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 1
  29
  30 #define CACHE_METADATA_CACHE_SIZE 64
  31
  32 /*
  33  *  3 for btree insert +
  34  *  2 for btree lookup used within space map
  35  */
  36 #define CACHE_MAX_CONCURRENT_LOCKS 5
  37 #define SPACE_MAP_ROOT_SIZE 128
  38
  39 enum superblock_flag_bits {
  40         /* for spotting crashes that would invalidate the dirty bitset */
  41         CLEAN_SHUTDOWN,
  42         /* metadata must be checked using the tools */
  43         NEEDS_CHECK,
  44 };
  45
  46 /*
  47  * Each mapping from cache block -> origin block carries a set of flags.
  48  */
  49 enum mapping_bits {
  50         /*
  51          * A valid mapping.  Because we're using an array we clear this
  52          * flag for an non existant mapping.
  53          */
  54         M_VALID = 1,
  55
  56         /*
  57          * The data on the cache is different from that on the origin.
  58          */
  59         M_DIRTY = 2
  60 };
  61
  62 struct cache_disk_superblock {
  63         __le32 csum;
  64         __le32 flags;
  65         __le64 blocknr;
  66
  67         __u8 uuid[16];
  68         __le64 magic;
  69         __le32 version;
  70
  71         __u8 policy_name[CACHE_POLICY_NAME_SIZE];
  72         __le32 policy_hint_size;
  73
  74         __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
  75         __le64 mapping_root;
  76         __le64 hint_root;
  77
  78         __le64 discard_root;
  79         __le64 discard_block_size;
  80         __le64 discard_nr_blocks;
  81
  82         __le32 data_block_size;
  83         __le32 metadata_block_size;
  84         __le32 cache_blocks;
  85
  86         __le32 compat_flags;
  87         __le32 compat_ro_flags;
  88         __le32 incompat_flags;
  89
  90         __le32 read_hits;
  91         __le32 read_misses;
  92         __le32 write_hits;
  93         __le32 write_misses;
  94
  95         __le32 policy_version[CACHE_POLICY_VERSION_SIZE];
  96 } __packed;
  97
  98 struct dm_cache_metadata {
  99         atomic_t ref_count;
 100         struct list_head list;
 101
 102         struct block_device *bdev;
 103         struct dm_block_manager *bm;
 104         struct dm_space_map *metadata_sm;
 105         struct dm_transaction_manager *tm;
 106
 107         struct dm_array_info info;
 108         struct dm_array_info hint_info;
 109         struct dm_disk_bitset discard_info;
 110
 111         struct rw_semaphore root_lock;
 112         unsigned long flags;
 113         dm_block_t root;
 114         dm_block_t hint_root;
 115         dm_block_t discard_root;
 116
 117         sector_t discard_block_size;
 118         dm_dblock_t discard_nr_blocks;
 119
 120         sector_t data_block_size;
 121         dm_cblock_t cache_blocks;
 122         bool changed:1;
 123         bool clean_when_opened:1;
 124
 125         char policy_name[CACHE_POLICY_NAME_SIZE];
 126         unsigned policy_version[CACHE_POLICY_VERSION_SIZE];
 127         size_t policy_hint_size;
 128         struct dm_cache_statistics stats;
 129
 130         /*
 131          * Reading the space map root can fail, so we read it into this
 132          * buffer before the superblock is locked and updated.
 133          */
 134         __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
 135
 136         /*
 137          * Set if a transaction has to be aborted but the attempt to roll
 138          * back to the previous (good) transaction failed.  The only
 139          * metadata operation permissible in this state is the closing of
 140          * the device.
 141          */
 142         bool fail_io:1;
 143 };
 144
 145 /*-------------------------------------------------------------------
 146  * superblock validator
 147  *-----------------------------------------------------------------*/
 148
 149 #define SUPERBLOCK_CSUM_XOR 9031977
 150
 151 static void sb_prepare_for_write(struct dm_block_validator *v,
 152                                  struct dm_block *b,
 153                                  size_t sb_block_size)
 154 {
 155         struct cache_disk_superblock *disk_super = dm_block_data(b);
 156
 157         disk_super->blocknr = cpu_to_le64(dm_block_location(b));
 158         disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
 159                                                       sb_block_size - sizeof(__le32),
 160                                                       SUPERBLOCK_CSUM_XOR));
 161 }
 162
 163 static int check_metadata_version(struct cache_disk_superblock *disk_super)
 164 {
 165         uint32_t metadata_version = le32_to_cpu(disk_super->version);
 166         if (metadata_version < MIN_CACHE_VERSION || metadata_version > MAX_CACHE_VERSION) {
 167                 DMERR("Cache metadata version %u found, but only versions between %u and %u supported.",
 168                       metadata_version, MIN_CACHE_VERSION, MAX_CACHE_VERSION);
 169                 return -EINVAL;
 170         }
 171
 172         return 0;
 173 }
 174
 175 static int sb_check(struct dm_block_validator *v,
 176                     struct dm_block *b,
 177                     size_t sb_block_size)
 178 {
 179         struct cache_disk_superblock *disk_super = dm_block_data(b);
 180         __le32 csum_le;
 181
 182         if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
 183                 DMERR("sb_check failed: blocknr %llu: wanted %llu",
 184                       le64_to_cpu(disk_super->blocknr),
 185                       (unsigned long long)dm_block_location(b));
 186                 return -ENOTBLK;
 187         }
 188
 189         if (le64_to_cpu(disk_super->magic) != CACHE_SUPERBLOCK_MAGIC) {
 190                 DMERR("sb_check failed: magic %llu: wanted %llu",
 191                       le64_to_cpu(disk_super->magic),
 192                       (unsigned long long)CACHE_SUPERBLOCK_MAGIC);
 193                 return -EILSEQ;
 194         }
 195
 196         csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
 197                                              sb_block_size - sizeof(__le32),
 198                                              SUPERBLOCK_CSUM_XOR));
 199         if (csum_le != disk_super->csum) {
 200                 DMERR("sb_check failed: csum %u: wanted %u",
 201                       le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
 202                 return -EILSEQ;
 203         }
 204
 205         return check_metadata_version(disk_super);
 206 }
 207
 208 static struct dm_block_validator sb_validator = {
 209         .name = "superblock",
 210         .prepare_for_write = sb_prepare_for_write,
 211         .check = sb_check
 212 };
 213
 214 /*----------------------------------------------------------------*/
 215
 216 static int superblock_read_lock(struct dm_cache_metadata *cmd,
 217                                 struct dm_block **sblock)
 218 {
 219         return dm_bm_read_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 220                                &sb_validator, sblock);
 221 }
 222
 223 static int superblock_lock_zero(struct dm_cache_metadata *cmd,
 224                                 struct dm_block **sblock)
 225 {
 226         return dm_bm_write_lock_zero(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 227                                      &sb_validator, sblock);
 228 }
 229
 230 static int superblock_lock(struct dm_cache_metadata *cmd,
 231                            struct dm_block **sblock)
 232 {
 233         return dm_bm_write_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 234                                 &sb_validator, sblock);
 235 }
 236
 237 /*----------------------------------------------------------------*/
 238
 239 static int __superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
 240 {
 241         int r;
 242         unsigned i;
 243         struct dm_block *b;
 244         __le64 *data_le, zero = cpu_to_le64(0);
 245         unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
 246
 247         /*
 248          * We can't use a validator here - it may be all zeroes.
 249          */
 250         r = dm_bm_read_lock(bm, CACHE_SUPERBLOCK_LOCATION, NULL, &b);
 251         if (r)
 252                 return r;
 253
 254         data_le = dm_block_data(b);
 255         *result = true;
 256         for (i = 0; i < sb_block_size; i++) {
 257                 if (data_le[i] != zero) {
 258                         *result = false;
 259                         break;
 260                 }
 261         }
 262
 263         dm_bm_unlock(b);
 264
 265         return 0;
 266 }
 267
 268 static void __setup_mapping_info(struct dm_cache_metadata *cmd)
 269 {
 270         struct dm_btree_value_type vt;
 271
 272         vt.context = NULL;
 273         vt.size = sizeof(__le64);
 274         vt.inc = NULL;
 275         vt.dec = NULL;
 276         vt.equal = NULL;
 277         dm_array_info_init(&cmd->info, cmd->tm, &vt);
 278
 279         if (cmd->policy_hint_size) {
 280                 vt.size = sizeof(__le32);
 281                 dm_array_info_init(&cmd->hint_info, cmd->tm, &vt);
 282         }
 283 }
 284
 285 static int __save_sm_root(struct dm_cache_metadata *cmd)
 286 {
 287         int r;
 288         size_t metadata_len;
 289
 290         r = dm_sm_root_size(cmd->metadata_sm, &metadata_len);
 291         if (r < 0)
 292                 return r;
 293
 294         return dm_sm_copy_root(cmd->metadata_sm, &cmd->metadata_space_map_root,
 295                                metadata_len);
 296 }
 297
 298 static void __copy_sm_root(struct dm_cache_metadata *cmd,
 299                            struct cache_disk_superblock *disk_super)
 300 {
 301         memcpy(&disk_super->metadata_space_map_root,
 302                &cmd->metadata_space_map_root,
 303                sizeof(cmd->metadata_space_map_root));
 304 }
 305
 306 static int __write_initial_superblock(struct dm_cache_metadata *cmd)
 307 {
 308         int r;
 309         struct dm_block *sblock;
 310         struct cache_disk_superblock *disk_super;
 311         sector_t bdev_size = i_size_read(cmd->bdev->bd_inode) >> SECTOR_SHIFT;
 312
 313         /* FIXME: see if we can lose the max sectors limit */
 314         if (bdev_size > DM_CACHE_METADATA_MAX_SECTORS)
 315                 bdev_size = DM_CACHE_METADATA_MAX_SECTORS;
 316
 317         r = dm_tm_pre_commit(cmd->tm);
 318         if (r < 0)
 319                 return r;
 320
 321         /*
 322          * dm_sm_copy_root() can fail.  So we need to do it before we start
 323          * updating the superblock.
 324          */
 325         r = __save_sm_root(cmd);
 326         if (r)
 327                 return r;
 328
 329         r = superblock_lock_zero(cmd, &sblock);
 330         if (r)
 331                 return r;
 332
 333         disk_super = dm_block_data(sblock);
 334         disk_super->flags = 0;
 335         memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
 336         disk_super->magic = cpu_to_le64(CACHE_SUPERBLOCK_MAGIC);
 337         disk_super->version = cpu_to_le32(MAX_CACHE_VERSION);
 338         memset(disk_super->policy_name, 0, sizeof(disk_super->policy_name));
 339         memset(disk_super->policy_version, 0, sizeof(disk_super->policy_version));
 340         disk_super->policy_hint_size = cpu_to_le32(0);
 341
 342         __copy_sm_root(cmd, disk_super);
 343
 344         disk_super->mapping_root = cpu_to_le64(cmd->root);
 345         disk_super->hint_root = cpu_to_le64(cmd->hint_root);
 346         disk_super->discard_root = cpu_to_le64(cmd->discard_root);
 347         disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
 348         disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
 349         disk_super->metadata_block_size = cpu_to_le32(DM_CACHE_METADATA_BLOCK_SIZE);
 350         disk_super->data_block_size = cpu_to_le32(cmd->data_block_size);
 351         disk_super->cache_blocks = cpu_to_le32(0);
 352
 353         disk_super->read_hits = cpu_to_le32(0);
 354         disk_super->read_misses = cpu_to_le32(0);
 355         disk_super->write_hits = cpu_to_le32(0);
 356         disk_super->write_misses = cpu_to_le32(0);
 357
 358         return dm_tm_commit(cmd->tm, sblock);
 359 }
 360
 361 static int __format_metadata(struct dm_cache_metadata *cmd)
 362 {
 363         int r;
 364
 365         r = dm_tm_create_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 366                                  &cmd->tm, &cmd->metadata_sm);
 367         if (r < 0) {
 368                 DMERR("tm_create_with_sm failed");
 369                 return r;
 370         }
 371
 372         __setup_mapping_info(cmd);
 373
 374         r = dm_array_empty(&cmd->info, &cmd->root);
 375         if (r < 0)
 376                 goto bad;
 377
 378         dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
 379
 380         r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root);
 381         if (r < 0)
 382                 goto bad;
 383
 384         cmd->discard_block_size = 0;
 385         cmd->discard_nr_blocks = 0;
 386
 387         r = __write_initial_superblock(cmd);
 388         if (r)
 389                 goto bad;
 390
 391         cmd->clean_when_opened = true;
 392         return 0;
 393
 394 bad:
 395         dm_tm_destroy(cmd->tm);
 396         dm_sm_destroy(cmd->metadata_sm);
 397
 398         return r;
 399 }
 400
 401 static int __check_incompat_features(struct cache_disk_superblock *disk_super,
 402                                      struct dm_cache_metadata *cmd)
 403 {
 404         uint32_t features;
 405
 406         features = le32_to_cpu(disk_super->incompat_flags) & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
 407         if (features) {
 408                 DMERR("could not access metadata due to unsupported optional features (%lx).",
 409                       (unsigned long)features);
 410                 return -EINVAL;
 411         }
 412
 413         /*
 414          * Check for read-only metadata to skip the following RDWR checks.
 415          */
 416         if (get_disk_ro(cmd->bdev->bd_disk))
 417                 return 0;
 418
 419         features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP;
 420         if (features) {
 421                 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
 422                       (unsigned long)features);
 423                 return -EINVAL;
 424         }
 425
 426         return 0;
 427 }
 428
 429 static int __open_metadata(struct dm_cache_metadata *cmd)
 430 {
 431         int r;
 432         struct dm_block *sblock;
 433         struct cache_disk_superblock *disk_super;
 434         unsigned long sb_flags;
 435
 436         r = superblock_read_lock(cmd, &sblock);
 437         if (r < 0) {
 438                 DMERR("couldn't read lock superblock");
 439                 return r;
 440         }
 441
 442         disk_super = dm_block_data(sblock);
 443
 444         /* Verify the data block size hasn't changed */
 445         if (le32_to_cpu(disk_super->data_block_size) != cmd->data_block_size) {
 446                 DMERR("changing the data block size (from %u to %llu) is not supported",
 447                       le32_to_cpu(disk_super->data_block_size),
 448                       (unsigned long long)cmd->data_block_size);
 449                 r = -EINVAL;
 450                 goto bad;
 451         }
 452
 453         r = __check_incompat_features(disk_super, cmd);
 454         if (r < 0)
 455                 goto bad;
 456
 457         r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 458                                disk_super->metadata_space_map_root,
 459                                sizeof(disk_super->metadata_space_map_root),
 460                                &cmd->tm, &cmd->metadata_sm);
 461         if (r < 0) {
 462                 DMERR("tm_open_with_sm failed");
 463                 goto bad;
 464         }
 465
 466         __setup_mapping_info(cmd);
 467         dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
 468         sb_flags = le32_to_cpu(disk_super->flags);
 469         cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags);
 470         dm_bm_unlock(sblock);
 471
 472         return 0;
 473
 474 bad:
 475         dm_bm_unlock(sblock);
 476         return r;
 477 }
 478
 479 static int __open_or_format_metadata(struct dm_cache_metadata *cmd,
 480                                      bool format_device)
 481 {
 482         int r;
 483         bool unformatted = false;
 484
 485         r = __superblock_all_zeroes(cmd->bm, &unformatted);
 486         if (r)
 487                 return r;
 488
 489         if (unformatted)
 490                 return format_device ? __format_metadata(cmd) : -EPERM;
 491
 492         return __open_metadata(cmd);
 493 }
 494
 495 static int __create_persistent_data_objects(struct dm_cache_metadata *cmd,
 496                                             bool may_format_device)
 497 {
 498         int r;
 499         cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
 500                                           CACHE_METADATA_CACHE_SIZE,
 501                                           CACHE_MAX_CONCURRENT_LOCKS);
 502         if (IS_ERR(cmd->bm)) {
 503                 DMERR("could not create block manager");
 504                 return PTR_ERR(cmd->bm);
 505         }
 506
 507         r = __open_or_format_metadata(cmd, may_format_device);
 508         if (r)
 509                 dm_block_manager_destroy(cmd->bm);
 510
 511         return r;
 512 }
 513
 514 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd)
 515 {
 516         dm_sm_destroy(cmd->metadata_sm);
 517         dm_tm_destroy(cmd->tm);
 518         dm_block_manager_destroy(cmd->bm);
 519 }
 520
 521 typedef unsigned long (*flags_mutator)(unsigned long);
 522
 523 static void update_flags(struct cache_disk_superblock *disk_super,
 524                          flags_mutator mutator)
 525 {
 526         uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
 527         disk_super->flags = cpu_to_le32(sb_flags);
 528 }
 529
 530 static unsigned long set_clean_shutdown(unsigned long flags)
 531 {
 532         set_bit(CLEAN_SHUTDOWN, &flags);
 533         return flags;
 534 }
 535
 536 static unsigned long clear_clean_shutdown(unsigned long flags)
 537 {
 538         clear_bit(CLEAN_SHUTDOWN, &flags);
 539         return flags;
 540 }
 541
 542 static void read_superblock_fields(struct dm_cache_metadata *cmd,
 543                                    struct cache_disk_superblock *disk_super)
 544 {
 545         cmd->flags = le32_to_cpu(disk_super->flags);
 546         cmd->root = le64_to_cpu(disk_super->mapping_root);
 547         cmd->hint_root = le64_to_cpu(disk_super->hint_root);
 548         cmd->discard_root = le64_to_cpu(disk_super->discard_root);
 549         cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
 550         cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
 551         cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
 552         cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
 553         strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
 554         cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
 555         cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
 556         cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
 557         cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
 558
 559         cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
 560         cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
 561         cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
 562         cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
 563
 564         cmd->changed = false;
 565 }
 566
 567 /*
 568  * The mutator updates the superblock flags.
 569  */
 570 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
 571                                      flags_mutator mutator)
 572 {
 573         int r;
 574         struct cache_disk_superblock *disk_super;
 575         struct dm_block *sblock;
 576
 577         r = superblock_lock(cmd, &sblock);
 578         if (r)
 579                 return r;
 580
 581         disk_super = dm_block_data(sblock);
 582         update_flags(disk_super, mutator);
 583         read_superblock_fields(cmd, disk_super);
 584         dm_bm_unlock(sblock);
 585
 586         return dm_bm_flush(cmd->bm);
 587 }
 588
 589 static int __begin_transaction(struct dm_cache_metadata *cmd)
 590 {
 591         int r;
 592         struct cache_disk_superblock *disk_super;
 593         struct dm_block *sblock;
 594
 595         /*
 596          * We re-read the superblock every time.  Shouldn't need to do this
 597          * really.
 598          */
 599         r = superblock_read_lock(cmd, &sblock);
 600         if (r)
 601                 return r;
 602
 603         disk_super = dm_block_data(sblock);
 604         read_superblock_fields(cmd, disk_super);
 605         dm_bm_unlock(sblock);
 606
 607         return 0;
 608 }
 609
 610 static int __commit_transaction(struct dm_cache_metadata *cmd,
 611                                 flags_mutator mutator)
 612 {
 613         int r;
 614         struct cache_disk_superblock *disk_super;
 615         struct dm_block *sblock;
 616
 617         /*
 618          * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
 619          */
 620         BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
 621
 622         r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
 623                             &cmd->discard_root);
 624         if (r)
 625                 return r;
 626
 627         r = dm_tm_pre_commit(cmd->tm);
 628         if (r < 0)
 629                 return r;
 630
 631         r = __save_sm_root(cmd);
 632         if (r)
 633                 return r;
 634
 635         r = superblock_lock(cmd, &sblock);
 636         if (r)
 637                 return r;
 638
 639         disk_super = dm_block_data(sblock);
 640
 641         disk_super->flags = cpu_to_le32(cmd->flags);
 642         if (mutator)
 643                 update_flags(disk_super, mutator);
 644
 645         disk_super->mapping_root = cpu_to_le64(cmd->root);
 646         disk_super->hint_root = cpu_to_le64(cmd->hint_root);
 647         disk_super->discard_root = cpu_to_le64(cmd->discard_root);
 648         disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
 649         disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
 650         disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
 651         strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
 652         disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
 653         disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
 654         disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
 655         disk_super->policy_hint_size = cpu_to_le32(cmd->policy_hint_size);
 656
 657         disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
 658         disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
 659         disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
 660         disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
 661         __copy_sm_root(cmd, disk_super);
 662
 663         return dm_tm_commit(cmd->tm, sblock);
 664 }
 665
 666 /*----------------------------------------------------------------*/
 667
 668 /*
 669  * The mappings are held in a dm-array that has 64-bit values stored in
 670  * little-endian format.  The index is the cblock, the high 48bits of the
 671  * value are the oblock and the low 16 bit the flags.
 672  */
 673 #define FLAGS_MASK ((1 << 16) - 1)
 674
 675 static __le64 pack_value(dm_oblock_t block, unsigned flags)
 676 {
 677         uint64_t value = from_oblock(block);
 678         value <<= 16;
 679         value = value | (flags & FLAGS_MASK);
 680         return cpu_to_le64(value);
 681 }
 682
 683 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
 684 {
 685         uint64_t value = le64_to_cpu(value_le);
 686         uint64_t b = value >> 16;
 687         *block = to_oblock(b);
 688         *flags = value & FLAGS_MASK;
 689 }
 690
 691 /*----------------------------------------------------------------*/
 692
 693 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
 694                                                sector_t data_block_size,
 695                                                bool may_format_device,
 696                                                size_t policy_hint_size)
 697 {
 698         int r;
 699         struct dm_cache_metadata *cmd;
 700
 701         cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
 702         if (!cmd) {
 703                 DMERR("could not allocate metadata struct");
 704                 return ERR_PTR(-ENOMEM);
 705         }
 706
 707         atomic_set(&cmd->ref_count, 1);
 708         init_rwsem(&cmd->root_lock);
 709         cmd->bdev = bdev;
 710         cmd->data_block_size = data_block_size;
 711         cmd->cache_blocks = 0;
 712         cmd->policy_hint_size = policy_hint_size;
 713         cmd->changed = true;
 714         cmd->fail_io = false;
 715
 716         r = __create_persistent_data_objects(cmd, may_format_device);
 717         if (r) {
 718                 kfree(cmd);
 719                 return ERR_PTR(r);
 720         }
 721
 722         r = __begin_transaction_flags(cmd, clear_clean_shutdown);
 723         if (r < 0) {
 724                 dm_cache_metadata_close(cmd);
 725                 return ERR_PTR(r);
 726         }
 727
 728         return cmd;
 729 }
 730
 731 /*
 732  * We keep a little list of ref counted metadata objects to prevent two
 733  * different target instances creating separate bufio instances.  This is
 734  * an issue if a table is reloaded before the suspend.
 735  */
 736 static DEFINE_MUTEX(table_lock);
 737 static LIST_HEAD(table);
 738
 739 static struct dm_cache_metadata *lookup(struct block_device *bdev)
 740 {
 741         struct dm_cache_metadata *cmd;
 742
 743         list_for_each_entry(cmd, &table, list)
 744                 if (cmd->bdev == bdev) {
 745                         atomic_inc(&cmd->ref_count);
 746                         return cmd;
 747                 }
 748
 749         return NULL;
 750 }
 751
 752 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
 753                                                 sector_t data_block_size,
 754                                                 bool may_format_device,
 755                                                 size_t policy_hint_size)
 756 {
 757         struct dm_cache_metadata *cmd, *cmd2;
 758
 759         mutex_lock(&table_lock);
 760         cmd = lookup(bdev);
 761         mutex_unlock(&table_lock);
 762
 763         if (cmd)
 764                 return cmd;
 765
 766         cmd = metadata_open(bdev, data_block_size, may_format_device, policy_hint_size);
 767         if (!IS_ERR(cmd)) {
 768                 mutex_lock(&table_lock);
 769                 cmd2 = lookup(bdev);
 770                 if (cmd2) {
 771                         mutex_unlock(&table_lock);
 772                         __destroy_persistent_data_objects(cmd);
 773                         kfree(cmd);
 774                         return cmd2;
 775                 }
 776                 list_add(&cmd->list, &table);
 777                 mutex_unlock(&table_lock);
 778         }
 779
 780         return cmd;
 781 }
 782
 783 static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
 784 {
 785         if (cmd->data_block_size != data_block_size) {
 786                 DMERR("data_block_size (%llu) different from that in metadata (%llu)\n",
 787                       (unsigned long long) data_block_size,
 788                       (unsigned long long) cmd->data_block_size);
 789                 return false;
 790         }
 791
 792         return true;
 793 }
 794
 795 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
 796                                                  sector_t data_block_size,
 797                                                  bool may_format_device,
 798                                                  size_t policy_hint_size)
 799 {
 800         struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size,
 801                                                        may_format_device, policy_hint_size);
 802
 803         if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
 804                 dm_cache_metadata_close(cmd);
 805                 return ERR_PTR(-EINVAL);
 806         }
 807
 808         return cmd;
 809 }
 810
 811 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
 812 {
 813         if (atomic_dec_and_test(&cmd->ref_count)) {
 814                 mutex_lock(&table_lock);
 815                 list_del(&cmd->list);
 816                 mutex_unlock(&table_lock);
 817
 818                 if (!cmd->fail_io)
 819                         __destroy_persistent_data_objects(cmd);
 820                 kfree(cmd);
 821         }
 822 }
 823
 824 /*
 825  * Checks that the given cache block is either unmapped or clean.
 826  */
 827 static int block_unmapped_or_clean(struct dm_cache_metadata *cmd, dm_cblock_t b,
 828                                    bool *result)
 829 {
 830         int r;
 831         __le64 value;
 832         dm_oblock_t ob;
 833         unsigned flags;
 834
 835         r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
 836         if (r) {
 837                 DMERR("block_unmapped_or_clean failed");
 838                 return r;
 839         }
 840
 841         unpack_value(value, &ob, &flags);
 842         *result = !((flags & M_VALID) && (flags & M_DIRTY));
 843
 844         return 0;
 845 }
 846
 847 static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
 848                                         dm_cblock_t begin, dm_cblock_t end,
 849                                         bool *result)
 850 {
 851         int r;
 852         *result = true;
 853
 854         while (begin != end) {
 855                 r = block_unmapped_or_clean(cmd, begin, result);
 856                 if (r)
 857                         return r;
 858
 859                 if (!*result) {
 860                         DMERR("cache block %llu is dirty",
 861                               (unsigned long long) from_cblock(begin));
 862                         return 0;
 863                 }
 864
 865                 begin = to_cblock(from_cblock(begin) + 1);
 866         }
 867
 868         return 0;
 869 }
 870
 871 static bool cmd_write_lock(struct dm_cache_metadata *cmd)
 872 {
 873         down_write(&cmd->root_lock);
 874         if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) {
 875                 up_write(&cmd->root_lock);
 876                 return false;
 877         }
 878         return true;
 879 }
 880
 881 #define WRITE_LOCK(cmd)                         \
 882         do {                                    \
 883                 if (!cmd_write_lock((cmd)))     \
 884                         return -EINVAL;         \
 885         } while(0)
 886
 887 #define WRITE_LOCK_VOID(cmd)                    \
 888         do {                                    \
 889                 if (!cmd_write_lock((cmd)))     \
 890                         return;                 \
 891         } while(0)
 892
 893 #define WRITE_UNLOCK(cmd) \
 894         up_write(&(cmd)->root_lock)
 895
 896 static bool cmd_read_lock(struct dm_cache_metadata *cmd)
 897 {
 898         down_read(&cmd->root_lock);
 899         if (cmd->fail_io) {
 900                 up_read(&cmd->root_lock);
 901                 return false;
 902         }
 903         return true;
 904 }
 905
 906 #define READ_LOCK(cmd)                          \
 907         do {                                    \
 908                 if (!cmd_read_lock((cmd)))      \
 909                         return -EINVAL;         \
 910         } while(0)
 911
 912 #define READ_LOCK_VOID(cmd)                     \
 913         do {                                    \
 914                 if (!cmd_read_lock((cmd)))      \
 915                         return;                 \
 916         } while(0)
 917
 918 #define READ_UNLOCK(cmd) \
 919         up_read(&(cmd)->root_lock)
 920
 921 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
 922 {
 923         int r;
 924         bool clean;
 925         __le64 null_mapping = pack_value(0, 0);
 926
 927         WRITE_LOCK(cmd);
 928         __dm_bless_for_disk(&null_mapping);
 929
 930         if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) {
 931                 r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean);
 932                 if (r) {
 933                         __dm_unbless_for_disk(&null_mapping);
 934                         goto out;
 935                 }
 936
 937                 if (!clean) {
 938                         DMERR("unable to shrink cache due to dirty blocks");
 939                         r = -EINVAL;
 940                         __dm_unbless_for_disk(&null_mapping);
 941                         goto out;
 942                 }
 943         }
 944
 945         r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
 946                             from_cblock(new_cache_size),
 947                             &null_mapping, &cmd->root);
 948         if (!r)
 949                 cmd->cache_blocks = new_cache_size;
 950         cmd->changed = true;
 951
 952 out:
 953         WRITE_UNLOCK(cmd);
 954
 955         return r;
 956 }
 957
 958 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
 959                                    sector_t discard_block_size,
 960                                    dm_dblock_t new_nr_entries)
 961 {
 962         int r;
 963
 964         WRITE_LOCK(cmd);
 965         r = dm_bitset_resize(&cmd->discard_info,
 966                              cmd->discard_root,
 967                              from_dblock(cmd->discard_nr_blocks),
 968                              from_dblock(new_nr_entries),
 969                              false, &cmd->discard_root);
 970         if (!r) {
 971                 cmd->discard_block_size = discard_block_size;
 972                 cmd->discard_nr_blocks = new_nr_entries;
 973         }
 974
 975         cmd->changed = true;
 976         WRITE_UNLOCK(cmd);
 977
 978         return r;
 979 }
 980
 981 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
 982 {
 983         return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
 984                                  from_dblock(b), &cmd->discard_root);
 985 }
 986
 987 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
 988 {
 989         return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
 990                                    from_dblock(b), &cmd->discard_root);
 991 }
 992
 993 static int __is_discarded(struct dm_cache_metadata *cmd, dm_dblock_t b,
 994                           bool *is_discarded)
 995 {
 996         return dm_bitset_test_bit(&cmd->discard_info, cmd->discard_root,
 997                                   from_dblock(b), &cmd->discard_root,
 998                                   is_discarded);
 999 }
1000
1001 static int __discard(struct dm_cache_metadata *cmd,
1002                      dm_dblock_t dblock, bool discard)
1003 {
1004         int r;
1005
1006         r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
1007         if (r)
1008                 return r;
1009
1010         cmd->changed = true;
1011         return 0;
1012 }
1013
1014 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
1015                          dm_dblock_t dblock, bool discard)
1016 {
1017         int r;
1018
1019         WRITE_LOCK(cmd);
1020         r = __discard(cmd, dblock, discard);
1021         WRITE_UNLOCK(cmd);
1022
1023         return r;
1024 }
1025
1026 static int __load_discards(struct dm_cache_metadata *cmd,
1027                            load_discard_fn fn, void *context)
1028 {
1029         int r = 0;
1030         dm_block_t b;
1031         bool discard;
1032
1033         for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1034                 dm_dblock_t dblock = to_dblock(b);
1035
1036                 if (cmd->clean_when_opened) {
1037                         r = __is_discarded(cmd, dblock, &discard);
1038                         if (r)
1039                                 return r;
1040                 } else
1041                         discard = false;
1042
1043                 r = fn(context, cmd->discard_block_size, dblock, discard);
1044                 if (r)
1045                         break;
1046         }
1047
1048         return r;
1049 }
1050
1051 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
1052                            load_discard_fn fn, void *context)
1053 {
1054         int r;
1055
1056         READ_LOCK(cmd);
1057         r = __load_discards(cmd, fn, context);
1058         READ_UNLOCK(cmd);
1059
1060         return r;
1061 }
1062
1063 int dm_cache_size(struct dm_cache_metadata *cmd, dm_cblock_t *result)
1064 {
1065         READ_LOCK(cmd);
1066         *result = cmd->cache_blocks;
1067         READ_UNLOCK(cmd);
1068
1069         return 0;
1070 }
1071
1072 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1073 {
1074         int r;
1075         __le64 value = pack_value(0, 0);
1076
1077         __dm_bless_for_disk(&value);
1078         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1079                                &value, &cmd->root);
1080         if (r)
1081                 return r;
1082
1083         cmd->changed = true;
1084         return 0;
1085 }
1086
1087 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1088 {
1089         int r;
1090
1091         WRITE_LOCK(cmd);
1092         r = __remove(cmd, cblock);
1093         WRITE_UNLOCK(cmd);
1094
1095         return r;
1096 }
1097
1098 static int __insert(struct dm_cache_metadata *cmd,
1099                     dm_cblock_t cblock, dm_oblock_t oblock)
1100 {
1101         int r;
1102         __le64 value = pack_value(oblock, M_VALID);
1103         __dm_bless_for_disk(&value);
1104
1105         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1106                                &value, &cmd->root);
1107         if (r)
1108                 return r;
1109
1110         cmd->changed = true;
1111         return 0;
1112 }
1113
1114 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
1115                             dm_cblock_t cblock, dm_oblock_t oblock)
1116 {
1117         int r;
1118
1119         WRITE_LOCK(cmd);
1120         r = __insert(cmd, cblock, oblock);
1121         WRITE_UNLOCK(cmd);
1122
1123         return r;
1124 }
1125
1126 struct thunk {
1127         load_mapping_fn fn;
1128         void *context;
1129
1130         struct dm_cache_metadata *cmd;
1131         bool respect_dirty_flags;
1132         bool hints_valid;
1133 };
1134
1135 static bool policy_unchanged(struct dm_cache_metadata *cmd,
1136                              struct dm_cache_policy *policy)
1137 {
1138         const char *policy_name = dm_cache_policy_get_name(policy);
1139         const unsigned *policy_version = dm_cache_policy_get_version(policy);
1140         size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
1141
1142         /*
1143          * Ensure policy names match.
1144          */
1145         if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
1146                 return false;
1147
1148         /*
1149          * Ensure policy major versions match.
1150          */
1151         if (cmd->policy_version[0] != policy_version[0])
1152                 return false;
1153
1154         /*
1155          * Ensure policy hint sizes match.
1156          */
1157         if (cmd->policy_hint_size != policy_hint_size)
1158                 return false;
1159
1160         return true;
1161 }
1162
1163 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1164 {
1165         return cmd->hint_root && cmd->policy_hint_size;
1166 }
1167
1168 static bool hints_array_available(struct dm_cache_metadata *cmd,
1169                                   struct dm_cache_policy *policy)
1170 {
1171         return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1172                 hints_array_initialized(cmd);
1173 }
1174
1175 static int __load_mapping(void *context, uint64_t cblock, void *leaf)
1176 {
1177         int r = 0;
1178         bool dirty;
1179         __le64 value;
1180         __le32 hint_value = 0;
1181         dm_oblock_t oblock;
1182         unsigned flags;
1183         struct thunk *thunk = context;
1184         struct dm_cache_metadata *cmd = thunk->cmd;
1185
1186         memcpy(&value, leaf, sizeof(value));
1187         unpack_value(value, &oblock, &flags);
1188
1189         if (flags & M_VALID) {
1190                 if (thunk->hints_valid) {
1191                         r = dm_array_get_value(&cmd->hint_info, cmd->hint_root,
1192                                                cblock, &hint_value);
1193                         if (r && r != -ENODATA)
1194                                 return r;
1195                 }
1196
1197                 dirty = thunk->respect_dirty_flags ? (flags & M_DIRTY) : true;
1198                 r = thunk->fn(thunk->context, oblock, to_cblock(cblock),
1199                               dirty, le32_to_cpu(hint_value), thunk->hints_valid);
1200         }
1201
1202         return r;
1203 }
1204
1205 static int __load_mappings(struct dm_cache_metadata *cmd,
1206                            struct dm_cache_policy *policy,
1207                            load_mapping_fn fn, void *context)
1208 {
1209         struct thunk thunk;
1210
1211         thunk.fn = fn;
1212         thunk.context = context;
1213
1214         thunk.cmd = cmd;
1215         thunk.respect_dirty_flags = cmd->clean_when_opened;
1216         thunk.hints_valid = hints_array_available(cmd, policy);
1217
1218         return dm_array_walk(&cmd->info, cmd->root, __load_mapping, &thunk);
1219 }
1220
1221 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1222                            struct dm_cache_policy *policy,
1223                            load_mapping_fn fn, void *context)
1224 {
1225         int r;
1226
1227         READ_LOCK(cmd);
1228         r = __load_mappings(cmd, policy, fn, context);
1229         READ_UNLOCK(cmd);
1230
1231         return r;
1232 }
1233
1234 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
1235 {
1236         int r = 0;
1237         __le64 value;
1238         dm_oblock_t oblock;
1239         unsigned flags;
1240
1241         memcpy(&value, leaf, sizeof(value));
1242         unpack_value(value, &oblock, &flags);
1243
1244         return r;
1245 }
1246
1247 static int __dump_mappings(struct dm_cache_metadata *cmd)
1248 {
1249         return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1250 }
1251
1252 void dm_cache_dump(struct dm_cache_metadata *cmd)
1253 {
1254         READ_LOCK_VOID(cmd);
1255         __dump_mappings(cmd);
1256         READ_UNLOCK(cmd);
1257 }
1258
1259 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1260 {
1261         int r;
1262
1263         READ_LOCK(cmd);
1264         r = cmd->changed;
1265         READ_UNLOCK(cmd);
1266
1267         return r;
1268 }
1269
1270 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1271 {
1272         int r;
1273         unsigned flags;
1274         dm_oblock_t oblock;
1275         __le64 value;
1276
1277         r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1278         if (r)
1279                 return r;
1280
1281         unpack_value(value, &oblock, &flags);
1282
1283         if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1284                 /* nothing to be done */
1285                 return 0;
1286
1287         value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1288         __dm_bless_for_disk(&value);
1289
1290         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1291                                &value, &cmd->root);
1292         if (r)
1293                 return r;
1294
1295         cmd->changed = true;
1296         return 0;
1297
1298 }
1299
1300 int dm_cache_set_dirty(struct dm_cache_metadata *cmd,
1301                        dm_cblock_t cblock, bool dirty)
1302 {
1303         int r;
1304
1305         WRITE_LOCK(cmd);
1306         r = __dirty(cmd, cblock, dirty);
1307         WRITE_UNLOCK(cmd);
1308
1309         return r;
1310 }
1311
1312 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1313                                  struct dm_cache_statistics *stats)
1314 {
1315         READ_LOCK_VOID(cmd);
1316         *stats = cmd->stats;
1317         READ_UNLOCK(cmd);
1318 }
1319
1320 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1321                                  struct dm_cache_statistics *stats)
1322 {
1323         WRITE_LOCK_VOID(cmd);
1324         cmd->stats = *stats;
1325         WRITE_UNLOCK(cmd);
1326 }
1327
1328 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1329 {
1330         int r = -EINVAL;
1331         flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1332                                  clear_clean_shutdown);
1333
1334         WRITE_LOCK(cmd);
1335         if (cmd->fail_io)
1336                 goto out;
1337
1338         r = __commit_transaction(cmd, mutator);
1339         if (r)
1340                 goto out;
1341
1342         r = __begin_transaction(cmd);
1343 out:
1344         WRITE_UNLOCK(cmd);
1345         return r;
1346 }
1347
1348 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1349                                            dm_block_t *result)
1350 {
1351         int r = -EINVAL;
1352
1353         READ_LOCK(cmd);
1354         if (!cmd->fail_io)
1355                 r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1356         READ_UNLOCK(cmd);
1357
1358         return r;
1359 }
1360
1361 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1362                                    dm_block_t *result)
1363 {
1364         int r = -EINVAL;
1365
1366         READ_LOCK(cmd);
1367         if (!cmd->fail_io)
1368                 r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1369         READ_UNLOCK(cmd);
1370
1371         return r;
1372 }
1373
1374 /*----------------------------------------------------------------*/
1375
1376 static int begin_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1377 {
1378         int r;
1379         __le32 value;
1380         size_t hint_size;
1381         const char *policy_name = dm_cache_policy_get_name(policy);
1382         const unsigned *policy_version = dm_cache_policy_get_version(policy);
1383
1384         if (!policy_name[0] ||
1385             (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1386                 return -EINVAL;
1387
1388         if (!policy_unchanged(cmd, policy)) {
1389                 strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1390                 memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1391
1392                 hint_size = dm_cache_policy_get_hint_size(policy);
1393                 if (!hint_size)
1394                         return 0; /* short-circuit hints initialization */
1395                 cmd->policy_hint_size = hint_size;
1396
1397                 if (cmd->hint_root) {
1398                         r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1399                         if (r)
1400                                 return r;
1401                 }
1402
1403                 r = dm_array_empty(&cmd->hint_info, &cmd->hint_root);
1404                 if (r)
1405                         return r;
1406
1407                 value = cpu_to_le32(0);
1408                 __dm_bless_for_disk(&value);
1409                 r = dm_array_resize(&cmd->hint_info, cmd->hint_root, 0,
1410                                     from_cblock(cmd->cache_blocks),
1411                                     &value, &cmd->hint_root);
1412                 if (r)
1413                         return r;
1414         }
1415
1416         return 0;
1417 }
1418
1419 static int save_hint(void *context, dm_cblock_t cblock, dm_oblock_t oblock, uint32_t hint)
1420 {
1421         struct dm_cache_metadata *cmd = context;
1422         __le32 value = cpu_to_le32(hint);
1423         int r;
1424
1425         __dm_bless_for_disk(&value);
1426
1427         r = dm_array_set_value(&cmd->hint_info, cmd->hint_root,
1428                                from_cblock(cblock), &value, &cmd->hint_root);
1429         cmd->changed = true;
1430
1431         return r;
1432 }
1433
1434 static int write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1435 {
1436         int r;
1437
1438         r = begin_hints(cmd, policy);
1439         if (r) {
1440                 DMERR("begin_hints failed");
1441                 return r;
1442         }
1443
1444         return policy_walk_mappings(policy, save_hint, cmd);
1445 }
1446
1447 int dm_cache_write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1448 {
1449         int r;
1450
1451         WRITE_LOCK(cmd);
1452         r = write_hints(cmd, policy);
1453         WRITE_UNLOCK(cmd);
1454
1455         return r;
1456 }
1457
1458 int dm_cache_metadata_all_clean(struct dm_cache_metadata *cmd, bool *result)
1459 {
1460         int r;
1461
1462         READ_LOCK(cmd);
1463         r = blocks_are_unmapped_or_clean(cmd, 0, cmd->cache_blocks, result);
1464         READ_UNLOCK(cmd);
1465
1466         return r;
1467 }
1468
1469 void dm_cache_metadata_set_read_only(struct dm_cache_metadata *cmd)
1470 {
1471         WRITE_LOCK_VOID(cmd);
1472         dm_bm_set_read_only(cmd->bm);
1473         WRITE_UNLOCK(cmd);
1474 }
1475
1476 void dm_cache_metadata_set_read_write(struct dm_cache_metadata *cmd)
1477 {
1478         WRITE_LOCK_VOID(cmd);
1479         dm_bm_set_read_write(cmd->bm);
1480         WRITE_UNLOCK(cmd);
1481 }
1482
1483 int dm_cache_metadata_set_needs_check(struct dm_cache_metadata *cmd)
1484 {
1485         int r;
1486         struct dm_block *sblock;
1487         struct cache_disk_superblock *disk_super;
1488
1489         WRITE_LOCK(cmd);
1490         set_bit(NEEDS_CHECK, &cmd->flags);
1491
1492         r = superblock_lock(cmd, &sblock);
1493         if (r) {
1494                 DMERR("couldn't read superblock");
1495                 goto out;
1496         }
1497
1498         disk_super = dm_block_data(sblock);
1499         disk_super->flags = cpu_to_le32(cmd->flags);
1500
1501         dm_bm_unlock(sblock);
1502
1503 out:
1504         WRITE_UNLOCK(cmd);
1505         return r;
1506 }
1507
1508 int dm_cache_metadata_needs_check(struct dm_cache_metadata *cmd, bool *result)
1509 {
1510         READ_LOCK(cmd);
1511         *result = !!test_bit(NEEDS_CHECK, &cmd->flags);
1512         READ_UNLOCK(cmd);
1513
1514         return 0;
1515 }
1516
1517 int dm_cache_metadata_abort(struct dm_cache_metadata *cmd)
1518 {
1519         int r;
1520
1521         WRITE_LOCK(cmd);
1522         __destroy_persistent_data_objects(cmd);
1523         r = __create_persistent_data_objects(cmd, false);
1524         if (r)
1525                 cmd->fail_io = true;
1526         WRITE_UNLOCK(cmd);
1527
1528         return r;
1529 }