1 /*
2  * Copyright (C) 2012 Red Hat, Inc.
3  *
4  * This file is released under the GPL.
5  */
6 
7 #include "dm-cache-metadata.h"
8 
9 #include "persistent-data/dm-array.h"
10 #include "persistent-data/dm-bitset.h"
11 #include "persistent-data/dm-space-map.h"
12 #include "persistent-data/dm-space-map-disk.h"
13 #include "persistent-data/dm-transaction-manager.h"
14 
15 #include <linux/device-mapper.h>
16 #include <linux/refcount.h>
17 
18 /*----------------------------------------------------------------*/
19 
20 #define DM_MSG_PREFIX   "cache metadata"
21 
22 #define CACHE_SUPERBLOCK_MAGIC 06142003
23 #define CACHE_SUPERBLOCK_LOCATION 0
24 
25 /*
26  * defines a range of metadata versions that this module can handle.
27  */
28 #define MIN_CACHE_VERSION 1
29 #define MAX_CACHE_VERSION 2
30 
31 /*
32  *  3 for btree insert +
33  *  2 for btree lookup used within space map
34  */
35 #define CACHE_MAX_CONCURRENT_LOCKS 5
36 #define SPACE_MAP_ROOT_SIZE 128
37 
38 enum superblock_flag_bits {
39 	/* for spotting crashes that would invalidate the dirty bitset */
40 	CLEAN_SHUTDOWN,
41 	/* metadata must be checked using the tools */
42 	NEEDS_CHECK,
43 };
44 
45 /*
46  * Each mapping from cache block -> origin block carries a set of flags.
47  */
48 enum mapping_bits {
49 	/*
50 	 * A valid mapping.  Because we're using an array we clear this
51 	 * flag for an non existant mapping.
52 	 */
53 	M_VALID = 1,
54 
55 	/*
56 	 * The data on the cache is different from that on the origin.
57 	 * This flag is only used by metadata format 1.
58 	 */
59 	M_DIRTY = 2
60 };
61 
62 struct cache_disk_superblock {
63 	__le32 csum;
64 	__le32 flags;
65 	__le64 blocknr;
66 
67 	__u8 uuid[16];
68 	__le64 magic;
69 	__le32 version;
70 
71 	__u8 policy_name[CACHE_POLICY_NAME_SIZE];
72 	__le32 policy_hint_size;
73 
74 	__u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
75 	__le64 mapping_root;
76 	__le64 hint_root;
77 
78 	__le64 discard_root;
79 	__le64 discard_block_size;
80 	__le64 discard_nr_blocks;
81 
82 	__le32 data_block_size;
83 	__le32 metadata_block_size;
84 	__le32 cache_blocks;
85 
86 	__le32 compat_flags;
87 	__le32 compat_ro_flags;
88 	__le32 incompat_flags;
89 
90 	__le32 read_hits;
91 	__le32 read_misses;
92 	__le32 write_hits;
93 	__le32 write_misses;
94 
95 	__le32 policy_version[CACHE_POLICY_VERSION_SIZE];
96 
97 	/*
98 	 * Metadata format 2 fields.
99 	 */
100 	__le64 dirty_root;
101 } __packed;
102 
103 struct dm_cache_metadata {
104 	refcount_t ref_count;
105 	struct list_head list;
106 
107 	unsigned version;
108 	struct block_device *bdev;
109 	struct dm_block_manager *bm;
110 	struct dm_space_map *metadata_sm;
111 	struct dm_transaction_manager *tm;
112 
113 	struct dm_array_info info;
114 	struct dm_array_info hint_info;
115 	struct dm_disk_bitset discard_info;
116 
117 	struct rw_semaphore root_lock;
118 	unsigned long flags;
119 	dm_block_t root;
120 	dm_block_t hint_root;
121 	dm_block_t discard_root;
122 
123 	sector_t discard_block_size;
124 	dm_dblock_t discard_nr_blocks;
125 
126 	sector_t data_block_size;
127 	dm_cblock_t cache_blocks;
128 	bool changed:1;
129 	bool clean_when_opened:1;
130 
131 	char policy_name[CACHE_POLICY_NAME_SIZE];
132 	unsigned policy_version[CACHE_POLICY_VERSION_SIZE];
133 	size_t policy_hint_size;
134 	struct dm_cache_statistics stats;
135 
136 	/*
137 	 * Reading the space map root can fail, so we read it into this
138 	 * buffer before the superblock is locked and updated.
139 	 */
140 	__u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
141 
142 	/*
143 	 * Set if a transaction has to be aborted but the attempt to roll
144 	 * back to the previous (good) transaction failed.  The only
145 	 * metadata operation permissible in this state is the closing of
146 	 * the device.
147 	 */
148 	bool fail_io:1;
149 
150 	/*
151 	 * Metadata format 2 fields.
152 	 */
153 	dm_block_t dirty_root;
154 	struct dm_disk_bitset dirty_info;
155 
156 	/*
157 	 * These structures are used when loading metadata.  They're too
158 	 * big to put on the stack.
159 	 */
160 	struct dm_array_cursor mapping_cursor;
161 	struct dm_array_cursor hint_cursor;
162 	struct dm_bitset_cursor dirty_cursor;
163 };
164 
165 /*-------------------------------------------------------------------
166  * superblock validator
167  *-----------------------------------------------------------------*/
168 
169 #define SUPERBLOCK_CSUM_XOR 9031977
170 
171 static void sb_prepare_for_write(struct dm_block_validator *v,
172 				 struct dm_block *b,
173 				 size_t sb_block_size)
174 {
175 	struct cache_disk_superblock *disk_super = dm_block_data(b);
176 
177 	disk_super->blocknr = cpu_to_le64(dm_block_location(b));
178 	disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
179 						      sb_block_size - sizeof(__le32),
180 						      SUPERBLOCK_CSUM_XOR));
181 }
182 
183 static int check_metadata_version(struct cache_disk_superblock *disk_super)
184 {
185 	uint32_t metadata_version = le32_to_cpu(disk_super->version);
186 
187 	if (metadata_version < MIN_CACHE_VERSION || metadata_version > MAX_CACHE_VERSION) {
188 		DMERR("Cache metadata version %u found, but only versions between %u and %u supported.",
189 		      metadata_version, MIN_CACHE_VERSION, MAX_CACHE_VERSION);
190 		return -EINVAL;
191 	}
192 
193 	return 0;
194 }
195 
196 static int sb_check(struct dm_block_validator *v,
197 		    struct dm_block *b,
198 		    size_t sb_block_size)
199 {
200 	struct cache_disk_superblock *disk_super = dm_block_data(b);
201 	__le32 csum_le;
202 
203 	if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
204 		DMERR("sb_check failed: blocknr %llu: wanted %llu",
205 		      le64_to_cpu(disk_super->blocknr),
206 		      (unsigned long long)dm_block_location(b));
207 		return -ENOTBLK;
208 	}
209 
210 	if (le64_to_cpu(disk_super->magic) != CACHE_SUPERBLOCK_MAGIC) {
211 		DMERR("sb_check failed: magic %llu: wanted %llu",
212 		      le64_to_cpu(disk_super->magic),
213 		      (unsigned long long)CACHE_SUPERBLOCK_MAGIC);
214 		return -EILSEQ;
215 	}
216 
217 	csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
218 					     sb_block_size - sizeof(__le32),
219 					     SUPERBLOCK_CSUM_XOR));
220 	if (csum_le != disk_super->csum) {
221 		DMERR("sb_check failed: csum %u: wanted %u",
222 		      le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
223 		return -EILSEQ;
224 	}
225 
226 	return check_metadata_version(disk_super);
227 }
228 
229 static struct dm_block_validator sb_validator = {
230 	.name = "superblock",
231 	.prepare_for_write = sb_prepare_for_write,
232 	.check = sb_check
233 };
234 
235 /*----------------------------------------------------------------*/
236 
237 static int superblock_read_lock(struct dm_cache_metadata *cmd,
238 				struct dm_block **sblock)
239 {
240 	return dm_bm_read_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
241 			       &sb_validator, sblock);
242 }
243 
244 static int superblock_lock_zero(struct dm_cache_metadata *cmd,
245 				struct dm_block **sblock)
246 {
247 	return dm_bm_write_lock_zero(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
248 				     &sb_validator, sblock);
249 }
250 
251 static int superblock_lock(struct dm_cache_metadata *cmd,
252 			   struct dm_block **sblock)
253 {
254 	return dm_bm_write_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
255 				&sb_validator, sblock);
256 }
257 
258 /*----------------------------------------------------------------*/
259 
260 static int __superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
261 {
262 	int r;
263 	unsigned i;
264 	struct dm_block *b;
265 	__le64 *data_le, zero = cpu_to_le64(0);
266 	unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
267 
268 	/*
269 	 * We can't use a validator here - it may be all zeroes.
270 	 */
271 	r = dm_bm_read_lock(bm, CACHE_SUPERBLOCK_LOCATION, NULL, &b);
272 	if (r)
273 		return r;
274 
275 	data_le = dm_block_data(b);
276 	*result = true;
277 	for (i = 0; i < sb_block_size; i++) {
278 		if (data_le[i] != zero) {
279 			*result = false;
280 			break;
281 		}
282 	}
283 
284 	dm_bm_unlock(b);
285 
286 	return 0;
287 }
288 
289 static void __setup_mapping_info(struct dm_cache_metadata *cmd)
290 {
291 	struct dm_btree_value_type vt;
292 
293 	vt.context = NULL;
294 	vt.size = sizeof(__le64);
295 	vt.inc = NULL;
296 	vt.dec = NULL;
297 	vt.equal = NULL;
298 	dm_array_info_init(&cmd->info, cmd->tm, &vt);
299 
300 	if (cmd->policy_hint_size) {
301 		vt.size = sizeof(__le32);
302 		dm_array_info_init(&cmd->hint_info, cmd->tm, &vt);
303 	}
304 }
305 
306 static int __save_sm_root(struct dm_cache_metadata *cmd)
307 {
308 	int r;
309 	size_t metadata_len;
310 
311 	r = dm_sm_root_size(cmd->metadata_sm, &metadata_len);
312 	if (r < 0)
313 		return r;
314 
315 	return dm_sm_copy_root(cmd->metadata_sm, &cmd->metadata_space_map_root,
316 			       metadata_len);
317 }
318 
319 static void __copy_sm_root(struct dm_cache_metadata *cmd,
320 			   struct cache_disk_superblock *disk_super)
321 {
322 	memcpy(&disk_super->metadata_space_map_root,
323 	       &cmd->metadata_space_map_root,
324 	       sizeof(cmd->metadata_space_map_root));
325 }
326 
327 static bool separate_dirty_bits(struct dm_cache_metadata *cmd)
328 {
329 	return cmd->version >= 2;
330 }
331 
332 static int __write_initial_superblock(struct dm_cache_metadata *cmd)
333 {
334 	int r;
335 	struct dm_block *sblock;
336 	struct cache_disk_superblock *disk_super;
337 	sector_t bdev_size = i_size_read(cmd->bdev->bd_inode) >> SECTOR_SHIFT;
338 
339 	/* FIXME: see if we can lose the max sectors limit */
340 	if (bdev_size > DM_CACHE_METADATA_MAX_SECTORS)
341 		bdev_size = DM_CACHE_METADATA_MAX_SECTORS;
342 
343 	r = dm_tm_pre_commit(cmd->tm);
344 	if (r < 0)
345 		return r;
346 
347 	/*
348 	 * dm_sm_copy_root() can fail.  So we need to do it before we start
349 	 * updating the superblock.
350 	 */
351 	r = __save_sm_root(cmd);
352 	if (r)
353 		return r;
354 
355 	r = superblock_lock_zero(cmd, &sblock);
356 	if (r)
357 		return r;
358 
359 	disk_super = dm_block_data(sblock);
360 	disk_super->flags = 0;
361 	memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
362 	disk_super->magic = cpu_to_le64(CACHE_SUPERBLOCK_MAGIC);
363 	disk_super->version = cpu_to_le32(cmd->version);
364 	memset(disk_super->policy_name, 0, sizeof(disk_super->policy_name));
365 	memset(disk_super->policy_version, 0, sizeof(disk_super->policy_version));
366 	disk_super->policy_hint_size = 0;
367 
368 	__copy_sm_root(cmd, disk_super);
369 
370 	disk_super->mapping_root = cpu_to_le64(cmd->root);
371 	disk_super->hint_root = cpu_to_le64(cmd->hint_root);
372 	disk_super->discard_root = cpu_to_le64(cmd->discard_root);
373 	disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
374 	disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
375 	disk_super->metadata_block_size = cpu_to_le32(DM_CACHE_METADATA_BLOCK_SIZE);
376 	disk_super->data_block_size = cpu_to_le32(cmd->data_block_size);
377 	disk_super->cache_blocks = cpu_to_le32(0);
378 
379 	disk_super->read_hits = cpu_to_le32(0);
380 	disk_super->read_misses = cpu_to_le32(0);
381 	disk_super->write_hits = cpu_to_le32(0);
382 	disk_super->write_misses = cpu_to_le32(0);
383 
384 	if (separate_dirty_bits(cmd))
385 		disk_super->dirty_root = cpu_to_le64(cmd->dirty_root);
386 
387 	return dm_tm_commit(cmd->tm, sblock);
388 }
389 
390 static int __format_metadata(struct dm_cache_metadata *cmd)
391 {
392 	int r;
393 
394 	r = dm_tm_create_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
395 				 &cmd->tm, &cmd->metadata_sm);
396 	if (r < 0) {
397 		DMERR("tm_create_with_sm failed");
398 		return r;
399 	}
400 
401 	__setup_mapping_info(cmd);
402 
403 	r = dm_array_empty(&cmd->info, &cmd->root);
404 	if (r < 0)
405 		goto bad;
406 
407 	if (separate_dirty_bits(cmd)) {
408 		dm_disk_bitset_init(cmd->tm, &cmd->dirty_info);
409 		r = dm_bitset_empty(&cmd->dirty_info, &cmd->dirty_root);
410 		if (r < 0)
411 			goto bad;
412 	}
413 
414 	dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
415 	r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root);
416 	if (r < 0)
417 		goto bad;
418 
419 	cmd->discard_block_size = 0;
420 	cmd->discard_nr_blocks = 0;
421 
422 	r = __write_initial_superblock(cmd);
423 	if (r)
424 		goto bad;
425 
426 	cmd->clean_when_opened = true;
427 	return 0;
428 
429 bad:
430 	dm_tm_destroy(cmd->tm);
431 	dm_sm_destroy(cmd->metadata_sm);
432 
433 	return r;
434 }
435 
436 static int __check_incompat_features(struct cache_disk_superblock *disk_super,
437 				     struct dm_cache_metadata *cmd)
438 {
439 	uint32_t incompat_flags, features;
440 
441 	incompat_flags = le32_to_cpu(disk_super->incompat_flags);
442 	features = incompat_flags & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
443 	if (features) {
444 		DMERR("could not access metadata due to unsupported optional features (%lx).",
445 		      (unsigned long)features);
446 		return -EINVAL;
447 	}
448 
449 	/*
450 	 * Check for read-only metadata to skip the following RDWR checks.
451 	 */
452 	if (get_disk_ro(cmd->bdev->bd_disk))
453 		return 0;
454 
455 	features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP;
456 	if (features) {
457 		DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
458 		      (unsigned long)features);
459 		return -EINVAL;
460 	}
461 
462 	return 0;
463 }
464 
465 static int __open_metadata(struct dm_cache_metadata *cmd)
466 {
467 	int r;
468 	struct dm_block *sblock;
469 	struct cache_disk_superblock *disk_super;
470 	unsigned long sb_flags;
471 
472 	r = superblock_read_lock(cmd, &sblock);
473 	if (r < 0) {
474 		DMERR("couldn't read lock superblock");
475 		return r;
476 	}
477 
478 	disk_super = dm_block_data(sblock);
479 
480 	/* Verify the data block size hasn't changed */
481 	if (le32_to_cpu(disk_super->data_block_size) != cmd->data_block_size) {
482 		DMERR("changing the data block size (from %u to %llu) is not supported",
483 		      le32_to_cpu(disk_super->data_block_size),
484 		      (unsigned long long)cmd->data_block_size);
485 		r = -EINVAL;
486 		goto bad;
487 	}
488 
489 	r = __check_incompat_features(disk_super, cmd);
490 	if (r < 0)
491 		goto bad;
492 
493 	r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
494 			       disk_super->metadata_space_map_root,
495 			       sizeof(disk_super->metadata_space_map_root),
496 			       &cmd->tm, &cmd->metadata_sm);
497 	if (r < 0) {
498 		DMERR("tm_open_with_sm failed");
499 		goto bad;
500 	}
501 
502 	__setup_mapping_info(cmd);
503 	dm_disk_bitset_init(cmd->tm, &cmd->dirty_info);
504 	dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
505 	sb_flags = le32_to_cpu(disk_super->flags);
506 	cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags);
507 	dm_bm_unlock(sblock);
508 
509 	return 0;
510 
511 bad:
512 	dm_bm_unlock(sblock);
513 	return r;
514 }
515 
516 static int __open_or_format_metadata(struct dm_cache_metadata *cmd,
517 				     bool format_device)
518 {
519 	int r;
520 	bool unformatted = false;
521 
522 	r = __superblock_all_zeroes(cmd->bm, &unformatted);
523 	if (r)
524 		return r;
525 
526 	if (unformatted)
527 		return format_device ? __format_metadata(cmd) : -EPERM;
528 
529 	return __open_metadata(cmd);
530 }
531 
532 static int __create_persistent_data_objects(struct dm_cache_metadata *cmd,
533 					    bool may_format_device)
534 {
535 	int r;
536 	cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
537 					  CACHE_MAX_CONCURRENT_LOCKS);
538 	if (IS_ERR(cmd->bm)) {
539 		DMERR("could not create block manager");
540 		return PTR_ERR(cmd->bm);
541 	}
542 
543 	r = __open_or_format_metadata(cmd, may_format_device);
544 	if (r)
545 		dm_block_manager_destroy(cmd->bm);
546 
547 	return r;
548 }
549 
550 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd)
551 {
552 	dm_sm_destroy(cmd->metadata_sm);
553 	dm_tm_destroy(cmd->tm);
554 	dm_block_manager_destroy(cmd->bm);
555 }
556 
557 typedef unsigned long (*flags_mutator)(unsigned long);
558 
559 static void update_flags(struct cache_disk_superblock *disk_super,
560 			 flags_mutator mutator)
561 {
562 	uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
563 	disk_super->flags = cpu_to_le32(sb_flags);
564 }
565 
566 static unsigned long set_clean_shutdown(unsigned long flags)
567 {
568 	set_bit(CLEAN_SHUTDOWN, &flags);
569 	return flags;
570 }
571 
572 static unsigned long clear_clean_shutdown(unsigned long flags)
573 {
574 	clear_bit(CLEAN_SHUTDOWN, &flags);
575 	return flags;
576 }
577 
578 static void read_superblock_fields(struct dm_cache_metadata *cmd,
579 				   struct cache_disk_superblock *disk_super)
580 {
581 	cmd->version = le32_to_cpu(disk_super->version);
582 	cmd->flags = le32_to_cpu(disk_super->flags);
583 	cmd->root = le64_to_cpu(disk_super->mapping_root);
584 	cmd->hint_root = le64_to_cpu(disk_super->hint_root);
585 	cmd->discard_root = le64_to_cpu(disk_super->discard_root);
586 	cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
587 	cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
588 	cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
589 	cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
590 	strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
591 	cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
592 	cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
593 	cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
594 	cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
595 
596 	cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
597 	cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
598 	cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
599 	cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
600 
601 	if (separate_dirty_bits(cmd))
602 		cmd->dirty_root = le64_to_cpu(disk_super->dirty_root);
603 
604 	cmd->changed = false;
605 }
606 
607 /*
608  * The mutator updates the superblock flags.
609  */
610 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
611 				     flags_mutator mutator)
612 {
613 	int r;
614 	struct cache_disk_superblock *disk_super;
615 	struct dm_block *sblock;
616 
617 	r = superblock_lock(cmd, &sblock);
618 	if (r)
619 		return r;
620 
621 	disk_super = dm_block_data(sblock);
622 	update_flags(disk_super, mutator);
623 	read_superblock_fields(cmd, disk_super);
624 	dm_bm_unlock(sblock);
625 
626 	return dm_bm_flush(cmd->bm);
627 }
628 
629 static int __begin_transaction(struct dm_cache_metadata *cmd)
630 {
631 	int r;
632 	struct cache_disk_superblock *disk_super;
633 	struct dm_block *sblock;
634 
635 	/*
636 	 * We re-read the superblock every time.  Shouldn't need to do this
637 	 * really.
638 	 */
639 	r = superblock_read_lock(cmd, &sblock);
640 	if (r)
641 		return r;
642 
643 	disk_super = dm_block_data(sblock);
644 	read_superblock_fields(cmd, disk_super);
645 	dm_bm_unlock(sblock);
646 
647 	return 0;
648 }
649 
650 static int __commit_transaction(struct dm_cache_metadata *cmd,
651 				flags_mutator mutator)
652 {
653 	int r;
654 	struct cache_disk_superblock *disk_super;
655 	struct dm_block *sblock;
656 
657 	/*
658 	 * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
659 	 */
660 	BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
661 
662 	if (separate_dirty_bits(cmd)) {
663 		r = dm_bitset_flush(&cmd->dirty_info, cmd->dirty_root,
664 				    &cmd->dirty_root);
665 		if (r)
666 			return r;
667 	}
668 
669 	r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
670 			    &cmd->discard_root);
671 	if (r)
672 		return r;
673 
674 	r = dm_tm_pre_commit(cmd->tm);
675 	if (r < 0)
676 		return r;
677 
678 	r = __save_sm_root(cmd);
679 	if (r)
680 		return r;
681 
682 	r = superblock_lock(cmd, &sblock);
683 	if (r)
684 		return r;
685 
686 	disk_super = dm_block_data(sblock);
687 
688 	disk_super->flags = cpu_to_le32(cmd->flags);
689 	if (mutator)
690 		update_flags(disk_super, mutator);
691 
692 	disk_super->mapping_root = cpu_to_le64(cmd->root);
693 	if (separate_dirty_bits(cmd))
694 		disk_super->dirty_root = cpu_to_le64(cmd->dirty_root);
695 	disk_super->hint_root = cpu_to_le64(cmd->hint_root);
696 	disk_super->discard_root = cpu_to_le64(cmd->discard_root);
697 	disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
698 	disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
699 	disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
700 	strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
701 	disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
702 	disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
703 	disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
704 
705 	disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
706 	disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
707 	disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
708 	disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
709 	__copy_sm_root(cmd, disk_super);
710 
711 	return dm_tm_commit(cmd->tm, sblock);
712 }
713 
714 /*----------------------------------------------------------------*/
715 
716 /*
717  * The mappings are held in a dm-array that has 64-bit values stored in
718  * little-endian format.  The index is the cblock, the high 48bits of the
719  * value are the oblock and the low 16 bit the flags.
720  */
721 #define FLAGS_MASK ((1 << 16) - 1)
722 
723 static __le64 pack_value(dm_oblock_t block, unsigned flags)
724 {
725 	uint64_t value = from_oblock(block);
726 	value <<= 16;
727 	value = value | (flags & FLAGS_MASK);
728 	return cpu_to_le64(value);
729 }
730 
731 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
732 {
733 	uint64_t value = le64_to_cpu(value_le);
734 	uint64_t b = value >> 16;
735 	*block = to_oblock(b);
736 	*flags = value & FLAGS_MASK;
737 }
738 
739 /*----------------------------------------------------------------*/
740 
741 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
742 					       sector_t data_block_size,
743 					       bool may_format_device,
744 					       size_t policy_hint_size,
745 					       unsigned metadata_version)
746 {
747 	int r;
748 	struct dm_cache_metadata *cmd;
749 
750 	cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
751 	if (!cmd) {
752 		DMERR("could not allocate metadata struct");
753 		return ERR_PTR(-ENOMEM);
754 	}
755 
756 	cmd->version = metadata_version;
757 	refcount_set(&cmd->ref_count, 1);
758 	init_rwsem(&cmd->root_lock);
759 	cmd->bdev = bdev;
760 	cmd->data_block_size = data_block_size;
761 	cmd->cache_blocks = 0;
762 	cmd->policy_hint_size = policy_hint_size;
763 	cmd->changed = true;
764 	cmd->fail_io = false;
765 
766 	r = __create_persistent_data_objects(cmd, may_format_device);
767 	if (r) {
768 		kfree(cmd);
769 		return ERR_PTR(r);
770 	}
771 
772 	r = __begin_transaction_flags(cmd, clear_clean_shutdown);
773 	if (r < 0) {
774 		dm_cache_metadata_close(cmd);
775 		return ERR_PTR(r);
776 	}
777 
778 	return cmd;
779 }
780 
781 /*
782  * We keep a little list of ref counted metadata objects to prevent two
783  * different target instances creating separate bufio instances.  This is
784  * an issue if a table is reloaded before the suspend.
785  */
786 static DEFINE_MUTEX(table_lock);
787 static LIST_HEAD(table);
788 
789 static struct dm_cache_metadata *lookup(struct block_device *bdev)
790 {
791 	struct dm_cache_metadata *cmd;
792 
793 	list_for_each_entry(cmd, &table, list)
794 		if (cmd->bdev == bdev) {
795 			refcount_inc(&cmd->ref_count);
796 			return cmd;
797 		}
798 
799 	return NULL;
800 }
801 
802 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
803 						sector_t data_block_size,
804 						bool may_format_device,
805 						size_t policy_hint_size,
806 						unsigned metadata_version)
807 {
808 	struct dm_cache_metadata *cmd, *cmd2;
809 
810 	mutex_lock(&table_lock);
811 	cmd = lookup(bdev);
812 	mutex_unlock(&table_lock);
813 
814 	if (cmd)
815 		return cmd;
816 
817 	cmd = metadata_open(bdev, data_block_size, may_format_device,
818 			    policy_hint_size, metadata_version);
819 	if (!IS_ERR(cmd)) {
820 		mutex_lock(&table_lock);
821 		cmd2 = lookup(bdev);
822 		if (cmd2) {
823 			mutex_unlock(&table_lock);
824 			__destroy_persistent_data_objects(cmd);
825 			kfree(cmd);
826 			return cmd2;
827 		}
828 		list_add(&cmd->list, &table);
829 		mutex_unlock(&table_lock);
830 	}
831 
832 	return cmd;
833 }
834 
835 static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
836 {
837 	if (cmd->data_block_size != data_block_size) {
838 		DMERR("data_block_size (%llu) different from that in metadata (%llu)",
839 		      (unsigned long long) data_block_size,
840 		      (unsigned long long) cmd->data_block_size);
841 		return false;
842 	}
843 
844 	return true;
845 }
846 
847 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
848 						 sector_t data_block_size,
849 						 bool may_format_device,
850 						 size_t policy_hint_size,
851 						 unsigned metadata_version)
852 {
853 	struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size, may_format_device,
854 						       policy_hint_size, metadata_version);
855 
856 	if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
857 		dm_cache_metadata_close(cmd);
858 		return ERR_PTR(-EINVAL);
859 	}
860 
861 	return cmd;
862 }
863 
864 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
865 {
866 	if (refcount_dec_and_test(&cmd->ref_count)) {
867 		mutex_lock(&table_lock);
868 		list_del(&cmd->list);
869 		mutex_unlock(&table_lock);
870 
871 		if (!cmd->fail_io)
872 			__destroy_persistent_data_objects(cmd);
873 		kfree(cmd);
874 	}
875 }
876 
877 /*
878  * Checks that the given cache block is either unmapped or clean.
879  */
880 static int block_clean_combined_dirty(struct dm_cache_metadata *cmd, dm_cblock_t b,
881 				      bool *result)
882 {
883 	int r;
884 	__le64 value;
885 	dm_oblock_t ob;
886 	unsigned flags;
887 
888 	r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
889 	if (r)
890 		return r;
891 
892 	unpack_value(value, &ob, &flags);
893 	*result = !((flags & M_VALID) && (flags & M_DIRTY));
894 
895 	return 0;
896 }
897 
898 static int blocks_are_clean_combined_dirty(struct dm_cache_metadata *cmd,
899 					   dm_cblock_t begin, dm_cblock_t end,
900 					   bool *result)
901 {
902 	int r;
903 	*result = true;
904 
905 	while (begin != end) {
906 		r = block_clean_combined_dirty(cmd, begin, result);
907 		if (r) {
908 			DMERR("block_clean_combined_dirty failed");
909 			return r;
910 		}
911 
912 		if (!*result) {
913 			DMERR("cache block %llu is dirty",
914 			      (unsigned long long) from_cblock(begin));
915 			return 0;
916 		}
917 
918 		begin = to_cblock(from_cblock(begin) + 1);
919 	}
920 
921 	return 0;
922 }
923 
924 static int blocks_are_clean_separate_dirty(struct dm_cache_metadata *cmd,
925 					   dm_cblock_t begin, dm_cblock_t end,
926 					   bool *result)
927 {
928 	int r;
929 	bool dirty_flag;
930 	*result = true;
931 
932 	r = dm_bitset_cursor_begin(&cmd->dirty_info, cmd->dirty_root,
933 				   from_cblock(cmd->cache_blocks), &cmd->dirty_cursor);
934 	if (r) {
935 		DMERR("%s: dm_bitset_cursor_begin for dirty failed", __func__);
936 		return r;
937 	}
938 
939 	r = dm_bitset_cursor_skip(&cmd->dirty_cursor, from_cblock(begin));
940 	if (r) {
941 		DMERR("%s: dm_bitset_cursor_skip for dirty failed", __func__);
942 		dm_bitset_cursor_end(&cmd->dirty_cursor);
943 		return r;
944 	}
945 
946 	while (begin != end) {
947 		/*
948 		 * We assume that unmapped blocks have their dirty bit
949 		 * cleared.
950 		 */
951 		dirty_flag = dm_bitset_cursor_get_value(&cmd->dirty_cursor);
952 		if (dirty_flag) {
953 			DMERR("%s: cache block %llu is dirty", __func__,
954 			      (unsigned long long) from_cblock(begin));
955 			dm_bitset_cursor_end(&cmd->dirty_cursor);
956 			*result = false;
957 			return 0;
958 		}
959 
960 		begin = to_cblock(from_cblock(begin) + 1);
961 		if (begin == end)
962 			break;
963 
964 		r = dm_bitset_cursor_next(&cmd->dirty_cursor);
965 		if (r) {
966 			DMERR("%s: dm_bitset_cursor_next for dirty failed", __func__);
967 			dm_bitset_cursor_end(&cmd->dirty_cursor);
968 			return r;
969 		}
970 	}
971 
972 	dm_bitset_cursor_end(&cmd->dirty_cursor);
973 
974 	return 0;
975 }
976 
977 static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
978 					dm_cblock_t begin, dm_cblock_t end,
979 					bool *result)
980 {
981 	if (separate_dirty_bits(cmd))
982 		return blocks_are_clean_separate_dirty(cmd, begin, end, result);
983 	else
984 		return blocks_are_clean_combined_dirty(cmd, begin, end, result);
985 }
986 
987 static bool cmd_write_lock(struct dm_cache_metadata *cmd)
988 {
989 	down_write(&cmd->root_lock);
990 	if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) {
991 		up_write(&cmd->root_lock);
992 		return false;
993 	}
994 	return true;
995 }
996 
997 #define WRITE_LOCK(cmd)				\
998 	do {					\
999 		if (!cmd_write_lock((cmd)))	\
1000 			return -EINVAL;		\
1001 	} while(0)
1002 
1003 #define WRITE_LOCK_VOID(cmd)			\
1004 	do {					\
1005 		if (!cmd_write_lock((cmd)))	\
1006 			return;			\
1007 	} while(0)
1008 
1009 #define WRITE_UNLOCK(cmd) \
1010 	up_write(&(cmd)->root_lock)
1011 
1012 static bool cmd_read_lock(struct dm_cache_metadata *cmd)
1013 {
1014 	down_read(&cmd->root_lock);
1015 	if (cmd->fail_io) {
1016 		up_read(&cmd->root_lock);
1017 		return false;
1018 	}
1019 	return true;
1020 }
1021 
1022 #define READ_LOCK(cmd)				\
1023 	do {					\
1024 		if (!cmd_read_lock((cmd)))	\
1025 			return -EINVAL;		\
1026 	} while(0)
1027 
1028 #define READ_LOCK_VOID(cmd)			\
1029 	do {					\
1030 		if (!cmd_read_lock((cmd)))	\
1031 			return;			\
1032 	} while(0)
1033 
1034 #define READ_UNLOCK(cmd) \
1035 	up_read(&(cmd)->root_lock)
1036 
1037 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
1038 {
1039 	int r;
1040 	bool clean;
1041 	__le64 null_mapping = pack_value(0, 0);
1042 
1043 	WRITE_LOCK(cmd);
1044 	__dm_bless_for_disk(&null_mapping);
1045 
1046 	if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) {
1047 		r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean);
1048 		if (r) {
1049 			__dm_unbless_for_disk(&null_mapping);
1050 			goto out;
1051 		}
1052 
1053 		if (!clean) {
1054 			DMERR("unable to shrink cache due to dirty blocks");
1055 			r = -EINVAL;
1056 			__dm_unbless_for_disk(&null_mapping);
1057 			goto out;
1058 		}
1059 	}
1060 
1061 	r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
1062 			    from_cblock(new_cache_size),
1063 			    &null_mapping, &cmd->root);
1064 	if (r)
1065 		goto out;
1066 
1067 	if (separate_dirty_bits(cmd)) {
1068 		r = dm_bitset_resize(&cmd->dirty_info, cmd->dirty_root,
1069 				     from_cblock(cmd->cache_blocks), from_cblock(new_cache_size),
1070 				     false, &cmd->dirty_root);
1071 		if (r)
1072 			goto out;
1073 	}
1074 
1075 	cmd->cache_blocks = new_cache_size;
1076 	cmd->changed = true;
1077 
1078 out:
1079 	WRITE_UNLOCK(cmd);
1080 
1081 	return r;
1082 }
1083 
1084 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
1085 				   sector_t discard_block_size,
1086 				   dm_dblock_t new_nr_entries)
1087 {
1088 	int r;
1089 
1090 	WRITE_LOCK(cmd);
1091 	r = dm_bitset_resize(&cmd->discard_info,
1092 			     cmd->discard_root,
1093 			     from_dblock(cmd->discard_nr_blocks),
1094 			     from_dblock(new_nr_entries),
1095 			     false, &cmd->discard_root);
1096 	if (!r) {
1097 		cmd->discard_block_size = discard_block_size;
1098 		cmd->discard_nr_blocks = new_nr_entries;
1099 	}
1100 
1101 	cmd->changed = true;
1102 	WRITE_UNLOCK(cmd);
1103 
1104 	return r;
1105 }
1106 
1107 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
1108 {
1109 	return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
1110 				 from_dblock(b), &cmd->discard_root);
1111 }
1112 
1113 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
1114 {
1115 	return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
1116 				   from_dblock(b), &cmd->discard_root);
1117 }
1118 
1119 static int __discard(struct dm_cache_metadata *cmd,
1120 		     dm_dblock_t dblock, bool discard)
1121 {
1122 	int r;
1123 
1124 	r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
1125 	if (r)
1126 		return r;
1127 
1128 	cmd->changed = true;
1129 	return 0;
1130 }
1131 
1132 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
1133 			 dm_dblock_t dblock, bool discard)
1134 {
1135 	int r;
1136 
1137 	WRITE_LOCK(cmd);
1138 	r = __discard(cmd, dblock, discard);
1139 	WRITE_UNLOCK(cmd);
1140 
1141 	return r;
1142 }
1143 
1144 static int __load_discards(struct dm_cache_metadata *cmd,
1145 			   load_discard_fn fn, void *context)
1146 {
1147 	int r = 0;
1148 	uint32_t b;
1149 	struct dm_bitset_cursor c;
1150 
1151 	if (from_dblock(cmd->discard_nr_blocks) == 0)
1152 		/* nothing to do */
1153 		return 0;
1154 
1155 	if (cmd->clean_when_opened) {
1156 		r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root, &cmd->discard_root);
1157 		if (r)
1158 			return r;
1159 
1160 		r = dm_bitset_cursor_begin(&cmd->discard_info, cmd->discard_root,
1161 					   from_dblock(cmd->discard_nr_blocks), &c);
1162 		if (r)
1163 			return r;
1164 
1165 		for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1166 			r = fn(context, cmd->discard_block_size, to_dblock(b),
1167 			       dm_bitset_cursor_get_value(&c));
1168 			if (r)
1169 				break;
1170 		}
1171 
1172 		dm_bitset_cursor_end(&c);
1173 
1174 	} else {
1175 		for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1176 			r = fn(context, cmd->discard_block_size, to_dblock(b), false);
1177 			if (r)
1178 				return r;
1179 		}
1180 	}
1181 
1182 	return r;
1183 }
1184 
1185 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
1186 			   load_discard_fn fn, void *context)
1187 {
1188 	int r;
1189 
1190 	READ_LOCK(cmd);
1191 	r = __load_discards(cmd, fn, context);
1192 	READ_UNLOCK(cmd);
1193 
1194 	return r;
1195 }
1196 
1197 int dm_cache_size(struct dm_cache_metadata *cmd, dm_cblock_t *result)
1198 {
1199 	READ_LOCK(cmd);
1200 	*result = cmd->cache_blocks;
1201 	READ_UNLOCK(cmd);
1202 
1203 	return 0;
1204 }
1205 
1206 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1207 {
1208 	int r;
1209 	__le64 value = pack_value(0, 0);
1210 
1211 	__dm_bless_for_disk(&value);
1212 	r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1213 			       &value, &cmd->root);
1214 	if (r)
1215 		return r;
1216 
1217 	cmd->changed = true;
1218 	return 0;
1219 }
1220 
1221 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1222 {
1223 	int r;
1224 
1225 	WRITE_LOCK(cmd);
1226 	r = __remove(cmd, cblock);
1227 	WRITE_UNLOCK(cmd);
1228 
1229 	return r;
1230 }
1231 
1232 static int __insert(struct dm_cache_metadata *cmd,
1233 		    dm_cblock_t cblock, dm_oblock_t oblock)
1234 {
1235 	int r;
1236 	__le64 value = pack_value(oblock, M_VALID);
1237 	__dm_bless_for_disk(&value);
1238 
1239 	r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1240 			       &value, &cmd->root);
1241 	if (r)
1242 		return r;
1243 
1244 	cmd->changed = true;
1245 	return 0;
1246 }
1247 
1248 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
1249 			    dm_cblock_t cblock, dm_oblock_t oblock)
1250 {
1251 	int r;
1252 
1253 	WRITE_LOCK(cmd);
1254 	r = __insert(cmd, cblock, oblock);
1255 	WRITE_UNLOCK(cmd);
1256 
1257 	return r;
1258 }
1259 
1260 struct thunk {
1261 	load_mapping_fn fn;
1262 	void *context;
1263 
1264 	struct dm_cache_metadata *cmd;
1265 	bool respect_dirty_flags;
1266 	bool hints_valid;
1267 };
1268 
1269 static bool policy_unchanged(struct dm_cache_metadata *cmd,
1270 			     struct dm_cache_policy *policy)
1271 {
1272 	const char *policy_name = dm_cache_policy_get_name(policy);
1273 	const unsigned *policy_version = dm_cache_policy_get_version(policy);
1274 	size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
1275 
1276 	/*
1277 	 * Ensure policy names match.
1278 	 */
1279 	if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
1280 		return false;
1281 
1282 	/*
1283 	 * Ensure policy major versions match.
1284 	 */
1285 	if (cmd->policy_version[0] != policy_version[0])
1286 		return false;
1287 
1288 	/*
1289 	 * Ensure policy hint sizes match.
1290 	 */
1291 	if (cmd->policy_hint_size != policy_hint_size)
1292 		return false;
1293 
1294 	return true;
1295 }
1296 
1297 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1298 {
1299 	return cmd->hint_root && cmd->policy_hint_size;
1300 }
1301 
1302 static bool hints_array_available(struct dm_cache_metadata *cmd,
1303 				  struct dm_cache_policy *policy)
1304 {
1305 	return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1306 		hints_array_initialized(cmd);
1307 }
1308 
1309 static int __load_mapping_v1(struct dm_cache_metadata *cmd,
1310 			     uint64_t cb, bool hints_valid,
1311 			     struct dm_array_cursor *mapping_cursor,
1312 			     struct dm_array_cursor *hint_cursor,
1313 			     load_mapping_fn fn, void *context)
1314 {
1315 	int r = 0;
1316 
1317 	__le64 mapping;
1318 	__le32 hint = 0;
1319 
1320 	__le64 *mapping_value_le;
1321 	__le32 *hint_value_le;
1322 
1323 	dm_oblock_t oblock;
1324 	unsigned flags;
1325 
1326 	dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1327 	memcpy(&mapping, mapping_value_le, sizeof(mapping));
1328 	unpack_value(mapping, &oblock, &flags);
1329 
1330 	if (flags & M_VALID) {
1331 		if (hints_valid) {
1332 			dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1333 			memcpy(&hint, hint_value_le, sizeof(hint));
1334 		}
1335 
1336 		r = fn(context, oblock, to_cblock(cb), flags & M_DIRTY,
1337 		       le32_to_cpu(hint), hints_valid);
1338 		if (r) {
1339 			DMERR("policy couldn't load cache block %llu",
1340 			      (unsigned long long) from_cblock(to_cblock(cb)));
1341 		}
1342 	}
1343 
1344 	return r;
1345 }
1346 
1347 static int __load_mapping_v2(struct dm_cache_metadata *cmd,
1348 			     uint64_t cb, bool hints_valid,
1349 			     struct dm_array_cursor *mapping_cursor,
1350 			     struct dm_array_cursor *hint_cursor,
1351 			     struct dm_bitset_cursor *dirty_cursor,
1352 			     load_mapping_fn fn, void *context)
1353 {
1354 	int r = 0;
1355 
1356 	__le64 mapping;
1357 	__le32 hint = 0;
1358 
1359 	__le64 *mapping_value_le;
1360 	__le32 *hint_value_le;
1361 
1362 	dm_oblock_t oblock;
1363 	unsigned flags;
1364 	bool dirty;
1365 
1366 	dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1367 	memcpy(&mapping, mapping_value_le, sizeof(mapping));
1368 	unpack_value(mapping, &oblock, &flags);
1369 
1370 	if (flags & M_VALID) {
1371 		if (hints_valid) {
1372 			dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1373 			memcpy(&hint, hint_value_le, sizeof(hint));
1374 		}
1375 
1376 		dirty = dm_bitset_cursor_get_value(dirty_cursor);
1377 		r = fn(context, oblock, to_cblock(cb), dirty,
1378 		       le32_to_cpu(hint), hints_valid);
1379 		if (r) {
1380 			DMERR("policy couldn't load cache block %llu",
1381 			      (unsigned long long) from_cblock(to_cblock(cb)));
1382 		}
1383 	}
1384 
1385 	return r;
1386 }
1387 
1388 static int __load_mappings(struct dm_cache_metadata *cmd,
1389 			   struct dm_cache_policy *policy,
1390 			   load_mapping_fn fn, void *context)
1391 {
1392 	int r;
1393 	uint64_t cb;
1394 
1395 	bool hints_valid = hints_array_available(cmd, policy);
1396 
1397 	if (from_cblock(cmd->cache_blocks) == 0)
1398 		/* Nothing to do */
1399 		return 0;
1400 
1401 	r = dm_array_cursor_begin(&cmd->info, cmd->root, &cmd->mapping_cursor);
1402 	if (r)
1403 		return r;
1404 
1405 	if (hints_valid) {
1406 		r = dm_array_cursor_begin(&cmd->hint_info, cmd->hint_root, &cmd->hint_cursor);
1407 		if (r) {
1408 			dm_array_cursor_end(&cmd->mapping_cursor);
1409 			return r;
1410 		}
1411 	}
1412 
1413 	if (separate_dirty_bits(cmd)) {
1414 		r = dm_bitset_cursor_begin(&cmd->dirty_info, cmd->dirty_root,
1415 					   from_cblock(cmd->cache_blocks),
1416 					   &cmd->dirty_cursor);
1417 		if (r) {
1418 			dm_array_cursor_end(&cmd->hint_cursor);
1419 			dm_array_cursor_end(&cmd->mapping_cursor);
1420 			return r;
1421 		}
1422 	}
1423 
1424 	for (cb = 0; ; cb++) {
1425 		if (separate_dirty_bits(cmd))
1426 			r = __load_mapping_v2(cmd, cb, hints_valid,
1427 					      &cmd->mapping_cursor,
1428 					      &cmd->hint_cursor,
1429 					      &cmd->dirty_cursor,
1430 					      fn, context);
1431 		else
1432 			r = __load_mapping_v1(cmd, cb, hints_valid,
1433 					      &cmd->mapping_cursor, &cmd->hint_cursor,
1434 					      fn, context);
1435 		if (r)
1436 			goto out;
1437 
1438 		/*
1439 		 * We need to break out before we move the cursors.
1440 		 */
1441 		if (cb >= (from_cblock(cmd->cache_blocks) - 1))
1442 			break;
1443 
1444 		r = dm_array_cursor_next(&cmd->mapping_cursor);
1445 		if (r) {
1446 			DMERR("dm_array_cursor_next for mapping failed");
1447 			goto out;
1448 		}
1449 
1450 		if (hints_valid) {
1451 			r = dm_array_cursor_next(&cmd->hint_cursor);
1452 			if (r) {
1453 				DMERR("dm_array_cursor_next for hint failed");
1454 				goto out;
1455 			}
1456 		}
1457 
1458 		if (separate_dirty_bits(cmd)) {
1459 			r = dm_bitset_cursor_next(&cmd->dirty_cursor);
1460 			if (r) {
1461 				DMERR("dm_bitset_cursor_next for dirty failed");
1462 				goto out;
1463 			}
1464 		}
1465 	}
1466 out:
1467 	dm_array_cursor_end(&cmd->mapping_cursor);
1468 	if (hints_valid)
1469 		dm_array_cursor_end(&cmd->hint_cursor);
1470 
1471 	if (separate_dirty_bits(cmd))
1472 		dm_bitset_cursor_end(&cmd->dirty_cursor);
1473 
1474 	return r;
1475 }
1476 
1477 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1478 			   struct dm_cache_policy *policy,
1479 			   load_mapping_fn fn, void *context)
1480 {
1481 	int r;
1482 
1483 	READ_LOCK(cmd);
1484 	r = __load_mappings(cmd, policy, fn, context);
1485 	READ_UNLOCK(cmd);
1486 
1487 	return r;
1488 }
1489 
1490 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
1491 {
1492 	int r = 0;
1493 	__le64 value;
1494 	dm_oblock_t oblock;
1495 	unsigned flags;
1496 
1497 	memcpy(&value, leaf, sizeof(value));
1498 	unpack_value(value, &oblock, &flags);
1499 
1500 	return r;
1501 }
1502 
1503 static int __dump_mappings(struct dm_cache_metadata *cmd)
1504 {
1505 	return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1506 }
1507 
1508 void dm_cache_dump(struct dm_cache_metadata *cmd)
1509 {
1510 	READ_LOCK_VOID(cmd);
1511 	__dump_mappings(cmd);
1512 	READ_UNLOCK(cmd);
1513 }
1514 
1515 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1516 {
1517 	int r;
1518 
1519 	READ_LOCK(cmd);
1520 	r = cmd->changed;
1521 	READ_UNLOCK(cmd);
1522 
1523 	return r;
1524 }
1525 
1526 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1527 {
1528 	int r;
1529 	unsigned flags;
1530 	dm_oblock_t oblock;
1531 	__le64 value;
1532 
1533 	r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1534 	if (r)
1535 		return r;
1536 
1537 	unpack_value(value, &oblock, &flags);
1538 
1539 	if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1540 		/* nothing to be done */
1541 		return 0;
1542 
1543 	value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1544 	__dm_bless_for_disk(&value);
1545 
1546 	r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1547 			       &value, &cmd->root);
1548 	if (r)
1549 		return r;
1550 
1551 	cmd->changed = true;
1552 	return 0;
1553 
1554 }
1555 
1556 static int __set_dirty_bits_v1(struct dm_cache_metadata *cmd, unsigned nr_bits, unsigned long *bits)
1557 {
1558 	int r;
1559 	unsigned i;
1560 	for (i = 0; i < nr_bits; i++) {
1561 		r = __dirty(cmd, to_cblock(i), test_bit(i, bits));
1562 		if (r)
1563 			return r;
1564 	}
1565 
1566 	return 0;
1567 }
1568 
1569 static int is_dirty_callback(uint32_t index, bool *value, void *context)
1570 {
1571 	unsigned long *bits = context;
1572 	*value = test_bit(index, bits);
1573 	return 0;
1574 }
1575 
1576 static int __set_dirty_bits_v2(struct dm_cache_metadata *cmd, unsigned nr_bits, unsigned long *bits)
1577 {
1578 	int r = 0;
1579 
1580 	/* nr_bits is really just a sanity check */
1581 	if (nr_bits != from_cblock(cmd->cache_blocks)) {
1582 		DMERR("dirty bitset is wrong size");
1583 		return -EINVAL;
1584 	}
1585 
1586 	r = dm_bitset_del(&cmd->dirty_info, cmd->dirty_root);
1587 	if (r)
1588 		return r;
1589 
1590 	cmd->changed = true;
1591 	return dm_bitset_new(&cmd->dirty_info, &cmd->dirty_root, nr_bits, is_dirty_callback, bits);
1592 }
1593 
1594 int dm_cache_set_dirty_bits(struct dm_cache_metadata *cmd,
1595 			    unsigned nr_bits,
1596 			    unsigned long *bits)
1597 {
1598 	int r;
1599 
1600 	WRITE_LOCK(cmd);
1601 	if (separate_dirty_bits(cmd))
1602 		r = __set_dirty_bits_v2(cmd, nr_bits, bits);
1603 	else
1604 		r = __set_dirty_bits_v1(cmd, nr_bits, bits);
1605 	WRITE_UNLOCK(cmd);
1606 
1607 	return r;
1608 }
1609 
1610 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1611 				 struct dm_cache_statistics *stats)
1612 {
1613 	READ_LOCK_VOID(cmd);
1614 	*stats = cmd->stats;
1615 	READ_UNLOCK(cmd);
1616 }
1617 
1618 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1619 				 struct dm_cache_statistics *stats)
1620 {
1621 	WRITE_LOCK_VOID(cmd);
1622 	cmd->stats = *stats;
1623 	WRITE_UNLOCK(cmd);
1624 }
1625 
1626 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1627 {
1628 	int r = -EINVAL;
1629 	flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1630 				 clear_clean_shutdown);
1631 
1632 	WRITE_LOCK(cmd);
1633 	if (cmd->fail_io)
1634 		goto out;
1635 
1636 	r = __commit_transaction(cmd, mutator);
1637 	if (r)
1638 		goto out;
1639 
1640 	r = __begin_transaction(cmd);
1641 out:
1642 	WRITE_UNLOCK(cmd);
1643 	return r;
1644 }
1645 
1646 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1647 					   dm_block_t *result)
1648 {
1649 	int r = -EINVAL;
1650 
1651 	READ_LOCK(cmd);
1652 	if (!cmd->fail_io)
1653 		r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1654 	READ_UNLOCK(cmd);
1655 
1656 	return r;
1657 }
1658 
1659 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1660 				   dm_block_t *result)
1661 {
1662 	int r = -EINVAL;
1663 
1664 	READ_LOCK(cmd);
1665 	if (!cmd->fail_io)
1666 		r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1667 	READ_UNLOCK(cmd);
1668 
1669 	return r;
1670 }
1671 
1672 /*----------------------------------------------------------------*/
1673 
1674 static int get_hint(uint32_t index, void *value_le, void *context)
1675 {
1676 	uint32_t value;
1677 	struct dm_cache_policy *policy = context;
1678 
1679 	value = policy_get_hint(policy, to_cblock(index));
1680 	*((__le32 *) value_le) = cpu_to_le32(value);
1681 
1682 	return 0;
1683 }
1684 
1685 /*
1686  * It's quicker to always delete the hint array, and recreate with
1687  * dm_array_new().
1688  */
1689 static int write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1690 {
1691 	int r;
1692 	size_t hint_size;
1693 	const char *policy_name = dm_cache_policy_get_name(policy);
1694 	const unsigned *policy_version = dm_cache_policy_get_version(policy);
1695 
1696 	if (!policy_name[0] ||
1697 	    (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1698 		return -EINVAL;
1699 
1700 	strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1701 	memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1702 
1703 	hint_size = dm_cache_policy_get_hint_size(policy);
1704 	if (!hint_size)
1705 		return 0; /* short-circuit hints initialization */
1706 	cmd->policy_hint_size = hint_size;
1707 
1708 	if (cmd->hint_root) {
1709 		r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1710 		if (r)
1711 			return r;
1712 	}
1713 
1714 	return dm_array_new(&cmd->hint_info, &cmd->hint_root,
1715 			    from_cblock(cmd->cache_blocks),
1716 			    get_hint, policy);
1717 }
1718 
1719 int dm_cache_write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1720 {
1721 	int r;
1722 
1723 	WRITE_LOCK(cmd);
1724 	r = write_hints(cmd, policy);
1725 	WRITE_UNLOCK(cmd);
1726 
1727 	return r;
1728 }
1729 
1730 int dm_cache_metadata_all_clean(struct dm_cache_metadata *cmd, bool *result)
1731 {
1732 	int r;
1733 
1734 	READ_LOCK(cmd);
1735 	r = blocks_are_unmapped_or_clean(cmd, 0, cmd->cache_blocks, result);
1736 	READ_UNLOCK(cmd);
1737 
1738 	return r;
1739 }
1740 
1741 void dm_cache_metadata_set_read_only(struct dm_cache_metadata *cmd)
1742 {
1743 	WRITE_LOCK_VOID(cmd);
1744 	dm_bm_set_read_only(cmd->bm);
1745 	WRITE_UNLOCK(cmd);
1746 }
1747 
1748 void dm_cache_metadata_set_read_write(struct dm_cache_metadata *cmd)
1749 {
1750 	WRITE_LOCK_VOID(cmd);
1751 	dm_bm_set_read_write(cmd->bm);
1752 	WRITE_UNLOCK(cmd);
1753 }
1754 
1755 int dm_cache_metadata_set_needs_check(struct dm_cache_metadata *cmd)
1756 {
1757 	int r;
1758 	struct dm_block *sblock;
1759 	struct cache_disk_superblock *disk_super;
1760 
1761 	WRITE_LOCK(cmd);
1762 	set_bit(NEEDS_CHECK, &cmd->flags);
1763 
1764 	r = superblock_lock(cmd, &sblock);
1765 	if (r) {
1766 		DMERR("couldn't read superblock");
1767 		goto out;
1768 	}
1769 
1770 	disk_super = dm_block_data(sblock);
1771 	disk_super->flags = cpu_to_le32(cmd->flags);
1772 
1773 	dm_bm_unlock(sblock);
1774 
1775 out:
1776 	WRITE_UNLOCK(cmd);
1777 	return r;
1778 }
1779 
1780 int dm_cache_metadata_needs_check(struct dm_cache_metadata *cmd, bool *result)
1781 {
1782 	READ_LOCK(cmd);
1783 	*result = !!test_bit(NEEDS_CHECK, &cmd->flags);
1784 	READ_UNLOCK(cmd);
1785 
1786 	return 0;
1787 }
1788 
1789 int dm_cache_metadata_abort(struct dm_cache_metadata *cmd)
1790 {
1791 	int r;
1792 
1793 	WRITE_LOCK(cmd);
1794 	__destroy_persistent_data_objects(cmd);
1795 	r = __create_persistent_data_objects(cmd, false);
1796 	if (r)
1797 		cmd->fail_io = true;
1798 	WRITE_UNLOCK(cmd);
1799 
1800 	return r;
1801 }
1802