dm-cache-metadata.c revision dd8b0c2096e53b336324e99455efcc498599ba0f
1/* 2 * Copyright (C) 2012 Red Hat, Inc. 3 * 4 * This file is released under the GPL. 5 */ 6 7#include "dm-cache-metadata.h" 8 9#include "persistent-data/dm-array.h" 10#include "persistent-data/dm-bitset.h" 11#include "persistent-data/dm-space-map.h" 12#include "persistent-data/dm-space-map-disk.h" 13#include "persistent-data/dm-transaction-manager.h" 14 15#include <linux/device-mapper.h> 16 17/*----------------------------------------------------------------*/ 18 19#define DM_MSG_PREFIX "cache metadata" 20 21#define CACHE_SUPERBLOCK_MAGIC 06142003 22#define CACHE_SUPERBLOCK_LOCATION 0 23#define CACHE_VERSION 1 24#define CACHE_METADATA_CACHE_SIZE 64 25 26/* 27 * 3 for btree insert + 28 * 2 for btree lookup used within space map 29 */ 30#define CACHE_MAX_CONCURRENT_LOCKS 5 31#define SPACE_MAP_ROOT_SIZE 128 32 33enum superblock_flag_bits { 34 /* for spotting crashes that would invalidate the dirty bitset */ 35 CLEAN_SHUTDOWN, 36}; 37 38/* 39 * Each mapping from cache block -> origin block carries a set of flags. 40 */ 41enum mapping_bits { 42 /* 43 * A valid mapping. Because we're using an array we clear this 44 * flag for an non existant mapping. 45 */ 46 M_VALID = 1, 47 48 /* 49 * The data on the cache is different from that on the origin. 50 */ 51 M_DIRTY = 2 52}; 53 54struct cache_disk_superblock { 55 __le32 csum; 56 __le32 flags; 57 __le64 blocknr; 58 59 __u8 uuid[16]; 60 __le64 magic; 61 __le32 version; 62 63 __u8 policy_name[CACHE_POLICY_NAME_SIZE]; 64 __le32 policy_hint_size; 65 66 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE]; 67 __le64 mapping_root; 68 __le64 hint_root; 69 70 __le64 discard_root; 71 __le64 discard_block_size; 72 __le64 discard_nr_blocks; 73 74 __le32 data_block_size; 75 __le32 metadata_block_size; 76 __le32 cache_blocks; 77 78 __le32 compat_flags; 79 __le32 compat_ro_flags; 80 __le32 incompat_flags; 81 82 __le32 read_hits; 83 __le32 read_misses; 84 __le32 write_hits; 85 __le32 write_misses; 86 87 __le32 policy_version[CACHE_POLICY_VERSION_SIZE]; 88} __packed; 89 90struct dm_cache_metadata { 91 struct block_device *bdev; 92 struct dm_block_manager *bm; 93 struct dm_space_map *metadata_sm; 94 struct dm_transaction_manager *tm; 95 96 struct dm_array_info info; 97 struct dm_array_info hint_info; 98 struct dm_disk_bitset discard_info; 99 100 struct rw_semaphore root_lock; 101 dm_block_t root; 102 dm_block_t hint_root; 103 dm_block_t discard_root; 104 105 sector_t discard_block_size; 106 dm_dblock_t discard_nr_blocks; 107 108 sector_t data_block_size; 109 dm_cblock_t cache_blocks; 110 bool changed:1; 111 bool clean_when_opened:1; 112 113 char policy_name[CACHE_POLICY_NAME_SIZE]; 114 unsigned policy_version[CACHE_POLICY_VERSION_SIZE]; 115 size_t policy_hint_size; 116 struct dm_cache_statistics stats; 117}; 118 119/*------------------------------------------------------------------- 120 * superblock validator 121 *-----------------------------------------------------------------*/ 122 123#define SUPERBLOCK_CSUM_XOR 9031977 124 125static void sb_prepare_for_write(struct dm_block_validator *v, 126 struct dm_block *b, 127 size_t sb_block_size) 128{ 129 struct cache_disk_superblock *disk_super = dm_block_data(b); 130 131 disk_super->blocknr = cpu_to_le64(dm_block_location(b)); 132 disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags, 133 sb_block_size - sizeof(__le32), 134 SUPERBLOCK_CSUM_XOR)); 135} 136 137static int sb_check(struct dm_block_validator *v, 138 struct dm_block *b, 139 size_t sb_block_size) 140{ 141 struct cache_disk_superblock *disk_super = dm_block_data(b); 142 __le32 csum_le; 143 144 if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) { 145 DMERR("sb_check failed: blocknr %llu: wanted %llu", 146 le64_to_cpu(disk_super->blocknr), 147 (unsigned long long)dm_block_location(b)); 148 return -ENOTBLK; 149 } 150 151 if (le64_to_cpu(disk_super->magic) != CACHE_SUPERBLOCK_MAGIC) { 152 DMERR("sb_check failed: magic %llu: wanted %llu", 153 le64_to_cpu(disk_super->magic), 154 (unsigned long long)CACHE_SUPERBLOCK_MAGIC); 155 return -EILSEQ; 156 } 157 158 csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags, 159 sb_block_size - sizeof(__le32), 160 SUPERBLOCK_CSUM_XOR)); 161 if (csum_le != disk_super->csum) { 162 DMERR("sb_check failed: csum %u: wanted %u", 163 le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum)); 164 return -EILSEQ; 165 } 166 167 return 0; 168} 169 170static struct dm_block_validator sb_validator = { 171 .name = "superblock", 172 .prepare_for_write = sb_prepare_for_write, 173 .check = sb_check 174}; 175 176/*----------------------------------------------------------------*/ 177 178static int superblock_read_lock(struct dm_cache_metadata *cmd, 179 struct dm_block **sblock) 180{ 181 return dm_bm_read_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION, 182 &sb_validator, sblock); 183} 184 185static int superblock_lock_zero(struct dm_cache_metadata *cmd, 186 struct dm_block **sblock) 187{ 188 return dm_bm_write_lock_zero(cmd->bm, CACHE_SUPERBLOCK_LOCATION, 189 &sb_validator, sblock); 190} 191 192static int superblock_lock(struct dm_cache_metadata *cmd, 193 struct dm_block **sblock) 194{ 195 return dm_bm_write_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION, 196 &sb_validator, sblock); 197} 198 199/*----------------------------------------------------------------*/ 200 201static int __superblock_all_zeroes(struct dm_block_manager *bm, bool *result) 202{ 203 int r; 204 unsigned i; 205 struct dm_block *b; 206 __le64 *data_le, zero = cpu_to_le64(0); 207 unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64); 208 209 /* 210 * We can't use a validator here - it may be all zeroes. 211 */ 212 r = dm_bm_read_lock(bm, CACHE_SUPERBLOCK_LOCATION, NULL, &b); 213 if (r) 214 return r; 215 216 data_le = dm_block_data(b); 217 *result = true; 218 for (i = 0; i < sb_block_size; i++) { 219 if (data_le[i] != zero) { 220 *result = false; 221 break; 222 } 223 } 224 225 return dm_bm_unlock(b); 226} 227 228static void __setup_mapping_info(struct dm_cache_metadata *cmd) 229{ 230 struct dm_btree_value_type vt; 231 232 vt.context = NULL; 233 vt.size = sizeof(__le64); 234 vt.inc = NULL; 235 vt.dec = NULL; 236 vt.equal = NULL; 237 dm_array_info_init(&cmd->info, cmd->tm, &vt); 238 239 if (cmd->policy_hint_size) { 240 vt.size = sizeof(__le32); 241 dm_array_info_init(&cmd->hint_info, cmd->tm, &vt); 242 } 243} 244 245static int __write_initial_superblock(struct dm_cache_metadata *cmd) 246{ 247 int r; 248 struct dm_block *sblock; 249 size_t metadata_len; 250 struct cache_disk_superblock *disk_super; 251 sector_t bdev_size = i_size_read(cmd->bdev->bd_inode) >> SECTOR_SHIFT; 252 253 /* FIXME: see if we can lose the max sectors limit */ 254 if (bdev_size > DM_CACHE_METADATA_MAX_SECTORS) 255 bdev_size = DM_CACHE_METADATA_MAX_SECTORS; 256 257 r = dm_sm_root_size(cmd->metadata_sm, &metadata_len); 258 if (r < 0) 259 return r; 260 261 r = dm_tm_pre_commit(cmd->tm); 262 if (r < 0) 263 return r; 264 265 r = superblock_lock_zero(cmd, &sblock); 266 if (r) 267 return r; 268 269 disk_super = dm_block_data(sblock); 270 disk_super->flags = 0; 271 memset(disk_super->uuid, 0, sizeof(disk_super->uuid)); 272 disk_super->magic = cpu_to_le64(CACHE_SUPERBLOCK_MAGIC); 273 disk_super->version = cpu_to_le32(CACHE_VERSION); 274 memset(disk_super->policy_name, 0, sizeof(disk_super->policy_name)); 275 memset(disk_super->policy_version, 0, sizeof(disk_super->policy_version)); 276 disk_super->policy_hint_size = 0; 277 278 r = dm_sm_copy_root(cmd->metadata_sm, &disk_super->metadata_space_map_root, 279 metadata_len); 280 if (r < 0) 281 goto bad_locked; 282 283 disk_super->mapping_root = cpu_to_le64(cmd->root); 284 disk_super->hint_root = cpu_to_le64(cmd->hint_root); 285 disk_super->discard_root = cpu_to_le64(cmd->discard_root); 286 disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size); 287 disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks)); 288 disk_super->metadata_block_size = cpu_to_le32(DM_CACHE_METADATA_BLOCK_SIZE >> SECTOR_SHIFT); 289 disk_super->data_block_size = cpu_to_le32(cmd->data_block_size); 290 disk_super->cache_blocks = cpu_to_le32(0); 291 292 disk_super->read_hits = cpu_to_le32(0); 293 disk_super->read_misses = cpu_to_le32(0); 294 disk_super->write_hits = cpu_to_le32(0); 295 disk_super->write_misses = cpu_to_le32(0); 296 297 return dm_tm_commit(cmd->tm, sblock); 298 299bad_locked: 300 dm_bm_unlock(sblock); 301 return r; 302} 303 304static int __format_metadata(struct dm_cache_metadata *cmd) 305{ 306 int r; 307 308 r = dm_tm_create_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION, 309 &cmd->tm, &cmd->metadata_sm); 310 if (r < 0) { 311 DMERR("tm_create_with_sm failed"); 312 return r; 313 } 314 315 __setup_mapping_info(cmd); 316 317 r = dm_array_empty(&cmd->info, &cmd->root); 318 if (r < 0) 319 goto bad; 320 321 dm_disk_bitset_init(cmd->tm, &cmd->discard_info); 322 323 r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root); 324 if (r < 0) 325 goto bad; 326 327 cmd->discard_block_size = 0; 328 cmd->discard_nr_blocks = 0; 329 330 r = __write_initial_superblock(cmd); 331 if (r) 332 goto bad; 333 334 cmd->clean_when_opened = true; 335 return 0; 336 337bad: 338 dm_tm_destroy(cmd->tm); 339 dm_sm_destroy(cmd->metadata_sm); 340 341 return r; 342} 343 344static int __check_incompat_features(struct cache_disk_superblock *disk_super, 345 struct dm_cache_metadata *cmd) 346{ 347 uint32_t features; 348 349 features = le32_to_cpu(disk_super->incompat_flags) & ~DM_CACHE_FEATURE_INCOMPAT_SUPP; 350 if (features) { 351 DMERR("could not access metadata due to unsupported optional features (%lx).", 352 (unsigned long)features); 353 return -EINVAL; 354 } 355 356 /* 357 * Check for read-only metadata to skip the following RDWR checks. 358 */ 359 if (get_disk_ro(cmd->bdev->bd_disk)) 360 return 0; 361 362 features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP; 363 if (features) { 364 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).", 365 (unsigned long)features); 366 return -EINVAL; 367 } 368 369 return 0; 370} 371 372static int __open_metadata(struct dm_cache_metadata *cmd) 373{ 374 int r; 375 struct dm_block *sblock; 376 struct cache_disk_superblock *disk_super; 377 unsigned long sb_flags; 378 379 r = superblock_read_lock(cmd, &sblock); 380 if (r < 0) { 381 DMERR("couldn't read lock superblock"); 382 return r; 383 } 384 385 disk_super = dm_block_data(sblock); 386 387 r = __check_incompat_features(disk_super, cmd); 388 if (r < 0) 389 goto bad; 390 391 r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION, 392 disk_super->metadata_space_map_root, 393 sizeof(disk_super->metadata_space_map_root), 394 &cmd->tm, &cmd->metadata_sm); 395 if (r < 0) { 396 DMERR("tm_open_with_sm failed"); 397 goto bad; 398 } 399 400 __setup_mapping_info(cmd); 401 dm_disk_bitset_init(cmd->tm, &cmd->discard_info); 402 sb_flags = le32_to_cpu(disk_super->flags); 403 cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags); 404 return dm_bm_unlock(sblock); 405 406bad: 407 dm_bm_unlock(sblock); 408 return r; 409} 410 411static int __open_or_format_metadata(struct dm_cache_metadata *cmd, 412 bool format_device) 413{ 414 int r; 415 bool unformatted = false; 416 417 r = __superblock_all_zeroes(cmd->bm, &unformatted); 418 if (r) 419 return r; 420 421 if (unformatted) 422 return format_device ? __format_metadata(cmd) : -EPERM; 423 424 return __open_metadata(cmd); 425} 426 427static int __create_persistent_data_objects(struct dm_cache_metadata *cmd, 428 bool may_format_device) 429{ 430 int r; 431 cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE, 432 CACHE_METADATA_CACHE_SIZE, 433 CACHE_MAX_CONCURRENT_LOCKS); 434 if (IS_ERR(cmd->bm)) { 435 DMERR("could not create block manager"); 436 return PTR_ERR(cmd->bm); 437 } 438 439 r = __open_or_format_metadata(cmd, may_format_device); 440 if (r) 441 dm_block_manager_destroy(cmd->bm); 442 443 return r; 444} 445 446static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd) 447{ 448 dm_sm_destroy(cmd->metadata_sm); 449 dm_tm_destroy(cmd->tm); 450 dm_block_manager_destroy(cmd->bm); 451} 452 453typedef unsigned long (*flags_mutator)(unsigned long); 454 455static void update_flags(struct cache_disk_superblock *disk_super, 456 flags_mutator mutator) 457{ 458 uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags)); 459 disk_super->flags = cpu_to_le32(sb_flags); 460} 461 462static unsigned long set_clean_shutdown(unsigned long flags) 463{ 464 set_bit(CLEAN_SHUTDOWN, &flags); 465 return flags; 466} 467 468static unsigned long clear_clean_shutdown(unsigned long flags) 469{ 470 clear_bit(CLEAN_SHUTDOWN, &flags); 471 return flags; 472} 473 474static void read_superblock_fields(struct dm_cache_metadata *cmd, 475 struct cache_disk_superblock *disk_super) 476{ 477 cmd->root = le64_to_cpu(disk_super->mapping_root); 478 cmd->hint_root = le64_to_cpu(disk_super->hint_root); 479 cmd->discard_root = le64_to_cpu(disk_super->discard_root); 480 cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size); 481 cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks)); 482 cmd->data_block_size = le32_to_cpu(disk_super->data_block_size); 483 cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks)); 484 strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name)); 485 cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]); 486 cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]); 487 cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]); 488 cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size); 489 490 cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits); 491 cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses); 492 cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits); 493 cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses); 494 495 cmd->changed = false; 496} 497 498/* 499 * The mutator updates the superblock flags. 500 */ 501static int __begin_transaction_flags(struct dm_cache_metadata *cmd, 502 flags_mutator mutator) 503{ 504 int r; 505 struct cache_disk_superblock *disk_super; 506 struct dm_block *sblock; 507 508 r = superblock_lock(cmd, &sblock); 509 if (r) 510 return r; 511 512 disk_super = dm_block_data(sblock); 513 update_flags(disk_super, mutator); 514 read_superblock_fields(cmd, disk_super); 515 516 return dm_bm_flush_and_unlock(cmd->bm, sblock); 517} 518 519static int __begin_transaction(struct dm_cache_metadata *cmd) 520{ 521 int r; 522 struct cache_disk_superblock *disk_super; 523 struct dm_block *sblock; 524 525 /* 526 * We re-read the superblock every time. Shouldn't need to do this 527 * really. 528 */ 529 r = superblock_read_lock(cmd, &sblock); 530 if (r) 531 return r; 532 533 disk_super = dm_block_data(sblock); 534 read_superblock_fields(cmd, disk_super); 535 dm_bm_unlock(sblock); 536 537 return 0; 538} 539 540static int __commit_transaction(struct dm_cache_metadata *cmd, 541 flags_mutator mutator) 542{ 543 int r; 544 size_t metadata_len; 545 struct cache_disk_superblock *disk_super; 546 struct dm_block *sblock; 547 548 /* 549 * We need to know if the cache_disk_superblock exceeds a 512-byte sector. 550 */ 551 BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512); 552 553 r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root, 554 &cmd->discard_root); 555 if (r) 556 return r; 557 558 r = dm_tm_pre_commit(cmd->tm); 559 if (r < 0) 560 return r; 561 562 r = dm_sm_root_size(cmd->metadata_sm, &metadata_len); 563 if (r < 0) 564 return r; 565 566 r = superblock_lock(cmd, &sblock); 567 if (r) 568 return r; 569 570 disk_super = dm_block_data(sblock); 571 572 if (mutator) 573 update_flags(disk_super, mutator); 574 575 disk_super->mapping_root = cpu_to_le64(cmd->root); 576 disk_super->hint_root = cpu_to_le64(cmd->hint_root); 577 disk_super->discard_root = cpu_to_le64(cmd->discard_root); 578 disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size); 579 disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks)); 580 disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks)); 581 strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name)); 582 disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]); 583 disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]); 584 disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]); 585 586 disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits); 587 disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses); 588 disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits); 589 disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses); 590 591 r = dm_sm_copy_root(cmd->metadata_sm, &disk_super->metadata_space_map_root, 592 metadata_len); 593 if (r < 0) { 594 dm_bm_unlock(sblock); 595 return r; 596 } 597 598 return dm_tm_commit(cmd->tm, sblock); 599} 600 601/*----------------------------------------------------------------*/ 602 603/* 604 * The mappings are held in a dm-array that has 64-bit values stored in 605 * little-endian format. The index is the cblock, the high 48bits of the 606 * value are the oblock and the low 16 bit the flags. 607 */ 608#define FLAGS_MASK ((1 << 16) - 1) 609 610static __le64 pack_value(dm_oblock_t block, unsigned flags) 611{ 612 uint64_t value = from_oblock(block); 613 value <<= 16; 614 value = value | (flags & FLAGS_MASK); 615 return cpu_to_le64(value); 616} 617 618static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags) 619{ 620 uint64_t value = le64_to_cpu(value_le); 621 uint64_t b = value >> 16; 622 *block = to_oblock(b); 623 *flags = value & FLAGS_MASK; 624} 625 626/*----------------------------------------------------------------*/ 627 628struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev, 629 sector_t data_block_size, 630 bool may_format_device, 631 size_t policy_hint_size) 632{ 633 int r; 634 struct dm_cache_metadata *cmd; 635 636 cmd = kzalloc(sizeof(*cmd), GFP_KERNEL); 637 if (!cmd) { 638 DMERR("could not allocate metadata struct"); 639 return NULL; 640 } 641 642 init_rwsem(&cmd->root_lock); 643 cmd->bdev = bdev; 644 cmd->data_block_size = data_block_size; 645 cmd->cache_blocks = 0; 646 cmd->policy_hint_size = policy_hint_size; 647 cmd->changed = true; 648 649 r = __create_persistent_data_objects(cmd, may_format_device); 650 if (r) { 651 kfree(cmd); 652 return ERR_PTR(r); 653 } 654 655 r = __begin_transaction_flags(cmd, clear_clean_shutdown); 656 if (r < 0) { 657 dm_cache_metadata_close(cmd); 658 return ERR_PTR(r); 659 } 660 661 return cmd; 662} 663 664void dm_cache_metadata_close(struct dm_cache_metadata *cmd) 665{ 666 __destroy_persistent_data_objects(cmd); 667 kfree(cmd); 668} 669 670int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size) 671{ 672 int r; 673 __le64 null_mapping = pack_value(0, 0); 674 675 down_write(&cmd->root_lock); 676 __dm_bless_for_disk(&null_mapping); 677 r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks), 678 from_cblock(new_cache_size), 679 &null_mapping, &cmd->root); 680 if (!r) 681 cmd->cache_blocks = new_cache_size; 682 cmd->changed = true; 683 up_write(&cmd->root_lock); 684 685 return r; 686} 687 688int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd, 689 sector_t discard_block_size, 690 dm_dblock_t new_nr_entries) 691{ 692 int r; 693 694 down_write(&cmd->root_lock); 695 r = dm_bitset_resize(&cmd->discard_info, 696 cmd->discard_root, 697 from_dblock(cmd->discard_nr_blocks), 698 from_dblock(new_nr_entries), 699 false, &cmd->discard_root); 700 if (!r) { 701 cmd->discard_block_size = discard_block_size; 702 cmd->discard_nr_blocks = new_nr_entries; 703 } 704 705 cmd->changed = true; 706 up_write(&cmd->root_lock); 707 708 return r; 709} 710 711static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b) 712{ 713 return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root, 714 from_dblock(b), &cmd->discard_root); 715} 716 717static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b) 718{ 719 return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root, 720 from_dblock(b), &cmd->discard_root); 721} 722 723static int __is_discarded(struct dm_cache_metadata *cmd, dm_dblock_t b, 724 bool *is_discarded) 725{ 726 return dm_bitset_test_bit(&cmd->discard_info, cmd->discard_root, 727 from_dblock(b), &cmd->discard_root, 728 is_discarded); 729} 730 731static int __discard(struct dm_cache_metadata *cmd, 732 dm_dblock_t dblock, bool discard) 733{ 734 int r; 735 736 r = (discard ? __set_discard : __clear_discard)(cmd, dblock); 737 if (r) 738 return r; 739 740 cmd->changed = true; 741 return 0; 742} 743 744int dm_cache_set_discard(struct dm_cache_metadata *cmd, 745 dm_dblock_t dblock, bool discard) 746{ 747 int r; 748 749 down_write(&cmd->root_lock); 750 r = __discard(cmd, dblock, discard); 751 up_write(&cmd->root_lock); 752 753 return r; 754} 755 756static int __load_discards(struct dm_cache_metadata *cmd, 757 load_discard_fn fn, void *context) 758{ 759 int r = 0; 760 dm_block_t b; 761 bool discard; 762 763 for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) { 764 dm_dblock_t dblock = to_dblock(b); 765 766 if (cmd->clean_when_opened) { 767 r = __is_discarded(cmd, dblock, &discard); 768 if (r) 769 return r; 770 } else 771 discard = false; 772 773 r = fn(context, cmd->discard_block_size, dblock, discard); 774 if (r) 775 break; 776 } 777 778 return r; 779} 780 781int dm_cache_load_discards(struct dm_cache_metadata *cmd, 782 load_discard_fn fn, void *context) 783{ 784 int r; 785 786 down_read(&cmd->root_lock); 787 r = __load_discards(cmd, fn, context); 788 up_read(&cmd->root_lock); 789 790 return r; 791} 792 793dm_cblock_t dm_cache_size(struct dm_cache_metadata *cmd) 794{ 795 dm_cblock_t r; 796 797 down_read(&cmd->root_lock); 798 r = cmd->cache_blocks; 799 up_read(&cmd->root_lock); 800 801 return r; 802} 803 804static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock) 805{ 806 int r; 807 __le64 value = pack_value(0, 0); 808 809 __dm_bless_for_disk(&value); 810 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock), 811 &value, &cmd->root); 812 if (r) 813 return r; 814 815 cmd->changed = true; 816 return 0; 817} 818 819int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock) 820{ 821 int r; 822 823 down_write(&cmd->root_lock); 824 r = __remove(cmd, cblock); 825 up_write(&cmd->root_lock); 826 827 return r; 828} 829 830static int __insert(struct dm_cache_metadata *cmd, 831 dm_cblock_t cblock, dm_oblock_t oblock) 832{ 833 int r; 834 __le64 value = pack_value(oblock, M_VALID); 835 __dm_bless_for_disk(&value); 836 837 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock), 838 &value, &cmd->root); 839 if (r) 840 return r; 841 842 cmd->changed = true; 843 return 0; 844} 845 846int dm_cache_insert_mapping(struct dm_cache_metadata *cmd, 847 dm_cblock_t cblock, dm_oblock_t oblock) 848{ 849 int r; 850 851 down_write(&cmd->root_lock); 852 r = __insert(cmd, cblock, oblock); 853 up_write(&cmd->root_lock); 854 855 return r; 856} 857 858struct thunk { 859 load_mapping_fn fn; 860 void *context; 861 862 struct dm_cache_metadata *cmd; 863 bool respect_dirty_flags; 864 bool hints_valid; 865}; 866 867static bool policy_unchanged(struct dm_cache_metadata *cmd, 868 struct dm_cache_policy *policy) 869{ 870 const char *policy_name = dm_cache_policy_get_name(policy); 871 const unsigned *policy_version = dm_cache_policy_get_version(policy); 872 size_t policy_hint_size = dm_cache_policy_get_hint_size(policy); 873 874 /* 875 * Ensure policy names match. 876 */ 877 if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name))) 878 return false; 879 880 /* 881 * Ensure policy major versions match. 882 */ 883 if (cmd->policy_version[0] != policy_version[0]) 884 return false; 885 886 /* 887 * Ensure policy hint sizes match. 888 */ 889 if (cmd->policy_hint_size != policy_hint_size) 890 return false; 891 892 return true; 893} 894 895static bool hints_array_initialized(struct dm_cache_metadata *cmd) 896{ 897 return cmd->hint_root && cmd->policy_hint_size; 898} 899 900static bool hints_array_available(struct dm_cache_metadata *cmd, 901 struct dm_cache_policy *policy) 902{ 903 return cmd->clean_when_opened && policy_unchanged(cmd, policy) && 904 hints_array_initialized(cmd); 905} 906 907static int __load_mapping(void *context, uint64_t cblock, void *leaf) 908{ 909 int r = 0; 910 bool dirty; 911 __le64 value; 912 __le32 hint_value = 0; 913 dm_oblock_t oblock; 914 unsigned flags; 915 struct thunk *thunk = context; 916 struct dm_cache_metadata *cmd = thunk->cmd; 917 918 memcpy(&value, leaf, sizeof(value)); 919 unpack_value(value, &oblock, &flags); 920 921 if (flags & M_VALID) { 922 if (thunk->hints_valid) { 923 r = dm_array_get_value(&cmd->hint_info, cmd->hint_root, 924 cblock, &hint_value); 925 if (r && r != -ENODATA) 926 return r; 927 } 928 929 dirty = thunk->respect_dirty_flags ? (flags & M_DIRTY) : true; 930 r = thunk->fn(thunk->context, oblock, to_cblock(cblock), 931 dirty, le32_to_cpu(hint_value), thunk->hints_valid); 932 } 933 934 return r; 935} 936 937static int __load_mappings(struct dm_cache_metadata *cmd, 938 struct dm_cache_policy *policy, 939 load_mapping_fn fn, void *context) 940{ 941 struct thunk thunk; 942 943 thunk.fn = fn; 944 thunk.context = context; 945 946 thunk.cmd = cmd; 947 thunk.respect_dirty_flags = cmd->clean_when_opened; 948 thunk.hints_valid = hints_array_available(cmd, policy); 949 950 return dm_array_walk(&cmd->info, cmd->root, __load_mapping, &thunk); 951} 952 953int dm_cache_load_mappings(struct dm_cache_metadata *cmd, 954 struct dm_cache_policy *policy, 955 load_mapping_fn fn, void *context) 956{ 957 int r; 958 959 down_read(&cmd->root_lock); 960 r = __load_mappings(cmd, policy, fn, context); 961 up_read(&cmd->root_lock); 962 963 return r; 964} 965 966static int __dump_mapping(void *context, uint64_t cblock, void *leaf) 967{ 968 int r = 0; 969 __le64 value; 970 dm_oblock_t oblock; 971 unsigned flags; 972 973 memcpy(&value, leaf, sizeof(value)); 974 unpack_value(value, &oblock, &flags); 975 976 return r; 977} 978 979static int __dump_mappings(struct dm_cache_metadata *cmd) 980{ 981 return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL); 982} 983 984void dm_cache_dump(struct dm_cache_metadata *cmd) 985{ 986 down_read(&cmd->root_lock); 987 __dump_mappings(cmd); 988 up_read(&cmd->root_lock); 989} 990 991int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd) 992{ 993 int r; 994 995 down_read(&cmd->root_lock); 996 r = cmd->changed; 997 up_read(&cmd->root_lock); 998 999 return r; 1000} 1001 1002static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty) 1003{ 1004 int r; 1005 unsigned flags; 1006 dm_oblock_t oblock; 1007 __le64 value; 1008 1009 r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value); 1010 if (r) 1011 return r; 1012 1013 unpack_value(value, &oblock, &flags); 1014 1015 if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty)) 1016 /* nothing to be done */ 1017 return 0; 1018 1019 value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0)); 1020 __dm_bless_for_disk(&value); 1021 1022 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock), 1023 &value, &cmd->root); 1024 if (r) 1025 return r; 1026 1027 cmd->changed = true; 1028 return 0; 1029 1030} 1031 1032int dm_cache_set_dirty(struct dm_cache_metadata *cmd, 1033 dm_cblock_t cblock, bool dirty) 1034{ 1035 int r; 1036 1037 down_write(&cmd->root_lock); 1038 r = __dirty(cmd, cblock, dirty); 1039 up_write(&cmd->root_lock); 1040 1041 return r; 1042} 1043 1044void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd, 1045 struct dm_cache_statistics *stats) 1046{ 1047 down_read(&cmd->root_lock); 1048 *stats = cmd->stats; 1049 up_read(&cmd->root_lock); 1050} 1051 1052void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd, 1053 struct dm_cache_statistics *stats) 1054{ 1055 down_write(&cmd->root_lock); 1056 cmd->stats = *stats; 1057 up_write(&cmd->root_lock); 1058} 1059 1060int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown) 1061{ 1062 int r; 1063 flags_mutator mutator = (clean_shutdown ? set_clean_shutdown : 1064 clear_clean_shutdown); 1065 1066 down_write(&cmd->root_lock); 1067 r = __commit_transaction(cmd, mutator); 1068 if (r) 1069 goto out; 1070 1071 r = __begin_transaction(cmd); 1072 1073out: 1074 up_write(&cmd->root_lock); 1075 return r; 1076} 1077 1078int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd, 1079 dm_block_t *result) 1080{ 1081 int r = -EINVAL; 1082 1083 down_read(&cmd->root_lock); 1084 r = dm_sm_get_nr_free(cmd->metadata_sm, result); 1085 up_read(&cmd->root_lock); 1086 1087 return r; 1088} 1089 1090int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd, 1091 dm_block_t *result) 1092{ 1093 int r = -EINVAL; 1094 1095 down_read(&cmd->root_lock); 1096 r = dm_sm_get_nr_blocks(cmd->metadata_sm, result); 1097 up_read(&cmd->root_lock); 1098 1099 return r; 1100} 1101 1102/*----------------------------------------------------------------*/ 1103 1104static int begin_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy) 1105{ 1106 int r; 1107 __le32 value; 1108 size_t hint_size; 1109 const char *policy_name = dm_cache_policy_get_name(policy); 1110 const unsigned *policy_version = dm_cache_policy_get_version(policy); 1111 1112 if (!policy_name[0] || 1113 (strlen(policy_name) > sizeof(cmd->policy_name) - 1)) 1114 return -EINVAL; 1115 1116 if (!policy_unchanged(cmd, policy)) { 1117 strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name)); 1118 memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version)); 1119 1120 hint_size = dm_cache_policy_get_hint_size(policy); 1121 if (!hint_size) 1122 return 0; /* short-circuit hints initialization */ 1123 cmd->policy_hint_size = hint_size; 1124 1125 if (cmd->hint_root) { 1126 r = dm_array_del(&cmd->hint_info, cmd->hint_root); 1127 if (r) 1128 return r; 1129 } 1130 1131 r = dm_array_empty(&cmd->hint_info, &cmd->hint_root); 1132 if (r) 1133 return r; 1134 1135 value = cpu_to_le32(0); 1136 __dm_bless_for_disk(&value); 1137 r = dm_array_resize(&cmd->hint_info, cmd->hint_root, 0, 1138 from_cblock(cmd->cache_blocks), 1139 &value, &cmd->hint_root); 1140 if (r) 1141 return r; 1142 } 1143 1144 return 0; 1145} 1146 1147int dm_cache_begin_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy) 1148{ 1149 int r; 1150 1151 down_write(&cmd->root_lock); 1152 r = begin_hints(cmd, policy); 1153 up_write(&cmd->root_lock); 1154 1155 return r; 1156} 1157 1158static int save_hint(struct dm_cache_metadata *cmd, dm_cblock_t cblock, 1159 uint32_t hint) 1160{ 1161 int r; 1162 __le32 value = cpu_to_le32(hint); 1163 __dm_bless_for_disk(&value); 1164 1165 r = dm_array_set_value(&cmd->hint_info, cmd->hint_root, 1166 from_cblock(cblock), &value, &cmd->hint_root); 1167 cmd->changed = true; 1168 1169 return r; 1170} 1171 1172int dm_cache_save_hint(struct dm_cache_metadata *cmd, dm_cblock_t cblock, 1173 uint32_t hint) 1174{ 1175 int r; 1176 1177 if (!hints_array_initialized(cmd)) 1178 return 0; 1179 1180 down_write(&cmd->root_lock); 1181 r = save_hint(cmd, cblock, hint); 1182 up_write(&cmd->root_lock); 1183 1184 return r; 1185} 1186