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 = cpu_to_le32(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 	disk_super->policy_hint_size = cpu_to_le32(cmd->policy_hint_size);
705 
706 	disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
707 	disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
708 	disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
709 	disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
710 	__copy_sm_root(cmd, disk_super);
711 
712 	return dm_tm_commit(cmd->tm, sblock);
713 }
714 
715 /*----------------------------------------------------------------*/
716 
717 /*
718  * The mappings are held in a dm-array that has 64-bit values stored in
719  * little-endian format.  The index is the cblock, the high 48bits of the
720  * value are the oblock and the low 16 bit the flags.
721  */
722 #define FLAGS_MASK ((1 << 16) - 1)
723 
724 static __le64 pack_value(dm_oblock_t block, unsigned flags)
725 {
726 	uint64_t value = from_oblock(block);
727 	value <<= 16;
728 	value = value | (flags & FLAGS_MASK);
729 	return cpu_to_le64(value);
730 }
731 
732 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
733 {
734 	uint64_t value = le64_to_cpu(value_le);
735 	uint64_t b = value >> 16;
736 	*block = to_oblock(b);
737 	*flags = value & FLAGS_MASK;
738 }
739 
740 /*----------------------------------------------------------------*/
741 
742 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
743 					       sector_t data_block_size,
744 					       bool may_format_device,
745 					       size_t policy_hint_size,
746 					       unsigned metadata_version)
747 {
748 	int r;
749 	struct dm_cache_metadata *cmd;
750 
751 	cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
752 	if (!cmd) {
753 		DMERR("could not allocate metadata struct");
754 		return ERR_PTR(-ENOMEM);
755 	}
756 
757 	cmd->version = metadata_version;
758 	refcount_set(&cmd->ref_count, 1);
759 	init_rwsem(&cmd->root_lock);
760 	cmd->bdev = bdev;
761 	cmd->data_block_size = data_block_size;
762 	cmd->cache_blocks = 0;
763 	cmd->policy_hint_size = policy_hint_size;
764 	cmd->changed = true;
765 	cmd->fail_io = false;
766 
767 	r = __create_persistent_data_objects(cmd, may_format_device);
768 	if (r) {
769 		kfree(cmd);
770 		return ERR_PTR(r);
771 	}
772 
773 	r = __begin_transaction_flags(cmd, clear_clean_shutdown);
774 	if (r < 0) {
775 		dm_cache_metadata_close(cmd);
776 		return ERR_PTR(r);
777 	}
778 
779 	return cmd;
780 }
781 
782 /*
783  * We keep a little list of ref counted metadata objects to prevent two
784  * different target instances creating separate bufio instances.  This is
785  * an issue if a table is reloaded before the suspend.
786  */
787 static DEFINE_MUTEX(table_lock);
788 static LIST_HEAD(table);
789 
790 static struct dm_cache_metadata *lookup(struct block_device *bdev)
791 {
792 	struct dm_cache_metadata *cmd;
793 
794 	list_for_each_entry(cmd, &table, list)
795 		if (cmd->bdev == bdev) {
796 			refcount_inc(&cmd->ref_count);
797 			return cmd;
798 		}
799 
800 	return NULL;
801 }
802 
803 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
804 						sector_t data_block_size,
805 						bool may_format_device,
806 						size_t policy_hint_size,
807 						unsigned metadata_version)
808 {
809 	struct dm_cache_metadata *cmd, *cmd2;
810 
811 	mutex_lock(&table_lock);
812 	cmd = lookup(bdev);
813 	mutex_unlock(&table_lock);
814 
815 	if (cmd)
816 		return cmd;
817 
818 	cmd = metadata_open(bdev, data_block_size, may_format_device,
819 			    policy_hint_size, metadata_version);
820 	if (!IS_ERR(cmd)) {
821 		mutex_lock(&table_lock);
822 		cmd2 = lookup(bdev);
823 		if (cmd2) {
824 			mutex_unlock(&table_lock);
825 			__destroy_persistent_data_objects(cmd);
826 			kfree(cmd);
827 			return cmd2;
828 		}
829 		list_add(&cmd->list, &table);
830 		mutex_unlock(&table_lock);
831 	}
832 
833 	return cmd;
834 }
835 
836 static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
837 {
838 	if (cmd->data_block_size != data_block_size) {
839 		DMERR("data_block_size (%llu) different from that in metadata (%llu)",
840 		      (unsigned long long) data_block_size,
841 		      (unsigned long long) cmd->data_block_size);
842 		return false;
843 	}
844 
845 	return true;
846 }
847 
848 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
849 						 sector_t data_block_size,
850 						 bool may_format_device,
851 						 size_t policy_hint_size,
852 						 unsigned metadata_version)
853 {
854 	struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size, may_format_device,
855 						       policy_hint_size, metadata_version);
856 
857 	if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
858 		dm_cache_metadata_close(cmd);
859 		return ERR_PTR(-EINVAL);
860 	}
861 
862 	return cmd;
863 }
864 
865 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
866 {
867 	if (refcount_dec_and_test(&cmd->ref_count)) {
868 		mutex_lock(&table_lock);
869 		list_del(&cmd->list);
870 		mutex_unlock(&table_lock);
871 
872 		if (!cmd->fail_io)
873 			__destroy_persistent_data_objects(cmd);
874 		kfree(cmd);
875 	}
876 }
877 
878 /*
879  * Checks that the given cache block is either unmapped or clean.
880  */
881 static int block_clean_combined_dirty(struct dm_cache_metadata *cmd, dm_cblock_t b,
882 				      bool *result)
883 {
884 	int r;
885 	__le64 value;
886 	dm_oblock_t ob;
887 	unsigned flags;
888 
889 	r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
890 	if (r)
891 		return r;
892 
893 	unpack_value(value, &ob, &flags);
894 	*result = !((flags & M_VALID) && (flags & M_DIRTY));
895 
896 	return 0;
897 }
898 
899 static int blocks_are_clean_combined_dirty(struct dm_cache_metadata *cmd,
900 					   dm_cblock_t begin, dm_cblock_t end,
901 					   bool *result)
902 {
903 	int r;
904 	*result = true;
905 
906 	while (begin != end) {
907 		r = block_clean_combined_dirty(cmd, begin, result);
908 		if (r) {
909 			DMERR("block_clean_combined_dirty failed");
910 			return r;
911 		}
912 
913 		if (!*result) {
914 			DMERR("cache block %llu is dirty",
915 			      (unsigned long long) from_cblock(begin));
916 			return 0;
917 		}
918 
919 		begin = to_cblock(from_cblock(begin) + 1);
920 	}
921 
922 	return 0;
923 }
924 
925 static int blocks_are_clean_separate_dirty(struct dm_cache_metadata *cmd,
926 					   dm_cblock_t begin, dm_cblock_t end,
927 					   bool *result)
928 {
929 	int r;
930 	bool dirty_flag;
931 	*result = true;
932 
933 	if (from_cblock(cmd->cache_blocks) == 0)
934 		/* Nothing to do */
935 		return 0;
936 
937 	r = dm_bitset_cursor_begin(&cmd->dirty_info, cmd->dirty_root,
938 				   from_cblock(cmd->cache_blocks), &cmd->dirty_cursor);
939 	if (r) {
940 		DMERR("%s: dm_bitset_cursor_begin for dirty failed", __func__);
941 		return r;
942 	}
943 
944 	r = dm_bitset_cursor_skip(&cmd->dirty_cursor, from_cblock(begin));
945 	if (r) {
946 		DMERR("%s: dm_bitset_cursor_skip for dirty failed", __func__);
947 		dm_bitset_cursor_end(&cmd->dirty_cursor);
948 		return r;
949 	}
950 
951 	while (begin != end) {
952 		/*
953 		 * We assume that unmapped blocks have their dirty bit
954 		 * cleared.
955 		 */
956 		dirty_flag = dm_bitset_cursor_get_value(&cmd->dirty_cursor);
957 		if (dirty_flag) {
958 			DMERR("%s: cache block %llu is dirty", __func__,
959 			      (unsigned long long) from_cblock(begin));
960 			dm_bitset_cursor_end(&cmd->dirty_cursor);
961 			*result = false;
962 			return 0;
963 		}
964 
965 		begin = to_cblock(from_cblock(begin) + 1);
966 		if (begin == end)
967 			break;
968 
969 		r = dm_bitset_cursor_next(&cmd->dirty_cursor);
970 		if (r) {
971 			DMERR("%s: dm_bitset_cursor_next for dirty failed", __func__);
972 			dm_bitset_cursor_end(&cmd->dirty_cursor);
973 			return r;
974 		}
975 	}
976 
977 	dm_bitset_cursor_end(&cmd->dirty_cursor);
978 
979 	return 0;
980 }
981 
982 static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
983 					dm_cblock_t begin, dm_cblock_t end,
984 					bool *result)
985 {
986 	if (separate_dirty_bits(cmd))
987 		return blocks_are_clean_separate_dirty(cmd, begin, end, result);
988 	else
989 		return blocks_are_clean_combined_dirty(cmd, begin, end, result);
990 }
991 
992 static bool cmd_write_lock(struct dm_cache_metadata *cmd)
993 {
994 	down_write(&cmd->root_lock);
995 	if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) {
996 		up_write(&cmd->root_lock);
997 		return false;
998 	}
999 	return true;
1000 }
1001 
1002 #define WRITE_LOCK(cmd)				\
1003 	do {					\
1004 		if (!cmd_write_lock((cmd)))	\
1005 			return -EINVAL;		\
1006 	} while(0)
1007 
1008 #define WRITE_LOCK_VOID(cmd)			\
1009 	do {					\
1010 		if (!cmd_write_lock((cmd)))	\
1011 			return;			\
1012 	} while(0)
1013 
1014 #define WRITE_UNLOCK(cmd) \
1015 	up_write(&(cmd)->root_lock)
1016 
1017 static bool cmd_read_lock(struct dm_cache_metadata *cmd)
1018 {
1019 	down_read(&cmd->root_lock);
1020 	if (cmd->fail_io) {
1021 		up_read(&cmd->root_lock);
1022 		return false;
1023 	}
1024 	return true;
1025 }
1026 
1027 #define READ_LOCK(cmd)				\
1028 	do {					\
1029 		if (!cmd_read_lock((cmd)))	\
1030 			return -EINVAL;		\
1031 	} while(0)
1032 
1033 #define READ_LOCK_VOID(cmd)			\
1034 	do {					\
1035 		if (!cmd_read_lock((cmd)))	\
1036 			return;			\
1037 	} while(0)
1038 
1039 #define READ_UNLOCK(cmd) \
1040 	up_read(&(cmd)->root_lock)
1041 
1042 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
1043 {
1044 	int r;
1045 	bool clean;
1046 	__le64 null_mapping = pack_value(0, 0);
1047 
1048 	WRITE_LOCK(cmd);
1049 	__dm_bless_for_disk(&null_mapping);
1050 
1051 	if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) {
1052 		r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean);
1053 		if (r) {
1054 			__dm_unbless_for_disk(&null_mapping);
1055 			goto out;
1056 		}
1057 
1058 		if (!clean) {
1059 			DMERR("unable to shrink cache due to dirty blocks");
1060 			r = -EINVAL;
1061 			__dm_unbless_for_disk(&null_mapping);
1062 			goto out;
1063 		}
1064 	}
1065 
1066 	r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
1067 			    from_cblock(new_cache_size),
1068 			    &null_mapping, &cmd->root);
1069 	if (r)
1070 		goto out;
1071 
1072 	if (separate_dirty_bits(cmd)) {
1073 		r = dm_bitset_resize(&cmd->dirty_info, cmd->dirty_root,
1074 				     from_cblock(cmd->cache_blocks), from_cblock(new_cache_size),
1075 				     false, &cmd->dirty_root);
1076 		if (r)
1077 			goto out;
1078 	}
1079 
1080 	cmd->cache_blocks = new_cache_size;
1081 	cmd->changed = true;
1082 
1083 out:
1084 	WRITE_UNLOCK(cmd);
1085 
1086 	return r;
1087 }
1088 
1089 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
1090 				   sector_t discard_block_size,
1091 				   dm_dblock_t new_nr_entries)
1092 {
1093 	int r;
1094 
1095 	WRITE_LOCK(cmd);
1096 	r = dm_bitset_resize(&cmd->discard_info,
1097 			     cmd->discard_root,
1098 			     from_dblock(cmd->discard_nr_blocks),
1099 			     from_dblock(new_nr_entries),
1100 			     false, &cmd->discard_root);
1101 	if (!r) {
1102 		cmd->discard_block_size = discard_block_size;
1103 		cmd->discard_nr_blocks = new_nr_entries;
1104 	}
1105 
1106 	cmd->changed = true;
1107 	WRITE_UNLOCK(cmd);
1108 
1109 	return r;
1110 }
1111 
1112 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
1113 {
1114 	return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
1115 				 from_dblock(b), &cmd->discard_root);
1116 }
1117 
1118 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
1119 {
1120 	return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
1121 				   from_dblock(b), &cmd->discard_root);
1122 }
1123 
1124 static int __discard(struct dm_cache_metadata *cmd,
1125 		     dm_dblock_t dblock, bool discard)
1126 {
1127 	int r;
1128 
1129 	r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
1130 	if (r)
1131 		return r;
1132 
1133 	cmd->changed = true;
1134 	return 0;
1135 }
1136 
1137 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
1138 			 dm_dblock_t dblock, bool discard)
1139 {
1140 	int r;
1141 
1142 	WRITE_LOCK(cmd);
1143 	r = __discard(cmd, dblock, discard);
1144 	WRITE_UNLOCK(cmd);
1145 
1146 	return r;
1147 }
1148 
1149 static int __load_discards(struct dm_cache_metadata *cmd,
1150 			   load_discard_fn fn, void *context)
1151 {
1152 	int r = 0;
1153 	uint32_t b;
1154 	struct dm_bitset_cursor c;
1155 
1156 	if (from_dblock(cmd->discard_nr_blocks) == 0)
1157 		/* nothing to do */
1158 		return 0;
1159 
1160 	if (cmd->clean_when_opened) {
1161 		r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root, &cmd->discard_root);
1162 		if (r)
1163 			return r;
1164 
1165 		r = dm_bitset_cursor_begin(&cmd->discard_info, cmd->discard_root,
1166 					   from_dblock(cmd->discard_nr_blocks), &c);
1167 		if (r)
1168 			return r;
1169 
1170 		for (b = 0; ; b++) {
1171 			r = fn(context, cmd->discard_block_size, to_dblock(b),
1172 			       dm_bitset_cursor_get_value(&c));
1173 			if (r)
1174 				break;
1175 
1176 			if (b >= (from_dblock(cmd->discard_nr_blocks) - 1))
1177 				break;
1178 
1179 			r = dm_bitset_cursor_next(&c);
1180 			if (r)
1181 				break;
1182 		}
1183 
1184 		dm_bitset_cursor_end(&c);
1185 
1186 	} else {
1187 		for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1188 			r = fn(context, cmd->discard_block_size, to_dblock(b), false);
1189 			if (r)
1190 				return r;
1191 		}
1192 	}
1193 
1194 	return r;
1195 }
1196 
1197 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
1198 			   load_discard_fn fn, void *context)
1199 {
1200 	int r;
1201 
1202 	READ_LOCK(cmd);
1203 	r = __load_discards(cmd, fn, context);
1204 	READ_UNLOCK(cmd);
1205 
1206 	return r;
1207 }
1208 
1209 int dm_cache_size(struct dm_cache_metadata *cmd, dm_cblock_t *result)
1210 {
1211 	READ_LOCK(cmd);
1212 	*result = cmd->cache_blocks;
1213 	READ_UNLOCK(cmd);
1214 
1215 	return 0;
1216 }
1217 
1218 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1219 {
1220 	int r;
1221 	__le64 value = pack_value(0, 0);
1222 
1223 	__dm_bless_for_disk(&value);
1224 	r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1225 			       &value, &cmd->root);
1226 	if (r)
1227 		return r;
1228 
1229 	cmd->changed = true;
1230 	return 0;
1231 }
1232 
1233 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1234 {
1235 	int r;
1236 
1237 	WRITE_LOCK(cmd);
1238 	r = __remove(cmd, cblock);
1239 	WRITE_UNLOCK(cmd);
1240 
1241 	return r;
1242 }
1243 
1244 static int __insert(struct dm_cache_metadata *cmd,
1245 		    dm_cblock_t cblock, dm_oblock_t oblock)
1246 {
1247 	int r;
1248 	__le64 value = pack_value(oblock, M_VALID);
1249 	__dm_bless_for_disk(&value);
1250 
1251 	r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1252 			       &value, &cmd->root);
1253 	if (r)
1254 		return r;
1255 
1256 	cmd->changed = true;
1257 	return 0;
1258 }
1259 
1260 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
1261 			    dm_cblock_t cblock, dm_oblock_t oblock)
1262 {
1263 	int r;
1264 
1265 	WRITE_LOCK(cmd);
1266 	r = __insert(cmd, cblock, oblock);
1267 	WRITE_UNLOCK(cmd);
1268 
1269 	return r;
1270 }
1271 
1272 struct thunk {
1273 	load_mapping_fn fn;
1274 	void *context;
1275 
1276 	struct dm_cache_metadata *cmd;
1277 	bool respect_dirty_flags;
1278 	bool hints_valid;
1279 };
1280 
1281 static bool policy_unchanged(struct dm_cache_metadata *cmd,
1282 			     struct dm_cache_policy *policy)
1283 {
1284 	const char *policy_name = dm_cache_policy_get_name(policy);
1285 	const unsigned *policy_version = dm_cache_policy_get_version(policy);
1286 	size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
1287 
1288 	/*
1289 	 * Ensure policy names match.
1290 	 */
1291 	if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
1292 		return false;
1293 
1294 	/*
1295 	 * Ensure policy major versions match.
1296 	 */
1297 	if (cmd->policy_version[0] != policy_version[0])
1298 		return false;
1299 
1300 	/*
1301 	 * Ensure policy hint sizes match.
1302 	 */
1303 	if (cmd->policy_hint_size != policy_hint_size)
1304 		return false;
1305 
1306 	return true;
1307 }
1308 
1309 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1310 {
1311 	return cmd->hint_root && cmd->policy_hint_size;
1312 }
1313 
1314 static bool hints_array_available(struct dm_cache_metadata *cmd,
1315 				  struct dm_cache_policy *policy)
1316 {
1317 	return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1318 		hints_array_initialized(cmd);
1319 }
1320 
1321 static int __load_mapping_v1(struct dm_cache_metadata *cmd,
1322 			     uint64_t cb, bool hints_valid,
1323 			     struct dm_array_cursor *mapping_cursor,
1324 			     struct dm_array_cursor *hint_cursor,
1325 			     load_mapping_fn fn, void *context)
1326 {
1327 	int r = 0;
1328 
1329 	__le64 mapping;
1330 	__le32 hint = 0;
1331 
1332 	__le64 *mapping_value_le;
1333 	__le32 *hint_value_le;
1334 
1335 	dm_oblock_t oblock;
1336 	unsigned flags;
1337 	bool dirty = true;
1338 
1339 	dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1340 	memcpy(&mapping, mapping_value_le, sizeof(mapping));
1341 	unpack_value(mapping, &oblock, &flags);
1342 
1343 	if (flags & M_VALID) {
1344 		if (hints_valid) {
1345 			dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1346 			memcpy(&hint, hint_value_le, sizeof(hint));
1347 		}
1348 		if (cmd->clean_when_opened)
1349 			dirty = flags & M_DIRTY;
1350 
1351 		r = fn(context, oblock, to_cblock(cb), dirty,
1352 		       le32_to_cpu(hint), hints_valid);
1353 		if (r) {
1354 			DMERR("policy couldn't load cache block %llu",
1355 			      (unsigned long long) from_cblock(to_cblock(cb)));
1356 		}
1357 	}
1358 
1359 	return r;
1360 }
1361 
1362 static int __load_mapping_v2(struct dm_cache_metadata *cmd,
1363 			     uint64_t cb, bool hints_valid,
1364 			     struct dm_array_cursor *mapping_cursor,
1365 			     struct dm_array_cursor *hint_cursor,
1366 			     struct dm_bitset_cursor *dirty_cursor,
1367 			     load_mapping_fn fn, void *context)
1368 {
1369 	int r = 0;
1370 
1371 	__le64 mapping;
1372 	__le32 hint = 0;
1373 
1374 	__le64 *mapping_value_le;
1375 	__le32 *hint_value_le;
1376 
1377 	dm_oblock_t oblock;
1378 	unsigned flags;
1379 	bool dirty = true;
1380 
1381 	dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1382 	memcpy(&mapping, mapping_value_le, sizeof(mapping));
1383 	unpack_value(mapping, &oblock, &flags);
1384 
1385 	if (flags & M_VALID) {
1386 		if (hints_valid) {
1387 			dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1388 			memcpy(&hint, hint_value_le, sizeof(hint));
1389 		}
1390 		if (cmd->clean_when_opened)
1391 			dirty = dm_bitset_cursor_get_value(dirty_cursor);
1392 
1393 		r = fn(context, oblock, to_cblock(cb), dirty,
1394 		       le32_to_cpu(hint), hints_valid);
1395 		if (r) {
1396 			DMERR("policy couldn't load cache block %llu",
1397 			      (unsigned long long) from_cblock(to_cblock(cb)));
1398 		}
1399 	}
1400 
1401 	return r;
1402 }
1403 
1404 static int __load_mappings(struct dm_cache_metadata *cmd,
1405 			   struct dm_cache_policy *policy,
1406 			   load_mapping_fn fn, void *context)
1407 {
1408 	int r;
1409 	uint64_t cb;
1410 
1411 	bool hints_valid = hints_array_available(cmd, policy);
1412 
1413 	if (from_cblock(cmd->cache_blocks) == 0)
1414 		/* Nothing to do */
1415 		return 0;
1416 
1417 	r = dm_array_cursor_begin(&cmd->info, cmd->root, &cmd->mapping_cursor);
1418 	if (r)
1419 		return r;
1420 
1421 	if (hints_valid) {
1422 		r = dm_array_cursor_begin(&cmd->hint_info, cmd->hint_root, &cmd->hint_cursor);
1423 		if (r) {
1424 			dm_array_cursor_end(&cmd->mapping_cursor);
1425 			return r;
1426 		}
1427 	}
1428 
1429 	if (separate_dirty_bits(cmd)) {
1430 		r = dm_bitset_cursor_begin(&cmd->dirty_info, cmd->dirty_root,
1431 					   from_cblock(cmd->cache_blocks),
1432 					   &cmd->dirty_cursor);
1433 		if (r) {
1434 			dm_array_cursor_end(&cmd->hint_cursor);
1435 			dm_array_cursor_end(&cmd->mapping_cursor);
1436 			return r;
1437 		}
1438 	}
1439 
1440 	for (cb = 0; ; cb++) {
1441 		if (separate_dirty_bits(cmd))
1442 			r = __load_mapping_v2(cmd, cb, hints_valid,
1443 					      &cmd->mapping_cursor,
1444 					      &cmd->hint_cursor,
1445 					      &cmd->dirty_cursor,
1446 					      fn, context);
1447 		else
1448 			r = __load_mapping_v1(cmd, cb, hints_valid,
1449 					      &cmd->mapping_cursor, &cmd->hint_cursor,
1450 					      fn, context);
1451 		if (r)
1452 			goto out;
1453 
1454 		/*
1455 		 * We need to break out before we move the cursors.
1456 		 */
1457 		if (cb >= (from_cblock(cmd->cache_blocks) - 1))
1458 			break;
1459 
1460 		r = dm_array_cursor_next(&cmd->mapping_cursor);
1461 		if (r) {
1462 			DMERR("dm_array_cursor_next for mapping failed");
1463 			goto out;
1464 		}
1465 
1466 		if (hints_valid) {
1467 			r = dm_array_cursor_next(&cmd->hint_cursor);
1468 			if (r) {
1469 				dm_array_cursor_end(&cmd->hint_cursor);
1470 				hints_valid = false;
1471 			}
1472 		}
1473 
1474 		if (separate_dirty_bits(cmd)) {
1475 			r = dm_bitset_cursor_next(&cmd->dirty_cursor);
1476 			if (r) {
1477 				DMERR("dm_bitset_cursor_next for dirty failed");
1478 				goto out;
1479 			}
1480 		}
1481 	}
1482 out:
1483 	dm_array_cursor_end(&cmd->mapping_cursor);
1484 	if (hints_valid)
1485 		dm_array_cursor_end(&cmd->hint_cursor);
1486 
1487 	if (separate_dirty_bits(cmd))
1488 		dm_bitset_cursor_end(&cmd->dirty_cursor);
1489 
1490 	return r;
1491 }
1492 
1493 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1494 			   struct dm_cache_policy *policy,
1495 			   load_mapping_fn fn, void *context)
1496 {
1497 	int r;
1498 
1499 	READ_LOCK(cmd);
1500 	r = __load_mappings(cmd, policy, fn, context);
1501 	READ_UNLOCK(cmd);
1502 
1503 	return r;
1504 }
1505 
1506 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
1507 {
1508 	int r = 0;
1509 	__le64 value;
1510 	dm_oblock_t oblock;
1511 	unsigned flags;
1512 
1513 	memcpy(&value, leaf, sizeof(value));
1514 	unpack_value(value, &oblock, &flags);
1515 
1516 	return r;
1517 }
1518 
1519 static int __dump_mappings(struct dm_cache_metadata *cmd)
1520 {
1521 	return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1522 }
1523 
1524 void dm_cache_dump(struct dm_cache_metadata *cmd)
1525 {
1526 	READ_LOCK_VOID(cmd);
1527 	__dump_mappings(cmd);
1528 	READ_UNLOCK(cmd);
1529 }
1530 
1531 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1532 {
1533 	int r;
1534 
1535 	READ_LOCK(cmd);
1536 	r = cmd->changed;
1537 	READ_UNLOCK(cmd);
1538 
1539 	return r;
1540 }
1541 
1542 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1543 {
1544 	int r;
1545 	unsigned flags;
1546 	dm_oblock_t oblock;
1547 	__le64 value;
1548 
1549 	r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1550 	if (r)
1551 		return r;
1552 
1553 	unpack_value(value, &oblock, &flags);
1554 
1555 	if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1556 		/* nothing to be done */
1557 		return 0;
1558 
1559 	value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1560 	__dm_bless_for_disk(&value);
1561 
1562 	r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1563 			       &value, &cmd->root);
1564 	if (r)
1565 		return r;
1566 
1567 	cmd->changed = true;
1568 	return 0;
1569 
1570 }
1571 
1572 static int __set_dirty_bits_v1(struct dm_cache_metadata *cmd, unsigned nr_bits, unsigned long *bits)
1573 {
1574 	int r;
1575 	unsigned i;
1576 	for (i = 0; i < nr_bits; i++) {
1577 		r = __dirty(cmd, to_cblock(i), test_bit(i, bits));
1578 		if (r)
1579 			return r;
1580 	}
1581 
1582 	return 0;
1583 }
1584 
1585 static int is_dirty_callback(uint32_t index, bool *value, void *context)
1586 {
1587 	unsigned long *bits = context;
1588 	*value = test_bit(index, bits);
1589 	return 0;
1590 }
1591 
1592 static int __set_dirty_bits_v2(struct dm_cache_metadata *cmd, unsigned nr_bits, unsigned long *bits)
1593 {
1594 	int r = 0;
1595 
1596 	/* nr_bits is really just a sanity check */
1597 	if (nr_bits != from_cblock(cmd->cache_blocks)) {
1598 		DMERR("dirty bitset is wrong size");
1599 		return -EINVAL;
1600 	}
1601 
1602 	r = dm_bitset_del(&cmd->dirty_info, cmd->dirty_root);
1603 	if (r)
1604 		return r;
1605 
1606 	cmd->changed = true;
1607 	return dm_bitset_new(&cmd->dirty_info, &cmd->dirty_root, nr_bits, is_dirty_callback, bits);
1608 }
1609 
1610 int dm_cache_set_dirty_bits(struct dm_cache_metadata *cmd,
1611 			    unsigned nr_bits,
1612 			    unsigned long *bits)
1613 {
1614 	int r;
1615 
1616 	WRITE_LOCK(cmd);
1617 	if (separate_dirty_bits(cmd))
1618 		r = __set_dirty_bits_v2(cmd, nr_bits, bits);
1619 	else
1620 		r = __set_dirty_bits_v1(cmd, nr_bits, bits);
1621 	WRITE_UNLOCK(cmd);
1622 
1623 	return r;
1624 }
1625 
1626 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1627 				 struct dm_cache_statistics *stats)
1628 {
1629 	READ_LOCK_VOID(cmd);
1630 	*stats = cmd->stats;
1631 	READ_UNLOCK(cmd);
1632 }
1633 
1634 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1635 				 struct dm_cache_statistics *stats)
1636 {
1637 	WRITE_LOCK_VOID(cmd);
1638 	cmd->stats = *stats;
1639 	WRITE_UNLOCK(cmd);
1640 }
1641 
1642 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1643 {
1644 	int r = -EINVAL;
1645 	flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1646 				 clear_clean_shutdown);
1647 
1648 	WRITE_LOCK(cmd);
1649 	if (cmd->fail_io)
1650 		goto out;
1651 
1652 	r = __commit_transaction(cmd, mutator);
1653 	if (r)
1654 		goto out;
1655 
1656 	r = __begin_transaction(cmd);
1657 out:
1658 	WRITE_UNLOCK(cmd);
1659 	return r;
1660 }
1661 
1662 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1663 					   dm_block_t *result)
1664 {
1665 	int r = -EINVAL;
1666 
1667 	READ_LOCK(cmd);
1668 	if (!cmd->fail_io)
1669 		r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1670 	READ_UNLOCK(cmd);
1671 
1672 	return r;
1673 }
1674 
1675 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1676 				   dm_block_t *result)
1677 {
1678 	int r = -EINVAL;
1679 
1680 	READ_LOCK(cmd);
1681 	if (!cmd->fail_io)
1682 		r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1683 	READ_UNLOCK(cmd);
1684 
1685 	return r;
1686 }
1687 
1688 /*----------------------------------------------------------------*/
1689 
1690 static int get_hint(uint32_t index, void *value_le, void *context)
1691 {
1692 	uint32_t value;
1693 	struct dm_cache_policy *policy = context;
1694 
1695 	value = policy_get_hint(policy, to_cblock(index));
1696 	*((__le32 *) value_le) = cpu_to_le32(value);
1697 
1698 	return 0;
1699 }
1700 
1701 /*
1702  * It's quicker to always delete the hint array, and recreate with
1703  * dm_array_new().
1704  */
1705 static int write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1706 {
1707 	int r;
1708 	size_t hint_size;
1709 	const char *policy_name = dm_cache_policy_get_name(policy);
1710 	const unsigned *policy_version = dm_cache_policy_get_version(policy);
1711 
1712 	if (!policy_name[0] ||
1713 	    (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1714 		return -EINVAL;
1715 
1716 	strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1717 	memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1718 
1719 	hint_size = dm_cache_policy_get_hint_size(policy);
1720 	if (!hint_size)
1721 		return 0; /* short-circuit hints initialization */
1722 	cmd->policy_hint_size = hint_size;
1723 
1724 	if (cmd->hint_root) {
1725 		r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1726 		if (r)
1727 			return r;
1728 	}
1729 
1730 	return dm_array_new(&cmd->hint_info, &cmd->hint_root,
1731 			    from_cblock(cmd->cache_blocks),
1732 			    get_hint, policy);
1733 }
1734 
1735 int dm_cache_write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1736 {
1737 	int r;
1738 
1739 	WRITE_LOCK(cmd);
1740 	r = write_hints(cmd, policy);
1741 	WRITE_UNLOCK(cmd);
1742 
1743 	return r;
1744 }
1745 
1746 int dm_cache_metadata_all_clean(struct dm_cache_metadata *cmd, bool *result)
1747 {
1748 	int r;
1749 
1750 	READ_LOCK(cmd);
1751 	r = blocks_are_unmapped_or_clean(cmd, 0, cmd->cache_blocks, result);
1752 	READ_UNLOCK(cmd);
1753 
1754 	return r;
1755 }
1756 
1757 void dm_cache_metadata_set_read_only(struct dm_cache_metadata *cmd)
1758 {
1759 	WRITE_LOCK_VOID(cmd);
1760 	dm_bm_set_read_only(cmd->bm);
1761 	WRITE_UNLOCK(cmd);
1762 }
1763 
1764 void dm_cache_metadata_set_read_write(struct dm_cache_metadata *cmd)
1765 {
1766 	WRITE_LOCK_VOID(cmd);
1767 	dm_bm_set_read_write(cmd->bm);
1768 	WRITE_UNLOCK(cmd);
1769 }
1770 
1771 int dm_cache_metadata_set_needs_check(struct dm_cache_metadata *cmd)
1772 {
1773 	int r;
1774 	struct dm_block *sblock;
1775 	struct cache_disk_superblock *disk_super;
1776 
1777 	WRITE_LOCK(cmd);
1778 	set_bit(NEEDS_CHECK, &cmd->flags);
1779 
1780 	r = superblock_lock(cmd, &sblock);
1781 	if (r) {
1782 		DMERR("couldn't read superblock");
1783 		goto out;
1784 	}
1785 
1786 	disk_super = dm_block_data(sblock);
1787 	disk_super->flags = cpu_to_le32(cmd->flags);
1788 
1789 	dm_bm_unlock(sblock);
1790 
1791 out:
1792 	WRITE_UNLOCK(cmd);
1793 	return r;
1794 }
1795 
1796 int dm_cache_metadata_needs_check(struct dm_cache_metadata *cmd, bool *result)
1797 {
1798 	READ_LOCK(cmd);
1799 	*result = !!test_bit(NEEDS_CHECK, &cmd->flags);
1800 	READ_UNLOCK(cmd);
1801 
1802 	return 0;
1803 }
1804 
1805 int dm_cache_metadata_abort(struct dm_cache_metadata *cmd)
1806 {
1807 	int r;
1808 
1809 	WRITE_LOCK(cmd);
1810 	__destroy_persistent_data_objects(cmd);
1811 	r = __create_persistent_data_objects(cmd, false);
1812 	if (r)
1813 		cmd->fail_io = true;
1814 	WRITE_UNLOCK(cmd);
1815 
1816 	return r;
1817 }
1818