drivers/md/dm-cache-metadata.c

   1 /*
   2  * Copyright (C) 2012 Red Hat, Inc.
   3  *
   4  * This file is released under the GPL.
   5  */
   6
   7 #include "dm-cache-metadata.h"
   8
   9 #include "persistent-data/dm-array.h"
  10 #include "persistent-data/dm-bitset.h"
  11 #include "persistent-data/dm-space-map.h"
  12 #include "persistent-data/dm-space-map-disk.h"
  13 #include "persistent-data/dm-transaction-manager.h"
  14
  15 #include <linux/device-mapper.h>
  16
  17 /*----------------------------------------------------------------*/
  18
  19 #define DM_MSG_PREFIX   "cache metadata"
  20
  21 #define CACHE_SUPERBLOCK_MAGIC 06142003
  22 #define CACHE_SUPERBLOCK_LOCATION 0
  23
  24 /*
  25  * defines a range of metadata versions that this module can handle.
  26  */
  27 #define MIN_CACHE_VERSION 1
  28 #define MAX_CACHE_VERSION 2
  29
  30 /*
  31  *  3 for btree insert +
  32  *  2 for btree lookup used within space map
  33  */
  34 #define CACHE_MAX_CONCURRENT_LOCKS 5
  35 #define SPACE_MAP_ROOT_SIZE 128
  36
  37 enum superblock_flag_bits {
  38         /* for spotting crashes that would invalidate the dirty bitset */
  39         CLEAN_SHUTDOWN,
  40         /* metadata must be checked using the tools */
  41         NEEDS_CHECK,
  42 };
  43
  44 /*
  45  * Each mapping from cache block -> origin block carries a set of flags.
  46  */
  47 enum mapping_bits {
  48         /*
  49          * A valid mapping.  Because we're using an array we clear this
  50          * flag for an non existant mapping.
  51          */
  52         M_VALID = 1,
  53
  54         /*
  55          * The data on the cache is different from that on the origin.
  56          * This flag is only used by metadata format 1.
  57          */
  58         M_DIRTY = 2
  59 };
  60
  61 struct cache_disk_superblock {
  62         __le32 csum;
  63         __le32 flags;
  64         __le64 blocknr;
  65
  66         __u8 uuid[16];
  67         __le64 magic;
  68         __le32 version;
  69
  70         __u8 policy_name[CACHE_POLICY_NAME_SIZE];
  71         __le32 policy_hint_size;
  72
  73         __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
  74         __le64 mapping_root;
  75         __le64 hint_root;
  76
  77         __le64 discard_root;
  78         __le64 discard_block_size;
  79         __le64 discard_nr_blocks;
  80
  81         __le32 data_block_size;
  82         __le32 metadata_block_size;
  83         __le32 cache_blocks;
  84
  85         __le32 compat_flags;
  86         __le32 compat_ro_flags;
  87         __le32 incompat_flags;
  88
  89         __le32 read_hits;
  90         __le32 read_misses;
  91         __le32 write_hits;
  92         __le32 write_misses;
  93
  94         __le32 policy_version[CACHE_POLICY_VERSION_SIZE];
  95
  96         /*
  97          * Metadata format 2 fields.
  98          */
  99         __le64 dirty_root;
 100 } __packed;
 101
 102 struct dm_cache_metadata {
 103         atomic_t ref_count;
 104         struct list_head list;
 105
 106         unsigned version;
 107         struct block_device *bdev;
 108         struct dm_block_manager *bm;
 109         struct dm_space_map *metadata_sm;
 110         struct dm_transaction_manager *tm;
 111
 112         struct dm_array_info info;
 113         struct dm_array_info hint_info;
 114         struct dm_disk_bitset discard_info;
 115
 116         struct rw_semaphore root_lock;
 117         unsigned long flags;
 118         dm_block_t root;
 119         dm_block_t hint_root;
 120         dm_block_t discard_root;
 121
 122         sector_t discard_block_size;
 123         dm_dblock_t discard_nr_blocks;
 124
 125         sector_t data_block_size;
 126         dm_cblock_t cache_blocks;
 127         bool changed:1;
 128         bool clean_when_opened:1;
 129
 130         char policy_name[CACHE_POLICY_NAME_SIZE];
 131         unsigned policy_version[CACHE_POLICY_VERSION_SIZE];
 132         size_t policy_hint_size;
 133         struct dm_cache_statistics stats;
 134
 135         /*
 136          * Reading the space map root can fail, so we read it into this
 137          * buffer before the superblock is locked and updated.
 138          */
 139         __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
 140
 141         /*
 142          * Set if a transaction has to be aborted but the attempt to roll
 143          * back to the previous (good) transaction failed.  The only
 144          * metadata operation permissible in this state is the closing of
 145          * the device.
 146          */
 147         bool fail_io:1;
 148
 149         /*
 150          * Metadata format 2 fields.
 151          */
 152         dm_block_t dirty_root;
 153         struct dm_disk_bitset dirty_info;
 154
 155         /*
 156          * These structures are used when loading metadata.  They're too
 157          * big to put on the stack.
 158          */
 159         struct dm_array_cursor mapping_cursor;
 160         struct dm_array_cursor hint_cursor;
 161         struct dm_bitset_cursor dirty_cursor;
 162 };
 163
 164 /*-------------------------------------------------------------------
 165  * superblock validator
 166  *-----------------------------------------------------------------*/
 167
 168 #define SUPERBLOCK_CSUM_XOR 9031977
 169
 170 static void sb_prepare_for_write(struct dm_block_validator *v,
 171                                  struct dm_block *b,
 172                                  size_t sb_block_size)
 173 {
 174         struct cache_disk_superblock *disk_super = dm_block_data(b);
 175
 176         disk_super->blocknr = cpu_to_le64(dm_block_location(b));
 177         disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
 178                                                       sb_block_size - sizeof(__le32),
 179                                                       SUPERBLOCK_CSUM_XOR));
 180 }
 181
 182 static int check_metadata_version(struct cache_disk_superblock *disk_super)
 183 {
 184         uint32_t metadata_version = le32_to_cpu(disk_super->version);
 185
 186         if (metadata_version < MIN_CACHE_VERSION || metadata_version > MAX_CACHE_VERSION) {
 187                 DMERR("Cache metadata version %u found, but only versions between %u and %u supported.",
 188                       metadata_version, MIN_CACHE_VERSION, MAX_CACHE_VERSION);
 189                 return -EINVAL;
 190         }
 191
 192         return 0;
 193 }
 194
 195 static int sb_check(struct dm_block_validator *v,
 196                     struct dm_block *b,
 197                     size_t sb_block_size)
 198 {
 199         struct cache_disk_superblock *disk_super = dm_block_data(b);
 200         __le32 csum_le;
 201
 202         if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
 203                 DMERR("sb_check failed: blocknr %llu: wanted %llu",
 204                       le64_to_cpu(disk_super->blocknr),
 205                       (unsigned long long)dm_block_location(b));
 206                 return -ENOTBLK;
 207         }
 208
 209         if (le64_to_cpu(disk_super->magic) != CACHE_SUPERBLOCK_MAGIC) {
 210                 DMERR("sb_check failed: magic %llu: wanted %llu",
 211                       le64_to_cpu(disk_super->magic),
 212                       (unsigned long long)CACHE_SUPERBLOCK_MAGIC);
 213                 return -EILSEQ;
 214         }
 215
 216         csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
 217                                              sb_block_size - sizeof(__le32),
 218                                              SUPERBLOCK_CSUM_XOR));
 219         if (csum_le != disk_super->csum) {
 220                 DMERR("sb_check failed: csum %u: wanted %u",
 221                       le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
 222                 return -EILSEQ;
 223         }
 224
 225         return check_metadata_version(disk_super);
 226 }
 227
 228 static struct dm_block_validator sb_validator = {
 229         .name = "superblock",
 230         .prepare_for_write = sb_prepare_for_write,
 231         .check = sb_check
 232 };
 233
 234 /*----------------------------------------------------------------*/
 235
 236 static int superblock_read_lock(struct dm_cache_metadata *cmd,
 237                                 struct dm_block **sblock)
 238 {
 239         return dm_bm_read_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 240                                &sb_validator, sblock);
 241 }
 242
 243 static int superblock_lock_zero(struct dm_cache_metadata *cmd,
 244                                 struct dm_block **sblock)
 245 {
 246         return dm_bm_write_lock_zero(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 247                                      &sb_validator, sblock);
 248 }
 249
 250 static int superblock_lock(struct dm_cache_metadata *cmd,
 251                            struct dm_block **sblock)
 252 {
 253         return dm_bm_write_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 254                                 &sb_validator, sblock);
 255 }
 256
 257 /*----------------------------------------------------------------*/
 258
 259 static int __superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
 260 {
 261         int r;
 262         unsigned i;
 263         struct dm_block *b;
 264         __le64 *data_le, zero = cpu_to_le64(0);
 265         unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
 266
 267         /*
 268          * We can't use a validator here - it may be all zeroes.
 269          */
 270         r = dm_bm_read_lock(bm, CACHE_SUPERBLOCK_LOCATION, NULL, &b);
 271         if (r)
 272                 return r;
 273
 274         data_le = dm_block_data(b);
 275         *result = true;
 276         for (i = 0; i < sb_block_size; i++) {
 277                 if (data_le[i] != zero) {
 278                         *result = false;
 279                         break;
 280                 }
 281         }
 282
 283         dm_bm_unlock(b);
 284
 285         return 0;
 286 }
 287
 288 static void __setup_mapping_info(struct dm_cache_metadata *cmd)
 289 {
 290         struct dm_btree_value_type vt;
 291
 292         vt.context = NULL;
 293         vt.size = sizeof(__le64);
 294         vt.inc = NULL;
 295         vt.dec = NULL;
 296         vt.equal = NULL;
 297         dm_array_info_init(&cmd->info, cmd->tm, &vt);
 298
 299         if (cmd->policy_hint_size) {
 300                 vt.size = sizeof(__le32);
 301                 dm_array_info_init(&cmd->hint_info, cmd->tm, &vt);
 302         }
 303 }
 304
 305 static int __save_sm_root(struct dm_cache_metadata *cmd)
 306 {
 307         int r;
 308         size_t metadata_len;
 309
 310         r = dm_sm_root_size(cmd->metadata_sm, &metadata_len);
 311         if (r < 0)
 312                 return r;
 313
 314         return dm_sm_copy_root(cmd->metadata_sm, &cmd->metadata_space_map_root,
 315                                metadata_len);
 316 }
 317
 318 static void __copy_sm_root(struct dm_cache_metadata *cmd,
 319                            struct cache_disk_superblock *disk_super)
 320 {
 321         memcpy(&disk_super->metadata_space_map_root,
 322                &cmd->metadata_space_map_root,
 323                sizeof(cmd->metadata_space_map_root));
 324 }
 325
 326 static bool separate_dirty_bits(struct dm_cache_metadata *cmd)
 327 {
 328         return cmd->version >= 2;
 329 }
 330
 331 static int __write_initial_superblock(struct dm_cache_metadata *cmd)
 332 {
 333         int r;
 334         struct dm_block *sblock;
 335         struct cache_disk_superblock *disk_super;
 336         sector_t bdev_size = i_size_read(cmd->bdev->bd_inode) >> SECTOR_SHIFT;
 337
 338         /* FIXME: see if we can lose the max sectors limit */
 339         if (bdev_size > DM_CACHE_METADATA_MAX_SECTORS)
 340                 bdev_size = DM_CACHE_METADATA_MAX_SECTORS;
 341
 342         r = dm_tm_pre_commit(cmd->tm);
 343         if (r < 0)
 344                 return r;
 345
 346         /*
 347          * dm_sm_copy_root() can fail.  So we need to do it before we start
 348          * updating the superblock.
 349          */
 350         r = __save_sm_root(cmd);
 351         if (r)
 352                 return r;
 353
 354         r = superblock_lock_zero(cmd, &sblock);
 355         if (r)
 356                 return r;
 357
 358         disk_super = dm_block_data(sblock);
 359         disk_super->flags = 0;
 360         memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
 361         disk_super->magic = cpu_to_le64(CACHE_SUPERBLOCK_MAGIC);
 362         disk_super->version = cpu_to_le32(cmd->version);
 363         memset(disk_super->policy_name, 0, sizeof(disk_super->policy_name));
 364         memset(disk_super->policy_version, 0, sizeof(disk_super->policy_version));
 365         disk_super->policy_hint_size = cpu_to_le32(0);
 366
 367         __copy_sm_root(cmd, disk_super);
 368
 369         disk_super->mapping_root = cpu_to_le64(cmd->root);
 370         disk_super->hint_root = cpu_to_le64(cmd->hint_root);
 371         disk_super->discard_root = cpu_to_le64(cmd->discard_root);
 372         disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
 373         disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
 374         disk_super->metadata_block_size = cpu_to_le32(DM_CACHE_METADATA_BLOCK_SIZE);
 375         disk_super->data_block_size = cpu_to_le32(cmd->data_block_size);
 376         disk_super->cache_blocks = cpu_to_le32(0);
 377
 378         disk_super->read_hits = cpu_to_le32(0);
 379         disk_super->read_misses = cpu_to_le32(0);
 380         disk_super->write_hits = cpu_to_le32(0);
 381         disk_super->write_misses = cpu_to_le32(0);
 382
 383         if (separate_dirty_bits(cmd))
 384                 disk_super->dirty_root = cpu_to_le64(cmd->dirty_root);
 385
 386         return dm_tm_commit(cmd->tm, sblock);
 387 }
 388
 389 static int __format_metadata(struct dm_cache_metadata *cmd)
 390 {
 391         int r;
 392
 393         r = dm_tm_create_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 394                                  &cmd->tm, &cmd->metadata_sm);
 395         if (r < 0) {
 396                 DMERR("tm_create_with_sm failed");
 397                 return r;
 398         }
 399
 400         __setup_mapping_info(cmd);
 401
 402         r = dm_array_empty(&cmd->info, &cmd->root);
 403         if (r < 0)
 404                 goto bad;
 405
 406         if (separate_dirty_bits(cmd)) {
 407                 dm_disk_bitset_init(cmd->tm, &cmd->dirty_info);
 408                 r = dm_bitset_empty(&cmd->dirty_info, &cmd->dirty_root);
 409                 if (r < 0)
 410                         goto bad;
 411         }
 412
 413         dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
 414         r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root);
 415         if (r < 0)
 416                 goto bad;
 417
 418         cmd->discard_block_size = 0;
 419         cmd->discard_nr_blocks = 0;
 420
 421         r = __write_initial_superblock(cmd);
 422         if (r)
 423                 goto bad;
 424
 425         cmd->clean_when_opened = true;
 426         return 0;
 427
 428 bad:
 429         dm_tm_destroy(cmd->tm);
 430         dm_sm_destroy(cmd->metadata_sm);
 431
 432         return r;
 433 }
 434
 435 static int __check_incompat_features(struct cache_disk_superblock *disk_super,
 436                                      struct dm_cache_metadata *cmd)
 437 {
 438         uint32_t incompat_flags, features;
 439
 440         incompat_flags = le32_to_cpu(disk_super->incompat_flags);
 441         features = incompat_flags & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
 442         if (features) {
 443                 DMERR("could not access metadata due to unsupported optional features (%lx).",
 444                       (unsigned long)features);
 445                 return -EINVAL;
 446         }
 447
 448         /*
 449          * Check for read-only metadata to skip the following RDWR checks.
 450          */
 451         if (get_disk_ro(cmd->bdev->bd_disk))
 452                 return 0;
 453
 454         features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP;
 455         if (features) {
 456                 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
 457                       (unsigned long)features);
 458                 return -EINVAL;
 459         }
 460
 461         return 0;
 462 }
 463
 464 static int __open_metadata(struct dm_cache_metadata *cmd)
 465 {
 466         int r;
 467         struct dm_block *sblock;
 468         struct cache_disk_superblock *disk_super;
 469         unsigned long sb_flags;
 470
 471         r = superblock_read_lock(cmd, &sblock);
 472         if (r < 0) {
 473                 DMERR("couldn't read lock superblock");
 474                 return r;
 475         }
 476
 477         disk_super = dm_block_data(sblock);
 478
 479         /* Verify the data block size hasn't changed */
 480         if (le32_to_cpu(disk_super->data_block_size) != cmd->data_block_size) {
 481                 DMERR("changing the data block size (from %u to %llu) is not supported",
 482                       le32_to_cpu(disk_super->data_block_size),
 483                       (unsigned long long)cmd->data_block_size);
 484                 r = -EINVAL;
 485                 goto bad;
 486         }
 487
 488         r = __check_incompat_features(disk_super, cmd);
 489         if (r < 0)
 490                 goto bad;
 491
 492         r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 493                                disk_super->metadata_space_map_root,
 494                                sizeof(disk_super->metadata_space_map_root),
 495                                &cmd->tm, &cmd->metadata_sm);
 496         if (r < 0) {
 497                 DMERR("tm_open_with_sm failed");
 498                 goto bad;
 499         }
 500
 501         __setup_mapping_info(cmd);
 502         dm_disk_bitset_init(cmd->tm, &cmd->dirty_info);
 503         dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
 504         sb_flags = le32_to_cpu(disk_super->flags);
 505         cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags);
 506         dm_bm_unlock(sblock);
 507
 508         return 0;
 509
 510 bad:
 511         dm_bm_unlock(sblock);
 512         return r;
 513 }
 514
 515 static int __open_or_format_metadata(struct dm_cache_metadata *cmd,
 516                                      bool format_device)
 517 {
 518         int r;
 519         bool unformatted = false;
 520
 521         r = __superblock_all_zeroes(cmd->bm, &unformatted);
 522         if (r)
 523                 return r;
 524
 525         if (unformatted)
 526                 return format_device ? __format_metadata(cmd) : -EPERM;
 527
 528         return __open_metadata(cmd);
 529 }
 530
 531 static int __create_persistent_data_objects(struct dm_cache_metadata *cmd,
 532                                             bool may_format_device)
 533 {
 534         int r;
 535         cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
 536                                           CACHE_MAX_CONCURRENT_LOCKS);
 537         if (IS_ERR(cmd->bm)) {
 538                 DMERR("could not create block manager");
 539                 r = PTR_ERR(cmd->bm);
 540                 cmd->bm = NULL;
 541                 return r;
 542         }
 543
 544         r = __open_or_format_metadata(cmd, may_format_device);
 545         if (r) {
 546                 dm_block_manager_destroy(cmd->bm);
 547                 cmd->bm = NULL;
 548         }
 549
 550         return r;
 551 }
 552
 553 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd)
 554 {
 555         dm_sm_destroy(cmd->metadata_sm);
 556         dm_tm_destroy(cmd->tm);
 557         dm_block_manager_destroy(cmd->bm);
 558 }
 559
 560 typedef unsigned long (*flags_mutator)(unsigned long);
 561
 562 static void update_flags(struct cache_disk_superblock *disk_super,
 563                          flags_mutator mutator)
 564 {
 565         uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
 566         disk_super->flags = cpu_to_le32(sb_flags);
 567 }
 568
 569 static unsigned long set_clean_shutdown(unsigned long flags)
 570 {
 571         set_bit(CLEAN_SHUTDOWN, &flags);
 572         return flags;
 573 }
 574
 575 static unsigned long clear_clean_shutdown(unsigned long flags)
 576 {
 577         clear_bit(CLEAN_SHUTDOWN, &flags);
 578         return flags;
 579 }
 580
 581 static void read_superblock_fields(struct dm_cache_metadata *cmd,
 582                                    struct cache_disk_superblock *disk_super)
 583 {
 584         cmd->version = le32_to_cpu(disk_super->version);
 585         cmd->flags = le32_to_cpu(disk_super->flags);
 586         cmd->root = le64_to_cpu(disk_super->mapping_root);
 587         cmd->hint_root = le64_to_cpu(disk_super->hint_root);
 588         cmd->discard_root = le64_to_cpu(disk_super->discard_root);
 589         cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
 590         cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
 591         cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
 592         cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
 593         strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
 594         cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
 595         cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
 596         cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
 597         cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
 598
 599         cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
 600         cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
 601         cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
 602         cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
 603
 604         if (separate_dirty_bits(cmd))
 605                 cmd->dirty_root = le64_to_cpu(disk_super->dirty_root);
 606
 607         cmd->changed = false;
 608 }
 609
 610 /*
 611  * The mutator updates the superblock flags.
 612  */
 613 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
 614                                      flags_mutator mutator)
 615 {
 616         int r;
 617         struct cache_disk_superblock *disk_super;
 618         struct dm_block *sblock;
 619
 620         r = superblock_lock(cmd, &sblock);
 621         if (r)
 622                 return r;
 623
 624         disk_super = dm_block_data(sblock);
 625         update_flags(disk_super, mutator);
 626         read_superblock_fields(cmd, disk_super);
 627         dm_bm_unlock(sblock);
 628
 629         return dm_bm_flush(cmd->bm);
 630 }
 631
 632 static int __begin_transaction(struct dm_cache_metadata *cmd)
 633 {
 634         int r;
 635         struct cache_disk_superblock *disk_super;
 636         struct dm_block *sblock;
 637
 638         /*
 639          * We re-read the superblock every time.  Shouldn't need to do this
 640          * really.
 641          */
 642         r = superblock_read_lock(cmd, &sblock);
 643         if (r)
 644                 return r;
 645
 646         disk_super = dm_block_data(sblock);
 647         read_superblock_fields(cmd, disk_super);
 648         dm_bm_unlock(sblock);
 649
 650         return 0;
 651 }
 652
 653 static int __commit_transaction(struct dm_cache_metadata *cmd,
 654                                 flags_mutator mutator)
 655 {
 656         int r;
 657         struct cache_disk_superblock *disk_super;
 658         struct dm_block *sblock;
 659
 660         /*
 661          * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
 662          */
 663         BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
 664
 665         if (separate_dirty_bits(cmd)) {
 666                 r = dm_bitset_flush(&cmd->dirty_info, cmd->dirty_root,
 667                                     &cmd->dirty_root);
 668                 if (r)
 669                         return r;
 670         }
 671
 672         r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
 673                             &cmd->discard_root);
 674         if (r)
 675                 return r;
 676
 677         r = dm_tm_pre_commit(cmd->tm);
 678         if (r < 0)
 679                 return r;
 680
 681         r = __save_sm_root(cmd);
 682         if (r)
 683                 return r;
 684
 685         r = superblock_lock(cmd, &sblock);
 686         if (r)
 687                 return r;
 688
 689         disk_super = dm_block_data(sblock);
 690
 691         disk_super->flags = cpu_to_le32(cmd->flags);
 692         if (mutator)
 693                 update_flags(disk_super, mutator);
 694
 695         disk_super->mapping_root = cpu_to_le64(cmd->root);
 696         if (separate_dirty_bits(cmd))
 697                 disk_super->dirty_root = cpu_to_le64(cmd->dirty_root);
 698         disk_super->hint_root = cpu_to_le64(cmd->hint_root);
 699         disk_super->discard_root = cpu_to_le64(cmd->discard_root);
 700         disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
 701         disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
 702         disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
 703         strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
 704         disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
 705         disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
 706         disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
 707         disk_super->policy_hint_size = cpu_to_le32(cmd->policy_hint_size);
 708
 709         disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
 710         disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
 711         disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
 712         disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
 713         __copy_sm_root(cmd, disk_super);
 714
 715         return dm_tm_commit(cmd->tm, sblock);
 716 }
 717
 718 /*----------------------------------------------------------------*/
 719
 720 /*
 721  * The mappings are held in a dm-array that has 64-bit values stored in
 722  * little-endian format.  The index is the cblock, the high 48bits of the
 723  * value are the oblock and the low 16 bit the flags.
 724  */
 725 #define FLAGS_MASK ((1 << 16) - 1)
 726
 727 static __le64 pack_value(dm_oblock_t block, unsigned flags)
 728 {
 729         uint64_t value = from_oblock(block);
 730         value <<= 16;
 731         value = value | (flags & FLAGS_MASK);
 732         return cpu_to_le64(value);
 733 }
 734
 735 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
 736 {
 737         uint64_t value = le64_to_cpu(value_le);
 738         uint64_t b = value >> 16;
 739         *block = to_oblock(b);
 740         *flags = value & FLAGS_MASK;
 741 }
 742
 743 /*----------------------------------------------------------------*/
 744
 745 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
 746                                                sector_t data_block_size,
 747                                                bool may_format_device,
 748                                                size_t policy_hint_size,
 749                                                unsigned metadata_version)
 750 {
 751         int r;
 752         struct dm_cache_metadata *cmd;
 753
 754         cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
 755         if (!cmd) {
 756                 DMERR("could not allocate metadata struct");
 757                 return ERR_PTR(-ENOMEM);
 758         }
 759
 760         cmd->version = metadata_version;
 761         atomic_set(&cmd->ref_count, 1);
 762         init_rwsem(&cmd->root_lock);
 763         cmd->bdev = bdev;
 764         cmd->data_block_size = data_block_size;
 765         cmd->cache_blocks = 0;
 766         cmd->policy_hint_size = policy_hint_size;
 767         cmd->changed = true;
 768         cmd->fail_io = false;
 769
 770         r = __create_persistent_data_objects(cmd, may_format_device);
 771         if (r) {
 772                 kfree(cmd);
 773                 return ERR_PTR(r);
 774         }
 775
 776         r = __begin_transaction_flags(cmd, clear_clean_shutdown);
 777         if (r < 0) {
 778                 dm_cache_metadata_close(cmd);
 779                 return ERR_PTR(r);
 780         }
 781
 782         return cmd;
 783 }
 784
 785 /*
 786  * We keep a little list of ref counted metadata objects to prevent two
 787  * different target instances creating separate bufio instances.  This is
 788  * an issue if a table is reloaded before the suspend.
 789  */
 790 static DEFINE_MUTEX(table_lock);
 791 static LIST_HEAD(table);
 792
 793 static struct dm_cache_metadata *lookup(struct block_device *bdev)
 794 {
 795         struct dm_cache_metadata *cmd;
 796
 797         list_for_each_entry(cmd, &table, list)
 798                 if (cmd->bdev == bdev) {
 799                         atomic_inc(&cmd->ref_count);
 800                         return cmd;
 801                 }
 802
 803         return NULL;
 804 }
 805
 806 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
 807                                                 sector_t data_block_size,
 808                                                 bool may_format_device,
 809                                                 size_t policy_hint_size,
 810                                                 unsigned metadata_version)
 811 {
 812         struct dm_cache_metadata *cmd, *cmd2;
 813
 814         mutex_lock(&table_lock);
 815         cmd = lookup(bdev);
 816         mutex_unlock(&table_lock);
 817
 818         if (cmd)
 819                 return cmd;
 820
 821         cmd = metadata_open(bdev, data_block_size, may_format_device,
 822                             policy_hint_size, metadata_version);
 823         if (!IS_ERR(cmd)) {
 824                 mutex_lock(&table_lock);
 825                 cmd2 = lookup(bdev);
 826                 if (cmd2) {
 827                         mutex_unlock(&table_lock);
 828                         __destroy_persistent_data_objects(cmd);
 829                         kfree(cmd);
 830                         return cmd2;
 831                 }
 832                 list_add(&cmd->list, &table);
 833                 mutex_unlock(&table_lock);
 834         }
 835
 836         return cmd;
 837 }
 838
 839 static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
 840 {
 841         if (cmd->data_block_size != data_block_size) {
 842                 DMERR("data_block_size (%llu) different from that in metadata (%llu)",
 843                       (unsigned long long) data_block_size,
 844                       (unsigned long long) cmd->data_block_size);
 845                 return false;
 846         }
 847
 848         return true;
 849 }
 850
 851 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
 852                                                  sector_t data_block_size,
 853                                                  bool may_format_device,
 854                                                  size_t policy_hint_size,
 855                                                  unsigned metadata_version)
 856 {
 857         struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size, may_format_device,
 858                                                        policy_hint_size, metadata_version);
 859
 860         if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
 861                 dm_cache_metadata_close(cmd);
 862                 return ERR_PTR(-EINVAL);
 863         }
 864
 865         return cmd;
 866 }
 867
 868 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
 869 {
 870         if (atomic_dec_and_test(&cmd->ref_count)) {
 871                 mutex_lock(&table_lock);
 872                 list_del(&cmd->list);
 873                 mutex_unlock(&table_lock);
 874
 875                 if (!cmd->fail_io)
 876                         __destroy_persistent_data_objects(cmd);
 877                 kfree(cmd);
 878         }
 879 }
 880
 881 /*
 882  * Checks that the given cache block is either unmapped or clean.
 883  */
 884 static int block_clean_combined_dirty(struct dm_cache_metadata *cmd, dm_cblock_t b,
 885                                       bool *result)
 886 {
 887         int r;
 888         __le64 value;
 889         dm_oblock_t ob;
 890         unsigned flags;
 891
 892         r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
 893         if (r)
 894                 return r;
 895
 896         unpack_value(value, &ob, &flags);
 897         *result = !((flags & M_VALID) && (flags & M_DIRTY));
 898
 899         return 0;
 900 }
 901
 902 static int blocks_are_clean_combined_dirty(struct dm_cache_metadata *cmd,
 903                                            dm_cblock_t begin, dm_cblock_t end,
 904                                            bool *result)
 905 {
 906         int r;
 907         *result = true;
 908
 909         while (begin != end) {
 910                 r = block_clean_combined_dirty(cmd, begin, result);
 911                 if (r) {
 912                         DMERR("block_clean_combined_dirty failed");
 913                         return r;
 914                 }
 915
 916                 if (!*result) {
 917                         DMERR("cache block %llu is dirty",
 918                               (unsigned long long) from_cblock(begin));
 919                         return 0;
 920                 }
 921
 922                 begin = to_cblock(from_cblock(begin) + 1);
 923         }
 924
 925         return 0;
 926 }
 927
 928 static int blocks_are_clean_separate_dirty(struct dm_cache_metadata *cmd,
 929                                            dm_cblock_t begin, dm_cblock_t end,
 930                                            bool *result)
 931 {
 932         int r;
 933         bool dirty_flag;
 934         *result = true;
 935
 936         if (from_cblock(cmd->cache_blocks) == 0)
 937                 /* Nothing to do */
 938                 return 0;
 939
 940         r = dm_bitset_cursor_begin(&cmd->dirty_info, cmd->dirty_root,
 941                                    from_cblock(cmd->cache_blocks), &cmd->dirty_cursor);
 942         if (r) {
 943                 DMERR("%s: dm_bitset_cursor_begin for dirty failed", __func__);
 944                 return r;
 945         }
 946
 947         r = dm_bitset_cursor_skip(&cmd->dirty_cursor, from_cblock(begin));
 948         if (r) {
 949                 DMERR("%s: dm_bitset_cursor_skip for dirty failed", __func__);
 950                 dm_bitset_cursor_end(&cmd->dirty_cursor);
 951                 return r;
 952         }
 953
 954         while (begin != end) {
 955                 /*
 956                  * We assume that unmapped blocks have their dirty bit
 957                  * cleared.
 958                  */
 959                 dirty_flag = dm_bitset_cursor_get_value(&cmd->dirty_cursor);
 960                 if (dirty_flag) {
 961                         DMERR("%s: cache block %llu is dirty", __func__,
 962                               (unsigned long long) from_cblock(begin));
 963                         dm_bitset_cursor_end(&cmd->dirty_cursor);
 964                         *result = false;
 965                         return 0;
 966                 }
 967
 968                 begin = to_cblock(from_cblock(begin) + 1);
 969                 if (begin == end)
 970                         break;
 971
 972                 r = dm_bitset_cursor_next(&cmd->dirty_cursor);
 973                 if (r) {
 974                         DMERR("%s: dm_bitset_cursor_next for dirty failed", __func__);
 975                         dm_bitset_cursor_end(&cmd->dirty_cursor);
 976                         return r;
 977                 }
 978         }
 979
 980         dm_bitset_cursor_end(&cmd->dirty_cursor);
 981
 982         return 0;
 983 }
 984
 985 static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
 986                                         dm_cblock_t begin, dm_cblock_t end,
 987                                         bool *result)
 988 {
 989         if (separate_dirty_bits(cmd))
 990                 return blocks_are_clean_separate_dirty(cmd, begin, end, result);
 991         else
 992                 return blocks_are_clean_combined_dirty(cmd, begin, end, result);
 993 }
 994
 995 static bool cmd_write_lock(struct dm_cache_metadata *cmd)
 996 {
 997         down_write(&cmd->root_lock);
 998         if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) {
 999                 up_write(&cmd->root_lock);
1000                 return false;
1001         }
1002         return true;
1003 }
1004
1005 #define WRITE_LOCK(cmd)                         \
1006         do {                                    \
1007                 if (!cmd_write_lock((cmd)))     \
1008                         return -EINVAL;         \
1009         } while(0)
1010
1011 #define WRITE_LOCK_VOID(cmd)                    \
1012         do {                                    \
1013                 if (!cmd_write_lock((cmd)))     \
1014                         return;                 \
1015         } while(0)
1016
1017 #define WRITE_UNLOCK(cmd) \
1018         up_write(&(cmd)->root_lock)
1019
1020 static bool cmd_read_lock(struct dm_cache_metadata *cmd)
1021 {
1022         down_read(&cmd->root_lock);
1023         if (cmd->fail_io) {
1024                 up_read(&cmd->root_lock);
1025                 return false;
1026         }
1027         return true;
1028 }
1029
1030 #define READ_LOCK(cmd)                          \
1031         do {                                    \
1032                 if (!cmd_read_lock((cmd)))      \
1033                         return -EINVAL;         \
1034         } while(0)
1035
1036 #define READ_LOCK_VOID(cmd)                     \
1037         do {                                    \
1038                 if (!cmd_read_lock((cmd)))      \
1039                         return;                 \
1040         } while(0)
1041
1042 #define READ_UNLOCK(cmd) \
1043         up_read(&(cmd)->root_lock)
1044
1045 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
1046 {
1047         int r;
1048         bool clean;
1049         __le64 null_mapping = pack_value(0, 0);
1050
1051         WRITE_LOCK(cmd);
1052         __dm_bless_for_disk(&null_mapping);
1053
1054         if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) {
1055                 r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean);
1056                 if (r) {
1057                         __dm_unbless_for_disk(&null_mapping);
1058                         goto out;
1059                 }
1060
1061                 if (!clean) {
1062                         DMERR("unable to shrink cache due to dirty blocks");
1063                         r = -EINVAL;
1064                         __dm_unbless_for_disk(&null_mapping);
1065                         goto out;
1066                 }
1067         }
1068
1069         r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
1070                             from_cblock(new_cache_size),
1071                             &null_mapping, &cmd->root);
1072         if (r)
1073                 goto out;
1074
1075         if (separate_dirty_bits(cmd)) {
1076                 r = dm_bitset_resize(&cmd->dirty_info, cmd->dirty_root,
1077                                      from_cblock(cmd->cache_blocks), from_cblock(new_cache_size),
1078                                      false, &cmd->dirty_root);
1079                 if (r)
1080                         goto out;
1081         }
1082
1083         cmd->cache_blocks = new_cache_size;
1084         cmd->changed = true;
1085
1086 out:
1087         WRITE_UNLOCK(cmd);
1088
1089         return r;
1090 }
1091
1092 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
1093                                    sector_t discard_block_size,
1094                                    dm_dblock_t new_nr_entries)
1095 {
1096         int r;
1097
1098         WRITE_LOCK(cmd);
1099         r = dm_bitset_resize(&cmd->discard_info,
1100                              cmd->discard_root,
1101                              from_dblock(cmd->discard_nr_blocks),
1102                              from_dblock(new_nr_entries),
1103                              false, &cmd->discard_root);
1104         if (!r) {
1105                 cmd->discard_block_size = discard_block_size;
1106                 cmd->discard_nr_blocks = new_nr_entries;
1107         }
1108
1109         cmd->changed = true;
1110         WRITE_UNLOCK(cmd);
1111
1112         return r;
1113 }
1114
1115 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
1116 {
1117         return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
1118                                  from_dblock(b), &cmd->discard_root);
1119 }
1120
1121 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
1122 {
1123         return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
1124                                    from_dblock(b), &cmd->discard_root);
1125 }
1126
1127 static int __discard(struct dm_cache_metadata *cmd,
1128                      dm_dblock_t dblock, bool discard)
1129 {
1130         int r;
1131
1132         r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
1133         if (r)
1134                 return r;
1135
1136         cmd->changed = true;
1137         return 0;
1138 }
1139
1140 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
1141                          dm_dblock_t dblock, bool discard)
1142 {
1143         int r;
1144
1145         WRITE_LOCK(cmd);
1146         r = __discard(cmd, dblock, discard);
1147         WRITE_UNLOCK(cmd);
1148
1149         return r;
1150 }
1151
1152 static int __load_discards(struct dm_cache_metadata *cmd,
1153                            load_discard_fn fn, void *context)
1154 {
1155         int r = 0;
1156         uint32_t b;
1157         struct dm_bitset_cursor c;
1158
1159         if (from_dblock(cmd->discard_nr_blocks) == 0)
1160                 /* nothing to do */
1161                 return 0;
1162
1163         if (cmd->clean_when_opened) {
1164                 r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root, &cmd->discard_root);
1165                 if (r)
1166                         return r;
1167
1168                 r = dm_bitset_cursor_begin(&cmd->discard_info, cmd->discard_root,
1169                                            from_dblock(cmd->discard_nr_blocks), &c);
1170                 if (r)
1171                         return r;
1172
1173                 for (b = 0; ; b++) {
1174                         r = fn(context, cmd->discard_block_size, to_dblock(b),
1175                                dm_bitset_cursor_get_value(&c));
1176                         if (r)
1177                                 break;
1178
1179                         if (b >= (from_dblock(cmd->discard_nr_blocks) - 1))
1180                                 break;
1181
1182                         r = dm_bitset_cursor_next(&c);
1183                         if (r)
1184                                 break;
1185                 }
1186
1187                 dm_bitset_cursor_end(&c);
1188
1189         } else {
1190                 for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1191                         r = fn(context, cmd->discard_block_size, to_dblock(b), false);
1192                         if (r)
1193                                 return r;
1194                 }
1195         }
1196
1197         return r;
1198 }
1199
1200 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
1201                            load_discard_fn fn, void *context)
1202 {
1203         int r;
1204
1205         READ_LOCK(cmd);
1206         r = __load_discards(cmd, fn, context);
1207         READ_UNLOCK(cmd);
1208
1209         return r;
1210 }
1211
1212 int dm_cache_size(struct dm_cache_metadata *cmd, dm_cblock_t *result)
1213 {
1214         READ_LOCK(cmd);
1215         *result = cmd->cache_blocks;
1216         READ_UNLOCK(cmd);
1217
1218         return 0;
1219 }
1220
1221 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1222 {
1223         int r;
1224         __le64 value = pack_value(0, 0);
1225
1226         __dm_bless_for_disk(&value);
1227         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1228                                &value, &cmd->root);
1229         if (r)
1230                 return r;
1231
1232         cmd->changed = true;
1233         return 0;
1234 }
1235
1236 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1237 {
1238         int r;
1239
1240         WRITE_LOCK(cmd);
1241         r = __remove(cmd, cblock);
1242         WRITE_UNLOCK(cmd);
1243
1244         return r;
1245 }
1246
1247 static int __insert(struct dm_cache_metadata *cmd,
1248                     dm_cblock_t cblock, dm_oblock_t oblock)
1249 {
1250         int r;
1251         __le64 value = pack_value(oblock, M_VALID);
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 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
1264                             dm_cblock_t cblock, dm_oblock_t oblock)
1265 {
1266         int r;
1267
1268         WRITE_LOCK(cmd);
1269         r = __insert(cmd, cblock, oblock);
1270         WRITE_UNLOCK(cmd);
1271
1272         return r;
1273 }
1274
1275 struct thunk {
1276         load_mapping_fn fn;
1277         void *context;
1278
1279         struct dm_cache_metadata *cmd;
1280         bool respect_dirty_flags;
1281         bool hints_valid;
1282 };
1283
1284 static bool policy_unchanged(struct dm_cache_metadata *cmd,
1285                              struct dm_cache_policy *policy)
1286 {
1287         const char *policy_name = dm_cache_policy_get_name(policy);
1288         const unsigned *policy_version = dm_cache_policy_get_version(policy);
1289         size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
1290
1291         /*
1292          * Ensure policy names match.
1293          */
1294         if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
1295                 return false;
1296
1297         /*
1298          * Ensure policy major versions match.
1299          */
1300         if (cmd->policy_version[0] != policy_version[0])
1301                 return false;
1302
1303         /*
1304          * Ensure policy hint sizes match.
1305          */
1306         if (cmd->policy_hint_size != policy_hint_size)
1307                 return false;
1308
1309         return true;
1310 }
1311
1312 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1313 {
1314         return cmd->hint_root && cmd->policy_hint_size;
1315 }
1316
1317 static bool hints_array_available(struct dm_cache_metadata *cmd,
1318                                   struct dm_cache_policy *policy)
1319 {
1320         return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1321                 hints_array_initialized(cmd);
1322 }
1323
1324 static int __load_mapping_v1(struct dm_cache_metadata *cmd,
1325                              uint64_t cb, bool hints_valid,
1326                              struct dm_array_cursor *mapping_cursor,
1327                              struct dm_array_cursor *hint_cursor,
1328                              load_mapping_fn fn, void *context)
1329 {
1330         int r = 0;
1331
1332         __le64 mapping;
1333         __le32 hint = 0;
1334
1335         __le64 *mapping_value_le;
1336         __le32 *hint_value_le;
1337
1338         dm_oblock_t oblock;
1339         unsigned flags;
1340         bool dirty = true;
1341
1342         dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1343         memcpy(&mapping, mapping_value_le, sizeof(mapping));
1344         unpack_value(mapping, &oblock, &flags);
1345
1346         if (flags & M_VALID) {
1347                 if (hints_valid) {
1348                         dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1349                         memcpy(&hint, hint_value_le, sizeof(hint));
1350                 }
1351                 if (cmd->clean_when_opened)
1352                         dirty = flags & M_DIRTY;
1353
1354                 r = fn(context, oblock, to_cblock(cb), dirty,
1355                        le32_to_cpu(hint), hints_valid);
1356                 if (r) {
1357                         DMERR("policy couldn't load cache block %llu",
1358                               (unsigned long long) from_cblock(to_cblock(cb)));
1359                 }
1360         }
1361
1362         return r;
1363 }
1364
1365 static int __load_mapping_v2(struct dm_cache_metadata *cmd,
1366                              uint64_t cb, bool hints_valid,
1367                              struct dm_array_cursor *mapping_cursor,
1368                              struct dm_array_cursor *hint_cursor,
1369                              struct dm_bitset_cursor *dirty_cursor,
1370                              load_mapping_fn fn, void *context)
1371 {
1372         int r = 0;
1373
1374         __le64 mapping;
1375         __le32 hint = 0;
1376
1377         __le64 *mapping_value_le;
1378         __le32 *hint_value_le;
1379
1380         dm_oblock_t oblock;
1381         unsigned flags;
1382         bool dirty = true;
1383
1384         dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1385         memcpy(&mapping, mapping_value_le, sizeof(mapping));
1386         unpack_value(mapping, &oblock, &flags);
1387
1388         if (flags & M_VALID) {
1389                 if (hints_valid) {
1390                         dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1391                         memcpy(&hint, hint_value_le, sizeof(hint));
1392                 }
1393                 if (cmd->clean_when_opened)
1394                         dirty = dm_bitset_cursor_get_value(dirty_cursor);
1395
1396                 r = fn(context, oblock, to_cblock(cb), dirty,
1397                        le32_to_cpu(hint), hints_valid);
1398                 if (r) {
1399                         DMERR("policy couldn't load cache block %llu",
1400                               (unsigned long long) from_cblock(to_cblock(cb)));
1401                 }
1402         }
1403
1404         return r;
1405 }
1406
1407 static int __load_mappings(struct dm_cache_metadata *cmd,
1408                            struct dm_cache_policy *policy,
1409                            load_mapping_fn fn, void *context)
1410 {
1411         int r;
1412         uint64_t cb;
1413
1414         bool hints_valid = hints_array_available(cmd, policy);
1415
1416         if (from_cblock(cmd->cache_blocks) == 0)
1417                 /* Nothing to do */
1418                 return 0;
1419
1420         r = dm_array_cursor_begin(&cmd->info, cmd->root, &cmd->mapping_cursor);
1421         if (r)
1422                 return r;
1423
1424         if (hints_valid) {
1425                 r = dm_array_cursor_begin(&cmd->hint_info, cmd->hint_root, &cmd->hint_cursor);
1426                 if (r) {
1427                         dm_array_cursor_end(&cmd->mapping_cursor);
1428                         return r;
1429                 }
1430         }
1431
1432         if (separate_dirty_bits(cmd)) {
1433                 r = dm_bitset_cursor_begin(&cmd->dirty_info, cmd->dirty_root,
1434                                            from_cblock(cmd->cache_blocks),
1435                                            &cmd->dirty_cursor);
1436                 if (r) {
1437                         dm_array_cursor_end(&cmd->hint_cursor);
1438                         dm_array_cursor_end(&cmd->mapping_cursor);
1439                         return r;
1440                 }
1441         }
1442
1443         for (cb = 0; ; cb++) {
1444                 if (separate_dirty_bits(cmd))
1445                         r = __load_mapping_v2(cmd, cb, hints_valid,
1446                                               &cmd->mapping_cursor,
1447                                               &cmd->hint_cursor,
1448                                               &cmd->dirty_cursor,
1449                                               fn, context);
1450                 else
1451                         r = __load_mapping_v1(cmd, cb, hints_valid,
1452                                               &cmd->mapping_cursor, &cmd->hint_cursor,
1453                                               fn, context);
1454                 if (r)
1455                         goto out;
1456
1457                 /*
1458                  * We need to break out before we move the cursors.
1459                  */
1460                 if (cb >= (from_cblock(cmd->cache_blocks) - 1))
1461                         break;
1462
1463                 r = dm_array_cursor_next(&cmd->mapping_cursor);
1464                 if (r) {
1465                         DMERR("dm_array_cursor_next for mapping failed");
1466                         goto out;
1467                 }
1468
1469                 if (hints_valid) {
1470                         r = dm_array_cursor_next(&cmd->hint_cursor);
1471                         if (r) {
1472                                 dm_array_cursor_end(&cmd->hint_cursor);
1473                                 hints_valid = false;
1474                         }
1475                 }
1476
1477                 if (separate_dirty_bits(cmd)) {
1478                         r = dm_bitset_cursor_next(&cmd->dirty_cursor);
1479                         if (r) {
1480                                 DMERR("dm_bitset_cursor_next for dirty failed");
1481                                 goto out;
1482                         }
1483                 }
1484         }
1485 out:
1486         dm_array_cursor_end(&cmd->mapping_cursor);
1487         if (hints_valid)
1488                 dm_array_cursor_end(&cmd->hint_cursor);
1489
1490         if (separate_dirty_bits(cmd))
1491                 dm_bitset_cursor_end(&cmd->dirty_cursor);
1492
1493         return r;
1494 }
1495
1496 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1497                            struct dm_cache_policy *policy,
1498                            load_mapping_fn fn, void *context)
1499 {
1500         int r;
1501
1502         READ_LOCK(cmd);
1503         r = __load_mappings(cmd, policy, fn, context);
1504         READ_UNLOCK(cmd);
1505
1506         return r;
1507 }
1508
1509 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
1510 {
1511         int r = 0;
1512         __le64 value;
1513         dm_oblock_t oblock;
1514         unsigned flags;
1515
1516         memcpy(&value, leaf, sizeof(value));
1517         unpack_value(value, &oblock, &flags);
1518
1519         return r;
1520 }
1521
1522 static int __dump_mappings(struct dm_cache_metadata *cmd)
1523 {
1524         return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1525 }
1526
1527 void dm_cache_dump(struct dm_cache_metadata *cmd)
1528 {
1529         READ_LOCK_VOID(cmd);
1530         __dump_mappings(cmd);
1531         READ_UNLOCK(cmd);
1532 }
1533
1534 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1535 {
1536         int r;
1537
1538         READ_LOCK(cmd);
1539         r = cmd->changed;
1540         READ_UNLOCK(cmd);
1541
1542         return r;
1543 }
1544
1545 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1546 {
1547         int r;
1548         unsigned flags;
1549         dm_oblock_t oblock;
1550         __le64 value;
1551
1552         r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1553         if (r)
1554                 return r;
1555
1556         unpack_value(value, &oblock, &flags);
1557
1558         if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1559                 /* nothing to be done */
1560                 return 0;
1561
1562         value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1563         __dm_bless_for_disk(&value);
1564
1565         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1566                                &value, &cmd->root);
1567         if (r)
1568                 return r;
1569
1570         cmd->changed = true;
1571         return 0;
1572
1573 }
1574
1575 static int __set_dirty_bits_v1(struct dm_cache_metadata *cmd, unsigned nr_bits, unsigned long *bits)
1576 {
1577         int r;
1578         unsigned i;
1579         for (i = 0; i < nr_bits; i++) {
1580                 r = __dirty(cmd, to_cblock(i), test_bit(i, bits));
1581                 if (r)
1582                         return r;
1583         }
1584
1585         return 0;
1586 }
1587
1588 static int is_dirty_callback(uint32_t index, bool *value, void *context)
1589 {
1590         unsigned long *bits = context;
1591         *value = test_bit(index, bits);
1592         return 0;
1593 }
1594
1595 static int __set_dirty_bits_v2(struct dm_cache_metadata *cmd, unsigned nr_bits, unsigned long *bits)
1596 {
1597         int r = 0;
1598
1599         /* nr_bits is really just a sanity check */
1600         if (nr_bits != from_cblock(cmd->cache_blocks)) {
1601                 DMERR("dirty bitset is wrong size");
1602                 return -EINVAL;
1603         }
1604
1605         r = dm_bitset_del(&cmd->dirty_info, cmd->dirty_root);
1606         if (r)
1607                 return r;
1608
1609         cmd->changed = true;
1610         return dm_bitset_new(&cmd->dirty_info, &cmd->dirty_root, nr_bits, is_dirty_callback, bits);
1611 }
1612
1613 int dm_cache_set_dirty_bits(struct dm_cache_metadata *cmd,
1614                             unsigned nr_bits,
1615                             unsigned long *bits)
1616 {
1617         int r;
1618
1619         WRITE_LOCK(cmd);
1620         if (separate_dirty_bits(cmd))
1621                 r = __set_dirty_bits_v2(cmd, nr_bits, bits);
1622         else
1623                 r = __set_dirty_bits_v1(cmd, nr_bits, bits);
1624         WRITE_UNLOCK(cmd);
1625
1626         return r;
1627 }
1628
1629 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1630                                  struct dm_cache_statistics *stats)
1631 {
1632         READ_LOCK_VOID(cmd);
1633         *stats = cmd->stats;
1634         READ_UNLOCK(cmd);
1635 }
1636
1637 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1638                                  struct dm_cache_statistics *stats)
1639 {
1640         WRITE_LOCK_VOID(cmd);
1641         cmd->stats = *stats;
1642         WRITE_UNLOCK(cmd);
1643 }
1644
1645 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1646 {
1647         int r = -EINVAL;
1648         flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1649                                  clear_clean_shutdown);
1650
1651         WRITE_LOCK(cmd);
1652         if (cmd->fail_io)
1653                 goto out;
1654
1655         r = __commit_transaction(cmd, mutator);
1656         if (r)
1657                 goto out;
1658
1659         r = __begin_transaction(cmd);
1660 out:
1661         WRITE_UNLOCK(cmd);
1662         return r;
1663 }
1664
1665 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1666                                            dm_block_t *result)
1667 {
1668         int r = -EINVAL;
1669
1670         READ_LOCK(cmd);
1671         if (!cmd->fail_io)
1672                 r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1673         READ_UNLOCK(cmd);
1674
1675         return r;
1676 }
1677
1678 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1679                                    dm_block_t *result)
1680 {
1681         int r = -EINVAL;
1682
1683         READ_LOCK(cmd);
1684         if (!cmd->fail_io)
1685                 r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1686         READ_UNLOCK(cmd);
1687
1688         return r;
1689 }
1690
1691 /*----------------------------------------------------------------*/
1692
1693 static int get_hint(uint32_t index, void *value_le, void *context)
1694 {
1695         uint32_t value;
1696         struct dm_cache_policy *policy = context;
1697
1698         value = policy_get_hint(policy, to_cblock(index));
1699         *((__le32 *) value_le) = cpu_to_le32(value);
1700
1701         return 0;
1702 }
1703
1704 /*
1705  * It's quicker to always delete the hint array, and recreate with
1706  * dm_array_new().
1707  */
1708 static int write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1709 {
1710         int r;
1711         size_t hint_size;
1712         const char *policy_name = dm_cache_policy_get_name(policy);
1713         const unsigned *policy_version = dm_cache_policy_get_version(policy);
1714
1715         if (!policy_name[0] ||
1716             (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1717                 return -EINVAL;
1718
1719         strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1720         memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1721
1722         hint_size = dm_cache_policy_get_hint_size(policy);
1723         if (!hint_size)
1724                 return 0; /* short-circuit hints initialization */
1725         cmd->policy_hint_size = hint_size;
1726
1727         if (cmd->hint_root) {
1728                 r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1729                 if (r)
1730                         return r;
1731         }
1732
1733         return dm_array_new(&cmd->hint_info, &cmd->hint_root,
1734                             from_cblock(cmd->cache_blocks),
1735                             get_hint, policy);
1736 }
1737
1738 int dm_cache_write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1739 {
1740         int r;
1741
1742         WRITE_LOCK(cmd);
1743         r = write_hints(cmd, policy);
1744         WRITE_UNLOCK(cmd);
1745
1746         return r;
1747 }
1748
1749 int dm_cache_metadata_all_clean(struct dm_cache_metadata *cmd, bool *result)
1750 {
1751         int r;
1752
1753         READ_LOCK(cmd);
1754         r = blocks_are_unmapped_or_clean(cmd, 0, cmd->cache_blocks, result);
1755         READ_UNLOCK(cmd);
1756
1757         return r;
1758 }
1759
1760 void dm_cache_metadata_set_read_only(struct dm_cache_metadata *cmd)
1761 {
1762         WRITE_LOCK_VOID(cmd);
1763         dm_bm_set_read_only(cmd->bm);
1764         WRITE_UNLOCK(cmd);
1765 }
1766
1767 void dm_cache_metadata_set_read_write(struct dm_cache_metadata *cmd)
1768 {
1769         WRITE_LOCK_VOID(cmd);
1770         dm_bm_set_read_write(cmd->bm);
1771         WRITE_UNLOCK(cmd);
1772 }
1773
1774 int dm_cache_metadata_set_needs_check(struct dm_cache_metadata *cmd)
1775 {
1776         int r;
1777         struct dm_block *sblock;
1778         struct cache_disk_superblock *disk_super;
1779
1780         WRITE_LOCK(cmd);
1781         set_bit(NEEDS_CHECK, &cmd->flags);
1782
1783         r = superblock_lock(cmd, &sblock);
1784         if (r) {
1785                 DMERR("couldn't read superblock");
1786                 goto out;
1787         }
1788
1789         disk_super = dm_block_data(sblock);
1790         disk_super->flags = cpu_to_le32(cmd->flags);
1791
1792         dm_bm_unlock(sblock);
1793
1794 out:
1795         WRITE_UNLOCK(cmd);
1796         return r;
1797 }
1798
1799 int dm_cache_metadata_needs_check(struct dm_cache_metadata *cmd, bool *result)
1800 {
1801         READ_LOCK(cmd);
1802         *result = !!test_bit(NEEDS_CHECK, &cmd->flags);
1803         READ_UNLOCK(cmd);
1804
1805         return 0;
1806 }
1807
1808 int dm_cache_metadata_abort(struct dm_cache_metadata *cmd)
1809 {
1810         int r;
1811
1812         WRITE_LOCK(cmd);
1813         __destroy_persistent_data_objects(cmd);
1814         r = __create_persistent_data_objects(cmd, false);
1815         if (r)
1816                 cmd->fail_io = true;
1817         WRITE_UNLOCK(cmd);
1818
1819         return r;
1820 }