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