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          * These structures are used when loading metadata.  They're too
 146          * big to put on the stack.
 147          */
 148         struct dm_array_cursor mapping_cursor;
 149         struct dm_array_cursor hint_cursor;
 150 };
 151
 152 /*-------------------------------------------------------------------
 153  * superblock validator
 154  *-----------------------------------------------------------------*/
 155
 156 #define SUPERBLOCK_CSUM_XOR 9031977
 157
 158 static void sb_prepare_for_write(struct dm_block_validator *v,
 159                                  struct dm_block *b,
 160                                  size_t sb_block_size)
 161 {
 162         struct cache_disk_superblock *disk_super = dm_block_data(b);
 163
 164         disk_super->blocknr = cpu_to_le64(dm_block_location(b));
 165         disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
 166                                                       sb_block_size - sizeof(__le32),
 167                                                       SUPERBLOCK_CSUM_XOR));
 168 }
 169
 170 static int check_metadata_version(struct cache_disk_superblock *disk_super)
 171 {
 172         uint32_t metadata_version = le32_to_cpu(disk_super->version);
 173         if (metadata_version < MIN_CACHE_VERSION || metadata_version > MAX_CACHE_VERSION) {
 174                 DMERR("Cache metadata version %u found, but only versions between %u and %u supported.",
 175                       metadata_version, MIN_CACHE_VERSION, MAX_CACHE_VERSION);
 176                 return -EINVAL;
 177         }
 178
 179         return 0;
 180 }
 181
 182 static int sb_check(struct dm_block_validator *v,
 183                     struct dm_block *b,
 184                     size_t sb_block_size)
 185 {
 186         struct cache_disk_superblock *disk_super = dm_block_data(b);
 187         __le32 csum_le;
 188
 189         if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
 190                 DMERR("sb_check failed: blocknr %llu: wanted %llu",
 191                       le64_to_cpu(disk_super->blocknr),
 192                       (unsigned long long)dm_block_location(b));
 193                 return -ENOTBLK;
 194         }
 195
 196         if (le64_to_cpu(disk_super->magic) != CACHE_SUPERBLOCK_MAGIC) {
 197                 DMERR("sb_check failed: magic %llu: wanted %llu",
 198                       le64_to_cpu(disk_super->magic),
 199                       (unsigned long long)CACHE_SUPERBLOCK_MAGIC);
 200                 return -EILSEQ;
 201         }
 202
 203         csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
 204                                              sb_block_size - sizeof(__le32),
 205                                              SUPERBLOCK_CSUM_XOR));
 206         if (csum_le != disk_super->csum) {
 207                 DMERR("sb_check failed: csum %u: wanted %u",
 208                       le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
 209                 return -EILSEQ;
 210         }
 211
 212         return check_metadata_version(disk_super);
 213 }
 214
 215 static struct dm_block_validator sb_validator = {
 216         .name = "superblock",
 217         .prepare_for_write = sb_prepare_for_write,
 218         .check = sb_check
 219 };
 220
 221 /*----------------------------------------------------------------*/
 222
 223 static int superblock_read_lock(struct dm_cache_metadata *cmd,
 224                                 struct dm_block **sblock)
 225 {
 226         return dm_bm_read_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 227                                &sb_validator, sblock);
 228 }
 229
 230 static int superblock_lock_zero(struct dm_cache_metadata *cmd,
 231                                 struct dm_block **sblock)
 232 {
 233         return dm_bm_write_lock_zero(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 234                                      &sb_validator, sblock);
 235 }
 236
 237 static int superblock_lock(struct dm_cache_metadata *cmd,
 238                            struct dm_block **sblock)
 239 {
 240         return dm_bm_write_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 241                                 &sb_validator, sblock);
 242 }
 243
 244 /*----------------------------------------------------------------*/
 245
 246 static int __superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
 247 {
 248         int r;
 249         unsigned i;
 250         struct dm_block *b;
 251         __le64 *data_le, zero = cpu_to_le64(0);
 252         unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
 253
 254         /*
 255          * We can't use a validator here - it may be all zeroes.
 256          */
 257         r = dm_bm_read_lock(bm, CACHE_SUPERBLOCK_LOCATION, NULL, &b);
 258         if (r)
 259                 return r;
 260
 261         data_le = dm_block_data(b);
 262         *result = true;
 263         for (i = 0; i < sb_block_size; i++) {
 264                 if (data_le[i] != zero) {
 265                         *result = false;
 266                         break;
 267                 }
 268         }
 269
 270         dm_bm_unlock(b);
 271
 272         return 0;
 273 }
 274
 275 static void __setup_mapping_info(struct dm_cache_metadata *cmd)
 276 {
 277         struct dm_btree_value_type vt;
 278
 279         vt.context = NULL;
 280         vt.size = sizeof(__le64);
 281         vt.inc = NULL;
 282         vt.dec = NULL;
 283         vt.equal = NULL;
 284         dm_array_info_init(&cmd->info, cmd->tm, &vt);
 285
 286         if (cmd->policy_hint_size) {
 287                 vt.size = sizeof(__le32);
 288                 dm_array_info_init(&cmd->hint_info, cmd->tm, &vt);
 289         }
 290 }
 291
 292 static int __save_sm_root(struct dm_cache_metadata *cmd)
 293 {
 294         int r;
 295         size_t metadata_len;
 296
 297         r = dm_sm_root_size(cmd->metadata_sm, &metadata_len);
 298         if (r < 0)
 299                 return r;
 300
 301         return dm_sm_copy_root(cmd->metadata_sm, &cmd->metadata_space_map_root,
 302                                metadata_len);
 303 }
 304
 305 static void __copy_sm_root(struct dm_cache_metadata *cmd,
 306                            struct cache_disk_superblock *disk_super)
 307 {
 308         memcpy(&disk_super->metadata_space_map_root,
 309                &cmd->metadata_space_map_root,
 310                sizeof(cmd->metadata_space_map_root));
 311 }
 312
 313 static int __write_initial_superblock(struct dm_cache_metadata *cmd)
 314 {
 315         int r;
 316         struct dm_block *sblock;
 317         struct cache_disk_superblock *disk_super;
 318         sector_t bdev_size = i_size_read(cmd->bdev->bd_inode) >> SECTOR_SHIFT;
 319
 320         /* FIXME: see if we can lose the max sectors limit */
 321         if (bdev_size > DM_CACHE_METADATA_MAX_SECTORS)
 322                 bdev_size = DM_CACHE_METADATA_MAX_SECTORS;
 323
 324         r = dm_tm_pre_commit(cmd->tm);
 325         if (r < 0)
 326                 return r;
 327
 328         /*
 329          * dm_sm_copy_root() can fail.  So we need to do it before we start
 330          * updating the superblock.
 331          */
 332         r = __save_sm_root(cmd);
 333         if (r)
 334                 return r;
 335
 336         r = superblock_lock_zero(cmd, &sblock);
 337         if (r)
 338                 return r;
 339
 340         disk_super = dm_block_data(sblock);
 341         disk_super->flags = 0;
 342         memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
 343         disk_super->magic = cpu_to_le64(CACHE_SUPERBLOCK_MAGIC);
 344         disk_super->version = cpu_to_le32(MAX_CACHE_VERSION);
 345         memset(disk_super->policy_name, 0, sizeof(disk_super->policy_name));
 346         memset(disk_super->policy_version, 0, sizeof(disk_super->policy_version));
 347         disk_super->policy_hint_size = 0;
 348
 349         __copy_sm_root(cmd, disk_super);
 350
 351         disk_super->mapping_root = cpu_to_le64(cmd->root);
 352         disk_super->hint_root = cpu_to_le64(cmd->hint_root);
 353         disk_super->discard_root = cpu_to_le64(cmd->discard_root);
 354         disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
 355         disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
 356         disk_super->metadata_block_size = cpu_to_le32(DM_CACHE_METADATA_BLOCK_SIZE);
 357         disk_super->data_block_size = cpu_to_le32(cmd->data_block_size);
 358         disk_super->cache_blocks = cpu_to_le32(0);
 359
 360         disk_super->read_hits = cpu_to_le32(0);
 361         disk_super->read_misses = cpu_to_le32(0);
 362         disk_super->write_hits = cpu_to_le32(0);
 363         disk_super->write_misses = cpu_to_le32(0);
 364
 365         return dm_tm_commit(cmd->tm, sblock);
 366 }
 367
 368 static int __format_metadata(struct dm_cache_metadata *cmd)
 369 {
 370         int r;
 371
 372         r = dm_tm_create_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 373                                  &cmd->tm, &cmd->metadata_sm);
 374         if (r < 0) {
 375                 DMERR("tm_create_with_sm failed");
 376                 return r;
 377         }
 378
 379         __setup_mapping_info(cmd);
 380
 381         r = dm_array_empty(&cmd->info, &cmd->root);
 382         if (r < 0)
 383                 goto bad;
 384
 385         dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
 386         r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root);
 387         if (r < 0)
 388                 goto bad;
 389
 390         cmd->discard_block_size = 0;
 391         cmd->discard_nr_blocks = 0;
 392
 393         r = __write_initial_superblock(cmd);
 394         if (r)
 395                 goto bad;
 396
 397         cmd->clean_when_opened = true;
 398         return 0;
 399
 400 bad:
 401         dm_tm_destroy(cmd->tm);
 402         dm_sm_destroy(cmd->metadata_sm);
 403
 404         return r;
 405 }
 406
 407 static int __check_incompat_features(struct cache_disk_superblock *disk_super,
 408                                      struct dm_cache_metadata *cmd)
 409 {
 410         uint32_t features;
 411
 412         features = le32_to_cpu(disk_super->incompat_flags) & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
 413         if (features) {
 414                 DMERR("could not access metadata due to unsupported optional features (%lx).",
 415                       (unsigned long)features);
 416                 return -EINVAL;
 417         }
 418
 419         /*
 420          * Check for read-only metadata to skip the following RDWR checks.
 421          */
 422         if (get_disk_ro(cmd->bdev->bd_disk))
 423                 return 0;
 424
 425         features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP;
 426         if (features) {
 427                 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
 428                       (unsigned long)features);
 429                 return -EINVAL;
 430         }
 431
 432         return 0;
 433 }
 434
 435 static int __open_metadata(struct dm_cache_metadata *cmd)
 436 {
 437         int r;
 438         struct dm_block *sblock;
 439         struct cache_disk_superblock *disk_super;
 440         unsigned long sb_flags;
 441
 442         r = superblock_read_lock(cmd, &sblock);
 443         if (r < 0) {
 444                 DMERR("couldn't read lock superblock");
 445                 return r;
 446         }
 447
 448         disk_super = dm_block_data(sblock);
 449
 450         /* Verify the data block size hasn't changed */
 451         if (le32_to_cpu(disk_super->data_block_size) != cmd->data_block_size) {
 452                 DMERR("changing the data block size (from %u to %llu) is not supported",
 453                       le32_to_cpu(disk_super->data_block_size),
 454                       (unsigned long long)cmd->data_block_size);
 455                 r = -EINVAL;
 456                 goto bad;
 457         }
 458
 459         r = __check_incompat_features(disk_super, cmd);
 460         if (r < 0)
 461                 goto bad;
 462
 463         r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 464                                disk_super->metadata_space_map_root,
 465                                sizeof(disk_super->metadata_space_map_root),
 466                                &cmd->tm, &cmd->metadata_sm);
 467         if (r < 0) {
 468                 DMERR("tm_open_with_sm failed");
 469                 goto bad;
 470         }
 471
 472         __setup_mapping_info(cmd);
 473         dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
 474         sb_flags = le32_to_cpu(disk_super->flags);
 475         cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags);
 476         dm_bm_unlock(sblock);
 477
 478         return 0;
 479
 480 bad:
 481         dm_bm_unlock(sblock);
 482         return r;
 483 }
 484
 485 static int __open_or_format_metadata(struct dm_cache_metadata *cmd,
 486                                      bool format_device)
 487 {
 488         int r;
 489         bool unformatted = false;
 490
 491         r = __superblock_all_zeroes(cmd->bm, &unformatted);
 492         if (r)
 493                 return r;
 494
 495         if (unformatted)
 496                 return format_device ? __format_metadata(cmd) : -EPERM;
 497
 498         return __open_metadata(cmd);
 499 }
 500
 501 static int __create_persistent_data_objects(struct dm_cache_metadata *cmd,
 502                                             bool may_format_device)
 503 {
 504         int r;
 505         cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
 506                                           CACHE_METADATA_CACHE_SIZE,
 507                                           CACHE_MAX_CONCURRENT_LOCKS);
 508         if (IS_ERR(cmd->bm)) {
 509                 DMERR("could not create block manager");
 510                 return PTR_ERR(cmd->bm);
 511         }
 512
 513         r = __open_or_format_metadata(cmd, may_format_device);
 514         if (r)
 515                 dm_block_manager_destroy(cmd->bm);
 516
 517         return r;
 518 }
 519
 520 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd)
 521 {
 522         dm_sm_destroy(cmd->metadata_sm);
 523         dm_tm_destroy(cmd->tm);
 524         dm_block_manager_destroy(cmd->bm);
 525 }
 526
 527 typedef unsigned long (*flags_mutator)(unsigned long);
 528
 529 static void update_flags(struct cache_disk_superblock *disk_super,
 530                          flags_mutator mutator)
 531 {
 532         uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
 533         disk_super->flags = cpu_to_le32(sb_flags);
 534 }
 535
 536 static unsigned long set_clean_shutdown(unsigned long flags)
 537 {
 538         set_bit(CLEAN_SHUTDOWN, &flags);
 539         return flags;
 540 }
 541
 542 static unsigned long clear_clean_shutdown(unsigned long flags)
 543 {
 544         clear_bit(CLEAN_SHUTDOWN, &flags);
 545         return flags;
 546 }
 547
 548 static void read_superblock_fields(struct dm_cache_metadata *cmd,
 549                                    struct cache_disk_superblock *disk_super)
 550 {
 551         cmd->flags = le32_to_cpu(disk_super->flags);
 552         cmd->root = le64_to_cpu(disk_super->mapping_root);
 553         cmd->hint_root = le64_to_cpu(disk_super->hint_root);
 554         cmd->discard_root = le64_to_cpu(disk_super->discard_root);
 555         cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
 556         cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
 557         cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
 558         cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
 559         strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
 560         cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
 561         cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
 562         cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
 563         cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
 564
 565         cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
 566         cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
 567         cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
 568         cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
 569
 570         cmd->changed = false;
 571 }
 572
 573 /*
 574  * The mutator updates the superblock flags.
 575  */
 576 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
 577                                      flags_mutator mutator)
 578 {
 579         int r;
 580         struct cache_disk_superblock *disk_super;
 581         struct dm_block *sblock;
 582
 583         r = superblock_lock(cmd, &sblock);
 584         if (r)
 585                 return r;
 586
 587         disk_super = dm_block_data(sblock);
 588         update_flags(disk_super, mutator);
 589         read_superblock_fields(cmd, disk_super);
 590         dm_bm_unlock(sblock);
 591
 592         return dm_bm_flush(cmd->bm);
 593 }
 594
 595 static int __begin_transaction(struct dm_cache_metadata *cmd)
 596 {
 597         int r;
 598         struct cache_disk_superblock *disk_super;
 599         struct dm_block *sblock;
 600
 601         /*
 602          * We re-read the superblock every time.  Shouldn't need to do this
 603          * really.
 604          */
 605         r = superblock_read_lock(cmd, &sblock);
 606         if (r)
 607                 return r;
 608
 609         disk_super = dm_block_data(sblock);
 610         read_superblock_fields(cmd, disk_super);
 611         dm_bm_unlock(sblock);
 612
 613         return 0;
 614 }
 615
 616 static int __commit_transaction(struct dm_cache_metadata *cmd,
 617                                 flags_mutator mutator)
 618 {
 619         int r;
 620         struct cache_disk_superblock *disk_super;
 621         struct dm_block *sblock;
 622
 623         /*
 624          * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
 625          */
 626         BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
 627
 628         r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
 629                             &cmd->discard_root);
 630         if (r)
 631                 return r;
 632
 633         r = dm_tm_pre_commit(cmd->tm);
 634         if (r < 0)
 635                 return r;
 636
 637         r = __save_sm_root(cmd);
 638         if (r)
 639                 return r;
 640
 641         r = superblock_lock(cmd, &sblock);
 642         if (r)
 643                 return r;
 644
 645         disk_super = dm_block_data(sblock);
 646
 647         disk_super->flags = cpu_to_le32(cmd->flags);
 648         if (mutator)
 649                 update_flags(disk_super, mutator);
 650
 651         disk_super->mapping_root = cpu_to_le64(cmd->root);
 652         disk_super->hint_root = cpu_to_le64(cmd->hint_root);
 653         disk_super->discard_root = cpu_to_le64(cmd->discard_root);
 654         disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
 655         disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
 656         disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
 657         strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
 658         disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
 659         disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
 660         disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
 661
 662         disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
 663         disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
 664         disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
 665         disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
 666         __copy_sm_root(cmd, disk_super);
 667
 668         return dm_tm_commit(cmd->tm, sblock);
 669 }
 670
 671 /*----------------------------------------------------------------*/
 672
 673 /*
 674  * The mappings are held in a dm-array that has 64-bit values stored in
 675  * little-endian format.  The index is the cblock, the high 48bits of the
 676  * value are the oblock and the low 16 bit the flags.
 677  */
 678 #define FLAGS_MASK ((1 << 16) - 1)
 679
 680 static __le64 pack_value(dm_oblock_t block, unsigned flags)
 681 {
 682         uint64_t value = from_oblock(block);
 683         value <<= 16;
 684         value = value | (flags & FLAGS_MASK);
 685         return cpu_to_le64(value);
 686 }
 687
 688 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
 689 {
 690         uint64_t value = le64_to_cpu(value_le);
 691         uint64_t b = value >> 16;
 692         *block = to_oblock(b);
 693         *flags = value & FLAGS_MASK;
 694 }
 695
 696 /*----------------------------------------------------------------*/
 697
 698 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
 699                                                sector_t data_block_size,
 700                                                bool may_format_device,
 701                                                size_t policy_hint_size)
 702 {
 703         int r;
 704         struct dm_cache_metadata *cmd;
 705
 706         cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
 707         if (!cmd) {
 708                 DMERR("could not allocate metadata struct");
 709                 return ERR_PTR(-ENOMEM);
 710         }
 711
 712         atomic_set(&cmd->ref_count, 1);
 713         init_rwsem(&cmd->root_lock);
 714         cmd->bdev = bdev;
 715         cmd->data_block_size = data_block_size;
 716         cmd->cache_blocks = 0;
 717         cmd->policy_hint_size = policy_hint_size;
 718         cmd->changed = true;
 719         cmd->fail_io = false;
 720
 721         r = __create_persistent_data_objects(cmd, may_format_device);
 722         if (r) {
 723                 kfree(cmd);
 724                 return ERR_PTR(r);
 725         }
 726
 727         r = __begin_transaction_flags(cmd, clear_clean_shutdown);
 728         if (r < 0) {
 729                 dm_cache_metadata_close(cmd);
 730                 return ERR_PTR(r);
 731         }
 732
 733         return cmd;
 734 }
 735
 736 /*
 737  * We keep a little list of ref counted metadata objects to prevent two
 738  * different target instances creating separate bufio instances.  This is
 739  * an issue if a table is reloaded before the suspend.
 740  */
 741 static DEFINE_MUTEX(table_lock);
 742 static LIST_HEAD(table);
 743
 744 static struct dm_cache_metadata *lookup(struct block_device *bdev)
 745 {
 746         struct dm_cache_metadata *cmd;
 747
 748         list_for_each_entry(cmd, &table, list)
 749                 if (cmd->bdev == bdev) {
 750                         atomic_inc(&cmd->ref_count);
 751                         return cmd;
 752                 }
 753
 754         return NULL;
 755 }
 756
 757 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
 758                                                 sector_t data_block_size,
 759                                                 bool may_format_device,
 760                                                 size_t policy_hint_size)
 761 {
 762         struct dm_cache_metadata *cmd, *cmd2;
 763
 764         mutex_lock(&table_lock);
 765         cmd = lookup(bdev);
 766         mutex_unlock(&table_lock);
 767
 768         if (cmd)
 769                 return cmd;
 770
 771         cmd = metadata_open(bdev, data_block_size, may_format_device, policy_hint_size);
 772         if (!IS_ERR(cmd)) {
 773                 mutex_lock(&table_lock);
 774                 cmd2 = lookup(bdev);
 775                 if (cmd2) {
 776                         mutex_unlock(&table_lock);
 777                         __destroy_persistent_data_objects(cmd);
 778                         kfree(cmd);
 779                         return cmd2;
 780                 }
 781                 list_add(&cmd->list, &table);
 782                 mutex_unlock(&table_lock);
 783         }
 784
 785         return cmd;
 786 }
 787
 788 static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
 789 {
 790         if (cmd->data_block_size != data_block_size) {
 791                 DMERR("data_block_size (%llu) different from that in metadata (%llu)",
 792                       (unsigned long long) data_block_size,
 793                       (unsigned long long) cmd->data_block_size);
 794                 return false;
 795         }
 796
 797         return true;
 798 }
 799
 800 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
 801                                                  sector_t data_block_size,
 802                                                  bool may_format_device,
 803                                                  size_t policy_hint_size)
 804 {
 805         struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size,
 806                                                        may_format_device, policy_hint_size);
 807
 808         if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
 809                 dm_cache_metadata_close(cmd);
 810                 return ERR_PTR(-EINVAL);
 811         }
 812
 813         return cmd;
 814 }
 815
 816 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
 817 {
 818         if (atomic_dec_and_test(&cmd->ref_count)) {
 819                 mutex_lock(&table_lock);
 820                 list_del(&cmd->list);
 821                 mutex_unlock(&table_lock);
 822
 823                 if (!cmd->fail_io)
 824                         __destroy_persistent_data_objects(cmd);
 825                 kfree(cmd);
 826         }
 827 }
 828
 829 /*
 830  * Checks that the given cache block is either unmapped or clean.
 831  */
 832 static int block_unmapped_or_clean(struct dm_cache_metadata *cmd, dm_cblock_t b,
 833                                    bool *result)
 834 {
 835         int r;
 836         __le64 value;
 837         dm_oblock_t ob;
 838         unsigned flags;
 839
 840         r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
 841         if (r) {
 842                 DMERR("block_unmapped_or_clean failed");
 843                 return r;
 844         }
 845
 846         unpack_value(value, &ob, &flags);
 847         *result = !((flags & M_VALID) && (flags & M_DIRTY));
 848
 849         return 0;
 850 }
 851
 852 static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
 853                                         dm_cblock_t begin, dm_cblock_t end,
 854                                         bool *result)
 855 {
 856         int r;
 857         *result = true;
 858
 859         while (begin != end) {
 860                 r = block_unmapped_or_clean(cmd, begin, result);
 861                 if (r)
 862                         return r;
 863
 864                 if (!*result) {
 865                         DMERR("cache block %llu is dirty",
 866                               (unsigned long long) from_cblock(begin));
 867                         return 0;
 868                 }
 869
 870                 begin = to_cblock(from_cblock(begin) + 1);
 871         }
 872
 873         return 0;
 874 }
 875
 876 static bool cmd_write_lock(struct dm_cache_metadata *cmd)
 877 {
 878         down_write(&cmd->root_lock);
 879         if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) {
 880                 up_write(&cmd->root_lock);
 881                 return false;
 882         }
 883         return true;
 884 }
 885
 886 #define WRITE_LOCK(cmd)                         \
 887         do {                                    \
 888                 if (!cmd_write_lock((cmd)))     \
 889                         return -EINVAL;         \
 890         } while(0)
 891
 892 #define WRITE_LOCK_VOID(cmd)                    \
 893         do {                                    \
 894                 if (!cmd_write_lock((cmd)))     \
 895                         return;                 \
 896         } while(0)
 897
 898 #define WRITE_UNLOCK(cmd) \
 899         up_write(&(cmd)->root_lock)
 900
 901 static bool cmd_read_lock(struct dm_cache_metadata *cmd)
 902 {
 903         down_read(&cmd->root_lock);
 904         if (cmd->fail_io) {
 905                 up_read(&cmd->root_lock);
 906                 return false;
 907         }
 908         return true;
 909 }
 910
 911 #define READ_LOCK(cmd)                          \
 912         do {                                    \
 913                 if (!cmd_read_lock((cmd)))      \
 914                         return -EINVAL;         \
 915         } while(0)
 916
 917 #define READ_LOCK_VOID(cmd)                     \
 918         do {                                    \
 919                 if (!cmd_read_lock((cmd)))      \
 920                         return;                 \
 921         } while(0)
 922
 923 #define READ_UNLOCK(cmd) \
 924         up_read(&(cmd)->root_lock)
 925
 926 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
 927 {
 928         int r;
 929         bool clean;
 930         __le64 null_mapping = pack_value(0, 0);
 931
 932         WRITE_LOCK(cmd);
 933         __dm_bless_for_disk(&null_mapping);
 934
 935         if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) {
 936                 r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean);
 937                 if (r) {
 938                         __dm_unbless_for_disk(&null_mapping);
 939                         goto out;
 940                 }
 941
 942                 if (!clean) {
 943                         DMERR("unable to shrink cache due to dirty blocks");
 944                         r = -EINVAL;
 945                         __dm_unbless_for_disk(&null_mapping);
 946                         goto out;
 947                 }
 948         }
 949
 950         r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
 951                             from_cblock(new_cache_size),
 952                             &null_mapping, &cmd->root);
 953         if (!r)
 954                 cmd->cache_blocks = new_cache_size;
 955         cmd->changed = true;
 956
 957 out:
 958         WRITE_UNLOCK(cmd);
 959
 960         return r;
 961 }
 962
 963 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
 964                                    sector_t discard_block_size,
 965                                    dm_dblock_t new_nr_entries)
 966 {
 967         int r;
 968
 969         WRITE_LOCK(cmd);
 970         r = dm_bitset_resize(&cmd->discard_info,
 971                              cmd->discard_root,
 972                              from_dblock(cmd->discard_nr_blocks),
 973                              from_dblock(new_nr_entries),
 974                              false, &cmd->discard_root);
 975         if (!r) {
 976                 cmd->discard_block_size = discard_block_size;
 977                 cmd->discard_nr_blocks = new_nr_entries;
 978         }
 979
 980         cmd->changed = true;
 981         WRITE_UNLOCK(cmd);
 982
 983         return r;
 984 }
 985
 986 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
 987 {
 988         return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
 989                                  from_dblock(b), &cmd->discard_root);
 990 }
 991
 992 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
 993 {
 994         return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
 995                                    from_dblock(b), &cmd->discard_root);
 996 }
 997
 998 static int __is_discarded(struct dm_cache_metadata *cmd, dm_dblock_t b,
 999                           bool *is_discarded)
1000 {
1001         return dm_bitset_test_bit(&cmd->discard_info, cmd->discard_root,
1002                                   from_dblock(b), &cmd->discard_root,
1003                                   is_discarded);
1004 }
1005
1006 static int __discard(struct dm_cache_metadata *cmd,
1007                      dm_dblock_t dblock, bool discard)
1008 {
1009         int r;
1010
1011         r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
1012         if (r)
1013                 return r;
1014
1015         cmd->changed = true;
1016         return 0;
1017 }
1018
1019 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
1020                          dm_dblock_t dblock, bool discard)
1021 {
1022         int r;
1023
1024         WRITE_LOCK(cmd);
1025         r = __discard(cmd, dblock, discard);
1026         WRITE_UNLOCK(cmd);
1027
1028         return r;
1029 }
1030
1031 static int __load_discards(struct dm_cache_metadata *cmd,
1032                            load_discard_fn fn, void *context)
1033 {
1034         int r = 0;
1035         dm_block_t b;
1036         bool discard;
1037
1038         for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1039                 dm_dblock_t dblock = to_dblock(b);
1040
1041                 if (cmd->clean_when_opened) {
1042                         r = __is_discarded(cmd, dblock, &discard);
1043                         if (r)
1044                                 return r;
1045                 } else
1046                         discard = false;
1047
1048                 r = fn(context, cmd->discard_block_size, dblock, discard);
1049                 if (r)
1050                         break;
1051         }
1052
1053         return r;
1054 }
1055
1056 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
1057                            load_discard_fn fn, void *context)
1058 {
1059         int r;
1060
1061         READ_LOCK(cmd);
1062         r = __load_discards(cmd, fn, context);
1063         READ_UNLOCK(cmd);
1064
1065         return r;
1066 }
1067
1068 int dm_cache_size(struct dm_cache_metadata *cmd, dm_cblock_t *result)
1069 {
1070         READ_LOCK(cmd);
1071         *result = cmd->cache_blocks;
1072         READ_UNLOCK(cmd);
1073
1074         return 0;
1075 }
1076
1077 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1078 {
1079         int r;
1080         __le64 value = pack_value(0, 0);
1081
1082         __dm_bless_for_disk(&value);
1083         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1084                                &value, &cmd->root);
1085         if (r)
1086                 return r;
1087
1088         cmd->changed = true;
1089         return 0;
1090 }
1091
1092 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1093 {
1094         int r;
1095
1096         WRITE_LOCK(cmd);
1097         r = __remove(cmd, cblock);
1098         WRITE_UNLOCK(cmd);
1099
1100         return r;
1101 }
1102
1103 static int __insert(struct dm_cache_metadata *cmd,
1104                     dm_cblock_t cblock, dm_oblock_t oblock)
1105 {
1106         int r;
1107         __le64 value = pack_value(oblock, M_VALID);
1108         __dm_bless_for_disk(&value);
1109
1110         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1111                                &value, &cmd->root);
1112         if (r)
1113                 return r;
1114
1115         cmd->changed = true;
1116         return 0;
1117 }
1118
1119 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
1120                             dm_cblock_t cblock, dm_oblock_t oblock)
1121 {
1122         int r;
1123
1124         WRITE_LOCK(cmd);
1125         r = __insert(cmd, cblock, oblock);
1126         WRITE_UNLOCK(cmd);
1127
1128         return r;
1129 }
1130
1131 struct thunk {
1132         load_mapping_fn fn;
1133         void *context;
1134
1135         struct dm_cache_metadata *cmd;
1136         bool respect_dirty_flags;
1137         bool hints_valid;
1138 };
1139
1140 static bool policy_unchanged(struct dm_cache_metadata *cmd,
1141                              struct dm_cache_policy *policy)
1142 {
1143         const char *policy_name = dm_cache_policy_get_name(policy);
1144         const unsigned *policy_version = dm_cache_policy_get_version(policy);
1145         size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
1146
1147         /*
1148          * Ensure policy names match.
1149          */
1150         if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
1151                 return false;
1152
1153         /*
1154          * Ensure policy major versions match.
1155          */
1156         if (cmd->policy_version[0] != policy_version[0])
1157                 return false;
1158
1159         /*
1160          * Ensure policy hint sizes match.
1161          */
1162         if (cmd->policy_hint_size != policy_hint_size)
1163                 return false;
1164
1165         return true;
1166 }
1167
1168 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1169 {
1170         return cmd->hint_root && cmd->policy_hint_size;
1171 }
1172
1173 static bool hints_array_available(struct dm_cache_metadata *cmd,
1174                                   struct dm_cache_policy *policy)
1175 {
1176         return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1177                 hints_array_initialized(cmd);
1178 }
1179
1180 static int __load_mapping(struct dm_cache_metadata *cmd,
1181                           uint64_t cb, bool hints_valid,
1182                           struct dm_array_cursor *mapping_cursor,
1183                           struct dm_array_cursor *hint_cursor,
1184                           load_mapping_fn fn, void *context)
1185 {
1186         int r = 0;
1187
1188         __le64 mapping;
1189         __le32 hint = 0;
1190
1191         __le64 *mapping_value_le;
1192         __le32 *hint_value_le;
1193
1194         dm_oblock_t oblock;
1195         unsigned flags;
1196
1197         dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1198         memcpy(&mapping, mapping_value_le, sizeof(mapping));
1199         unpack_value(mapping, &oblock, &flags);
1200
1201         if (flags & M_VALID) {
1202                 if (hints_valid) {
1203                         dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1204                         memcpy(&hint, hint_value_le, sizeof(hint));
1205                 }
1206
1207                 r = fn(context, oblock, to_cblock(cb), flags & M_DIRTY,
1208                        le32_to_cpu(hint), hints_valid);
1209                 if (r)
1210                         DMERR("policy couldn't load cblock");
1211         }
1212
1213         return r;
1214 }
1215
1216 static int __load_mappings(struct dm_cache_metadata *cmd,
1217                            struct dm_cache_policy *policy,
1218                            load_mapping_fn fn, void *context)
1219 {
1220         int r;
1221         uint64_t cb;
1222
1223         bool hints_valid = hints_array_available(cmd, policy);
1224
1225         if (from_cblock(cmd->cache_blocks) == 0)
1226                 /* Nothing to do */
1227                 return 0;
1228
1229         r = dm_array_cursor_begin(&cmd->info, cmd->root, &cmd->mapping_cursor);
1230         if (r)
1231                 return r;
1232
1233         if (hints_valid) {
1234                 r = dm_array_cursor_begin(&cmd->hint_info, cmd->hint_root, &cmd->hint_cursor);
1235                 if (r) {
1236                         dm_array_cursor_end(&cmd->mapping_cursor);
1237                         return r;
1238                 }
1239         }
1240
1241         for (cb = 0; ; cb++) {
1242                 r = __load_mapping(cmd, cb, hints_valid,
1243                                    &cmd->mapping_cursor, &cmd->hint_cursor,
1244                                    fn, context);
1245                 if (r)
1246                         goto out;
1247
1248                 /*
1249                  * We need to break out before we move the cursors.
1250                  */
1251                 if (cb >= (from_cblock(cmd->cache_blocks) - 1))
1252                         break;
1253
1254                 r = dm_array_cursor_next(&cmd->mapping_cursor);
1255                 if (r) {
1256                         DMERR("dm_array_cursor_next for mapping failed");
1257                         goto out;
1258                 }
1259
1260                 if (hints_valid) {
1261                         r = dm_array_cursor_next(&cmd->hint_cursor);
1262                         if (r) {
1263                                 DMERR("dm_array_cursor_next for hint failed");
1264                                 goto out;
1265                         }
1266                 }
1267         }
1268 out:
1269         dm_array_cursor_end(&cmd->mapping_cursor);
1270         if (hints_valid)
1271                 dm_array_cursor_end(&cmd->hint_cursor);
1272
1273         return r;
1274 }
1275
1276 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1277                            struct dm_cache_policy *policy,
1278                            load_mapping_fn fn, void *context)
1279 {
1280         int r;
1281
1282         READ_LOCK(cmd);
1283         r = __load_mappings(cmd, policy, fn, context);
1284         READ_UNLOCK(cmd);
1285
1286         return r;
1287 }
1288
1289 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
1290 {
1291         int r = 0;
1292         __le64 value;
1293         dm_oblock_t oblock;
1294         unsigned flags;
1295
1296         memcpy(&value, leaf, sizeof(value));
1297         unpack_value(value, &oblock, &flags);
1298
1299         return r;
1300 }
1301
1302 static int __dump_mappings(struct dm_cache_metadata *cmd)
1303 {
1304         return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1305 }
1306
1307 void dm_cache_dump(struct dm_cache_metadata *cmd)
1308 {
1309         READ_LOCK_VOID(cmd);
1310         __dump_mappings(cmd);
1311         READ_UNLOCK(cmd);
1312 }
1313
1314 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1315 {
1316         int r;
1317
1318         READ_LOCK(cmd);
1319         r = cmd->changed;
1320         READ_UNLOCK(cmd);
1321
1322         return r;
1323 }
1324
1325 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1326 {
1327         int r;
1328         unsigned flags;
1329         dm_oblock_t oblock;
1330         __le64 value;
1331
1332         r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1333         if (r)
1334                 return r;
1335
1336         unpack_value(value, &oblock, &flags);
1337
1338         if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1339                 /* nothing to be done */
1340                 return 0;
1341
1342         value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1343         __dm_bless_for_disk(&value);
1344
1345         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1346                                &value, &cmd->root);
1347         if (r)
1348                 return r;
1349
1350         cmd->changed = true;
1351         return 0;
1352
1353 }
1354
1355 int dm_cache_set_dirty(struct dm_cache_metadata *cmd,
1356                        dm_cblock_t cblock, bool dirty)
1357 {
1358         int r;
1359
1360         WRITE_LOCK(cmd);
1361         r = __dirty(cmd, cblock, dirty);
1362         WRITE_UNLOCK(cmd);
1363
1364         return r;
1365 }
1366
1367 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1368                                  struct dm_cache_statistics *stats)
1369 {
1370         READ_LOCK_VOID(cmd);
1371         *stats = cmd->stats;
1372         READ_UNLOCK(cmd);
1373 }
1374
1375 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1376                                  struct dm_cache_statistics *stats)
1377 {
1378         WRITE_LOCK_VOID(cmd);
1379         cmd->stats = *stats;
1380         WRITE_UNLOCK(cmd);
1381 }
1382
1383 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1384 {
1385         int r;
1386         flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1387                                  clear_clean_shutdown);
1388
1389         WRITE_LOCK(cmd);
1390         r = __commit_transaction(cmd, mutator);
1391         if (r)
1392                 goto out;
1393
1394         r = __begin_transaction(cmd);
1395
1396 out:
1397         WRITE_UNLOCK(cmd);
1398         return r;
1399 }
1400
1401 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1402                                            dm_block_t *result)
1403 {
1404         int r = -EINVAL;
1405
1406         READ_LOCK(cmd);
1407         r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1408         READ_UNLOCK(cmd);
1409
1410         return r;
1411 }
1412
1413 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1414                                    dm_block_t *result)
1415 {
1416         int r = -EINVAL;
1417
1418         READ_LOCK(cmd);
1419         r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1420         READ_UNLOCK(cmd);
1421
1422         return r;
1423 }
1424
1425 /*----------------------------------------------------------------*/
1426
1427 static int get_hint(uint32_t index, void *value_le, void *context)
1428 {
1429         uint32_t value;
1430         struct dm_cache_policy *policy = context;
1431
1432         value = policy_get_hint(policy, to_cblock(index));
1433         *((__le32 *) value_le) = cpu_to_le32(value);
1434
1435         return 0;
1436 }
1437
1438 /*
1439  * It's quicker to always delete the hint array, and recreate with
1440  * dm_array_new().
1441  */
1442 static int write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1443 {
1444         int r;
1445         size_t hint_size;
1446         const char *policy_name = dm_cache_policy_get_name(policy);
1447         const unsigned *policy_version = dm_cache_policy_get_version(policy);
1448
1449         if (!policy_name[0] ||
1450             (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1451                 return -EINVAL;
1452
1453         strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1454         memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1455
1456         hint_size = dm_cache_policy_get_hint_size(policy);
1457         if (!hint_size)
1458                 return 0; /* short-circuit hints initialization */
1459         cmd->policy_hint_size = hint_size;
1460
1461         if (cmd->hint_root) {
1462                 r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1463                 if (r)
1464                         return r;
1465         }
1466
1467         return dm_array_new(&cmd->hint_info, &cmd->hint_root,
1468                             from_cblock(cmd->cache_blocks),
1469                             get_hint, policy);
1470 }
1471
1472 int dm_cache_write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1473 {
1474         int r;
1475
1476         WRITE_LOCK(cmd);
1477         r = write_hints(cmd, policy);
1478         WRITE_UNLOCK(cmd);
1479
1480         return r;
1481 }
1482
1483 int dm_cache_metadata_all_clean(struct dm_cache_metadata *cmd, bool *result)
1484 {
1485         int r;
1486
1487         READ_LOCK(cmd);
1488         r = blocks_are_unmapped_or_clean(cmd, 0, cmd->cache_blocks, result);
1489         READ_UNLOCK(cmd);
1490
1491         return r;
1492 }
1493
1494 void dm_cache_metadata_set_read_only(struct dm_cache_metadata *cmd)
1495 {
1496         WRITE_LOCK_VOID(cmd);
1497         dm_bm_set_read_only(cmd->bm);
1498         WRITE_UNLOCK(cmd);
1499 }
1500
1501 void dm_cache_metadata_set_read_write(struct dm_cache_metadata *cmd)
1502 {
1503         WRITE_LOCK_VOID(cmd);
1504         dm_bm_set_read_write(cmd->bm);
1505         WRITE_UNLOCK(cmd);
1506 }
1507
1508 int dm_cache_metadata_set_needs_check(struct dm_cache_metadata *cmd)
1509 {
1510         int r;
1511         struct dm_block *sblock;
1512         struct cache_disk_superblock *disk_super;
1513
1514         WRITE_LOCK(cmd);
1515         set_bit(NEEDS_CHECK, &cmd->flags);
1516
1517         r = superblock_lock(cmd, &sblock);
1518         if (r) {
1519                 DMERR("couldn't read superblock");
1520                 goto out;
1521         }
1522
1523         disk_super = dm_block_data(sblock);
1524         disk_super->flags = cpu_to_le32(cmd->flags);
1525
1526         dm_bm_unlock(sblock);
1527
1528 out:
1529         WRITE_UNLOCK(cmd);
1530         return r;
1531 }
1532
1533 int dm_cache_metadata_needs_check(struct dm_cache_metadata *cmd, bool *result)
1534 {
1535         READ_LOCK(cmd);
1536         *result = !!test_bit(NEEDS_CHECK, &cmd->flags);
1537         READ_UNLOCK(cmd);
1538
1539         return 0;
1540 }
1541
1542 int dm_cache_metadata_abort(struct dm_cache_metadata *cmd)
1543 {
1544         int r;
1545
1546         WRITE_LOCK(cmd);
1547         __destroy_persistent_data_objects(cmd);
1548         r = __create_persistent_data_objects(cmd, false);
1549         if (r)
1550                 cmd->fail_io = true;
1551         WRITE_UNLOCK(cmd);
1552
1553         return r;
1554 }