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                 r = PTR_ERR(cmd->bm);
 505                 cmd->bm = NULL;
 506                 return r;
 507         }
 508
 509         r = __open_or_format_metadata(cmd, may_format_device);
 510         if (r) {
 511                 dm_block_manager_destroy(cmd->bm);
 512                 cmd->bm = NULL;
 513         }
 514
 515         return r;
 516 }
 517
 518 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd)
 519 {
 520         dm_sm_destroy(cmd->metadata_sm);
 521         dm_tm_destroy(cmd->tm);
 522         dm_block_manager_destroy(cmd->bm);
 523 }
 524
 525 typedef unsigned long (*flags_mutator)(unsigned long);
 526
 527 static void update_flags(struct cache_disk_superblock *disk_super,
 528                          flags_mutator mutator)
 529 {
 530         uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
 531         disk_super->flags = cpu_to_le32(sb_flags);
 532 }
 533
 534 static unsigned long set_clean_shutdown(unsigned long flags)
 535 {
 536         set_bit(CLEAN_SHUTDOWN, &flags);
 537         return flags;
 538 }
 539
 540 static unsigned long clear_clean_shutdown(unsigned long flags)
 541 {
 542         clear_bit(CLEAN_SHUTDOWN, &flags);
 543         return flags;
 544 }
 545
 546 static void read_superblock_fields(struct dm_cache_metadata *cmd,
 547                                    struct cache_disk_superblock *disk_super)
 548 {
 549         cmd->flags = le32_to_cpu(disk_super->flags);
 550         cmd->root = le64_to_cpu(disk_super->mapping_root);
 551         cmd->hint_root = le64_to_cpu(disk_super->hint_root);
 552         cmd->discard_root = le64_to_cpu(disk_super->discard_root);
 553         cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
 554         cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
 555         cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
 556         cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
 557         strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
 558         cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
 559         cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
 560         cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
 561         cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
 562
 563         cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
 564         cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
 565         cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
 566         cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
 567
 568         cmd->changed = false;
 569 }
 570
 571 /*
 572  * The mutator updates the superblock flags.
 573  */
 574 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
 575                                      flags_mutator mutator)
 576 {
 577         int r;
 578         struct cache_disk_superblock *disk_super;
 579         struct dm_block *sblock;
 580
 581         r = superblock_lock(cmd, &sblock);
 582         if (r)
 583                 return r;
 584
 585         disk_super = dm_block_data(sblock);
 586         update_flags(disk_super, mutator);
 587         read_superblock_fields(cmd, disk_super);
 588         dm_bm_unlock(sblock);
 589
 590         return dm_bm_flush(cmd->bm);
 591 }
 592
 593 static int __begin_transaction(struct dm_cache_metadata *cmd)
 594 {
 595         int r;
 596         struct cache_disk_superblock *disk_super;
 597         struct dm_block *sblock;
 598
 599         /*
 600          * We re-read the superblock every time.  Shouldn't need to do this
 601          * really.
 602          */
 603         r = superblock_read_lock(cmd, &sblock);
 604         if (r)
 605                 return r;
 606
 607         disk_super = dm_block_data(sblock);
 608         read_superblock_fields(cmd, disk_super);
 609         dm_bm_unlock(sblock);
 610
 611         return 0;
 612 }
 613
 614 static int __commit_transaction(struct dm_cache_metadata *cmd,
 615                                 flags_mutator mutator)
 616 {
 617         int r;
 618         struct cache_disk_superblock *disk_super;
 619         struct dm_block *sblock;
 620
 621         /*
 622          * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
 623          */
 624         BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
 625
 626         r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
 627                             &cmd->discard_root);
 628         if (r)
 629                 return r;
 630
 631         r = dm_tm_pre_commit(cmd->tm);
 632         if (r < 0)
 633                 return r;
 634
 635         r = __save_sm_root(cmd);
 636         if (r)
 637                 return r;
 638
 639         r = superblock_lock(cmd, &sblock);
 640         if (r)
 641                 return r;
 642
 643         disk_super = dm_block_data(sblock);
 644
 645         disk_super->flags = cpu_to_le32(cmd->flags);
 646         if (mutator)
 647                 update_flags(disk_super, mutator);
 648
 649         disk_super->mapping_root = cpu_to_le64(cmd->root);
 650         disk_super->hint_root = cpu_to_le64(cmd->hint_root);
 651         disk_super->discard_root = cpu_to_le64(cmd->discard_root);
 652         disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
 653         disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
 654         disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
 655         strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
 656         disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
 657         disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
 658         disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
 659         disk_super->policy_hint_size = cpu_to_le32(cmd->policy_hint_size);
 660
 661         disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
 662         disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
 663         disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
 664         disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
 665         __copy_sm_root(cmd, disk_super);
 666
 667         return dm_tm_commit(cmd->tm, sblock);
 668 }
 669
 670 /*----------------------------------------------------------------*/
 671
 672 /*
 673  * The mappings are held in a dm-array that has 64-bit values stored in
 674  * little-endian format.  The index is the cblock, the high 48bits of the
 675  * value are the oblock and the low 16 bit the flags.
 676  */
 677 #define FLAGS_MASK ((1 << 16) - 1)
 678
 679 static __le64 pack_value(dm_oblock_t block, unsigned flags)
 680 {
 681         uint64_t value = from_oblock(block);
 682         value <<= 16;
 683         value = value | (flags & FLAGS_MASK);
 684         return cpu_to_le64(value);
 685 }
 686
 687 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
 688 {
 689         uint64_t value = le64_to_cpu(value_le);
 690         uint64_t b = value >> 16;
 691         *block = to_oblock(b);
 692         *flags = value & FLAGS_MASK;
 693 }
 694
 695 /*----------------------------------------------------------------*/
 696
 697 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
 698                                                sector_t data_block_size,
 699                                                bool may_format_device,
 700                                                size_t policy_hint_size)
 701 {
 702         int r;
 703         struct dm_cache_metadata *cmd;
 704
 705         cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
 706         if (!cmd) {
 707                 DMERR("could not allocate metadata struct");
 708                 return ERR_PTR(-ENOMEM);
 709         }
 710
 711         atomic_set(&cmd->ref_count, 1);
 712         init_rwsem(&cmd->root_lock);
 713         cmd->bdev = bdev;
 714         cmd->data_block_size = data_block_size;
 715         cmd->cache_blocks = 0;
 716         cmd->policy_hint_size = policy_hint_size;
 717         cmd->changed = true;
 718         cmd->fail_io = false;
 719
 720         r = __create_persistent_data_objects(cmd, may_format_device);
 721         if (r) {
 722                 kfree(cmd);
 723                 return ERR_PTR(r);
 724         }
 725
 726         r = __begin_transaction_flags(cmd, clear_clean_shutdown);
 727         if (r < 0) {
 728                 dm_cache_metadata_close(cmd);
 729                 return ERR_PTR(r);
 730         }
 731
 732         return cmd;
 733 }
 734
 735 /*
 736  * We keep a little list of ref counted metadata objects to prevent two
 737  * different target instances creating separate bufio instances.  This is
 738  * an issue if a table is reloaded before the suspend.
 739  */
 740 static DEFINE_MUTEX(table_lock);
 741 static LIST_HEAD(table);
 742
 743 static struct dm_cache_metadata *lookup(struct block_device *bdev)
 744 {
 745         struct dm_cache_metadata *cmd;
 746
 747         list_for_each_entry(cmd, &table, list)
 748                 if (cmd->bdev == bdev) {
 749                         atomic_inc(&cmd->ref_count);
 750                         return cmd;
 751                 }
 752
 753         return NULL;
 754 }
 755
 756 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
 757                                                 sector_t data_block_size,
 758                                                 bool may_format_device,
 759                                                 size_t policy_hint_size)
 760 {
 761         struct dm_cache_metadata *cmd, *cmd2;
 762
 763         mutex_lock(&table_lock);
 764         cmd = lookup(bdev);
 765         mutex_unlock(&table_lock);
 766
 767         if (cmd)
 768                 return cmd;
 769
 770         cmd = metadata_open(bdev, data_block_size, may_format_device, policy_hint_size);
 771         if (!IS_ERR(cmd)) {
 772                 mutex_lock(&table_lock);
 773                 cmd2 = lookup(bdev);
 774                 if (cmd2) {
 775                         mutex_unlock(&table_lock);
 776                         __destroy_persistent_data_objects(cmd);
 777                         kfree(cmd);
 778                         return cmd2;
 779                 }
 780                 list_add(&cmd->list, &table);
 781                 mutex_unlock(&table_lock);
 782         }
 783
 784         return cmd;
 785 }
 786
 787 static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
 788 {
 789         if (cmd->data_block_size != data_block_size) {
 790                 DMERR("data_block_size (%llu) different from that in metadata (%llu)\n",
 791                       (unsigned long long) data_block_size,
 792                       (unsigned long long) cmd->data_block_size);
 793                 return false;
 794         }
 795
 796         return true;
 797 }
 798
 799 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
 800                                                  sector_t data_block_size,
 801                                                  bool may_format_device,
 802                                                  size_t policy_hint_size)
 803 {
 804         struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size,
 805                                                        may_format_device, policy_hint_size);
 806
 807         if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
 808                 dm_cache_metadata_close(cmd);
 809                 return ERR_PTR(-EINVAL);
 810         }
 811
 812         return cmd;
 813 }
 814
 815 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
 816 {
 817         if (atomic_dec_and_test(&cmd->ref_count)) {
 818                 mutex_lock(&table_lock);
 819                 list_del(&cmd->list);
 820                 mutex_unlock(&table_lock);
 821
 822                 if (!cmd->fail_io)
 823                         __destroy_persistent_data_objects(cmd);
 824                 kfree(cmd);
 825         }
 826 }
 827
 828 /*
 829  * Checks that the given cache block is either unmapped or clean.
 830  */
 831 static int block_unmapped_or_clean(struct dm_cache_metadata *cmd, dm_cblock_t b,
 832                                    bool *result)
 833 {
 834         int r;
 835         __le64 value;
 836         dm_oblock_t ob;
 837         unsigned flags;
 838
 839         r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
 840         if (r) {
 841                 DMERR("block_unmapped_or_clean failed");
 842                 return r;
 843         }
 844
 845         unpack_value(value, &ob, &flags);
 846         *result = !((flags & M_VALID) && (flags & M_DIRTY));
 847
 848         return 0;
 849 }
 850
 851 static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
 852                                         dm_cblock_t begin, dm_cblock_t end,
 853                                         bool *result)
 854 {
 855         int r;
 856         *result = true;
 857
 858         while (begin != end) {
 859                 r = block_unmapped_or_clean(cmd, begin, result);
 860                 if (r)
 861                         return r;
 862
 863                 if (!*result) {
 864                         DMERR("cache block %llu is dirty",
 865                               (unsigned long long) from_cblock(begin));
 866                         return 0;
 867                 }
 868
 869                 begin = to_cblock(from_cblock(begin) + 1);
 870         }
 871
 872         return 0;
 873 }
 874
 875 static bool cmd_write_lock(struct dm_cache_metadata *cmd)
 876 {
 877         down_write(&cmd->root_lock);
 878         if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) {
 879                 up_write(&cmd->root_lock);
 880                 return false;
 881         }
 882         return true;
 883 }
 884
 885 #define WRITE_LOCK(cmd)                         \
 886         do {                                    \
 887                 if (!cmd_write_lock((cmd)))     \
 888                         return -EINVAL;         \
 889         } while(0)
 890
 891 #define WRITE_LOCK_VOID(cmd)                    \
 892         do {                                    \
 893                 if (!cmd_write_lock((cmd)))     \
 894                         return;                 \
 895         } while(0)
 896
 897 #define WRITE_UNLOCK(cmd) \
 898         up_write(&(cmd)->root_lock)
 899
 900 static bool cmd_read_lock(struct dm_cache_metadata *cmd)
 901 {
 902         down_read(&cmd->root_lock);
 903         if (cmd->fail_io) {
 904                 up_read(&cmd->root_lock);
 905                 return false;
 906         }
 907         return true;
 908 }
 909
 910 #define READ_LOCK(cmd)                          \
 911         do {                                    \
 912                 if (!cmd_read_lock((cmd)))      \
 913                         return -EINVAL;         \
 914         } while(0)
 915
 916 #define READ_LOCK_VOID(cmd)                     \
 917         do {                                    \
 918                 if (!cmd_read_lock((cmd)))      \
 919                         return;                 \
 920         } while(0)
 921
 922 #define READ_UNLOCK(cmd) \
 923         up_read(&(cmd)->root_lock)
 924
 925 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
 926 {
 927         int r;
 928         bool clean;
 929         __le64 null_mapping = pack_value(0, 0);
 930
 931         WRITE_LOCK(cmd);
 932         __dm_bless_for_disk(&null_mapping);
 933
 934         if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) {
 935                 r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean);
 936                 if (r) {
 937                         __dm_unbless_for_disk(&null_mapping);
 938                         goto out;
 939                 }
 940
 941                 if (!clean) {
 942                         DMERR("unable to shrink cache due to dirty blocks");
 943                         r = -EINVAL;
 944                         __dm_unbless_for_disk(&null_mapping);
 945                         goto out;
 946                 }
 947         }
 948
 949         r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
 950                             from_cblock(new_cache_size),
 951                             &null_mapping, &cmd->root);
 952         if (!r)
 953                 cmd->cache_blocks = new_cache_size;
 954         cmd->changed = true;
 955
 956 out:
 957         WRITE_UNLOCK(cmd);
 958
 959         return r;
 960 }
 961
 962 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
 963                                    sector_t discard_block_size,
 964                                    dm_dblock_t new_nr_entries)
 965 {
 966         int r;
 967
 968         WRITE_LOCK(cmd);
 969         r = dm_bitset_resize(&cmd->discard_info,
 970                              cmd->discard_root,
 971                              from_dblock(cmd->discard_nr_blocks),
 972                              from_dblock(new_nr_entries),
 973                              false, &cmd->discard_root);
 974         if (!r) {
 975                 cmd->discard_block_size = discard_block_size;
 976                 cmd->discard_nr_blocks = new_nr_entries;
 977         }
 978
 979         cmd->changed = true;
 980         WRITE_UNLOCK(cmd);
 981
 982         return r;
 983 }
 984
 985 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
 986 {
 987         return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
 988                                  from_dblock(b), &cmd->discard_root);
 989 }
 990
 991 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
 992 {
 993         return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
 994                                    from_dblock(b), &cmd->discard_root);
 995 }
 996
 997 static int __is_discarded(struct dm_cache_metadata *cmd, dm_dblock_t b,
 998                           bool *is_discarded)
 999 {
1000         return dm_bitset_test_bit(&cmd->discard_info, cmd->discard_root,
1001                                   from_dblock(b), &cmd->discard_root,
1002                                   is_discarded);
1003 }
1004
1005 static int __discard(struct dm_cache_metadata *cmd,
1006                      dm_dblock_t dblock, bool discard)
1007 {
1008         int r;
1009
1010         r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
1011         if (r)
1012                 return r;
1013
1014         cmd->changed = true;
1015         return 0;
1016 }
1017
1018 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
1019                          dm_dblock_t dblock, bool discard)
1020 {
1021         int r;
1022
1023         WRITE_LOCK(cmd);
1024         r = __discard(cmd, dblock, discard);
1025         WRITE_UNLOCK(cmd);
1026
1027         return r;
1028 }
1029
1030 static int __load_discards(struct dm_cache_metadata *cmd,
1031                            load_discard_fn fn, void *context)
1032 {
1033         int r = 0;
1034         dm_block_t b;
1035         bool discard;
1036
1037         for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1038                 dm_dblock_t dblock = to_dblock(b);
1039
1040                 if (cmd->clean_when_opened) {
1041                         r = __is_discarded(cmd, dblock, &discard);
1042                         if (r)
1043                                 return r;
1044                 } else
1045                         discard = false;
1046
1047                 r = fn(context, cmd->discard_block_size, dblock, discard);
1048                 if (r)
1049                         break;
1050         }
1051
1052         return r;
1053 }
1054
1055 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
1056                            load_discard_fn fn, void *context)
1057 {
1058         int r;
1059
1060         READ_LOCK(cmd);
1061         r = __load_discards(cmd, fn, context);
1062         READ_UNLOCK(cmd);
1063
1064         return r;
1065 }
1066
1067 int dm_cache_size(struct dm_cache_metadata *cmd, dm_cblock_t *result)
1068 {
1069         READ_LOCK(cmd);
1070         *result = cmd->cache_blocks;
1071         READ_UNLOCK(cmd);
1072
1073         return 0;
1074 }
1075
1076 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1077 {
1078         int r;
1079         __le64 value = pack_value(0, 0);
1080
1081         __dm_bless_for_disk(&value);
1082         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1083                                &value, &cmd->root);
1084         if (r)
1085                 return r;
1086
1087         cmd->changed = true;
1088         return 0;
1089 }
1090
1091 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1092 {
1093         int r;
1094
1095         WRITE_LOCK(cmd);
1096         r = __remove(cmd, cblock);
1097         WRITE_UNLOCK(cmd);
1098
1099         return r;
1100 }
1101
1102 static int __insert(struct dm_cache_metadata *cmd,
1103                     dm_cblock_t cblock, dm_oblock_t oblock)
1104 {
1105         int r;
1106         __le64 value = pack_value(oblock, M_VALID);
1107         __dm_bless_for_disk(&value);
1108
1109         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1110                                &value, &cmd->root);
1111         if (r)
1112                 return r;
1113
1114         cmd->changed = true;
1115         return 0;
1116 }
1117
1118 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
1119                             dm_cblock_t cblock, dm_oblock_t oblock)
1120 {
1121         int r;
1122
1123         WRITE_LOCK(cmd);
1124         r = __insert(cmd, cblock, oblock);
1125         WRITE_UNLOCK(cmd);
1126
1127         return r;
1128 }
1129
1130 struct thunk {
1131         load_mapping_fn fn;
1132         void *context;
1133
1134         struct dm_cache_metadata *cmd;
1135         bool respect_dirty_flags;
1136         bool hints_valid;
1137 };
1138
1139 static bool policy_unchanged(struct dm_cache_metadata *cmd,
1140                              struct dm_cache_policy *policy)
1141 {
1142         const char *policy_name = dm_cache_policy_get_name(policy);
1143         const unsigned *policy_version = dm_cache_policy_get_version(policy);
1144         size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
1145
1146         /*
1147          * Ensure policy names match.
1148          */
1149         if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
1150                 return false;
1151
1152         /*
1153          * Ensure policy major versions match.
1154          */
1155         if (cmd->policy_version[0] != policy_version[0])
1156                 return false;
1157
1158         /*
1159          * Ensure policy hint sizes match.
1160          */
1161         if (cmd->policy_hint_size != policy_hint_size)
1162                 return false;
1163
1164         return true;
1165 }
1166
1167 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1168 {
1169         return cmd->hint_root && cmd->policy_hint_size;
1170 }
1171
1172 static bool hints_array_available(struct dm_cache_metadata *cmd,
1173                                   struct dm_cache_policy *policy)
1174 {
1175         return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1176                 hints_array_initialized(cmd);
1177 }
1178
1179 static int __load_mapping(void *context, uint64_t cblock, void *leaf)
1180 {
1181         int r = 0;
1182         bool dirty;
1183         __le64 value;
1184         __le32 hint_value = 0;
1185         dm_oblock_t oblock;
1186         unsigned flags;
1187         struct thunk *thunk = context;
1188         struct dm_cache_metadata *cmd = thunk->cmd;
1189
1190         memcpy(&value, leaf, sizeof(value));
1191         unpack_value(value, &oblock, &flags);
1192
1193         if (flags & M_VALID) {
1194                 if (thunk->hints_valid) {
1195                         r = dm_array_get_value(&cmd->hint_info, cmd->hint_root,
1196                                                cblock, &hint_value);
1197                         if (r && r != -ENODATA)
1198                                 return r;
1199                 }
1200
1201                 dirty = thunk->respect_dirty_flags ? (flags & M_DIRTY) : true;
1202                 r = thunk->fn(thunk->context, oblock, to_cblock(cblock),
1203                               dirty, le32_to_cpu(hint_value), thunk->hints_valid);
1204         }
1205
1206         return r;
1207 }
1208
1209 static int __load_mappings(struct dm_cache_metadata *cmd,
1210                            struct dm_cache_policy *policy,
1211                            load_mapping_fn fn, void *context)
1212 {
1213         struct thunk thunk;
1214
1215         thunk.fn = fn;
1216         thunk.context = context;
1217
1218         thunk.cmd = cmd;
1219         thunk.respect_dirty_flags = cmd->clean_when_opened;
1220         thunk.hints_valid = hints_array_available(cmd, policy);
1221
1222         return dm_array_walk(&cmd->info, cmd->root, __load_mapping, &thunk);
1223 }
1224
1225 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1226                            struct dm_cache_policy *policy,
1227                            load_mapping_fn fn, void *context)
1228 {
1229         int r;
1230
1231         READ_LOCK(cmd);
1232         r = __load_mappings(cmd, policy, fn, context);
1233         READ_UNLOCK(cmd);
1234
1235         return r;
1236 }
1237
1238 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
1239 {
1240         int r = 0;
1241         __le64 value;
1242         dm_oblock_t oblock;
1243         unsigned flags;
1244
1245         memcpy(&value, leaf, sizeof(value));
1246         unpack_value(value, &oblock, &flags);
1247
1248         return r;
1249 }
1250
1251 static int __dump_mappings(struct dm_cache_metadata *cmd)
1252 {
1253         return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1254 }
1255
1256 void dm_cache_dump(struct dm_cache_metadata *cmd)
1257 {
1258         READ_LOCK_VOID(cmd);
1259         __dump_mappings(cmd);
1260         READ_UNLOCK(cmd);
1261 }
1262
1263 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1264 {
1265         int r;
1266
1267         READ_LOCK(cmd);
1268         r = cmd->changed;
1269         READ_UNLOCK(cmd);
1270
1271         return r;
1272 }
1273
1274 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1275 {
1276         int r;
1277         unsigned flags;
1278         dm_oblock_t oblock;
1279         __le64 value;
1280
1281         r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1282         if (r)
1283                 return r;
1284
1285         unpack_value(value, &oblock, &flags);
1286
1287         if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1288                 /* nothing to be done */
1289                 return 0;
1290
1291         value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1292         __dm_bless_for_disk(&value);
1293
1294         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1295                                &value, &cmd->root);
1296         if (r)
1297                 return r;
1298
1299         cmd->changed = true;
1300         return 0;
1301
1302 }
1303
1304 int dm_cache_set_dirty(struct dm_cache_metadata *cmd,
1305                        dm_cblock_t cblock, bool dirty)
1306 {
1307         int r;
1308
1309         WRITE_LOCK(cmd);
1310         r = __dirty(cmd, cblock, dirty);
1311         WRITE_UNLOCK(cmd);
1312
1313         return r;
1314 }
1315
1316 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1317                                  struct dm_cache_statistics *stats)
1318 {
1319         READ_LOCK_VOID(cmd);
1320         *stats = cmd->stats;
1321         READ_UNLOCK(cmd);
1322 }
1323
1324 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1325                                  struct dm_cache_statistics *stats)
1326 {
1327         WRITE_LOCK_VOID(cmd);
1328         cmd->stats = *stats;
1329         WRITE_UNLOCK(cmd);
1330 }
1331
1332 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1333 {
1334         int r = -EINVAL;
1335         flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1336                                  clear_clean_shutdown);
1337
1338         WRITE_LOCK(cmd);
1339         if (cmd->fail_io)
1340                 goto out;
1341
1342         r = __commit_transaction(cmd, mutator);
1343         if (r)
1344                 goto out;
1345
1346         r = __begin_transaction(cmd);
1347 out:
1348         WRITE_UNLOCK(cmd);
1349         return r;
1350 }
1351
1352 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1353                                            dm_block_t *result)
1354 {
1355         int r = -EINVAL;
1356
1357         READ_LOCK(cmd);
1358         if (!cmd->fail_io)
1359                 r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1360         READ_UNLOCK(cmd);
1361
1362         return r;
1363 }
1364
1365 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1366                                    dm_block_t *result)
1367 {
1368         int r = -EINVAL;
1369
1370         READ_LOCK(cmd);
1371         if (!cmd->fail_io)
1372                 r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1373         READ_UNLOCK(cmd);
1374
1375         return r;
1376 }
1377
1378 /*----------------------------------------------------------------*/
1379
1380 static int begin_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1381 {
1382         int r;
1383         __le32 value;
1384         size_t hint_size;
1385         const char *policy_name = dm_cache_policy_get_name(policy);
1386         const unsigned *policy_version = dm_cache_policy_get_version(policy);
1387
1388         if (!policy_name[0] ||
1389             (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1390                 return -EINVAL;
1391
1392         if (!policy_unchanged(cmd, policy)) {
1393                 strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1394                 memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1395
1396                 hint_size = dm_cache_policy_get_hint_size(policy);
1397                 if (!hint_size)
1398                         return 0; /* short-circuit hints initialization */
1399                 cmd->policy_hint_size = hint_size;
1400
1401                 if (cmd->hint_root) {
1402                         r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1403                         if (r)
1404                                 return r;
1405                 }
1406
1407                 r = dm_array_empty(&cmd->hint_info, &cmd->hint_root);
1408                 if (r)
1409                         return r;
1410
1411                 value = cpu_to_le32(0);
1412                 __dm_bless_for_disk(&value);
1413                 r = dm_array_resize(&cmd->hint_info, cmd->hint_root, 0,
1414                                     from_cblock(cmd->cache_blocks),
1415                                     &value, &cmd->hint_root);
1416                 if (r)
1417                         return r;
1418         }
1419
1420         return 0;
1421 }
1422
1423 static int save_hint(void *context, dm_cblock_t cblock, dm_oblock_t oblock, uint32_t hint)
1424 {
1425         struct dm_cache_metadata *cmd = context;
1426         __le32 value = cpu_to_le32(hint);
1427         int r;
1428
1429         __dm_bless_for_disk(&value);
1430
1431         r = dm_array_set_value(&cmd->hint_info, cmd->hint_root,
1432                                from_cblock(cblock), &value, &cmd->hint_root);
1433         cmd->changed = true;
1434
1435         return r;
1436 }
1437
1438 static int write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1439 {
1440         int r;
1441
1442         r = begin_hints(cmd, policy);
1443         if (r) {
1444                 DMERR("begin_hints failed");
1445                 return r;
1446         }
1447
1448         return policy_walk_mappings(policy, save_hint, cmd);
1449 }
1450
1451 int dm_cache_write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1452 {
1453         int r;
1454
1455         WRITE_LOCK(cmd);
1456         r = write_hints(cmd, policy);
1457         WRITE_UNLOCK(cmd);
1458
1459         return r;
1460 }
1461
1462 int dm_cache_metadata_all_clean(struct dm_cache_metadata *cmd, bool *result)
1463 {
1464         int r;
1465
1466         READ_LOCK(cmd);
1467         r = blocks_are_unmapped_or_clean(cmd, 0, cmd->cache_blocks, result);
1468         READ_UNLOCK(cmd);
1469
1470         return r;
1471 }
1472
1473 void dm_cache_metadata_set_read_only(struct dm_cache_metadata *cmd)
1474 {
1475         WRITE_LOCK_VOID(cmd);
1476         dm_bm_set_read_only(cmd->bm);
1477         WRITE_UNLOCK(cmd);
1478 }
1479
1480 void dm_cache_metadata_set_read_write(struct dm_cache_metadata *cmd)
1481 {
1482         WRITE_LOCK_VOID(cmd);
1483         dm_bm_set_read_write(cmd->bm);
1484         WRITE_UNLOCK(cmd);
1485 }
1486
1487 int dm_cache_metadata_set_needs_check(struct dm_cache_metadata *cmd)
1488 {
1489         int r;
1490         struct dm_block *sblock;
1491         struct cache_disk_superblock *disk_super;
1492
1493         WRITE_LOCK(cmd);
1494         set_bit(NEEDS_CHECK, &cmd->flags);
1495
1496         r = superblock_lock(cmd, &sblock);
1497         if (r) {
1498                 DMERR("couldn't read superblock");
1499                 goto out;
1500         }
1501
1502         disk_super = dm_block_data(sblock);
1503         disk_super->flags = cpu_to_le32(cmd->flags);
1504
1505         dm_bm_unlock(sblock);
1506
1507 out:
1508         WRITE_UNLOCK(cmd);
1509         return r;
1510 }
1511
1512 int dm_cache_metadata_needs_check(struct dm_cache_metadata *cmd, bool *result)
1513 {
1514         READ_LOCK(cmd);
1515         *result = !!test_bit(NEEDS_CHECK, &cmd->flags);
1516         READ_UNLOCK(cmd);
1517
1518         return 0;
1519 }
1520
1521 int dm_cache_metadata_abort(struct dm_cache_metadata *cmd)
1522 {
1523         int r;
1524
1525         WRITE_LOCK(cmd);
1526         __destroy_persistent_data_objects(cmd);
1527         r = __create_persistent_data_objects(cmd, false);
1528         if (r)
1529                 cmd->fail_io = true;
1530         WRITE_UNLOCK(cmd);
1531
1532         return r;
1533 }