Commit | Line | Data |
---|---|---|
c6b4fcba JT |
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 | |
53d49819 JT |
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 | ||
c6b4fcba JT |
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, | |
028ae9f7 JT |
42 | /* metadata must be checked using the tools */ |
43 | NEEDS_CHECK, | |
c6b4fcba JT |
44 | }; |
45 | ||
46 | /* | |
47 | * Each mapping from cache block -> origin block carries a set of flags. | |
48 | */ | |
49 | enum mapping_bits { | |
50 | /* | |
51 | * A valid mapping. Because we're using an array we clear this | |
52 | * flag for an non existant mapping. | |
53 | */ | |
54 | M_VALID = 1, | |
55 | ||
56 | /* | |
57 | * The data on the cache is different from that on the origin. | |
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; | |
4e7f506f MS |
94 | |
95 | __le32 policy_version[CACHE_POLICY_VERSION_SIZE]; | |
c6b4fcba JT |
96 | } __packed; |
97 | ||
98 | struct dm_cache_metadata { | |
9b1cc9f2 JT |
99 | atomic_t ref_count; |
100 | struct list_head list; | |
101 | ||
c6b4fcba JT |
102 | struct block_device *bdev; |
103 | struct dm_block_manager *bm; | |
104 | struct dm_space_map *metadata_sm; | |
105 | struct dm_transaction_manager *tm; | |
106 | ||
107 | struct dm_array_info info; | |
108 | struct dm_array_info hint_info; | |
109 | struct dm_disk_bitset discard_info; | |
110 | ||
111 | struct rw_semaphore root_lock; | |
028ae9f7 | 112 | unsigned long flags; |
c6b4fcba JT |
113 | dm_block_t root; |
114 | dm_block_t hint_root; | |
115 | dm_block_t discard_root; | |
116 | ||
117 | sector_t discard_block_size; | |
1bad9bc4 | 118 | dm_dblock_t discard_nr_blocks; |
c6b4fcba JT |
119 | |
120 | sector_t data_block_size; | |
121 | dm_cblock_t cache_blocks; | |
122 | bool changed:1; | |
123 | bool clean_when_opened:1; | |
124 | ||
125 | char policy_name[CACHE_POLICY_NAME_SIZE]; | |
4e7f506f | 126 | unsigned policy_version[CACHE_POLICY_VERSION_SIZE]; |
c6b4fcba JT |
127 | size_t policy_hint_size; |
128 | struct dm_cache_statistics stats; | |
5a32083d JT |
129 | |
130 | /* | |
131 | * Reading the space map root can fail, so we read it into this | |
132 | * buffer before the superblock is locked and updated. | |
133 | */ | |
134 | __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE]; | |
028ae9f7 JT |
135 | |
136 | /* | |
137 | * Set if a transaction has to be aborted but the attempt to roll | |
138 | * back to the previous (good) transaction failed. The only | |
139 | * metadata operation permissible in this state is the closing of | |
140 | * the device. | |
141 | */ | |
142 | bool fail_io:1; | |
c6b4fcba JT |
143 | }; |
144 | ||
145 | /*------------------------------------------------------------------- | |
146 | * superblock validator | |
147 | *-----------------------------------------------------------------*/ | |
148 | ||
149 | #define SUPERBLOCK_CSUM_XOR 9031977 | |
150 | ||
151 | static void sb_prepare_for_write(struct dm_block_validator *v, | |
152 | struct dm_block *b, | |
153 | size_t sb_block_size) | |
154 | { | |
155 | struct cache_disk_superblock *disk_super = dm_block_data(b); | |
156 | ||
157 | disk_super->blocknr = cpu_to_le64(dm_block_location(b)); | |
158 | disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags, | |
159 | sb_block_size - sizeof(__le32), | |
160 | SUPERBLOCK_CSUM_XOR)); | |
161 | } | |
162 | ||
53d49819 JT |
163 | static int check_metadata_version(struct cache_disk_superblock *disk_super) |
164 | { | |
165 | uint32_t metadata_version = le32_to_cpu(disk_super->version); | |
166 | if (metadata_version < MIN_CACHE_VERSION || metadata_version > MAX_CACHE_VERSION) { | |
167 | DMERR("Cache metadata version %u found, but only versions between %u and %u supported.", | |
168 | metadata_version, MIN_CACHE_VERSION, MAX_CACHE_VERSION); | |
169 | return -EINVAL; | |
170 | } | |
171 | ||
172 | return 0; | |
173 | } | |
174 | ||
c6b4fcba JT |
175 | static int sb_check(struct dm_block_validator *v, |
176 | struct dm_block *b, | |
177 | size_t sb_block_size) | |
178 | { | |
179 | struct cache_disk_superblock *disk_super = dm_block_data(b); | |
180 | __le32 csum_le; | |
181 | ||
182 | if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) { | |
183 | DMERR("sb_check failed: blocknr %llu: wanted %llu", | |
184 | le64_to_cpu(disk_super->blocknr), | |
185 | (unsigned long long)dm_block_location(b)); | |
186 | return -ENOTBLK; | |
187 | } | |
188 | ||
189 | if (le64_to_cpu(disk_super->magic) != CACHE_SUPERBLOCK_MAGIC) { | |
190 | DMERR("sb_check failed: magic %llu: wanted %llu", | |
191 | le64_to_cpu(disk_super->magic), | |
192 | (unsigned long long)CACHE_SUPERBLOCK_MAGIC); | |
193 | return -EILSEQ; | |
194 | } | |
195 | ||
196 | csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags, | |
197 | sb_block_size - sizeof(__le32), | |
198 | SUPERBLOCK_CSUM_XOR)); | |
199 | if (csum_le != disk_super->csum) { | |
200 | DMERR("sb_check failed: csum %u: wanted %u", | |
201 | le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum)); | |
202 | return -EILSEQ; | |
203 | } | |
204 | ||
53d49819 | 205 | return check_metadata_version(disk_super); |
c6b4fcba JT |
206 | } |
207 | ||
208 | static struct dm_block_validator sb_validator = { | |
209 | .name = "superblock", | |
210 | .prepare_for_write = sb_prepare_for_write, | |
211 | .check = sb_check | |
212 | }; | |
213 | ||
214 | /*----------------------------------------------------------------*/ | |
215 | ||
216 | static int superblock_read_lock(struct dm_cache_metadata *cmd, | |
217 | struct dm_block **sblock) | |
218 | { | |
219 | return dm_bm_read_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION, | |
220 | &sb_validator, sblock); | |
221 | } | |
222 | ||
223 | static int superblock_lock_zero(struct dm_cache_metadata *cmd, | |
224 | struct dm_block **sblock) | |
225 | { | |
226 | return dm_bm_write_lock_zero(cmd->bm, CACHE_SUPERBLOCK_LOCATION, | |
227 | &sb_validator, sblock); | |
228 | } | |
229 | ||
230 | static int superblock_lock(struct dm_cache_metadata *cmd, | |
231 | struct dm_block **sblock) | |
232 | { | |
233 | return dm_bm_write_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION, | |
234 | &sb_validator, sblock); | |
235 | } | |
236 | ||
237 | /*----------------------------------------------------------------*/ | |
238 | ||
dd8b0c20 | 239 | static int __superblock_all_zeroes(struct dm_block_manager *bm, bool *result) |
c6b4fcba JT |
240 | { |
241 | int r; | |
242 | unsigned i; | |
243 | struct dm_block *b; | |
244 | __le64 *data_le, zero = cpu_to_le64(0); | |
245 | unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64); | |
246 | ||
247 | /* | |
248 | * We can't use a validator here - it may be all zeroes. | |
249 | */ | |
250 | r = dm_bm_read_lock(bm, CACHE_SUPERBLOCK_LOCATION, NULL, &b); | |
251 | if (r) | |
252 | return r; | |
253 | ||
254 | data_le = dm_block_data(b); | |
dd8b0c20 | 255 | *result = true; |
c6b4fcba JT |
256 | for (i = 0; i < sb_block_size; i++) { |
257 | if (data_le[i] != zero) { | |
dd8b0c20 | 258 | *result = false; |
c6b4fcba JT |
259 | break; |
260 | } | |
261 | } | |
262 | ||
4c7da06f MP |
263 | dm_bm_unlock(b); |
264 | ||
265 | return 0; | |
c6b4fcba JT |
266 | } |
267 | ||
268 | static void __setup_mapping_info(struct dm_cache_metadata *cmd) | |
269 | { | |
270 | struct dm_btree_value_type vt; | |
271 | ||
272 | vt.context = NULL; | |
273 | vt.size = sizeof(__le64); | |
274 | vt.inc = NULL; | |
275 | vt.dec = NULL; | |
276 | vt.equal = NULL; | |
277 | dm_array_info_init(&cmd->info, cmd->tm, &vt); | |
278 | ||
279 | if (cmd->policy_hint_size) { | |
280 | vt.size = sizeof(__le32); | |
281 | dm_array_info_init(&cmd->hint_info, cmd->tm, &vt); | |
282 | } | |
283 | } | |
284 | ||
5a32083d JT |
285 | static int __save_sm_root(struct dm_cache_metadata *cmd) |
286 | { | |
287 | int r; | |
288 | size_t metadata_len; | |
289 | ||
290 | r = dm_sm_root_size(cmd->metadata_sm, &metadata_len); | |
291 | if (r < 0) | |
292 | return r; | |
293 | ||
294 | return dm_sm_copy_root(cmd->metadata_sm, &cmd->metadata_space_map_root, | |
295 | metadata_len); | |
296 | } | |
297 | ||
298 | static void __copy_sm_root(struct dm_cache_metadata *cmd, | |
299 | struct cache_disk_superblock *disk_super) | |
300 | { | |
301 | memcpy(&disk_super->metadata_space_map_root, | |
302 | &cmd->metadata_space_map_root, | |
303 | sizeof(cmd->metadata_space_map_root)); | |
304 | } | |
305 | ||
c6b4fcba JT |
306 | static int __write_initial_superblock(struct dm_cache_metadata *cmd) |
307 | { | |
308 | int r; | |
309 | struct dm_block *sblock; | |
c6b4fcba JT |
310 | struct cache_disk_superblock *disk_super; |
311 | sector_t bdev_size = i_size_read(cmd->bdev->bd_inode) >> SECTOR_SHIFT; | |
312 | ||
313 | /* FIXME: see if we can lose the max sectors limit */ | |
314 | if (bdev_size > DM_CACHE_METADATA_MAX_SECTORS) | |
315 | bdev_size = DM_CACHE_METADATA_MAX_SECTORS; | |
316 | ||
5a32083d | 317 | r = dm_tm_pre_commit(cmd->tm); |
c6b4fcba JT |
318 | if (r < 0) |
319 | return r; | |
320 | ||
5a32083d JT |
321 | /* |
322 | * dm_sm_copy_root() can fail. So we need to do it before we start | |
323 | * updating the superblock. | |
324 | */ | |
325 | r = __save_sm_root(cmd); | |
326 | if (r) | |
c6b4fcba JT |
327 | return r; |
328 | ||
329 | r = superblock_lock_zero(cmd, &sblock); | |
330 | if (r) | |
331 | return r; | |
332 | ||
333 | disk_super = dm_block_data(sblock); | |
334 | disk_super->flags = 0; | |
335 | memset(disk_super->uuid, 0, sizeof(disk_super->uuid)); | |
336 | disk_super->magic = cpu_to_le64(CACHE_SUPERBLOCK_MAGIC); | |
53d49819 | 337 | disk_super->version = cpu_to_le32(MAX_CACHE_VERSION); |
4e7f506f MS |
338 | memset(disk_super->policy_name, 0, sizeof(disk_super->policy_name)); |
339 | memset(disk_super->policy_version, 0, sizeof(disk_super->policy_version)); | |
c6b4fcba JT |
340 | disk_super->policy_hint_size = 0; |
341 | ||
5a32083d | 342 | __copy_sm_root(cmd, disk_super); |
c6b4fcba JT |
343 | |
344 | disk_super->mapping_root = cpu_to_le64(cmd->root); | |
345 | disk_super->hint_root = cpu_to_le64(cmd->hint_root); | |
346 | disk_super->discard_root = cpu_to_le64(cmd->discard_root); | |
347 | disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size); | |
1bad9bc4 | 348 | disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks)); |
895b47d7 | 349 | disk_super->metadata_block_size = cpu_to_le32(DM_CACHE_METADATA_BLOCK_SIZE); |
c6b4fcba JT |
350 | disk_super->data_block_size = cpu_to_le32(cmd->data_block_size); |
351 | disk_super->cache_blocks = cpu_to_le32(0); | |
c6b4fcba JT |
352 | |
353 | disk_super->read_hits = cpu_to_le32(0); | |
354 | disk_super->read_misses = cpu_to_le32(0); | |
355 | disk_super->write_hits = cpu_to_le32(0); | |
356 | disk_super->write_misses = cpu_to_le32(0); | |
357 | ||
358 | return dm_tm_commit(cmd->tm, sblock); | |
c6b4fcba JT |
359 | } |
360 | ||
361 | static int __format_metadata(struct dm_cache_metadata *cmd) | |
362 | { | |
363 | int r; | |
364 | ||
365 | r = dm_tm_create_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION, | |
366 | &cmd->tm, &cmd->metadata_sm); | |
367 | if (r < 0) { | |
368 | DMERR("tm_create_with_sm failed"); | |
369 | return r; | |
370 | } | |
371 | ||
372 | __setup_mapping_info(cmd); | |
373 | ||
374 | r = dm_array_empty(&cmd->info, &cmd->root); | |
375 | if (r < 0) | |
376 | goto bad; | |
377 | ||
378 | dm_disk_bitset_init(cmd->tm, &cmd->discard_info); | |
379 | ||
380 | r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root); | |
381 | if (r < 0) | |
382 | goto bad; | |
383 | ||
384 | cmd->discard_block_size = 0; | |
385 | cmd->discard_nr_blocks = 0; | |
386 | ||
387 | r = __write_initial_superblock(cmd); | |
388 | if (r) | |
389 | goto bad; | |
390 | ||
391 | cmd->clean_when_opened = true; | |
392 | return 0; | |
393 | ||
394 | bad: | |
395 | dm_tm_destroy(cmd->tm); | |
396 | dm_sm_destroy(cmd->metadata_sm); | |
397 | ||
398 | return r; | |
399 | } | |
400 | ||
401 | static int __check_incompat_features(struct cache_disk_superblock *disk_super, | |
402 | struct dm_cache_metadata *cmd) | |
403 | { | |
404 | uint32_t features; | |
405 | ||
406 | features = le32_to_cpu(disk_super->incompat_flags) & ~DM_CACHE_FEATURE_INCOMPAT_SUPP; | |
407 | if (features) { | |
408 | DMERR("could not access metadata due to unsupported optional features (%lx).", | |
409 | (unsigned long)features); | |
410 | return -EINVAL; | |
411 | } | |
412 | ||
413 | /* | |
414 | * Check for read-only metadata to skip the following RDWR checks. | |
415 | */ | |
416 | if (get_disk_ro(cmd->bdev->bd_disk)) | |
417 | return 0; | |
418 | ||
419 | features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP; | |
420 | if (features) { | |
421 | DMERR("could not access metadata RDWR due to unsupported optional features (%lx).", | |
422 | (unsigned long)features); | |
423 | return -EINVAL; | |
424 | } | |
425 | ||
426 | return 0; | |
427 | } | |
428 | ||
429 | static int __open_metadata(struct dm_cache_metadata *cmd) | |
430 | { | |
431 | int r; | |
432 | struct dm_block *sblock; | |
433 | struct cache_disk_superblock *disk_super; | |
434 | unsigned long sb_flags; | |
435 | ||
436 | r = superblock_read_lock(cmd, &sblock); | |
437 | if (r < 0) { | |
438 | DMERR("couldn't read lock superblock"); | |
439 | return r; | |
440 | } | |
441 | ||
442 | disk_super = dm_block_data(sblock); | |
443 | ||
048e5a07 MS |
444 | /* Verify the data block size hasn't changed */ |
445 | if (le32_to_cpu(disk_super->data_block_size) != cmd->data_block_size) { | |
446 | DMERR("changing the data block size (from %u to %llu) is not supported", | |
447 | le32_to_cpu(disk_super->data_block_size), | |
448 | (unsigned long long)cmd->data_block_size); | |
449 | r = -EINVAL; | |
450 | goto bad; | |
451 | } | |
452 | ||
c6b4fcba JT |
453 | r = __check_incompat_features(disk_super, cmd); |
454 | if (r < 0) | |
455 | goto bad; | |
456 | ||
457 | r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION, | |
458 | disk_super->metadata_space_map_root, | |
459 | sizeof(disk_super->metadata_space_map_root), | |
460 | &cmd->tm, &cmd->metadata_sm); | |
461 | if (r < 0) { | |
462 | DMERR("tm_open_with_sm failed"); | |
463 | goto bad; | |
464 | } | |
465 | ||
466 | __setup_mapping_info(cmd); | |
467 | dm_disk_bitset_init(cmd->tm, &cmd->discard_info); | |
468 | sb_flags = le32_to_cpu(disk_super->flags); | |
469 | cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags); | |
4c7da06f MP |
470 | dm_bm_unlock(sblock); |
471 | ||
472 | return 0; | |
c6b4fcba JT |
473 | |
474 | bad: | |
475 | dm_bm_unlock(sblock); | |
476 | return r; | |
477 | } | |
478 | ||
479 | static int __open_or_format_metadata(struct dm_cache_metadata *cmd, | |
480 | bool format_device) | |
481 | { | |
dd8b0c20 JT |
482 | int r; |
483 | bool unformatted = false; | |
c6b4fcba JT |
484 | |
485 | r = __superblock_all_zeroes(cmd->bm, &unformatted); | |
486 | if (r) | |
487 | return r; | |
488 | ||
489 | if (unformatted) | |
490 | return format_device ? __format_metadata(cmd) : -EPERM; | |
491 | ||
492 | return __open_metadata(cmd); | |
493 | } | |
494 | ||
495 | static int __create_persistent_data_objects(struct dm_cache_metadata *cmd, | |
496 | bool may_format_device) | |
497 | { | |
498 | int r; | |
895b47d7 | 499 | cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE << SECTOR_SHIFT, |
c6b4fcba JT |
500 | CACHE_METADATA_CACHE_SIZE, |
501 | CACHE_MAX_CONCURRENT_LOCKS); | |
502 | if (IS_ERR(cmd->bm)) { | |
503 | DMERR("could not create block manager"); | |
504 | return PTR_ERR(cmd->bm); | |
505 | } | |
506 | ||
507 | r = __open_or_format_metadata(cmd, may_format_device); | |
508 | if (r) | |
509 | dm_block_manager_destroy(cmd->bm); | |
510 | ||
511 | return r; | |
512 | } | |
513 | ||
514 | static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd) | |
515 | { | |
516 | dm_sm_destroy(cmd->metadata_sm); | |
517 | dm_tm_destroy(cmd->tm); | |
518 | dm_block_manager_destroy(cmd->bm); | |
519 | } | |
520 | ||
521 | typedef unsigned long (*flags_mutator)(unsigned long); | |
522 | ||
523 | static void update_flags(struct cache_disk_superblock *disk_super, | |
524 | flags_mutator mutator) | |
525 | { | |
526 | uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags)); | |
527 | disk_super->flags = cpu_to_le32(sb_flags); | |
528 | } | |
529 | ||
530 | static unsigned long set_clean_shutdown(unsigned long flags) | |
531 | { | |
532 | set_bit(CLEAN_SHUTDOWN, &flags); | |
533 | return flags; | |
534 | } | |
535 | ||
536 | static unsigned long clear_clean_shutdown(unsigned long flags) | |
537 | { | |
538 | clear_bit(CLEAN_SHUTDOWN, &flags); | |
539 | return flags; | |
540 | } | |
541 | ||
542 | static void read_superblock_fields(struct dm_cache_metadata *cmd, | |
543 | struct cache_disk_superblock *disk_super) | |
544 | { | |
028ae9f7 | 545 | cmd->flags = le32_to_cpu(disk_super->flags); |
c6b4fcba JT |
546 | cmd->root = le64_to_cpu(disk_super->mapping_root); |
547 | cmd->hint_root = le64_to_cpu(disk_super->hint_root); | |
548 | cmd->discard_root = le64_to_cpu(disk_super->discard_root); | |
549 | cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size); | |
1bad9bc4 | 550 | cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks)); |
c6b4fcba JT |
551 | cmd->data_block_size = le32_to_cpu(disk_super->data_block_size); |
552 | cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks)); | |
553 | strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name)); | |
4e7f506f MS |
554 | cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]); |
555 | cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]); | |
556 | cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]); | |
c6b4fcba JT |
557 | cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size); |
558 | ||
559 | cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits); | |
560 | cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses); | |
561 | cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits); | |
562 | cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses); | |
563 | ||
564 | cmd->changed = false; | |
565 | } | |
566 | ||
567 | /* | |
568 | * The mutator updates the superblock flags. | |
569 | */ | |
570 | static int __begin_transaction_flags(struct dm_cache_metadata *cmd, | |
571 | flags_mutator mutator) | |
572 | { | |
573 | int r; | |
574 | struct cache_disk_superblock *disk_super; | |
575 | struct dm_block *sblock; | |
576 | ||
577 | r = superblock_lock(cmd, &sblock); | |
578 | if (r) | |
579 | return r; | |
580 | ||
581 | disk_super = dm_block_data(sblock); | |
582 | update_flags(disk_super, mutator); | |
583 | read_superblock_fields(cmd, disk_super); | |
a9d45396 | 584 | dm_bm_unlock(sblock); |
c6b4fcba | 585 | |
a9d45396 | 586 | return dm_bm_flush(cmd->bm); |
c6b4fcba JT |
587 | } |
588 | ||
589 | static int __begin_transaction(struct dm_cache_metadata *cmd) | |
590 | { | |
591 | int r; | |
592 | struct cache_disk_superblock *disk_super; | |
593 | struct dm_block *sblock; | |
594 | ||
595 | /* | |
596 | * We re-read the superblock every time. Shouldn't need to do this | |
597 | * really. | |
598 | */ | |
599 | r = superblock_read_lock(cmd, &sblock); | |
600 | if (r) | |
601 | return r; | |
602 | ||
603 | disk_super = dm_block_data(sblock); | |
604 | read_superblock_fields(cmd, disk_super); | |
605 | dm_bm_unlock(sblock); | |
606 | ||
607 | return 0; | |
608 | } | |
609 | ||
610 | static int __commit_transaction(struct dm_cache_metadata *cmd, | |
611 | flags_mutator mutator) | |
612 | { | |
613 | int r; | |
c6b4fcba JT |
614 | struct cache_disk_superblock *disk_super; |
615 | struct dm_block *sblock; | |
616 | ||
617 | /* | |
618 | * We need to know if the cache_disk_superblock exceeds a 512-byte sector. | |
619 | */ | |
620 | BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512); | |
621 | ||
622 | r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root, | |
623 | &cmd->discard_root); | |
624 | if (r) | |
625 | return r; | |
626 | ||
627 | r = dm_tm_pre_commit(cmd->tm); | |
628 | if (r < 0) | |
629 | return r; | |
630 | ||
5a32083d JT |
631 | r = __save_sm_root(cmd); |
632 | if (r) | |
c6b4fcba JT |
633 | return r; |
634 | ||
635 | r = superblock_lock(cmd, &sblock); | |
636 | if (r) | |
637 | return r; | |
638 | ||
639 | disk_super = dm_block_data(sblock); | |
640 | ||
3201ac45 | 641 | disk_super->flags = cpu_to_le32(cmd->flags); |
c6b4fcba JT |
642 | if (mutator) |
643 | update_flags(disk_super, mutator); | |
644 | ||
645 | disk_super->mapping_root = cpu_to_le64(cmd->root); | |
646 | disk_super->hint_root = cpu_to_le64(cmd->hint_root); | |
647 | disk_super->discard_root = cpu_to_le64(cmd->discard_root); | |
648 | disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size); | |
1bad9bc4 | 649 | disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks)); |
c6b4fcba JT |
650 | disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks)); |
651 | strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name)); | |
4e7f506f MS |
652 | disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]); |
653 | disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]); | |
654 | disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]); | |
c6b4fcba JT |
655 | |
656 | disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits); | |
657 | disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses); | |
658 | disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits); | |
659 | disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses); | |
5a32083d | 660 | __copy_sm_root(cmd, disk_super); |
c6b4fcba JT |
661 | |
662 | return dm_tm_commit(cmd->tm, sblock); | |
663 | } | |
664 | ||
665 | /*----------------------------------------------------------------*/ | |
666 | ||
667 | /* | |
668 | * The mappings are held in a dm-array that has 64-bit values stored in | |
669 | * little-endian format. The index is the cblock, the high 48bits of the | |
670 | * value are the oblock and the low 16 bit the flags. | |
671 | */ | |
672 | #define FLAGS_MASK ((1 << 16) - 1) | |
673 | ||
674 | static __le64 pack_value(dm_oblock_t block, unsigned flags) | |
675 | { | |
676 | uint64_t value = from_oblock(block); | |
677 | value <<= 16; | |
678 | value = value | (flags & FLAGS_MASK); | |
679 | return cpu_to_le64(value); | |
680 | } | |
681 | ||
682 | static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags) | |
683 | { | |
684 | uint64_t value = le64_to_cpu(value_le); | |
685 | uint64_t b = value >> 16; | |
686 | *block = to_oblock(b); | |
687 | *flags = value & FLAGS_MASK; | |
688 | } | |
689 | ||
690 | /*----------------------------------------------------------------*/ | |
691 | ||
9b1cc9f2 JT |
692 | static struct dm_cache_metadata *metadata_open(struct block_device *bdev, |
693 | sector_t data_block_size, | |
694 | bool may_format_device, | |
695 | size_t policy_hint_size) | |
c6b4fcba JT |
696 | { |
697 | int r; | |
698 | struct dm_cache_metadata *cmd; | |
699 | ||
700 | cmd = kzalloc(sizeof(*cmd), GFP_KERNEL); | |
701 | if (!cmd) { | |
702 | DMERR("could not allocate metadata struct"); | |
766a7888 | 703 | return ERR_PTR(-ENOMEM); |
c6b4fcba JT |
704 | } |
705 | ||
9b1cc9f2 | 706 | atomic_set(&cmd->ref_count, 1); |
c6b4fcba JT |
707 | init_rwsem(&cmd->root_lock); |
708 | cmd->bdev = bdev; | |
709 | cmd->data_block_size = data_block_size; | |
710 | cmd->cache_blocks = 0; | |
711 | cmd->policy_hint_size = policy_hint_size; | |
712 | cmd->changed = true; | |
028ae9f7 | 713 | cmd->fail_io = false; |
c6b4fcba JT |
714 | |
715 | r = __create_persistent_data_objects(cmd, may_format_device); | |
716 | if (r) { | |
717 | kfree(cmd); | |
718 | return ERR_PTR(r); | |
719 | } | |
720 | ||
721 | r = __begin_transaction_flags(cmd, clear_clean_shutdown); | |
722 | if (r < 0) { | |
723 | dm_cache_metadata_close(cmd); | |
724 | return ERR_PTR(r); | |
725 | } | |
726 | ||
727 | return cmd; | |
728 | } | |
729 | ||
9b1cc9f2 JT |
730 | /* |
731 | * We keep a little list of ref counted metadata objects to prevent two | |
732 | * different target instances creating separate bufio instances. This is | |
733 | * an issue if a table is reloaded before the suspend. | |
734 | */ | |
735 | static DEFINE_MUTEX(table_lock); | |
736 | static LIST_HEAD(table); | |
737 | ||
738 | static struct dm_cache_metadata *lookup(struct block_device *bdev) | |
739 | { | |
740 | struct dm_cache_metadata *cmd; | |
741 | ||
742 | list_for_each_entry(cmd, &table, list) | |
743 | if (cmd->bdev == bdev) { | |
744 | atomic_inc(&cmd->ref_count); | |
745 | return cmd; | |
746 | } | |
747 | ||
748 | return NULL; | |
749 | } | |
750 | ||
751 | static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev, | |
752 | sector_t data_block_size, | |
753 | bool may_format_device, | |
754 | size_t policy_hint_size) | |
755 | { | |
756 | struct dm_cache_metadata *cmd, *cmd2; | |
757 | ||
758 | mutex_lock(&table_lock); | |
759 | cmd = lookup(bdev); | |
760 | mutex_unlock(&table_lock); | |
761 | ||
762 | if (cmd) | |
763 | return cmd; | |
764 | ||
765 | cmd = metadata_open(bdev, data_block_size, may_format_device, policy_hint_size); | |
766a7888 | 766 | if (!IS_ERR(cmd)) { |
9b1cc9f2 JT |
767 | mutex_lock(&table_lock); |
768 | cmd2 = lookup(bdev); | |
769 | if (cmd2) { | |
770 | mutex_unlock(&table_lock); | |
771 | __destroy_persistent_data_objects(cmd); | |
772 | kfree(cmd); | |
773 | return cmd2; | |
774 | } | |
775 | list_add(&cmd->list, &table); | |
776 | mutex_unlock(&table_lock); | |
777 | } | |
778 | ||
779 | return cmd; | |
780 | } | |
781 | ||
782 | static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size) | |
783 | { | |
784 | if (cmd->data_block_size != data_block_size) { | |
785 | DMERR("data_block_size (%llu) different from that in metadata (%llu)\n", | |
786 | (unsigned long long) data_block_size, | |
787 | (unsigned long long) cmd->data_block_size); | |
788 | return false; | |
789 | } | |
790 | ||
791 | return true; | |
792 | } | |
793 | ||
794 | struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev, | |
795 | sector_t data_block_size, | |
796 | bool may_format_device, | |
797 | size_t policy_hint_size) | |
798 | { | |
799 | struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size, | |
800 | may_format_device, policy_hint_size); | |
766a7888 JT |
801 | |
802 | if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) { | |
9b1cc9f2 | 803 | dm_cache_metadata_close(cmd); |
766a7888 | 804 | return ERR_PTR(-EINVAL); |
9b1cc9f2 JT |
805 | } |
806 | ||
807 | return cmd; | |
808 | } | |
809 | ||
c6b4fcba JT |
810 | void dm_cache_metadata_close(struct dm_cache_metadata *cmd) |
811 | { | |
9b1cc9f2 JT |
812 | if (atomic_dec_and_test(&cmd->ref_count)) { |
813 | mutex_lock(&table_lock); | |
814 | list_del(&cmd->list); | |
815 | mutex_unlock(&table_lock); | |
816 | ||
028ae9f7 JT |
817 | if (!cmd->fail_io) |
818 | __destroy_persistent_data_objects(cmd); | |
9b1cc9f2 JT |
819 | kfree(cmd); |
820 | } | |
c6b4fcba JT |
821 | } |
822 | ||
f494a9c6 JT |
823 | /* |
824 | * Checks that the given cache block is either unmapped or clean. | |
825 | */ | |
826 | static int block_unmapped_or_clean(struct dm_cache_metadata *cmd, dm_cblock_t b, | |
827 | bool *result) | |
828 | { | |
829 | int r; | |
830 | __le64 value; | |
831 | dm_oblock_t ob; | |
832 | unsigned flags; | |
833 | ||
834 | r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value); | |
835 | if (r) { | |
836 | DMERR("block_unmapped_or_clean failed"); | |
837 | return r; | |
838 | } | |
839 | ||
840 | unpack_value(value, &ob, &flags); | |
841 | *result = !((flags & M_VALID) && (flags & M_DIRTY)); | |
842 | ||
843 | return 0; | |
844 | } | |
845 | ||
846 | static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd, | |
847 | dm_cblock_t begin, dm_cblock_t end, | |
848 | bool *result) | |
849 | { | |
850 | int r; | |
851 | *result = true; | |
852 | ||
853 | while (begin != end) { | |
854 | r = block_unmapped_or_clean(cmd, begin, result); | |
855 | if (r) | |
856 | return r; | |
857 | ||
858 | if (!*result) { | |
859 | DMERR("cache block %llu is dirty", | |
860 | (unsigned long long) from_cblock(begin)); | |
861 | return 0; | |
862 | } | |
863 | ||
864 | begin = to_cblock(from_cblock(begin) + 1); | |
865 | } | |
866 | ||
867 | return 0; | |
868 | } | |
869 | ||
028ae9f7 JT |
870 | #define WRITE_LOCK(cmd) \ |
871 | if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) \ | |
872 | return -EINVAL; \ | |
873 | down_write(&cmd->root_lock) | |
874 | ||
875 | #define WRITE_LOCK_VOID(cmd) \ | |
876 | if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) \ | |
877 | return; \ | |
878 | down_write(&cmd->root_lock) | |
879 | ||
880 | #define WRITE_UNLOCK(cmd) \ | |
881 | up_write(&cmd->root_lock) | |
882 | ||
c6b4fcba JT |
883 | int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size) |
884 | { | |
885 | int r; | |
f494a9c6 | 886 | bool clean; |
c6b4fcba JT |
887 | __le64 null_mapping = pack_value(0, 0); |
888 | ||
028ae9f7 | 889 | WRITE_LOCK(cmd); |
c6b4fcba | 890 | __dm_bless_for_disk(&null_mapping); |
f494a9c6 JT |
891 | |
892 | if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) { | |
893 | r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean); | |
894 | if (r) { | |
895 | __dm_unbless_for_disk(&null_mapping); | |
896 | goto out; | |
897 | } | |
898 | ||
899 | if (!clean) { | |
900 | DMERR("unable to shrink cache due to dirty blocks"); | |
901 | r = -EINVAL; | |
902 | __dm_unbless_for_disk(&null_mapping); | |
903 | goto out; | |
904 | } | |
905 | } | |
906 | ||
c6b4fcba JT |
907 | r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks), |
908 | from_cblock(new_cache_size), | |
909 | &null_mapping, &cmd->root); | |
910 | if (!r) | |
911 | cmd->cache_blocks = new_cache_size; | |
912 | cmd->changed = true; | |
f494a9c6 JT |
913 | |
914 | out: | |
028ae9f7 | 915 | WRITE_UNLOCK(cmd); |
c6b4fcba JT |
916 | |
917 | return r; | |
918 | } | |
919 | ||
920 | int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd, | |
921 | sector_t discard_block_size, | |
1bad9bc4 | 922 | dm_dblock_t new_nr_entries) |
c6b4fcba JT |
923 | { |
924 | int r; | |
925 | ||
028ae9f7 | 926 | WRITE_LOCK(cmd); |
c6b4fcba JT |
927 | r = dm_bitset_resize(&cmd->discard_info, |
928 | cmd->discard_root, | |
1bad9bc4 JT |
929 | from_dblock(cmd->discard_nr_blocks), |
930 | from_dblock(new_nr_entries), | |
c6b4fcba JT |
931 | false, &cmd->discard_root); |
932 | if (!r) { | |
933 | cmd->discard_block_size = discard_block_size; | |
934 | cmd->discard_nr_blocks = new_nr_entries; | |
935 | } | |
936 | ||
937 | cmd->changed = true; | |
028ae9f7 | 938 | WRITE_UNLOCK(cmd); |
c6b4fcba JT |
939 | |
940 | return r; | |
941 | } | |
942 | ||
1bad9bc4 | 943 | static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b) |
c6b4fcba JT |
944 | { |
945 | return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root, | |
1bad9bc4 | 946 | from_dblock(b), &cmd->discard_root); |
c6b4fcba JT |
947 | } |
948 | ||
1bad9bc4 | 949 | static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b) |
c6b4fcba JT |
950 | { |
951 | return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root, | |
1bad9bc4 | 952 | from_dblock(b), &cmd->discard_root); |
c6b4fcba JT |
953 | } |
954 | ||
1bad9bc4 | 955 | static int __is_discarded(struct dm_cache_metadata *cmd, dm_dblock_t b, |
c6b4fcba JT |
956 | bool *is_discarded) |
957 | { | |
958 | return dm_bitset_test_bit(&cmd->discard_info, cmd->discard_root, | |
1bad9bc4 | 959 | from_dblock(b), &cmd->discard_root, |
c6b4fcba JT |
960 | is_discarded); |
961 | } | |
962 | ||
963 | static int __discard(struct dm_cache_metadata *cmd, | |
1bad9bc4 | 964 | dm_dblock_t dblock, bool discard) |
c6b4fcba JT |
965 | { |
966 | int r; | |
967 | ||
968 | r = (discard ? __set_discard : __clear_discard)(cmd, dblock); | |
969 | if (r) | |
970 | return r; | |
971 | ||
972 | cmd->changed = true; | |
973 | return 0; | |
974 | } | |
975 | ||
976 | int dm_cache_set_discard(struct dm_cache_metadata *cmd, | |
1bad9bc4 | 977 | dm_dblock_t dblock, bool discard) |
c6b4fcba JT |
978 | { |
979 | int r; | |
980 | ||
028ae9f7 | 981 | WRITE_LOCK(cmd); |
c6b4fcba | 982 | r = __discard(cmd, dblock, discard); |
028ae9f7 | 983 | WRITE_UNLOCK(cmd); |
c6b4fcba JT |
984 | |
985 | return r; | |
986 | } | |
987 | ||
988 | static int __load_discards(struct dm_cache_metadata *cmd, | |
989 | load_discard_fn fn, void *context) | |
990 | { | |
991 | int r = 0; | |
992 | dm_block_t b; | |
993 | bool discard; | |
994 | ||
1bad9bc4 JT |
995 | for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) { |
996 | dm_dblock_t dblock = to_dblock(b); | |
c6b4fcba JT |
997 | |
998 | if (cmd->clean_when_opened) { | |
999 | r = __is_discarded(cmd, dblock, &discard); | |
1000 | if (r) | |
1001 | return r; | |
1002 | } else | |
1003 | discard = false; | |
1004 | ||
1005 | r = fn(context, cmd->discard_block_size, dblock, discard); | |
1006 | if (r) | |
1007 | break; | |
1008 | } | |
1009 | ||
1010 | return r; | |
1011 | } | |
1012 | ||
1013 | int dm_cache_load_discards(struct dm_cache_metadata *cmd, | |
1014 | load_discard_fn fn, void *context) | |
1015 | { | |
1016 | int r; | |
1017 | ||
1018 | down_read(&cmd->root_lock); | |
1019 | r = __load_discards(cmd, fn, context); | |
1020 | up_read(&cmd->root_lock); | |
1021 | ||
1022 | return r; | |
1023 | } | |
1024 | ||
1025 | dm_cblock_t dm_cache_size(struct dm_cache_metadata *cmd) | |
1026 | { | |
1027 | dm_cblock_t r; | |
1028 | ||
1029 | down_read(&cmd->root_lock); | |
1030 | r = cmd->cache_blocks; | |
1031 | up_read(&cmd->root_lock); | |
1032 | ||
1033 | return r; | |
1034 | } | |
1035 | ||
1036 | static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock) | |
1037 | { | |
1038 | int r; | |
1039 | __le64 value = pack_value(0, 0); | |
1040 | ||
1041 | __dm_bless_for_disk(&value); | |
1042 | r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock), | |
1043 | &value, &cmd->root); | |
1044 | if (r) | |
1045 | return r; | |
1046 | ||
1047 | cmd->changed = true; | |
1048 | return 0; | |
1049 | } | |
1050 | ||
1051 | int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock) | |
1052 | { | |
1053 | int r; | |
1054 | ||
028ae9f7 | 1055 | WRITE_LOCK(cmd); |
c6b4fcba | 1056 | r = __remove(cmd, cblock); |
028ae9f7 | 1057 | WRITE_UNLOCK(cmd); |
c6b4fcba JT |
1058 | |
1059 | return r; | |
1060 | } | |
1061 | ||
1062 | static int __insert(struct dm_cache_metadata *cmd, | |
1063 | dm_cblock_t cblock, dm_oblock_t oblock) | |
1064 | { | |
1065 | int r; | |
1066 | __le64 value = pack_value(oblock, M_VALID); | |
1067 | __dm_bless_for_disk(&value); | |
1068 | ||
1069 | r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock), | |
1070 | &value, &cmd->root); | |
1071 | if (r) | |
1072 | return r; | |
1073 | ||
1074 | cmd->changed = true; | |
1075 | return 0; | |
1076 | } | |
1077 | ||
1078 | int dm_cache_insert_mapping(struct dm_cache_metadata *cmd, | |
1079 | dm_cblock_t cblock, dm_oblock_t oblock) | |
1080 | { | |
1081 | int r; | |
1082 | ||
028ae9f7 | 1083 | WRITE_LOCK(cmd); |
c6b4fcba | 1084 | r = __insert(cmd, cblock, oblock); |
028ae9f7 | 1085 | WRITE_UNLOCK(cmd); |
c6b4fcba JT |
1086 | |
1087 | return r; | |
1088 | } | |
1089 | ||
1090 | struct thunk { | |
1091 | load_mapping_fn fn; | |
1092 | void *context; | |
1093 | ||
1094 | struct dm_cache_metadata *cmd; | |
1095 | bool respect_dirty_flags; | |
1096 | bool hints_valid; | |
1097 | }; | |
1098 | ||
ea2dd8c1 MS |
1099 | static bool policy_unchanged(struct dm_cache_metadata *cmd, |
1100 | struct dm_cache_policy *policy) | |
1101 | { | |
1102 | const char *policy_name = dm_cache_policy_get_name(policy); | |
1103 | const unsigned *policy_version = dm_cache_policy_get_version(policy); | |
1104 | size_t policy_hint_size = dm_cache_policy_get_hint_size(policy); | |
1105 | ||
1106 | /* | |
1107 | * Ensure policy names match. | |
1108 | */ | |
1109 | if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name))) | |
1110 | return false; | |
1111 | ||
1112 | /* | |
1113 | * Ensure policy major versions match. | |
1114 | */ | |
1115 | if (cmd->policy_version[0] != policy_version[0]) | |
1116 | return false; | |
1117 | ||
1118 | /* | |
1119 | * Ensure policy hint sizes match. | |
1120 | */ | |
1121 | if (cmd->policy_hint_size != policy_hint_size) | |
1122 | return false; | |
1123 | ||
1124 | return true; | |
1125 | } | |
1126 | ||
c6b4fcba JT |
1127 | static bool hints_array_initialized(struct dm_cache_metadata *cmd) |
1128 | { | |
1129 | return cmd->hint_root && cmd->policy_hint_size; | |
1130 | } | |
1131 | ||
1132 | static bool hints_array_available(struct dm_cache_metadata *cmd, | |
ea2dd8c1 | 1133 | struct dm_cache_policy *policy) |
c6b4fcba | 1134 | { |
ea2dd8c1 | 1135 | return cmd->clean_when_opened && policy_unchanged(cmd, policy) && |
c6b4fcba JT |
1136 | hints_array_initialized(cmd); |
1137 | } | |
1138 | ||
1139 | static int __load_mapping(void *context, uint64_t cblock, void *leaf) | |
1140 | { | |
1141 | int r = 0; | |
1142 | bool dirty; | |
1143 | __le64 value; | |
1144 | __le32 hint_value = 0; | |
1145 | dm_oblock_t oblock; | |
1146 | unsigned flags; | |
1147 | struct thunk *thunk = context; | |
1148 | struct dm_cache_metadata *cmd = thunk->cmd; | |
1149 | ||
1150 | memcpy(&value, leaf, sizeof(value)); | |
1151 | unpack_value(value, &oblock, &flags); | |
1152 | ||
1153 | if (flags & M_VALID) { | |
1154 | if (thunk->hints_valid) { | |
1155 | r = dm_array_get_value(&cmd->hint_info, cmd->hint_root, | |
1156 | cblock, &hint_value); | |
1157 | if (r && r != -ENODATA) | |
1158 | return r; | |
1159 | } | |
1160 | ||
1161 | dirty = thunk->respect_dirty_flags ? (flags & M_DIRTY) : true; | |
1162 | r = thunk->fn(thunk->context, oblock, to_cblock(cblock), | |
1163 | dirty, le32_to_cpu(hint_value), thunk->hints_valid); | |
1164 | } | |
1165 | ||
1166 | return r; | |
1167 | } | |
1168 | ||
ea2dd8c1 MS |
1169 | static int __load_mappings(struct dm_cache_metadata *cmd, |
1170 | struct dm_cache_policy *policy, | |
c6b4fcba JT |
1171 | load_mapping_fn fn, void *context) |
1172 | { | |
1173 | struct thunk thunk; | |
1174 | ||
1175 | thunk.fn = fn; | |
1176 | thunk.context = context; | |
1177 | ||
1178 | thunk.cmd = cmd; | |
1179 | thunk.respect_dirty_flags = cmd->clean_when_opened; | |
ea2dd8c1 | 1180 | thunk.hints_valid = hints_array_available(cmd, policy); |
c6b4fcba JT |
1181 | |
1182 | return dm_array_walk(&cmd->info, cmd->root, __load_mapping, &thunk); | |
1183 | } | |
1184 | ||
ea2dd8c1 MS |
1185 | int dm_cache_load_mappings(struct dm_cache_metadata *cmd, |
1186 | struct dm_cache_policy *policy, | |
c6b4fcba JT |
1187 | load_mapping_fn fn, void *context) |
1188 | { | |
1189 | int r; | |
1190 | ||
1191 | down_read(&cmd->root_lock); | |
ea2dd8c1 | 1192 | r = __load_mappings(cmd, policy, fn, context); |
c6b4fcba JT |
1193 | up_read(&cmd->root_lock); |
1194 | ||
1195 | return r; | |
1196 | } | |
1197 | ||
1198 | static int __dump_mapping(void *context, uint64_t cblock, void *leaf) | |
1199 | { | |
1200 | int r = 0; | |
1201 | __le64 value; | |
1202 | dm_oblock_t oblock; | |
1203 | unsigned flags; | |
1204 | ||
1205 | memcpy(&value, leaf, sizeof(value)); | |
1206 | unpack_value(value, &oblock, &flags); | |
1207 | ||
1208 | return r; | |
1209 | } | |
1210 | ||
1211 | static int __dump_mappings(struct dm_cache_metadata *cmd) | |
1212 | { | |
1213 | return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL); | |
1214 | } | |
1215 | ||
1216 | void dm_cache_dump(struct dm_cache_metadata *cmd) | |
1217 | { | |
1218 | down_read(&cmd->root_lock); | |
1219 | __dump_mappings(cmd); | |
1220 | up_read(&cmd->root_lock); | |
1221 | } | |
1222 | ||
1223 | int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd) | |
1224 | { | |
1225 | int r; | |
1226 | ||
1227 | down_read(&cmd->root_lock); | |
1228 | r = cmd->changed; | |
1229 | up_read(&cmd->root_lock); | |
1230 | ||
1231 | return r; | |
1232 | } | |
1233 | ||
1234 | static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty) | |
1235 | { | |
1236 | int r; | |
1237 | unsigned flags; | |
1238 | dm_oblock_t oblock; | |
1239 | __le64 value; | |
1240 | ||
1241 | r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value); | |
1242 | if (r) | |
1243 | return r; | |
1244 | ||
1245 | unpack_value(value, &oblock, &flags); | |
1246 | ||
1247 | if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty)) | |
1248 | /* nothing to be done */ | |
1249 | return 0; | |
1250 | ||
79ed9caf | 1251 | value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0)); |
c6b4fcba JT |
1252 | __dm_bless_for_disk(&value); |
1253 | ||
1254 | r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock), | |
1255 | &value, &cmd->root); | |
1256 | if (r) | |
1257 | return r; | |
1258 | ||
1259 | cmd->changed = true; | |
1260 | return 0; | |
1261 | ||
1262 | } | |
1263 | ||
1264 | int dm_cache_set_dirty(struct dm_cache_metadata *cmd, | |
1265 | dm_cblock_t cblock, bool dirty) | |
1266 | { | |
1267 | int r; | |
1268 | ||
028ae9f7 | 1269 | WRITE_LOCK(cmd); |
c6b4fcba | 1270 | r = __dirty(cmd, cblock, dirty); |
028ae9f7 | 1271 | WRITE_UNLOCK(cmd); |
c6b4fcba JT |
1272 | |
1273 | return r; | |
1274 | } | |
1275 | ||
1276 | void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd, | |
1277 | struct dm_cache_statistics *stats) | |
1278 | { | |
1279 | down_read(&cmd->root_lock); | |
8c5008fa | 1280 | *stats = cmd->stats; |
c6b4fcba JT |
1281 | up_read(&cmd->root_lock); |
1282 | } | |
1283 | ||
1284 | void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd, | |
1285 | struct dm_cache_statistics *stats) | |
1286 | { | |
028ae9f7 | 1287 | WRITE_LOCK_VOID(cmd); |
8c5008fa | 1288 | cmd->stats = *stats; |
028ae9f7 | 1289 | WRITE_UNLOCK(cmd); |
c6b4fcba JT |
1290 | } |
1291 | ||
1292 | int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown) | |
1293 | { | |
1294 | int r; | |
1295 | flags_mutator mutator = (clean_shutdown ? set_clean_shutdown : | |
1296 | clear_clean_shutdown); | |
1297 | ||
028ae9f7 | 1298 | WRITE_LOCK(cmd); |
c6b4fcba JT |
1299 | r = __commit_transaction(cmd, mutator); |
1300 | if (r) | |
1301 | goto out; | |
1302 | ||
1303 | r = __begin_transaction(cmd); | |
1304 | ||
1305 | out: | |
028ae9f7 | 1306 | WRITE_UNLOCK(cmd); |
c6b4fcba JT |
1307 | return r; |
1308 | } | |
1309 | ||
1310 | int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd, | |
1311 | dm_block_t *result) | |
1312 | { | |
1313 | int r = -EINVAL; | |
1314 | ||
1315 | down_read(&cmd->root_lock); | |
1316 | r = dm_sm_get_nr_free(cmd->metadata_sm, result); | |
1317 | up_read(&cmd->root_lock); | |
1318 | ||
1319 | return r; | |
1320 | } | |
1321 | ||
1322 | int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd, | |
1323 | dm_block_t *result) | |
1324 | { | |
1325 | int r = -EINVAL; | |
1326 | ||
1327 | down_read(&cmd->root_lock); | |
1328 | r = dm_sm_get_nr_blocks(cmd->metadata_sm, result); | |
1329 | up_read(&cmd->root_lock); | |
1330 | ||
1331 | return r; | |
1332 | } | |
1333 | ||
1334 | /*----------------------------------------------------------------*/ | |
1335 | ||
1336 | static int begin_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy) | |
1337 | { | |
1338 | int r; | |
1339 | __le32 value; | |
1340 | size_t hint_size; | |
1341 | const char *policy_name = dm_cache_policy_get_name(policy); | |
4e7f506f | 1342 | const unsigned *policy_version = dm_cache_policy_get_version(policy); |
c6b4fcba JT |
1343 | |
1344 | if (!policy_name[0] || | |
1345 | (strlen(policy_name) > sizeof(cmd->policy_name) - 1)) | |
1346 | return -EINVAL; | |
1347 | ||
ea2dd8c1 | 1348 | if (!policy_unchanged(cmd, policy)) { |
c6b4fcba | 1349 | strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name)); |
4e7f506f | 1350 | memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version)); |
c6b4fcba JT |
1351 | |
1352 | hint_size = dm_cache_policy_get_hint_size(policy); | |
1353 | if (!hint_size) | |
1354 | return 0; /* short-circuit hints initialization */ | |
1355 | cmd->policy_hint_size = hint_size; | |
1356 | ||
1357 | if (cmd->hint_root) { | |
1358 | r = dm_array_del(&cmd->hint_info, cmd->hint_root); | |
1359 | if (r) | |
1360 | return r; | |
1361 | } | |
1362 | ||
1363 | r = dm_array_empty(&cmd->hint_info, &cmd->hint_root); | |
1364 | if (r) | |
1365 | return r; | |
1366 | ||
1367 | value = cpu_to_le32(0); | |
1368 | __dm_bless_for_disk(&value); | |
1369 | r = dm_array_resize(&cmd->hint_info, cmd->hint_root, 0, | |
1370 | from_cblock(cmd->cache_blocks), | |
1371 | &value, &cmd->hint_root); | |
1372 | if (r) | |
1373 | return r; | |
1374 | } | |
1375 | ||
1376 | return 0; | |
1377 | } | |
1378 | ||
0596661f | 1379 | static int save_hint(void *context, dm_cblock_t cblock, dm_oblock_t oblock, uint32_t hint) |
c6b4fcba | 1380 | { |
0596661f JT |
1381 | struct dm_cache_metadata *cmd = context; |
1382 | __le32 value = cpu_to_le32(hint); | |
c6b4fcba JT |
1383 | int r; |
1384 | ||
c6b4fcba JT |
1385 | __dm_bless_for_disk(&value); |
1386 | ||
1387 | r = dm_array_set_value(&cmd->hint_info, cmd->hint_root, | |
1388 | from_cblock(cblock), &value, &cmd->hint_root); | |
1389 | cmd->changed = true; | |
1390 | ||
1391 | return r; | |
1392 | } | |
1393 | ||
0596661f | 1394 | static int write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy) |
c6b4fcba JT |
1395 | { |
1396 | int r; | |
1397 | ||
0596661f JT |
1398 | r = begin_hints(cmd, policy); |
1399 | if (r) { | |
1400 | DMERR("begin_hints failed"); | |
1401 | return r; | |
1402 | } | |
1403 | ||
1404 | return policy_walk_mappings(policy, save_hint, cmd); | |
1405 | } | |
1406 | ||
1407 | int dm_cache_write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy) | |
1408 | { | |
1409 | int r; | |
c6b4fcba | 1410 | |
028ae9f7 | 1411 | WRITE_LOCK(cmd); |
0596661f | 1412 | r = write_hints(cmd, policy); |
028ae9f7 | 1413 | WRITE_UNLOCK(cmd); |
c6b4fcba JT |
1414 | |
1415 | return r; | |
1416 | } | |
2ee57d58 JT |
1417 | |
1418 | int dm_cache_metadata_all_clean(struct dm_cache_metadata *cmd, bool *result) | |
1419 | { | |
1420 | return blocks_are_unmapped_or_clean(cmd, 0, cmd->cache_blocks, result); | |
1421 | } | |
028ae9f7 JT |
1422 | |
1423 | void dm_cache_metadata_set_read_only(struct dm_cache_metadata *cmd) | |
1424 | { | |
1425 | WRITE_LOCK_VOID(cmd); | |
1426 | dm_bm_set_read_only(cmd->bm); | |
1427 | WRITE_UNLOCK(cmd); | |
1428 | } | |
1429 | ||
1430 | void dm_cache_metadata_set_read_write(struct dm_cache_metadata *cmd) | |
1431 | { | |
1432 | WRITE_LOCK_VOID(cmd); | |
1433 | dm_bm_set_read_write(cmd->bm); | |
1434 | WRITE_UNLOCK(cmd); | |
1435 | } | |
1436 | ||
1437 | int dm_cache_metadata_set_needs_check(struct dm_cache_metadata *cmd) | |
1438 | { | |
1439 | int r; | |
1440 | struct dm_block *sblock; | |
1441 | struct cache_disk_superblock *disk_super; | |
1442 | ||
1443 | /* | |
1444 | * We ignore fail_io for this function. | |
1445 | */ | |
1446 | down_write(&cmd->root_lock); | |
1447 | set_bit(NEEDS_CHECK, &cmd->flags); | |
1448 | ||
1449 | r = superblock_lock(cmd, &sblock); | |
1450 | if (r) { | |
1451 | DMERR("couldn't read superblock"); | |
1452 | goto out; | |
1453 | } | |
1454 | ||
1455 | disk_super = dm_block_data(sblock); | |
1456 | disk_super->flags = cpu_to_le32(cmd->flags); | |
1457 | ||
1458 | dm_bm_unlock(sblock); | |
1459 | ||
1460 | out: | |
1461 | up_write(&cmd->root_lock); | |
1462 | return r; | |
1463 | } | |
1464 | ||
1465 | bool dm_cache_metadata_needs_check(struct dm_cache_metadata *cmd) | |
1466 | { | |
1467 | bool needs_check; | |
1468 | ||
1469 | down_read(&cmd->root_lock); | |
1470 | needs_check = !!test_bit(NEEDS_CHECK, &cmd->flags); | |
1471 | up_read(&cmd->root_lock); | |
1472 | ||
1473 | return needs_check; | |
1474 | } | |
1475 | ||
1476 | int dm_cache_metadata_abort(struct dm_cache_metadata *cmd) | |
1477 | { | |
1478 | int r; | |
1479 | ||
1480 | WRITE_LOCK(cmd); | |
1481 | __destroy_persistent_data_objects(cmd); | |
1482 | r = __create_persistent_data_objects(cmd, false); | |
1483 | if (r) | |
1484 | cmd->fail_io = true; | |
1485 | WRITE_UNLOCK(cmd); | |
1486 | ||
1487 | return r; | |
1488 | } |