xref: /linux/drivers/md/dm-cache-metadata.c (revision 3932b9ca55b0be314a36d3e84faff3e823c081f5)
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 
24 /*
25  * defines a range of metadata versions that this module can handle.
26  */
27 #define MIN_CACHE_VERSION 1
28 #define MAX_CACHE_VERSION 1
29 
30 #define CACHE_METADATA_CACHE_SIZE 64
31 
32 /*
33  *  3 for btree insert +
34  *  2 for btree lookup used within space map
35  */
36 #define CACHE_MAX_CONCURRENT_LOCKS 5
37 #define SPACE_MAP_ROOT_SIZE 128
38 
39 enum superblock_flag_bits {
40 	/* for spotting crashes that would invalidate the dirty bitset */
41 	CLEAN_SHUTDOWN,
42 };
43 
44 /*
45  * Each mapping from cache block -> origin block carries a set of flags.
46  */
47 enum mapping_bits {
48 	/*
49 	 * A valid mapping.  Because we're using an array we clear this
50 	 * flag for an non existant mapping.
51 	 */
52 	M_VALID = 1,
53 
54 	/*
55 	 * The data on the cache is different from that on the origin.
56 	 */
57 	M_DIRTY = 2
58 };
59 
60 struct cache_disk_superblock {
61 	__le32 csum;
62 	__le32 flags;
63 	__le64 blocknr;
64 
65 	__u8 uuid[16];
66 	__le64 magic;
67 	__le32 version;
68 
69 	__u8 policy_name[CACHE_POLICY_NAME_SIZE];
70 	__le32 policy_hint_size;
71 
72 	__u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
73 	__le64 mapping_root;
74 	__le64 hint_root;
75 
76 	__le64 discard_root;
77 	__le64 discard_block_size;
78 	__le64 discard_nr_blocks;
79 
80 	__le32 data_block_size;
81 	__le32 metadata_block_size;
82 	__le32 cache_blocks;
83 
84 	__le32 compat_flags;
85 	__le32 compat_ro_flags;
86 	__le32 incompat_flags;
87 
88 	__le32 read_hits;
89 	__le32 read_misses;
90 	__le32 write_hits;
91 	__le32 write_misses;
92 
93 	__le32 policy_version[CACHE_POLICY_VERSION_SIZE];
94 } __packed;
95 
96 struct dm_cache_metadata {
97 	struct block_device *bdev;
98 	struct dm_block_manager *bm;
99 	struct dm_space_map *metadata_sm;
100 	struct dm_transaction_manager *tm;
101 
102 	struct dm_array_info info;
103 	struct dm_array_info hint_info;
104 	struct dm_disk_bitset discard_info;
105 
106 	struct rw_semaphore root_lock;
107 	dm_block_t root;
108 	dm_block_t hint_root;
109 	dm_block_t discard_root;
110 
111 	sector_t discard_block_size;
112 	dm_oblock_t discard_nr_blocks;
113 
114 	sector_t data_block_size;
115 	dm_cblock_t cache_blocks;
116 	bool changed:1;
117 	bool clean_when_opened:1;
118 
119 	char policy_name[CACHE_POLICY_NAME_SIZE];
120 	unsigned policy_version[CACHE_POLICY_VERSION_SIZE];
121 	size_t policy_hint_size;
122 	struct dm_cache_statistics stats;
123 
124 	/*
125 	 * Reading the space map root can fail, so we read it into this
126 	 * buffer before the superblock is locked and updated.
127 	 */
128 	__u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
129 };
130 
131 /*-------------------------------------------------------------------
132  * superblock validator
133  *-----------------------------------------------------------------*/
134 
135 #define SUPERBLOCK_CSUM_XOR 9031977
136 
137 static void sb_prepare_for_write(struct dm_block_validator *v,
138 				 struct dm_block *b,
139 				 size_t sb_block_size)
140 {
141 	struct cache_disk_superblock *disk_super = dm_block_data(b);
142 
143 	disk_super->blocknr = cpu_to_le64(dm_block_location(b));
144 	disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
145 						      sb_block_size - sizeof(__le32),
146 						      SUPERBLOCK_CSUM_XOR));
147 }
148 
149 static int check_metadata_version(struct cache_disk_superblock *disk_super)
150 {
151 	uint32_t metadata_version = le32_to_cpu(disk_super->version);
152 	if (metadata_version < MIN_CACHE_VERSION || metadata_version > MAX_CACHE_VERSION) {
153 		DMERR("Cache metadata version %u found, but only versions between %u and %u supported.",
154 		      metadata_version, MIN_CACHE_VERSION, MAX_CACHE_VERSION);
155 		return -EINVAL;
156 	}
157 
158 	return 0;
159 }
160 
161 static int sb_check(struct dm_block_validator *v,
162 		    struct dm_block *b,
163 		    size_t sb_block_size)
164 {
165 	struct cache_disk_superblock *disk_super = dm_block_data(b);
166 	__le32 csum_le;
167 
168 	if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
169 		DMERR("sb_check failed: blocknr %llu: wanted %llu",
170 		      le64_to_cpu(disk_super->blocknr),
171 		      (unsigned long long)dm_block_location(b));
172 		return -ENOTBLK;
173 	}
174 
175 	if (le64_to_cpu(disk_super->magic) != CACHE_SUPERBLOCK_MAGIC) {
176 		DMERR("sb_check failed: magic %llu: wanted %llu",
177 		      le64_to_cpu(disk_super->magic),
178 		      (unsigned long long)CACHE_SUPERBLOCK_MAGIC);
179 		return -EILSEQ;
180 	}
181 
182 	csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
183 					     sb_block_size - sizeof(__le32),
184 					     SUPERBLOCK_CSUM_XOR));
185 	if (csum_le != disk_super->csum) {
186 		DMERR("sb_check failed: csum %u: wanted %u",
187 		      le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
188 		return -EILSEQ;
189 	}
190 
191 	return check_metadata_version(disk_super);
192 }
193 
194 static struct dm_block_validator sb_validator = {
195 	.name = "superblock",
196 	.prepare_for_write = sb_prepare_for_write,
197 	.check = sb_check
198 };
199 
200 /*----------------------------------------------------------------*/
201 
202 static int superblock_read_lock(struct dm_cache_metadata *cmd,
203 				struct dm_block **sblock)
204 {
205 	return dm_bm_read_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
206 			       &sb_validator, sblock);
207 }
208 
209 static int superblock_lock_zero(struct dm_cache_metadata *cmd,
210 				struct dm_block **sblock)
211 {
212 	return dm_bm_write_lock_zero(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
213 				     &sb_validator, sblock);
214 }
215 
216 static int superblock_lock(struct dm_cache_metadata *cmd,
217 			   struct dm_block **sblock)
218 {
219 	return dm_bm_write_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
220 				&sb_validator, sblock);
221 }
222 
223 /*----------------------------------------------------------------*/
224 
225 static int __superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
226 {
227 	int r;
228 	unsigned i;
229 	struct dm_block *b;
230 	__le64 *data_le, zero = cpu_to_le64(0);
231 	unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
232 
233 	/*
234 	 * We can't use a validator here - it may be all zeroes.
235 	 */
236 	r = dm_bm_read_lock(bm, CACHE_SUPERBLOCK_LOCATION, NULL, &b);
237 	if (r)
238 		return r;
239 
240 	data_le = dm_block_data(b);
241 	*result = true;
242 	for (i = 0; i < sb_block_size; i++) {
243 		if (data_le[i] != zero) {
244 			*result = false;
245 			break;
246 		}
247 	}
248 
249 	return dm_bm_unlock(b);
250 }
251 
252 static void __setup_mapping_info(struct dm_cache_metadata *cmd)
253 {
254 	struct dm_btree_value_type vt;
255 
256 	vt.context = NULL;
257 	vt.size = sizeof(__le64);
258 	vt.inc = NULL;
259 	vt.dec = NULL;
260 	vt.equal = NULL;
261 	dm_array_info_init(&cmd->info, cmd->tm, &vt);
262 
263 	if (cmd->policy_hint_size) {
264 		vt.size = sizeof(__le32);
265 		dm_array_info_init(&cmd->hint_info, cmd->tm, &vt);
266 	}
267 }
268 
269 static int __save_sm_root(struct dm_cache_metadata *cmd)
270 {
271 	int r;
272 	size_t metadata_len;
273 
274 	r = dm_sm_root_size(cmd->metadata_sm, &metadata_len);
275 	if (r < 0)
276 		return r;
277 
278 	return dm_sm_copy_root(cmd->metadata_sm, &cmd->metadata_space_map_root,
279 			       metadata_len);
280 }
281 
282 static void __copy_sm_root(struct dm_cache_metadata *cmd,
283 			   struct cache_disk_superblock *disk_super)
284 {
285 	memcpy(&disk_super->metadata_space_map_root,
286 	       &cmd->metadata_space_map_root,
287 	       sizeof(cmd->metadata_space_map_root));
288 }
289 
290 static int __write_initial_superblock(struct dm_cache_metadata *cmd)
291 {
292 	int r;
293 	struct dm_block *sblock;
294 	struct cache_disk_superblock *disk_super;
295 	sector_t bdev_size = i_size_read(cmd->bdev->bd_inode) >> SECTOR_SHIFT;
296 
297 	/* FIXME: see if we can lose the max sectors limit */
298 	if (bdev_size > DM_CACHE_METADATA_MAX_SECTORS)
299 		bdev_size = DM_CACHE_METADATA_MAX_SECTORS;
300 
301 	r = dm_tm_pre_commit(cmd->tm);
302 	if (r < 0)
303 		return r;
304 
305 	/*
306 	 * dm_sm_copy_root() can fail.  So we need to do it before we start
307 	 * updating the superblock.
308 	 */
309 	r = __save_sm_root(cmd);
310 	if (r)
311 		return r;
312 
313 	r = superblock_lock_zero(cmd, &sblock);
314 	if (r)
315 		return r;
316 
317 	disk_super = dm_block_data(sblock);
318 	disk_super->flags = 0;
319 	memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
320 	disk_super->magic = cpu_to_le64(CACHE_SUPERBLOCK_MAGIC);
321 	disk_super->version = cpu_to_le32(MAX_CACHE_VERSION);
322 	memset(disk_super->policy_name, 0, sizeof(disk_super->policy_name));
323 	memset(disk_super->policy_version, 0, sizeof(disk_super->policy_version));
324 	disk_super->policy_hint_size = 0;
325 
326 	__copy_sm_root(cmd, disk_super);
327 
328 	disk_super->mapping_root = cpu_to_le64(cmd->root);
329 	disk_super->hint_root = cpu_to_le64(cmd->hint_root);
330 	disk_super->discard_root = cpu_to_le64(cmd->discard_root);
331 	disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
332 	disk_super->discard_nr_blocks = cpu_to_le64(from_oblock(cmd->discard_nr_blocks));
333 	disk_super->metadata_block_size = cpu_to_le32(DM_CACHE_METADATA_BLOCK_SIZE);
334 	disk_super->data_block_size = cpu_to_le32(cmd->data_block_size);
335 	disk_super->cache_blocks = cpu_to_le32(0);
336 
337 	disk_super->read_hits = cpu_to_le32(0);
338 	disk_super->read_misses = cpu_to_le32(0);
339 	disk_super->write_hits = cpu_to_le32(0);
340 	disk_super->write_misses = cpu_to_le32(0);
341 
342 	return dm_tm_commit(cmd->tm, sblock);
343 }
344 
345 static int __format_metadata(struct dm_cache_metadata *cmd)
346 {
347 	int r;
348 
349 	r = dm_tm_create_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
350 				 &cmd->tm, &cmd->metadata_sm);
351 	if (r < 0) {
352 		DMERR("tm_create_with_sm failed");
353 		return r;
354 	}
355 
356 	__setup_mapping_info(cmd);
357 
358 	r = dm_array_empty(&cmd->info, &cmd->root);
359 	if (r < 0)
360 		goto bad;
361 
362 	dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
363 
364 	r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root);
365 	if (r < 0)
366 		goto bad;
367 
368 	cmd->discard_block_size = 0;
369 	cmd->discard_nr_blocks = 0;
370 
371 	r = __write_initial_superblock(cmd);
372 	if (r)
373 		goto bad;
374 
375 	cmd->clean_when_opened = true;
376 	return 0;
377 
378 bad:
379 	dm_tm_destroy(cmd->tm);
380 	dm_sm_destroy(cmd->metadata_sm);
381 
382 	return r;
383 }
384 
385 static int __check_incompat_features(struct cache_disk_superblock *disk_super,
386 				     struct dm_cache_metadata *cmd)
387 {
388 	uint32_t features;
389 
390 	features = le32_to_cpu(disk_super->incompat_flags) & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
391 	if (features) {
392 		DMERR("could not access metadata due to unsupported optional features (%lx).",
393 		      (unsigned long)features);
394 		return -EINVAL;
395 	}
396 
397 	/*
398 	 * Check for read-only metadata to skip the following RDWR checks.
399 	 */
400 	if (get_disk_ro(cmd->bdev->bd_disk))
401 		return 0;
402 
403 	features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP;
404 	if (features) {
405 		DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
406 		      (unsigned long)features);
407 		return -EINVAL;
408 	}
409 
410 	return 0;
411 }
412 
413 static int __open_metadata(struct dm_cache_metadata *cmd)
414 {
415 	int r;
416 	struct dm_block *sblock;
417 	struct cache_disk_superblock *disk_super;
418 	unsigned long sb_flags;
419 
420 	r = superblock_read_lock(cmd, &sblock);
421 	if (r < 0) {
422 		DMERR("couldn't read lock superblock");
423 		return r;
424 	}
425 
426 	disk_super = dm_block_data(sblock);
427 
428 	/* Verify the data block size hasn't changed */
429 	if (le32_to_cpu(disk_super->data_block_size) != cmd->data_block_size) {
430 		DMERR("changing the data block size (from %u to %llu) is not supported",
431 		      le32_to_cpu(disk_super->data_block_size),
432 		      (unsigned long long)cmd->data_block_size);
433 		r = -EINVAL;
434 		goto bad;
435 	}
436 
437 	r = __check_incompat_features(disk_super, cmd);
438 	if (r < 0)
439 		goto bad;
440 
441 	r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
442 			       disk_super->metadata_space_map_root,
443 			       sizeof(disk_super->metadata_space_map_root),
444 			       &cmd->tm, &cmd->metadata_sm);
445 	if (r < 0) {
446 		DMERR("tm_open_with_sm failed");
447 		goto bad;
448 	}
449 
450 	__setup_mapping_info(cmd);
451 	dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
452 	sb_flags = le32_to_cpu(disk_super->flags);
453 	cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags);
454 	return dm_bm_unlock(sblock);
455 
456 bad:
457 	dm_bm_unlock(sblock);
458 	return r;
459 }
460 
461 static int __open_or_format_metadata(struct dm_cache_metadata *cmd,
462 				     bool format_device)
463 {
464 	int r;
465 	bool unformatted = false;
466 
467 	r = __superblock_all_zeroes(cmd->bm, &unformatted);
468 	if (r)
469 		return r;
470 
471 	if (unformatted)
472 		return format_device ? __format_metadata(cmd) : -EPERM;
473 
474 	return __open_metadata(cmd);
475 }
476 
477 static int __create_persistent_data_objects(struct dm_cache_metadata *cmd,
478 					    bool may_format_device)
479 {
480 	int r;
481 	cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
482 					  CACHE_METADATA_CACHE_SIZE,
483 					  CACHE_MAX_CONCURRENT_LOCKS);
484 	if (IS_ERR(cmd->bm)) {
485 		DMERR("could not create block manager");
486 		return PTR_ERR(cmd->bm);
487 	}
488 
489 	r = __open_or_format_metadata(cmd, may_format_device);
490 	if (r)
491 		dm_block_manager_destroy(cmd->bm);
492 
493 	return r;
494 }
495 
496 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd)
497 {
498 	dm_sm_destroy(cmd->metadata_sm);
499 	dm_tm_destroy(cmd->tm);
500 	dm_block_manager_destroy(cmd->bm);
501 }
502 
503 typedef unsigned long (*flags_mutator)(unsigned long);
504 
505 static void update_flags(struct cache_disk_superblock *disk_super,
506 			 flags_mutator mutator)
507 {
508 	uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
509 	disk_super->flags = cpu_to_le32(sb_flags);
510 }
511 
512 static unsigned long set_clean_shutdown(unsigned long flags)
513 {
514 	set_bit(CLEAN_SHUTDOWN, &flags);
515 	return flags;
516 }
517 
518 static unsigned long clear_clean_shutdown(unsigned long flags)
519 {
520 	clear_bit(CLEAN_SHUTDOWN, &flags);
521 	return flags;
522 }
523 
524 static void read_superblock_fields(struct dm_cache_metadata *cmd,
525 				   struct cache_disk_superblock *disk_super)
526 {
527 	cmd->root = le64_to_cpu(disk_super->mapping_root);
528 	cmd->hint_root = le64_to_cpu(disk_super->hint_root);
529 	cmd->discard_root = le64_to_cpu(disk_super->discard_root);
530 	cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
531 	cmd->discard_nr_blocks = to_oblock(le64_to_cpu(disk_super->discard_nr_blocks));
532 	cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
533 	cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
534 	strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
535 	cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
536 	cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
537 	cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
538 	cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
539 
540 	cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
541 	cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
542 	cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
543 	cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
544 
545 	cmd->changed = false;
546 }
547 
548 /*
549  * The mutator updates the superblock flags.
550  */
551 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
552 				     flags_mutator mutator)
553 {
554 	int r;
555 	struct cache_disk_superblock *disk_super;
556 	struct dm_block *sblock;
557 
558 	r = superblock_lock(cmd, &sblock);
559 	if (r)
560 		return r;
561 
562 	disk_super = dm_block_data(sblock);
563 	update_flags(disk_super, mutator);
564 	read_superblock_fields(cmd, disk_super);
565 	dm_bm_unlock(sblock);
566 
567 	return dm_bm_flush(cmd->bm);
568 }
569 
570 static int __begin_transaction(struct dm_cache_metadata *cmd)
571 {
572 	int r;
573 	struct cache_disk_superblock *disk_super;
574 	struct dm_block *sblock;
575 
576 	/*
577 	 * We re-read the superblock every time.  Shouldn't need to do this
578 	 * really.
579 	 */
580 	r = superblock_read_lock(cmd, &sblock);
581 	if (r)
582 		return r;
583 
584 	disk_super = dm_block_data(sblock);
585 	read_superblock_fields(cmd, disk_super);
586 	dm_bm_unlock(sblock);
587 
588 	return 0;
589 }
590 
591 static int __commit_transaction(struct dm_cache_metadata *cmd,
592 				flags_mutator mutator)
593 {
594 	int r;
595 	struct cache_disk_superblock *disk_super;
596 	struct dm_block *sblock;
597 
598 	/*
599 	 * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
600 	 */
601 	BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
602 
603 	r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
604 			    &cmd->discard_root);
605 	if (r)
606 		return r;
607 
608 	r = dm_tm_pre_commit(cmd->tm);
609 	if (r < 0)
610 		return r;
611 
612 	r = __save_sm_root(cmd);
613 	if (r)
614 		return r;
615 
616 	r = superblock_lock(cmd, &sblock);
617 	if (r)
618 		return r;
619 
620 	disk_super = dm_block_data(sblock);
621 
622 	if (mutator)
623 		update_flags(disk_super, mutator);
624 
625 	disk_super->mapping_root = cpu_to_le64(cmd->root);
626 	disk_super->hint_root = cpu_to_le64(cmd->hint_root);
627 	disk_super->discard_root = cpu_to_le64(cmd->discard_root);
628 	disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
629 	disk_super->discard_nr_blocks = cpu_to_le64(from_oblock(cmd->discard_nr_blocks));
630 	disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
631 	strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
632 	disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
633 	disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
634 	disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
635 
636 	disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
637 	disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
638 	disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
639 	disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
640 	__copy_sm_root(cmd, disk_super);
641 
642 	return dm_tm_commit(cmd->tm, sblock);
643 }
644 
645 /*----------------------------------------------------------------*/
646 
647 /*
648  * The mappings are held in a dm-array that has 64-bit values stored in
649  * little-endian format.  The index is the cblock, the high 48bits of the
650  * value are the oblock and the low 16 bit the flags.
651  */
652 #define FLAGS_MASK ((1 << 16) - 1)
653 
654 static __le64 pack_value(dm_oblock_t block, unsigned flags)
655 {
656 	uint64_t value = from_oblock(block);
657 	value <<= 16;
658 	value = value | (flags & FLAGS_MASK);
659 	return cpu_to_le64(value);
660 }
661 
662 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
663 {
664 	uint64_t value = le64_to_cpu(value_le);
665 	uint64_t b = value >> 16;
666 	*block = to_oblock(b);
667 	*flags = value & FLAGS_MASK;
668 }
669 
670 /*----------------------------------------------------------------*/
671 
672 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
673 						 sector_t data_block_size,
674 						 bool may_format_device,
675 						 size_t policy_hint_size)
676 {
677 	int r;
678 	struct dm_cache_metadata *cmd;
679 
680 	cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
681 	if (!cmd) {
682 		DMERR("could not allocate metadata struct");
683 		return NULL;
684 	}
685 
686 	init_rwsem(&cmd->root_lock);
687 	cmd->bdev = bdev;
688 	cmd->data_block_size = data_block_size;
689 	cmd->cache_blocks = 0;
690 	cmd->policy_hint_size = policy_hint_size;
691 	cmd->changed = true;
692 
693 	r = __create_persistent_data_objects(cmd, may_format_device);
694 	if (r) {
695 		kfree(cmd);
696 		return ERR_PTR(r);
697 	}
698 
699 	r = __begin_transaction_flags(cmd, clear_clean_shutdown);
700 	if (r < 0) {
701 		dm_cache_metadata_close(cmd);
702 		return ERR_PTR(r);
703 	}
704 
705 	return cmd;
706 }
707 
708 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
709 {
710 	__destroy_persistent_data_objects(cmd);
711 	kfree(cmd);
712 }
713 
714 /*
715  * Checks that the given cache block is either unmapped or clean.
716  */
717 static int block_unmapped_or_clean(struct dm_cache_metadata *cmd, dm_cblock_t b,
718 				   bool *result)
719 {
720 	int r;
721 	__le64 value;
722 	dm_oblock_t ob;
723 	unsigned flags;
724 
725 	r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
726 	if (r) {
727 		DMERR("block_unmapped_or_clean failed");
728 		return r;
729 	}
730 
731 	unpack_value(value, &ob, &flags);
732 	*result = !((flags & M_VALID) && (flags & M_DIRTY));
733 
734 	return 0;
735 }
736 
737 static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
738 					dm_cblock_t begin, dm_cblock_t end,
739 					bool *result)
740 {
741 	int r;
742 	*result = true;
743 
744 	while (begin != end) {
745 		r = block_unmapped_or_clean(cmd, begin, result);
746 		if (r)
747 			return r;
748 
749 		if (!*result) {
750 			DMERR("cache block %llu is dirty",
751 			      (unsigned long long) from_cblock(begin));
752 			return 0;
753 		}
754 
755 		begin = to_cblock(from_cblock(begin) + 1);
756 	}
757 
758 	return 0;
759 }
760 
761 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
762 {
763 	int r;
764 	bool clean;
765 	__le64 null_mapping = pack_value(0, 0);
766 
767 	down_write(&cmd->root_lock);
768 	__dm_bless_for_disk(&null_mapping);
769 
770 	if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) {
771 		r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean);
772 		if (r) {
773 			__dm_unbless_for_disk(&null_mapping);
774 			goto out;
775 		}
776 
777 		if (!clean) {
778 			DMERR("unable to shrink cache due to dirty blocks");
779 			r = -EINVAL;
780 			__dm_unbless_for_disk(&null_mapping);
781 			goto out;
782 		}
783 	}
784 
785 	r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
786 			    from_cblock(new_cache_size),
787 			    &null_mapping, &cmd->root);
788 	if (!r)
789 		cmd->cache_blocks = new_cache_size;
790 	cmd->changed = true;
791 
792 out:
793 	up_write(&cmd->root_lock);
794 
795 	return r;
796 }
797 
798 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
799 				   sector_t discard_block_size,
800 				   dm_oblock_t new_nr_entries)
801 {
802 	int r;
803 
804 	down_write(&cmd->root_lock);
805 	r = dm_bitset_resize(&cmd->discard_info,
806 			     cmd->discard_root,
807 			     from_oblock(cmd->discard_nr_blocks),
808 			     from_oblock(new_nr_entries),
809 			     false, &cmd->discard_root);
810 	if (!r) {
811 		cmd->discard_block_size = discard_block_size;
812 		cmd->discard_nr_blocks = new_nr_entries;
813 	}
814 
815 	cmd->changed = true;
816 	up_write(&cmd->root_lock);
817 
818 	return r;
819 }
820 
821 static int __set_discard(struct dm_cache_metadata *cmd, dm_oblock_t b)
822 {
823 	return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
824 				 from_oblock(b), &cmd->discard_root);
825 }
826 
827 static int __clear_discard(struct dm_cache_metadata *cmd, dm_oblock_t b)
828 {
829 	return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
830 				   from_oblock(b), &cmd->discard_root);
831 }
832 
833 static int __is_discarded(struct dm_cache_metadata *cmd, dm_oblock_t b,
834 			  bool *is_discarded)
835 {
836 	return dm_bitset_test_bit(&cmd->discard_info, cmd->discard_root,
837 				  from_oblock(b), &cmd->discard_root,
838 				  is_discarded);
839 }
840 
841 static int __discard(struct dm_cache_metadata *cmd,
842 		     dm_oblock_t dblock, bool discard)
843 {
844 	int r;
845 
846 	r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
847 	if (r)
848 		return r;
849 
850 	cmd->changed = true;
851 	return 0;
852 }
853 
854 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
855 			 dm_oblock_t dblock, bool discard)
856 {
857 	int r;
858 
859 	down_write(&cmd->root_lock);
860 	r = __discard(cmd, dblock, discard);
861 	up_write(&cmd->root_lock);
862 
863 	return r;
864 }
865 
866 static int __load_discards(struct dm_cache_metadata *cmd,
867 			   load_discard_fn fn, void *context)
868 {
869 	int r = 0;
870 	dm_block_t b;
871 	bool discard;
872 
873 	for (b = 0; b < from_oblock(cmd->discard_nr_blocks); b++) {
874 		dm_oblock_t dblock = to_oblock(b);
875 
876 		if (cmd->clean_when_opened) {
877 			r = __is_discarded(cmd, dblock, &discard);
878 			if (r)
879 				return r;
880 		} else
881 			discard = false;
882 
883 		r = fn(context, cmd->discard_block_size, dblock, discard);
884 		if (r)
885 			break;
886 	}
887 
888 	return r;
889 }
890 
891 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
892 			   load_discard_fn fn, void *context)
893 {
894 	int r;
895 
896 	down_read(&cmd->root_lock);
897 	r = __load_discards(cmd, fn, context);
898 	up_read(&cmd->root_lock);
899 
900 	return r;
901 }
902 
903 dm_cblock_t dm_cache_size(struct dm_cache_metadata *cmd)
904 {
905 	dm_cblock_t r;
906 
907 	down_read(&cmd->root_lock);
908 	r = cmd->cache_blocks;
909 	up_read(&cmd->root_lock);
910 
911 	return r;
912 }
913 
914 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
915 {
916 	int r;
917 	__le64 value = pack_value(0, 0);
918 
919 	__dm_bless_for_disk(&value);
920 	r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
921 			       &value, &cmd->root);
922 	if (r)
923 		return r;
924 
925 	cmd->changed = true;
926 	return 0;
927 }
928 
929 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
930 {
931 	int r;
932 
933 	down_write(&cmd->root_lock);
934 	r = __remove(cmd, cblock);
935 	up_write(&cmd->root_lock);
936 
937 	return r;
938 }
939 
940 static int __insert(struct dm_cache_metadata *cmd,
941 		    dm_cblock_t cblock, dm_oblock_t oblock)
942 {
943 	int r;
944 	__le64 value = pack_value(oblock, M_VALID);
945 	__dm_bless_for_disk(&value);
946 
947 	r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
948 			       &value, &cmd->root);
949 	if (r)
950 		return r;
951 
952 	cmd->changed = true;
953 	return 0;
954 }
955 
956 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
957 			    dm_cblock_t cblock, dm_oblock_t oblock)
958 {
959 	int r;
960 
961 	down_write(&cmd->root_lock);
962 	r = __insert(cmd, cblock, oblock);
963 	up_write(&cmd->root_lock);
964 
965 	return r;
966 }
967 
968 struct thunk {
969 	load_mapping_fn fn;
970 	void *context;
971 
972 	struct dm_cache_metadata *cmd;
973 	bool respect_dirty_flags;
974 	bool hints_valid;
975 };
976 
977 static bool policy_unchanged(struct dm_cache_metadata *cmd,
978 			     struct dm_cache_policy *policy)
979 {
980 	const char *policy_name = dm_cache_policy_get_name(policy);
981 	const unsigned *policy_version = dm_cache_policy_get_version(policy);
982 	size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
983 
984 	/*
985 	 * Ensure policy names match.
986 	 */
987 	if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
988 		return false;
989 
990 	/*
991 	 * Ensure policy major versions match.
992 	 */
993 	if (cmd->policy_version[0] != policy_version[0])
994 		return false;
995 
996 	/*
997 	 * Ensure policy hint sizes match.
998 	 */
999 	if (cmd->policy_hint_size != policy_hint_size)
1000 		return false;
1001 
1002 	return true;
1003 }
1004 
1005 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1006 {
1007 	return cmd->hint_root && cmd->policy_hint_size;
1008 }
1009 
1010 static bool hints_array_available(struct dm_cache_metadata *cmd,
1011 				  struct dm_cache_policy *policy)
1012 {
1013 	return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1014 		hints_array_initialized(cmd);
1015 }
1016 
1017 static int __load_mapping(void *context, uint64_t cblock, void *leaf)
1018 {
1019 	int r = 0;
1020 	bool dirty;
1021 	__le64 value;
1022 	__le32 hint_value = 0;
1023 	dm_oblock_t oblock;
1024 	unsigned flags;
1025 	struct thunk *thunk = context;
1026 	struct dm_cache_metadata *cmd = thunk->cmd;
1027 
1028 	memcpy(&value, leaf, sizeof(value));
1029 	unpack_value(value, &oblock, &flags);
1030 
1031 	if (flags & M_VALID) {
1032 		if (thunk->hints_valid) {
1033 			r = dm_array_get_value(&cmd->hint_info, cmd->hint_root,
1034 					       cblock, &hint_value);
1035 			if (r && r != -ENODATA)
1036 				return r;
1037 		}
1038 
1039 		dirty = thunk->respect_dirty_flags ? (flags & M_DIRTY) : true;
1040 		r = thunk->fn(thunk->context, oblock, to_cblock(cblock),
1041 			      dirty, le32_to_cpu(hint_value), thunk->hints_valid);
1042 	}
1043 
1044 	return r;
1045 }
1046 
1047 static int __load_mappings(struct dm_cache_metadata *cmd,
1048 			   struct dm_cache_policy *policy,
1049 			   load_mapping_fn fn, void *context)
1050 {
1051 	struct thunk thunk;
1052 
1053 	thunk.fn = fn;
1054 	thunk.context = context;
1055 
1056 	thunk.cmd = cmd;
1057 	thunk.respect_dirty_flags = cmd->clean_when_opened;
1058 	thunk.hints_valid = hints_array_available(cmd, policy);
1059 
1060 	return dm_array_walk(&cmd->info, cmd->root, __load_mapping, &thunk);
1061 }
1062 
1063 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1064 			   struct dm_cache_policy *policy,
1065 			   load_mapping_fn fn, void *context)
1066 {
1067 	int r;
1068 
1069 	down_read(&cmd->root_lock);
1070 	r = __load_mappings(cmd, policy, fn, context);
1071 	up_read(&cmd->root_lock);
1072 
1073 	return r;
1074 }
1075 
1076 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
1077 {
1078 	int r = 0;
1079 	__le64 value;
1080 	dm_oblock_t oblock;
1081 	unsigned flags;
1082 
1083 	memcpy(&value, leaf, sizeof(value));
1084 	unpack_value(value, &oblock, &flags);
1085 
1086 	return r;
1087 }
1088 
1089 static int __dump_mappings(struct dm_cache_metadata *cmd)
1090 {
1091 	return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1092 }
1093 
1094 void dm_cache_dump(struct dm_cache_metadata *cmd)
1095 {
1096 	down_read(&cmd->root_lock);
1097 	__dump_mappings(cmd);
1098 	up_read(&cmd->root_lock);
1099 }
1100 
1101 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1102 {
1103 	int r;
1104 
1105 	down_read(&cmd->root_lock);
1106 	r = cmd->changed;
1107 	up_read(&cmd->root_lock);
1108 
1109 	return r;
1110 }
1111 
1112 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1113 {
1114 	int r;
1115 	unsigned flags;
1116 	dm_oblock_t oblock;
1117 	__le64 value;
1118 
1119 	r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1120 	if (r)
1121 		return r;
1122 
1123 	unpack_value(value, &oblock, &flags);
1124 
1125 	if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1126 		/* nothing to be done */
1127 		return 0;
1128 
1129 	value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1130 	__dm_bless_for_disk(&value);
1131 
1132 	r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1133 			       &value, &cmd->root);
1134 	if (r)
1135 		return r;
1136 
1137 	cmd->changed = true;
1138 	return 0;
1139 
1140 }
1141 
1142 int dm_cache_set_dirty(struct dm_cache_metadata *cmd,
1143 		       dm_cblock_t cblock, bool dirty)
1144 {
1145 	int r;
1146 
1147 	down_write(&cmd->root_lock);
1148 	r = __dirty(cmd, cblock, dirty);
1149 	up_write(&cmd->root_lock);
1150 
1151 	return r;
1152 }
1153 
1154 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1155 				 struct dm_cache_statistics *stats)
1156 {
1157 	down_read(&cmd->root_lock);
1158 	*stats = cmd->stats;
1159 	up_read(&cmd->root_lock);
1160 }
1161 
1162 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1163 				 struct dm_cache_statistics *stats)
1164 {
1165 	down_write(&cmd->root_lock);
1166 	cmd->stats = *stats;
1167 	up_write(&cmd->root_lock);
1168 }
1169 
1170 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1171 {
1172 	int r;
1173 	flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1174 				 clear_clean_shutdown);
1175 
1176 	down_write(&cmd->root_lock);
1177 	r = __commit_transaction(cmd, mutator);
1178 	if (r)
1179 		goto out;
1180 
1181 	r = __begin_transaction(cmd);
1182 
1183 out:
1184 	up_write(&cmd->root_lock);
1185 	return r;
1186 }
1187 
1188 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1189 					   dm_block_t *result)
1190 {
1191 	int r = -EINVAL;
1192 
1193 	down_read(&cmd->root_lock);
1194 	r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1195 	up_read(&cmd->root_lock);
1196 
1197 	return r;
1198 }
1199 
1200 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1201 				   dm_block_t *result)
1202 {
1203 	int r = -EINVAL;
1204 
1205 	down_read(&cmd->root_lock);
1206 	r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1207 	up_read(&cmd->root_lock);
1208 
1209 	return r;
1210 }
1211 
1212 /*----------------------------------------------------------------*/
1213 
1214 static int begin_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1215 {
1216 	int r;
1217 	__le32 value;
1218 	size_t hint_size;
1219 	const char *policy_name = dm_cache_policy_get_name(policy);
1220 	const unsigned *policy_version = dm_cache_policy_get_version(policy);
1221 
1222 	if (!policy_name[0] ||
1223 	    (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1224 		return -EINVAL;
1225 
1226 	if (!policy_unchanged(cmd, policy)) {
1227 		strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1228 		memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1229 
1230 		hint_size = dm_cache_policy_get_hint_size(policy);
1231 		if (!hint_size)
1232 			return 0; /* short-circuit hints initialization */
1233 		cmd->policy_hint_size = hint_size;
1234 
1235 		if (cmd->hint_root) {
1236 			r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1237 			if (r)
1238 				return r;
1239 		}
1240 
1241 		r = dm_array_empty(&cmd->hint_info, &cmd->hint_root);
1242 		if (r)
1243 			return r;
1244 
1245 		value = cpu_to_le32(0);
1246 		__dm_bless_for_disk(&value);
1247 		r = dm_array_resize(&cmd->hint_info, cmd->hint_root, 0,
1248 				    from_cblock(cmd->cache_blocks),
1249 				    &value, &cmd->hint_root);
1250 		if (r)
1251 			return r;
1252 	}
1253 
1254 	return 0;
1255 }
1256 
1257 static int save_hint(void *context, dm_cblock_t cblock, dm_oblock_t oblock, uint32_t hint)
1258 {
1259 	struct dm_cache_metadata *cmd = context;
1260 	__le32 value = cpu_to_le32(hint);
1261 	int r;
1262 
1263 	__dm_bless_for_disk(&value);
1264 
1265 	r = dm_array_set_value(&cmd->hint_info, cmd->hint_root,
1266 			       from_cblock(cblock), &value, &cmd->hint_root);
1267 	cmd->changed = true;
1268 
1269 	return r;
1270 }
1271 
1272 static int write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1273 {
1274 	int r;
1275 
1276 	r = begin_hints(cmd, policy);
1277 	if (r) {
1278 		DMERR("begin_hints failed");
1279 		return r;
1280 	}
1281 
1282 	return policy_walk_mappings(policy, save_hint, cmd);
1283 }
1284 
1285 int dm_cache_write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1286 {
1287 	int r;
1288 
1289 	down_write(&cmd->root_lock);
1290 	r = write_hints(cmd, policy);
1291 	up_write(&cmd->root_lock);
1292 
1293 	return r;
1294 }
1295 
1296 int dm_cache_metadata_all_clean(struct dm_cache_metadata *cmd, bool *result)
1297 {
1298 	return blocks_are_unmapped_or_clean(cmd, 0, cmd->cache_blocks, result);
1299 }
1300