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 
17 /*----------------------------------------------------------------*/
18 
19 #define DM_MSG_PREFIX   "cache metadata"
20 
21 #define CACHE_SUPERBLOCK_MAGIC 06142003
22 #define CACHE_SUPERBLOCK_LOCATION 0
23 #define CACHE_VERSION 1
24 #define CACHE_METADATA_CACHE_SIZE 64
25 
26 /*
27  *  3 for btree insert +
28  *  2 for btree lookup used within space map
29  */
30 #define CACHE_MAX_CONCURRENT_LOCKS 5
31 #define SPACE_MAP_ROOT_SIZE 128
32 
33 enum superblock_flag_bits {
34 	/* for spotting crashes that would invalidate the dirty bitset */
35 	CLEAN_SHUTDOWN,
36 };
37 
38 /*
39  * Each mapping from cache block -> origin block carries a set of flags.
40  */
41 enum mapping_bits {
42 	/*
43 	 * A valid mapping.  Because we're using an array we clear this
44 	 * flag for an non existant mapping.
45 	 */
46 	M_VALID = 1,
47 
48 	/*
49 	 * The data on the cache is different from that on the origin.
50 	 */
51 	M_DIRTY = 2
52 };
53 
54 struct cache_disk_superblock {
55 	__le32 csum;
56 	__le32 flags;
57 	__le64 blocknr;
58 
59 	__u8 uuid[16];
60 	__le64 magic;
61 	__le32 version;
62 
63 	__u8 policy_name[CACHE_POLICY_NAME_SIZE];
64 	__le32 policy_hint_size;
65 
66 	__u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
67 	__le64 mapping_root;
68 	__le64 hint_root;
69 
70 	__le64 discard_root;
71 	__le64 discard_block_size;
72 	__le64 discard_nr_blocks;
73 
74 	__le32 data_block_size;
75 	__le32 metadata_block_size;
76 	__le32 cache_blocks;
77 
78 	__le32 compat_flags;
79 	__le32 compat_ro_flags;
80 	__le32 incompat_flags;
81 
82 	__le32 read_hits;
83 	__le32 read_misses;
84 	__le32 write_hits;
85 	__le32 write_misses;
86 } __packed;
87 
88 struct dm_cache_metadata {
89 	struct block_device *bdev;
90 	struct dm_block_manager *bm;
91 	struct dm_space_map *metadata_sm;
92 	struct dm_transaction_manager *tm;
93 
94 	struct dm_array_info info;
95 	struct dm_array_info hint_info;
96 	struct dm_disk_bitset discard_info;
97 
98 	struct rw_semaphore root_lock;
99 	dm_block_t root;
100 	dm_block_t hint_root;
101 	dm_block_t discard_root;
102 
103 	sector_t discard_block_size;
104 	dm_dblock_t discard_nr_blocks;
105 
106 	sector_t data_block_size;
107 	dm_cblock_t cache_blocks;
108 	bool changed:1;
109 	bool clean_when_opened:1;
110 
111 	char policy_name[CACHE_POLICY_NAME_SIZE];
112 	size_t policy_hint_size;
113 	struct dm_cache_statistics stats;
114 };
115 
116 /*-------------------------------------------------------------------
117  * superblock validator
118  *-----------------------------------------------------------------*/
119 
120 #define SUPERBLOCK_CSUM_XOR 9031977
121 
122 static void sb_prepare_for_write(struct dm_block_validator *v,
123 				 struct dm_block *b,
124 				 size_t sb_block_size)
125 {
126 	struct cache_disk_superblock *disk_super = dm_block_data(b);
127 
128 	disk_super->blocknr = cpu_to_le64(dm_block_location(b));
129 	disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
130 						      sb_block_size - sizeof(__le32),
131 						      SUPERBLOCK_CSUM_XOR));
132 }
133 
134 static int sb_check(struct dm_block_validator *v,
135 		    struct dm_block *b,
136 		    size_t sb_block_size)
137 {
138 	struct cache_disk_superblock *disk_super = dm_block_data(b);
139 	__le32 csum_le;
140 
141 	if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
142 		DMERR("sb_check failed: blocknr %llu: wanted %llu",
143 		      le64_to_cpu(disk_super->blocknr),
144 		      (unsigned long long)dm_block_location(b));
145 		return -ENOTBLK;
146 	}
147 
148 	if (le64_to_cpu(disk_super->magic) != CACHE_SUPERBLOCK_MAGIC) {
149 		DMERR("sb_check failed: magic %llu: wanted %llu",
150 		      le64_to_cpu(disk_super->magic),
151 		      (unsigned long long)CACHE_SUPERBLOCK_MAGIC);
152 		return -EILSEQ;
153 	}
154 
155 	csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
156 					     sb_block_size - sizeof(__le32),
157 					     SUPERBLOCK_CSUM_XOR));
158 	if (csum_le != disk_super->csum) {
159 		DMERR("sb_check failed: csum %u: wanted %u",
160 		      le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
161 		return -EILSEQ;
162 	}
163 
164 	return 0;
165 }
166 
167 static struct dm_block_validator sb_validator = {
168 	.name = "superblock",
169 	.prepare_for_write = sb_prepare_for_write,
170 	.check = sb_check
171 };
172 
173 /*----------------------------------------------------------------*/
174 
175 static int superblock_read_lock(struct dm_cache_metadata *cmd,
176 				struct dm_block **sblock)
177 {
178 	return dm_bm_read_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
179 			       &sb_validator, sblock);
180 }
181 
182 static int superblock_lock_zero(struct dm_cache_metadata *cmd,
183 				struct dm_block **sblock)
184 {
185 	return dm_bm_write_lock_zero(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
186 				     &sb_validator, sblock);
187 }
188 
189 static int superblock_lock(struct dm_cache_metadata *cmd,
190 			   struct dm_block **sblock)
191 {
192 	return dm_bm_write_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
193 				&sb_validator, sblock);
194 }
195 
196 /*----------------------------------------------------------------*/
197 
198 static int __superblock_all_zeroes(struct dm_block_manager *bm, int *result)
199 {
200 	int r;
201 	unsigned i;
202 	struct dm_block *b;
203 	__le64 *data_le, zero = cpu_to_le64(0);
204 	unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
205 
206 	/*
207 	 * We can't use a validator here - it may be all zeroes.
208 	 */
209 	r = dm_bm_read_lock(bm, CACHE_SUPERBLOCK_LOCATION, NULL, &b);
210 	if (r)
211 		return r;
212 
213 	data_le = dm_block_data(b);
214 	*result = 1;
215 	for (i = 0; i < sb_block_size; i++) {
216 		if (data_le[i] != zero) {
217 			*result = 0;
218 			break;
219 		}
220 	}
221 
222 	return dm_bm_unlock(b);
223 }
224 
225 static void __setup_mapping_info(struct dm_cache_metadata *cmd)
226 {
227 	struct dm_btree_value_type vt;
228 
229 	vt.context = NULL;
230 	vt.size = sizeof(__le64);
231 	vt.inc = NULL;
232 	vt.dec = NULL;
233 	vt.equal = NULL;
234 	dm_array_info_init(&cmd->info, cmd->tm, &vt);
235 
236 	if (cmd->policy_hint_size) {
237 		vt.size = sizeof(__le32);
238 		dm_array_info_init(&cmd->hint_info, cmd->tm, &vt);
239 	}
240 }
241 
242 static int __write_initial_superblock(struct dm_cache_metadata *cmd)
243 {
244 	int r;
245 	struct dm_block *sblock;
246 	size_t metadata_len;
247 	struct cache_disk_superblock *disk_super;
248 	sector_t bdev_size = i_size_read(cmd->bdev->bd_inode) >> SECTOR_SHIFT;
249 
250 	/* FIXME: see if we can lose the max sectors limit */
251 	if (bdev_size > DM_CACHE_METADATA_MAX_SECTORS)
252 		bdev_size = DM_CACHE_METADATA_MAX_SECTORS;
253 
254 	r = dm_sm_root_size(cmd->metadata_sm, &metadata_len);
255 	if (r < 0)
256 		return r;
257 
258 	r = dm_tm_pre_commit(cmd->tm);
259 	if (r < 0)
260 		return r;
261 
262 	r = superblock_lock_zero(cmd, &sblock);
263 	if (r)
264 		return r;
265 
266 	disk_super = dm_block_data(sblock);
267 	disk_super->flags = 0;
268 	memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
269 	disk_super->magic = cpu_to_le64(CACHE_SUPERBLOCK_MAGIC);
270 	disk_super->version = cpu_to_le32(CACHE_VERSION);
271 	memset(disk_super->policy_name, 0, CACHE_POLICY_NAME_SIZE);
272 	disk_super->policy_hint_size = 0;
273 
274 	r = dm_sm_copy_root(cmd->metadata_sm, &disk_super->metadata_space_map_root,
275 			    metadata_len);
276 	if (r < 0)
277 		goto bad_locked;
278 
279 	disk_super->mapping_root = cpu_to_le64(cmd->root);
280 	disk_super->hint_root = cpu_to_le64(cmd->hint_root);
281 	disk_super->discard_root = cpu_to_le64(cmd->discard_root);
282 	disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
283 	disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
284 	disk_super->metadata_block_size = cpu_to_le32(DM_CACHE_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
285 	disk_super->data_block_size = cpu_to_le32(cmd->data_block_size);
286 	disk_super->cache_blocks = cpu_to_le32(0);
287 	memset(disk_super->policy_name, 0, sizeof(disk_super->policy_name));
288 
289 	disk_super->read_hits = cpu_to_le32(0);
290 	disk_super->read_misses = cpu_to_le32(0);
291 	disk_super->write_hits = cpu_to_le32(0);
292 	disk_super->write_misses = cpu_to_le32(0);
293 
294 	return dm_tm_commit(cmd->tm, sblock);
295 
296 bad_locked:
297 	dm_bm_unlock(sblock);
298 	return r;
299 }
300 
301 static int __format_metadata(struct dm_cache_metadata *cmd)
302 {
303 	int r;
304 
305 	r = dm_tm_create_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
306 				 &cmd->tm, &cmd->metadata_sm);
307 	if (r < 0) {
308 		DMERR("tm_create_with_sm failed");
309 		return r;
310 	}
311 
312 	__setup_mapping_info(cmd);
313 
314 	r = dm_array_empty(&cmd->info, &cmd->root);
315 	if (r < 0)
316 		goto bad;
317 
318 	dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
319 
320 	r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root);
321 	if (r < 0)
322 		goto bad;
323 
324 	cmd->discard_block_size = 0;
325 	cmd->discard_nr_blocks = 0;
326 
327 	r = __write_initial_superblock(cmd);
328 	if (r)
329 		goto bad;
330 
331 	cmd->clean_when_opened = true;
332 	return 0;
333 
334 bad:
335 	dm_tm_destroy(cmd->tm);
336 	dm_sm_destroy(cmd->metadata_sm);
337 
338 	return r;
339 }
340 
341 static int __check_incompat_features(struct cache_disk_superblock *disk_super,
342 				     struct dm_cache_metadata *cmd)
343 {
344 	uint32_t features;
345 
346 	features = le32_to_cpu(disk_super->incompat_flags) & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
347 	if (features) {
348 		DMERR("could not access metadata due to unsupported optional features (%lx).",
349 		      (unsigned long)features);
350 		return -EINVAL;
351 	}
352 
353 	/*
354 	 * Check for read-only metadata to skip the following RDWR checks.
355 	 */
356 	if (get_disk_ro(cmd->bdev->bd_disk))
357 		return 0;
358 
359 	features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP;
360 	if (features) {
361 		DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
362 		      (unsigned long)features);
363 		return -EINVAL;
364 	}
365 
366 	return 0;
367 }
368 
369 static int __open_metadata(struct dm_cache_metadata *cmd)
370 {
371 	int r;
372 	struct dm_block *sblock;
373 	struct cache_disk_superblock *disk_super;
374 	unsigned long sb_flags;
375 
376 	r = superblock_read_lock(cmd, &sblock);
377 	if (r < 0) {
378 		DMERR("couldn't read lock superblock");
379 		return r;
380 	}
381 
382 	disk_super = dm_block_data(sblock);
383 
384 	r = __check_incompat_features(disk_super, cmd);
385 	if (r < 0)
386 		goto bad;
387 
388 	r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
389 			       disk_super->metadata_space_map_root,
390 			       sizeof(disk_super->metadata_space_map_root),
391 			       &cmd->tm, &cmd->metadata_sm);
392 	if (r < 0) {
393 		DMERR("tm_open_with_sm failed");
394 		goto bad;
395 	}
396 
397 	__setup_mapping_info(cmd);
398 	dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
399 	sb_flags = le32_to_cpu(disk_super->flags);
400 	cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags);
401 	return dm_bm_unlock(sblock);
402 
403 bad:
404 	dm_bm_unlock(sblock);
405 	return r;
406 }
407 
408 static int __open_or_format_metadata(struct dm_cache_metadata *cmd,
409 				     bool format_device)
410 {
411 	int r, unformatted;
412 
413 	r = __superblock_all_zeroes(cmd->bm, &unformatted);
414 	if (r)
415 		return r;
416 
417 	if (unformatted)
418 		return format_device ? __format_metadata(cmd) : -EPERM;
419 
420 	return __open_metadata(cmd);
421 }
422 
423 static int __create_persistent_data_objects(struct dm_cache_metadata *cmd,
424 					    bool may_format_device)
425 {
426 	int r;
427 	cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE,
428 					  CACHE_METADATA_CACHE_SIZE,
429 					  CACHE_MAX_CONCURRENT_LOCKS);
430 	if (IS_ERR(cmd->bm)) {
431 		DMERR("could not create block manager");
432 		return PTR_ERR(cmd->bm);
433 	}
434 
435 	r = __open_or_format_metadata(cmd, may_format_device);
436 	if (r)
437 		dm_block_manager_destroy(cmd->bm);
438 
439 	return r;
440 }
441 
442 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd)
443 {
444 	dm_sm_destroy(cmd->metadata_sm);
445 	dm_tm_destroy(cmd->tm);
446 	dm_block_manager_destroy(cmd->bm);
447 }
448 
449 typedef unsigned long (*flags_mutator)(unsigned long);
450 
451 static void update_flags(struct cache_disk_superblock *disk_super,
452 			 flags_mutator mutator)
453 {
454 	uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
455 	disk_super->flags = cpu_to_le32(sb_flags);
456 }
457 
458 static unsigned long set_clean_shutdown(unsigned long flags)
459 {
460 	set_bit(CLEAN_SHUTDOWN, &flags);
461 	return flags;
462 }
463 
464 static unsigned long clear_clean_shutdown(unsigned long flags)
465 {
466 	clear_bit(CLEAN_SHUTDOWN, &flags);
467 	return flags;
468 }
469 
470 static void read_superblock_fields(struct dm_cache_metadata *cmd,
471 				   struct cache_disk_superblock *disk_super)
472 {
473 	cmd->root = le64_to_cpu(disk_super->mapping_root);
474 	cmd->hint_root = le64_to_cpu(disk_super->hint_root);
475 	cmd->discard_root = le64_to_cpu(disk_super->discard_root);
476 	cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
477 	cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
478 	cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
479 	cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
480 	strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
481 	cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
482 
483 	cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
484 	cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
485 	cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
486 	cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
487 
488 	cmd->changed = false;
489 }
490 
491 /*
492  * The mutator updates the superblock flags.
493  */
494 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
495 				     flags_mutator mutator)
496 {
497 	int r;
498 	struct cache_disk_superblock *disk_super;
499 	struct dm_block *sblock;
500 
501 	r = superblock_lock(cmd, &sblock);
502 	if (r)
503 		return r;
504 
505 	disk_super = dm_block_data(sblock);
506 	update_flags(disk_super, mutator);
507 	read_superblock_fields(cmd, disk_super);
508 
509 	return dm_bm_flush_and_unlock(cmd->bm, sblock);
510 }
511 
512 static int __begin_transaction(struct dm_cache_metadata *cmd)
513 {
514 	int r;
515 	struct cache_disk_superblock *disk_super;
516 	struct dm_block *sblock;
517 
518 	/*
519 	 * We re-read the superblock every time.  Shouldn't need to do this
520 	 * really.
521 	 */
522 	r = superblock_read_lock(cmd, &sblock);
523 	if (r)
524 		return r;
525 
526 	disk_super = dm_block_data(sblock);
527 	read_superblock_fields(cmd, disk_super);
528 	dm_bm_unlock(sblock);
529 
530 	return 0;
531 }
532 
533 static int __commit_transaction(struct dm_cache_metadata *cmd,
534 				flags_mutator mutator)
535 {
536 	int r;
537 	size_t metadata_len;
538 	struct cache_disk_superblock *disk_super;
539 	struct dm_block *sblock;
540 
541 	/*
542 	 * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
543 	 */
544 	BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
545 
546 	r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
547 			    &cmd->discard_root);
548 	if (r)
549 		return r;
550 
551 	r = dm_tm_pre_commit(cmd->tm);
552 	if (r < 0)
553 		return r;
554 
555 	r = dm_sm_root_size(cmd->metadata_sm, &metadata_len);
556 	if (r < 0)
557 		return r;
558 
559 	r = superblock_lock(cmd, &sblock);
560 	if (r)
561 		return r;
562 
563 	disk_super = dm_block_data(sblock);
564 
565 	if (mutator)
566 		update_flags(disk_super, mutator);
567 
568 	disk_super->mapping_root = cpu_to_le64(cmd->root);
569 	disk_super->hint_root = cpu_to_le64(cmd->hint_root);
570 	disk_super->discard_root = cpu_to_le64(cmd->discard_root);
571 	disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
572 	disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
573 	disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
574 	strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
575 
576 	disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
577 	disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
578 	disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
579 	disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
580 
581 	r = dm_sm_copy_root(cmd->metadata_sm, &disk_super->metadata_space_map_root,
582 			    metadata_len);
583 	if (r < 0) {
584 		dm_bm_unlock(sblock);
585 		return r;
586 	}
587 
588 	return dm_tm_commit(cmd->tm, sblock);
589 }
590 
591 /*----------------------------------------------------------------*/
592 
593 /*
594  * The mappings are held in a dm-array that has 64-bit values stored in
595  * little-endian format.  The index is the cblock, the high 48bits of the
596  * value are the oblock and the low 16 bit the flags.
597  */
598 #define FLAGS_MASK ((1 << 16) - 1)
599 
600 static __le64 pack_value(dm_oblock_t block, unsigned flags)
601 {
602 	uint64_t value = from_oblock(block);
603 	value <<= 16;
604 	value = value | (flags & FLAGS_MASK);
605 	return cpu_to_le64(value);
606 }
607 
608 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
609 {
610 	uint64_t value = le64_to_cpu(value_le);
611 	uint64_t b = value >> 16;
612 	*block = to_oblock(b);
613 	*flags = value & FLAGS_MASK;
614 }
615 
616 /*----------------------------------------------------------------*/
617 
618 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
619 						 sector_t data_block_size,
620 						 bool may_format_device,
621 						 size_t policy_hint_size)
622 {
623 	int r;
624 	struct dm_cache_metadata *cmd;
625 
626 	cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
627 	if (!cmd) {
628 		DMERR("could not allocate metadata struct");
629 		return NULL;
630 	}
631 
632 	init_rwsem(&cmd->root_lock);
633 	cmd->bdev = bdev;
634 	cmd->data_block_size = data_block_size;
635 	cmd->cache_blocks = 0;
636 	cmd->policy_hint_size = policy_hint_size;
637 	cmd->changed = true;
638 
639 	r = __create_persistent_data_objects(cmd, may_format_device);
640 	if (r) {
641 		kfree(cmd);
642 		return ERR_PTR(r);
643 	}
644 
645 	r = __begin_transaction_flags(cmd, clear_clean_shutdown);
646 	if (r < 0) {
647 		dm_cache_metadata_close(cmd);
648 		return ERR_PTR(r);
649 	}
650 
651 	return cmd;
652 }
653 
654 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
655 {
656 	__destroy_persistent_data_objects(cmd);
657 	kfree(cmd);
658 }
659 
660 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
661 {
662 	int r;
663 	__le64 null_mapping = pack_value(0, 0);
664 
665 	down_write(&cmd->root_lock);
666 	__dm_bless_for_disk(&null_mapping);
667 	r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
668 			    from_cblock(new_cache_size),
669 			    &null_mapping, &cmd->root);
670 	if (!r)
671 		cmd->cache_blocks = new_cache_size;
672 	cmd->changed = true;
673 	up_write(&cmd->root_lock);
674 
675 	return r;
676 }
677 
678 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
679 				   sector_t discard_block_size,
680 				   dm_dblock_t new_nr_entries)
681 {
682 	int r;
683 
684 	down_write(&cmd->root_lock);
685 	r = dm_bitset_resize(&cmd->discard_info,
686 			     cmd->discard_root,
687 			     from_dblock(cmd->discard_nr_blocks),
688 			     from_dblock(new_nr_entries),
689 			     false, &cmd->discard_root);
690 	if (!r) {
691 		cmd->discard_block_size = discard_block_size;
692 		cmd->discard_nr_blocks = new_nr_entries;
693 	}
694 
695 	cmd->changed = true;
696 	up_write(&cmd->root_lock);
697 
698 	return r;
699 }
700 
701 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
702 {
703 	return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
704 				 from_dblock(b), &cmd->discard_root);
705 }
706 
707 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
708 {
709 	return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
710 				   from_dblock(b), &cmd->discard_root);
711 }
712 
713 static int __is_discarded(struct dm_cache_metadata *cmd, dm_dblock_t b,
714 			  bool *is_discarded)
715 {
716 	return dm_bitset_test_bit(&cmd->discard_info, cmd->discard_root,
717 				  from_dblock(b), &cmd->discard_root,
718 				  is_discarded);
719 }
720 
721 static int __discard(struct dm_cache_metadata *cmd,
722 		     dm_dblock_t dblock, bool discard)
723 {
724 	int r;
725 
726 	r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
727 	if (r)
728 		return r;
729 
730 	cmd->changed = true;
731 	return 0;
732 }
733 
734 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
735 			 dm_dblock_t dblock, bool discard)
736 {
737 	int r;
738 
739 	down_write(&cmd->root_lock);
740 	r = __discard(cmd, dblock, discard);
741 	up_write(&cmd->root_lock);
742 
743 	return r;
744 }
745 
746 static int __load_discards(struct dm_cache_metadata *cmd,
747 			   load_discard_fn fn, void *context)
748 {
749 	int r = 0;
750 	dm_block_t b;
751 	bool discard;
752 
753 	for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
754 		dm_dblock_t dblock = to_dblock(b);
755 
756 		if (cmd->clean_when_opened) {
757 			r = __is_discarded(cmd, dblock, &discard);
758 			if (r)
759 				return r;
760 		} else
761 			discard = false;
762 
763 		r = fn(context, cmd->discard_block_size, dblock, discard);
764 		if (r)
765 			break;
766 	}
767 
768 	return r;
769 }
770 
771 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
772 			   load_discard_fn fn, void *context)
773 {
774 	int r;
775 
776 	down_read(&cmd->root_lock);
777 	r = __load_discards(cmd, fn, context);
778 	up_read(&cmd->root_lock);
779 
780 	return r;
781 }
782 
783 dm_cblock_t dm_cache_size(struct dm_cache_metadata *cmd)
784 {
785 	dm_cblock_t r;
786 
787 	down_read(&cmd->root_lock);
788 	r = cmd->cache_blocks;
789 	up_read(&cmd->root_lock);
790 
791 	return r;
792 }
793 
794 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
795 {
796 	int r;
797 	__le64 value = pack_value(0, 0);
798 
799 	__dm_bless_for_disk(&value);
800 	r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
801 			       &value, &cmd->root);
802 	if (r)
803 		return r;
804 
805 	cmd->changed = true;
806 	return 0;
807 }
808 
809 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
810 {
811 	int r;
812 
813 	down_write(&cmd->root_lock);
814 	r = __remove(cmd, cblock);
815 	up_write(&cmd->root_lock);
816 
817 	return r;
818 }
819 
820 static int __insert(struct dm_cache_metadata *cmd,
821 		    dm_cblock_t cblock, dm_oblock_t oblock)
822 {
823 	int r;
824 	__le64 value = pack_value(oblock, M_VALID);
825 	__dm_bless_for_disk(&value);
826 
827 	r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
828 			       &value, &cmd->root);
829 	if (r)
830 		return r;
831 
832 	cmd->changed = true;
833 	return 0;
834 }
835 
836 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
837 			    dm_cblock_t cblock, dm_oblock_t oblock)
838 {
839 	int r;
840 
841 	down_write(&cmd->root_lock);
842 	r = __insert(cmd, cblock, oblock);
843 	up_write(&cmd->root_lock);
844 
845 	return r;
846 }
847 
848 struct thunk {
849 	load_mapping_fn fn;
850 	void *context;
851 
852 	struct dm_cache_metadata *cmd;
853 	bool respect_dirty_flags;
854 	bool hints_valid;
855 };
856 
857 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
858 {
859 	return cmd->hint_root && cmd->policy_hint_size;
860 }
861 
862 static bool hints_array_available(struct dm_cache_metadata *cmd,
863 				  const char *policy_name)
864 {
865 	bool policy_names_match = !strncmp(cmd->policy_name, policy_name,
866 					   sizeof(cmd->policy_name));
867 
868 	return cmd->clean_when_opened && policy_names_match &&
869 		hints_array_initialized(cmd);
870 }
871 
872 static int __load_mapping(void *context, uint64_t cblock, void *leaf)
873 {
874 	int r = 0;
875 	bool dirty;
876 	__le64 value;
877 	__le32 hint_value = 0;
878 	dm_oblock_t oblock;
879 	unsigned flags;
880 	struct thunk *thunk = context;
881 	struct dm_cache_metadata *cmd = thunk->cmd;
882 
883 	memcpy(&value, leaf, sizeof(value));
884 	unpack_value(value, &oblock, &flags);
885 
886 	if (flags & M_VALID) {
887 		if (thunk->hints_valid) {
888 			r = dm_array_get_value(&cmd->hint_info, cmd->hint_root,
889 					       cblock, &hint_value);
890 			if (r && r != -ENODATA)
891 				return r;
892 		}
893 
894 		dirty = thunk->respect_dirty_flags ? (flags & M_DIRTY) : true;
895 		r = thunk->fn(thunk->context, oblock, to_cblock(cblock),
896 			      dirty, le32_to_cpu(hint_value), thunk->hints_valid);
897 	}
898 
899 	return r;
900 }
901 
902 static int __load_mappings(struct dm_cache_metadata *cmd, const char *policy_name,
903 			   load_mapping_fn fn, void *context)
904 {
905 	struct thunk thunk;
906 
907 	thunk.fn = fn;
908 	thunk.context = context;
909 
910 	thunk.cmd = cmd;
911 	thunk.respect_dirty_flags = cmd->clean_when_opened;
912 	thunk.hints_valid = hints_array_available(cmd, policy_name);
913 
914 	return dm_array_walk(&cmd->info, cmd->root, __load_mapping, &thunk);
915 }
916 
917 int dm_cache_load_mappings(struct dm_cache_metadata *cmd, const char *policy_name,
918 			   load_mapping_fn fn, void *context)
919 {
920 	int r;
921 
922 	down_read(&cmd->root_lock);
923 	r = __load_mappings(cmd, policy_name, fn, context);
924 	up_read(&cmd->root_lock);
925 
926 	return r;
927 }
928 
929 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
930 {
931 	int r = 0;
932 	__le64 value;
933 	dm_oblock_t oblock;
934 	unsigned flags;
935 
936 	memcpy(&value, leaf, sizeof(value));
937 	unpack_value(value, &oblock, &flags);
938 
939 	return r;
940 }
941 
942 static int __dump_mappings(struct dm_cache_metadata *cmd)
943 {
944 	return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
945 }
946 
947 void dm_cache_dump(struct dm_cache_metadata *cmd)
948 {
949 	down_read(&cmd->root_lock);
950 	__dump_mappings(cmd);
951 	up_read(&cmd->root_lock);
952 }
953 
954 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
955 {
956 	int r;
957 
958 	down_read(&cmd->root_lock);
959 	r = cmd->changed;
960 	up_read(&cmd->root_lock);
961 
962 	return r;
963 }
964 
965 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
966 {
967 	int r;
968 	unsigned flags;
969 	dm_oblock_t oblock;
970 	__le64 value;
971 
972 	r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
973 	if (r)
974 		return r;
975 
976 	unpack_value(value, &oblock, &flags);
977 
978 	if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
979 		/* nothing to be done */
980 		return 0;
981 
982 	value = pack_value(oblock, flags | (dirty ? M_DIRTY : 0));
983 	__dm_bless_for_disk(&value);
984 
985 	r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
986 			       &value, &cmd->root);
987 	if (r)
988 		return r;
989 
990 	cmd->changed = true;
991 	return 0;
992 
993 }
994 
995 int dm_cache_set_dirty(struct dm_cache_metadata *cmd,
996 		       dm_cblock_t cblock, bool dirty)
997 {
998 	int r;
999 
1000 	down_write(&cmd->root_lock);
1001 	r = __dirty(cmd, cblock, dirty);
1002 	up_write(&cmd->root_lock);
1003 
1004 	return r;
1005 }
1006 
1007 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1008 				 struct dm_cache_statistics *stats)
1009 {
1010 	down_read(&cmd->root_lock);
1011 	memcpy(stats, &cmd->stats, sizeof(*stats));
1012 	up_read(&cmd->root_lock);
1013 }
1014 
1015 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1016 				 struct dm_cache_statistics *stats)
1017 {
1018 	down_write(&cmd->root_lock);
1019 	memcpy(&cmd->stats, stats, sizeof(*stats));
1020 	up_write(&cmd->root_lock);
1021 }
1022 
1023 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1024 {
1025 	int r;
1026 	flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1027 				 clear_clean_shutdown);
1028 
1029 	down_write(&cmd->root_lock);
1030 	r = __commit_transaction(cmd, mutator);
1031 	if (r)
1032 		goto out;
1033 
1034 	r = __begin_transaction(cmd);
1035 
1036 out:
1037 	up_write(&cmd->root_lock);
1038 	return r;
1039 }
1040 
1041 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1042 					   dm_block_t *result)
1043 {
1044 	int r = -EINVAL;
1045 
1046 	down_read(&cmd->root_lock);
1047 	r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1048 	up_read(&cmd->root_lock);
1049 
1050 	return r;
1051 }
1052 
1053 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1054 				   dm_block_t *result)
1055 {
1056 	int r = -EINVAL;
1057 
1058 	down_read(&cmd->root_lock);
1059 	r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1060 	up_read(&cmd->root_lock);
1061 
1062 	return r;
1063 }
1064 
1065 /*----------------------------------------------------------------*/
1066 
1067 static int begin_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1068 {
1069 	int r;
1070 	__le32 value;
1071 	size_t hint_size;
1072 	const char *policy_name = dm_cache_policy_get_name(policy);
1073 
1074 	if (!policy_name[0] ||
1075 	    (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1076 		return -EINVAL;
1077 
1078 	if (strcmp(cmd->policy_name, policy_name)) {
1079 		strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1080 
1081 		hint_size = dm_cache_policy_get_hint_size(policy);
1082 		if (!hint_size)
1083 			return 0; /* short-circuit hints initialization */
1084 		cmd->policy_hint_size = hint_size;
1085 
1086 		if (cmd->hint_root) {
1087 			r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1088 			if (r)
1089 				return r;
1090 		}
1091 
1092 		r = dm_array_empty(&cmd->hint_info, &cmd->hint_root);
1093 		if (r)
1094 			return r;
1095 
1096 		value = cpu_to_le32(0);
1097 		__dm_bless_for_disk(&value);
1098 		r = dm_array_resize(&cmd->hint_info, cmd->hint_root, 0,
1099 				    from_cblock(cmd->cache_blocks),
1100 				    &value, &cmd->hint_root);
1101 		if (r)
1102 			return r;
1103 	}
1104 
1105 	return 0;
1106 }
1107 
1108 int dm_cache_begin_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1109 {
1110 	int r;
1111 
1112 	down_write(&cmd->root_lock);
1113 	r = begin_hints(cmd, policy);
1114 	up_write(&cmd->root_lock);
1115 
1116 	return r;
1117 }
1118 
1119 static int save_hint(struct dm_cache_metadata *cmd, dm_cblock_t cblock,
1120 		     uint32_t hint)
1121 {
1122 	int r;
1123 	__le32 value = cpu_to_le32(hint);
1124 	__dm_bless_for_disk(&value);
1125 
1126 	r = dm_array_set_value(&cmd->hint_info, cmd->hint_root,
1127 			       from_cblock(cblock), &value, &cmd->hint_root);
1128 	cmd->changed = true;
1129 
1130 	return r;
1131 }
1132 
1133 int dm_cache_save_hint(struct dm_cache_metadata *cmd, dm_cblock_t cblock,
1134 		       uint32_t hint)
1135 {
1136 	int r;
1137 
1138 	if (!hints_array_initialized(cmd))
1139 		return 0;
1140 
1141 	down_write(&cmd->root_lock);
1142 	r = save_hint(cmd, cblock, hint);
1143 	up_write(&cmd->root_lock);
1144 
1145 	return r;
1146 }
1147