	} else if (era->digest.step)
1239 		wake_worker(era);
1240 }
1241 
1242 static void process_deferred_bios(struct era *era)
1243 {
1244 	int r;
1245 	struct bio_list deferred_bios, marked_bios;
1246 	struct bio *bio;
1247 	struct blk_plug plug;
1248 	bool commit_needed = false;
1249 	bool failed = false;
1250 	struct writeset *ws = era->md->current_writeset;
1251 
1252 	bio_list_init(&deferred_bios);
1253 	bio_list_init(&marked_bios);
1254 
1255 	spin_lock(&era->deferred_lock);
1256 	bio_list_merge(&deferred_bios, &era->deferred_bios);
1257 	bio_list_init(&era->deferred_bios);
1258 	spin_unlock(&era->deferred_lock);
1259 
1260 	if (bio_list_empty(&deferred_bios))
1261 		return;
1262 
1263 	while ((bio = bio_list_pop(&deferred_bios))) {
1264 		r = writeset_test_and_set(&era->md->bitset_info, ws,
1265 					  get_block(era, bio));
1266 		if (r < 0) {
1267 			/*
1268 			 * This is bad news, we need to rollback.
1269 			 * FIXME: finish.
1270 			 */
1271 			failed = true;
1272 		} else if (r == 0)
1273 			commit_needed = true;
1274 
1275 		bio_list_add(&marked_bios, bio);
1276 	}
1277 
1278 	if (commit_needed) {
1279 		r = metadata_commit(era->md);
1280 		if (r)
1281 			failed = true;
1282 	}
1283 
1284 	if (failed)
1285 		while ((bio = bio_list_pop(&marked_bios)))
1286 			bio_io_error(bio);
1287 	else {
1288 		blk_start_plug(&plug);
1289 		while ((bio = bio_list_pop(&marked_bios))) {
1290 			/*
1291 			 * Only update the in-core writeset if the on-disk one
1292 			 * was updated too.
1293 			 */
1294 			if (commit_needed)
1295 				set_bit(get_block(era, bio), ws->bits);
1296 			submit_bio_noacct(bio);
1297 		}
1298 		blk_finish_plug(&plug);
1299 	}
1300 }
1301 
1302 static void process_rpc_calls(struct era *era)
1303 {
1304 	int r;
1305 	bool need_commit = false;
1306 	struct list_head calls;
1307 	struct rpc *rpc, *tmp;
1308 
1309 	INIT_LIST_HEAD(&calls);
1310 	spin_lock(&era->rpc_lock);
1311 	list_splice_init(&era->rpc_calls, &calls);
1312 	spin_unlock(&era->rpc_lock);
1313 
1314 	list_for_each_entry_safe(rpc, tmp, &calls, list) {
1315 		rpc->result = rpc->fn0 ? rpc->fn0(era->md) : rpc->fn1(era->md, rpc->arg);
1316 		need_commit = true;
1317 	}
1318 
1319 	if (need_commit) {
1320 		r = metadata_commit(era->md);
1321 		if (r)
1322 			list_for_each_entry_safe(rpc, tmp, &calls, list)
1323 				rpc->result = r;
1324 	}
1325 
1326 	list_for_each_entry_safe(rpc, tmp, &calls, list)
1327 		complete(&rpc->complete);
1328 }
1329 
1330 static void kick_off_digest(struct era *era)
1331 {
1332 	if (era->md->archived_writesets) {
1333 		era->md->archived_writesets = false;
1334 		metadata_digest_start(era->md, &era->digest);
1335 	}
1336 }
1337 
1338 static void do_work(struct work_struct *ws)
1339 {
1340 	struct era *era = container_of(ws, struct era, worker);
1341 
1342 	kick_off_digest(era);
1343 	process_old_eras(era);
1344 	process_deferred_bios(era);
1345 	process_rpc_calls(era);
1346 }
1347 
1348 static void defer_bio(struct era *era, struct bio *bio)
1349 {
1350 	spin_lock(&era->deferred_lock);
1351 	bio_list_add(&era->deferred_bios, bio);
1352 	spin_unlock(&era->deferred_lock);
1353 
1354 	wake_worker(era);
1355 }
1356 
1357 /*
1358  * Make an rpc call to the worker to change the metadata.
1359  */
1360 static int perform_rpc(struct era *era, struct rpc *rpc)
1361 {
1362 	rpc->result = 0;
1363 	init_completion(&rpc->complete);
1364 
1365 	spin_lock(&era->rpc_lock);
1366 	list_add(&rpc->list, &era->rpc_calls);
1367 	spin_unlock(&era->rpc_lock);
1368 
1369 	wake_worker(era);
1370 	wait_for_completion(&rpc->complete);
1371 
1372 	return rpc->result;
1373 }
1374 
1375 static int in_worker0(struct era *era, int (*fn)(struct era_metadata *))
1376 {
1377 	struct rpc rpc;
1378 	rpc.fn0 = fn;
1379 	rpc.fn1 = NULL;
1380 
1381 	return perform_rpc(era, &rpc);
1382 }
1383 
1384 static int in_worker1(struct era *era,
1385 		      int (*fn)(struct era_metadata *, void *), void *arg)
1386 {
1387 	struct rpc rpc;
1388 	rpc.fn0 = NULL;
1389 	rpc.fn1 = fn;
1390 	rpc.arg = arg;
1391 
1392 	return perform_rpc(era, &rpc);
1393 }
1394 
1395 static void start_worker(struct era *era)
1396 {
1397 	atomic_set(&era->suspended, 0);
1398 }
1399 
1400 static void stop_worker(struct era *era)
1401 {
1402 	atomic_set(&era->suspended, 1);
1403 	flush_workqueue(era->wq);
1404 }
1405 
1406 /*----------------------------------------------------------------
1407  * Target methods
1408  *--------------------------------------------------------------*/
1409 static void era_destroy(struct era *era)
1410 {
1411 	if (era->md)
1412 		metadata_close(era->md);
1413 
1414 	if (era->wq)
1415 		destroy_workqueue(era->wq);
1416 
1417 	if (era->origin_dev)
1418 		dm_put_device(era->ti, era->origin_dev);
1419 
1420 	if (era->metadata_dev)
1421 		dm_put_device(era->ti, era->metadata_dev);
1422 
1423 	kfree(era);
1424 }
1425 
1426 static dm_block_t calc_nr_blocks(struct era *era)
1427 {
1428 	return dm_sector_div_up(era->ti->len, era->sectors_per_block);
1429 }
1430 
1431 static bool valid_block_size(dm_block_t block_size)
1432 {
1433 	bool greater_than_zero = block_size > 0;
1434 	bool multiple_of_min_block_size = (block_size & (MIN_BLOCK_SIZE - 1)) == 0;
1435 
1436 	return greater_than_zero && multiple_of_min_block_size;
1437 }
1438 
1439 /*
1440  * <metadata dev> <data dev> <data block size (sectors)>
1441  */
1442 static int era_ctr(struct dm_target *ti, unsigned argc, char **argv)
1443 {
1444 	int r;
1445 	char dummy;
1446 	struct era *era;
1447 	struct era_metadata *md;
1448 
1449 	if (argc != 3) {
1450 		ti->error = "Invalid argument count";
1451 		return -EINVAL;
1452 	}
1453 
1454 	era = kzalloc(sizeof(*era), GFP_KERNEL);
1455 	if (!era) {
1456 		ti->error = "Error allocating era structure";
1457 		return -ENOMEM;
1458 	}
1459 
1460 	era->ti = ti;
1461 
1462 	r = dm_get_device(ti, argv[0], FMODE_READ | FMODE_WRITE, &era->metadata_dev);
1463 	if (r) {
1464 		ti->error = "Error opening metadata device";
1465 		era_destroy(era);
1466 		return -EINVAL;
1467 	}
1468 
1469 	r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &era->origin_dev);
1470 	if (r) {
1471 		ti->error = "Error opening data device";
1472 		era_destroy(era);
1473 		return -EINVAL;
1474 	}
1475 
1476 	r = sscanf(argv[2], "%u%c", &era->sectors_per_block, &dummy);
1477 	if (r != 1) {
1478 		ti->error = "Error parsing block size";
1479 		era_destroy(era);
1480 		return -EINVAL;
1481 	}
1482 
1483 	r = dm_set_target_max_io_len(ti, era->sectors_per_block);
1484 	if (r) {
1485 		ti->error = "could not set max io len";
1486 		era_destroy(era);
1487 		return -EINVAL;
1488 	}
1489 
1490 	if (!valid_block_size(era->sectors_per_block)) {
1491 		ti->error = "Invalid block size";
1492 		era_destroy(era);
1493 		return -EINVAL;
1494 	}
1495 	if (era->sectors_per_block & (era->sectors_per_block - 1))
1496 		era->sectors_per_block_shift = -1;
1497 	else
1498 		era->sectors_per_block_shift = __ffs(era->sectors_per_block);
1499 
1500 	md = metadata_open(era->metadata_dev->bdev, era->sectors_per_block, true);
1501 	if (IS_ERR(md)) {
1502 		ti->error = "Error reading metadata";
1503 		era_destroy(era);
1504 		return PTR_ERR(md);
1505 	}
1506 	era->md = md;
1507 
1508 	era->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
1509 	if (!era->wq) {
1510 		ti->error = "could not create workqueue for metadata object";
1511 		era_destroy(era);
1512 		return -ENOMEM;
1513 	}
1514 	INIT_WORK(&era->worker, do_work);
1515 
1516 	spin_lock_init(&era->deferred_lock);
1517 	bio_list_init(&era->deferred_bios);
1518 
1519 	spin_lock_init(&era->rpc_lock);
1520 	INIT_LIST_HEAD(&era->rpc_calls);
1521 
1522 	ti->private = era;
1523 	ti->num_flush_bios = 1;
1524 	ti->flush_supported = true;
1525 
1526 	ti->num_discard_bios = 1;
1527 
1528 	return 0;
1529 }
1530 
1531 static void era_dtr(struct dm_target *ti)
1532 {
1533 	era_destroy(ti->private);
1534 }
1535 
1536 static int era_map(struct dm_target *ti, struct bio *bio)
1537 {
1538 	struct era *era = ti->private;
1539 	dm_block_t block = get_block(era, bio);
1540 
1541 	/*
1542 	 * All bios get remapped to the origin device.  We do this now, but
1543 	 * it may not get issued until later.  Depending on whether the
1544 	 * block is marked in this era.
1545 	 */
1546 	remap_to_origin(era, bio);
1547 
1548 	/*
1549 	 * REQ_PREFLUSH bios carry no data, so we're not interested in them.
1550 	 */
1551 	if (!(bio->bi_opf & REQ_PREFLUSH) &&
1552 	    (bio_data_dir(bio) == WRITE) &&
1553 	    !metadata_current_marked(era->md, block)) {
1554 		defer_bio(era, bio);
1555 		return DM_MAPIO_SUBMITTED;
1556 	}
1557 
1558 	return DM_MAPIO_REMAPPED;
1559 }
1560 
1561 static void era_postsuspend(struct dm_target *ti)
1562 {
1563 	int r;
1564 	struct era *era = ti->private;
1565 
1566 	r = in_worker0(era, metadata_era_archive);
1567 	if (r) {
1568 		DMERR("%s: couldn't archive current era", __func__);
1569 		/* FIXME: fail mode */
1570 	}
1571 
1572 	stop_worker(era);
1573 }
1574 
1575 static int era_preresume(struct dm_target *ti)
1576 {
1577 	int r;
1578 	struct era *era = ti->private;
1579 	dm_block_t new_size = calc_nr_blocks(era);
1580 
1581 	if (era->nr_blocks != new_size) {
1582 		r = metadata_resize(era->md, &new_size);
1583 		if (r) {
1584 			DMERR("%s: metadata_resize failed", __func__);
1585 			return r;
1586 		}
1587 
1588 		r = metadata_commit(era->md);
1589 		if (r) {
1590 			DMERR("%s: metadata_commit failed", __func__);
1591 			return r;
1592 		}
1593 
1594 		era->nr_blocks = new_size;
1595 	}
1596 
1597 	start_worker(era);
1598 
1599 	r = in_worker0(era, metadata_era_rollover);
1600 	if (r) {
1601 		DMERR("%s: metadata_era_rollover failed", __func__);
1602 		return r;
1603 	}
1604 
1605 	return 0;
1606 }
1607 
1608 /*
1609  * Status format:
1610  *
1611  * <metadata block size> <#used metadata blocks>/<#total metadata blocks>
1612  * <current era> <held metadata root | '-'>
1613  */
1614 static void era_status(struct dm_target *ti, status_type_t type,
1615 		       unsigned status_flags, char *result, unsigned maxlen)
1616 {
1617 	int r;
1618 	struct era *era = ti->private;
1619 	ssize_t sz = 0;
1620 	struct metadata_stats stats;
1621 	char buf[BDEVNAME_SIZE];
1622 
1623 	switch (type) {
1624 	case STATUSTYPE_INFO:
1625 		r = in_worker1(era, metadata_get_stats, &stats);
1626 		if (r)
1627 			goto err;
1628 
1629 		DMEMIT("%u %llu/%llu %u",
1630 		       (unsigned) (DM_ERA_METADATA_BLOCK_SIZE >> SECTOR_SHIFT),
1631 		       (unsigned long long) stats.used,
1632 		       (unsigned long long) stats.total,
1633 		       (unsigned) stats.era);
1634 
1635 		if (stats.snap != SUPERBLOCK_LOCATION)
1636 			DMEMIT(" %llu", stats.snap);
1637 		else
1638 			DMEMIT(" -");
1639 		break;
1640 
1641 	case STATUSTYPE_TABLE:
1642 		format_dev_t(buf, era->metadata_dev->bdev->bd_dev);
1643 		DMEMIT("%s ", buf);
1644 		format_dev_t(buf, era->origin_dev->bdev->bd_dev);
1645 		DMEMIT("%s %u", buf, era->sectors_per_block);
1646 		break;
1647 
1648 	case STATUSTYPE_IMA:
1649 		*result = '\0';
1650 		break;
1651 	}
1652 
1653 	return;
1654 
1655 err:
1656 	DMEMIT("Error");
1657 }
1658 
1659 static int era_message(struct dm_target *ti, unsigned argc, char **argv,
1660 		       char *result, unsigned maxlen)
1661 {
1662 	struct era *era = ti->private;
1663 
1664 	if (argc != 1) {
1665 		DMERR("incorrect number of message arguments");
1666 		return -EINVAL;
1667 	}
1668 
1669 	if (!strcasecmp(argv[0], "checkpoint"))
1670 		return in_worker0(era, metadata_checkpoint);
1671 
1672 	if (!strcasecmp(argv[0], "take_metadata_snap"))
1673 		return in_worker0(era, metadata_take_snap);
1674 
1675 	if (!strcasecmp(argv[0], "drop_metadata_snap"))
1676 		return in_worker0(era, metadata_drop_snap);
1677 
1678 	DMERR("unsupported message '%s'", argv[0]);
1679 	return -EINVAL;
1680 }
1681 
1682 static sector_t get_dev_size(struct dm_dev *dev)
1683 {
1684 	return bdev_nr_sectors(dev->bdev);
1685 }
1686 
1687 static int era_iterate_devices(struct dm_target *ti,
1688 			       iterate_devices_callout_fn fn, void *data)
1689 {
1690 	struct era *era = ti->private;
1691 	return fn(ti, era->origin_dev, 0, get_dev_size(era->origin_dev), data);
1692 }
1693 
1694 static void era_io_hints(struct dm_target *ti, struct queue_limits *limits)
1695 {
1696 	struct era *era = ti->private;
1697 	uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
1698 
1699 	/*
1700 	 * If the system-determined stacked limits are compatible with the
1701 	 * era device's blocksize (io_opt is a factor) do not override them.
1702 	 */
1703 	if (io_opt_sectors < era->sectors_per_block ||
1704 	    do_div(io_opt_sectors, era->sectors_per_block)) {
1705 		blk_limits_io_min(limits, 0);
1706 		blk_limits_io_opt(limits, era->sectors_per_block << SECTOR_SHIFT);
1707 	}
1708 }
1709 
1710 /*----------------------------------------------------------------*/
1711 
1712 static struct target_type era_target = {
1713 	.name = "era",
1714 	.version = {1, 0, 0},
1715 	.module = THIS_MODULE,
1716 	.ctr = era_ctr,
1717 	.dtr = era_dtr,
1718 	.map = era_map,
1719 	.postsuspend = era_postsuspend,
1720 	.preresume = era_preresume,
1721 	.status = era_status,
1722 	.message = era_message,
1723 	.iterate_devices = era_iterate_devices,
1724 	.io_hints = era_io_hints
1725 };
1726 
1727 static int __init dm_era_init(void)
1728 {
1729 	int r;
1730 
1731 	r = dm_register_target(&era_target);
1732 	if (r) {
1733 		DMERR("era target registration failed: %d", r);
1734 		return r;
1735 	}
1736 
1737 	return 0;
1738 }
1739 
1740 static void __exit dm_era_exit(void)
1741 {
1742 	dm_unregister_target(&era_target);
1743 }
1744 
1745 module_init(dm_era_init);
1746 module_exit(dm_era_exit);
1747 
1748 MODULE_DESCRIPTION(DM_NAME " era target");
1749 MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>");
1750 MODULE_LICENSE("GPL");
1751