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 = cpu_to_le32(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
 387         r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root);
 388         if (r < 0)
 389                 goto bad;
 390
 391         cmd->discard_block_size = 0;
 392         cmd->discard_nr_blocks = 0;
 393
 394         r = __write_initial_superblock(cmd);
 395         if (r)
 396                 goto bad;
 397
 398         cmd->clean_when_opened = true;
 399         return 0;
 400
 401 bad:
 402         dm_tm_destroy(cmd->tm);
 403         dm_sm_destroy(cmd->metadata_sm);
 404
 405         return r;
 406 }
 407
 408 static int __check_incompat_features(struct cache_disk_superblock *disk_super,
 409                                      struct dm_cache_metadata *cmd)
 410 {
 411         uint32_t features;
 412
 413         features = le32_to_cpu(disk_super->incompat_flags) & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
 414         if (features) {
 415                 DMERR("could not access metadata due to unsupported optional features (%lx).",
 416                       (unsigned long)features);
 417                 return -EINVAL;
 418         }
 419
 420         /*
 421          * Check for read-only metadata to skip the following RDWR checks.
 422          */
 423         if (get_disk_ro(cmd->bdev->bd_disk))
 424                 return 0;
 425
 426         features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP;
 427         if (features) {
 428                 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
 429                       (unsigned long)features);
 430                 return -EINVAL;
 431         }
 432
 433         return 0;
 434 }
 435
 436 static int __open_metadata(struct dm_cache_metadata *cmd)
 437 {
 438         int r;
 439         struct dm_block *sblock;
 440         struct cache_disk_superblock *disk_super;
 441         unsigned long sb_flags;
 442
 443         r = superblock_read_lock(cmd, &sblock);
 444         if (r < 0) {
 445                 DMERR("couldn't read lock superblock");
 446                 return r;
 447         }
 448
 449         disk_super = dm_block_data(sblock);
 450
 451         /* Verify the data block size hasn't changed */
 452         if (le32_to_cpu(disk_super->data_block_size) != cmd->data_block_size) {
 453                 DMERR("changing the data block size (from %u to %llu) is not supported",
 454                       le32_to_cpu(disk_super->data_block_size),
 455                       (unsigned long long)cmd->data_block_size);
 456                 r = -EINVAL;
 457                 goto bad;
 458         }
 459
 460         r = __check_incompat_features(disk_super, cmd);
 461         if (r < 0)
 462                 goto bad;
 463
 464         r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 465                                disk_super->metadata_space_map_root,
 466                                sizeof(disk_super->metadata_space_map_root),
 467                                &cmd->tm, &cmd->metadata_sm);
 468         if (r < 0) {
 469                 DMERR("tm_open_with_sm failed");
 470                 goto bad;
 471         }
 472
 473         __setup_mapping_info(cmd);
 474         dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
 475         sb_flags = le32_to_cpu(disk_super->flags);
 476         cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags);
 477         dm_bm_unlock(sblock);
 478
 479         return 0;
 480
 481 bad:
 482         dm_bm_unlock(sblock);
 483         return r;
 484 }
 485
 486 static int __open_or_format_metadata(struct dm_cache_metadata *cmd,
 487                                      bool format_device)
 488 {
 489         int r;
 490         bool unformatted = false;
 491
 492         r = __superblock_all_zeroes(cmd->bm, &unformatted);
 493         if (r)
 494                 return r;
 495
 496         if (unformatted)
 497                 return format_device ? __format_metadata(cmd) : -EPERM;
 498
 499         return __open_metadata(cmd);
 500 }
 501
 502 static int __create_persistent_data_objects(struct dm_cache_metadata *cmd,
 503                                             bool may_format_device)
 504 {
 505         int r;
 506         cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
 507                                           CACHE_METADATA_CACHE_SIZE,
 508                                           CACHE_MAX_CONCURRENT_LOCKS);
 509         if (IS_ERR(cmd->bm)) {
 510                 DMERR("could not create block manager");
 511                 r = PTR_ERR(cmd->bm);
 512                 cmd->bm = NULL;
 513                 return r;
 514         }
 515
 516         r = __open_or_format_metadata(cmd, may_format_device);
 517         if (r) {
 518                 dm_block_manager_destroy(cmd->bm);
 519                 cmd->bm = NULL;
 520         }
 521
 522         return r;
 523 }
 524
 525 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd)
 526 {
 527         dm_sm_destroy(cmd->metadata_sm);
 528         dm_tm_destroy(cmd->tm);
 529         dm_block_manager_destroy(cmd->bm);
 530 }
 531
 532 typedef unsigned long (*flags_mutator)(unsigned long);
 533
 534 static void update_flags(struct cache_disk_superblock *disk_super,
 535                          flags_mutator mutator)
 536 {
 537         uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
 538         disk_super->flags = cpu_to_le32(sb_flags);
 539 }
 540
 541 static unsigned long set_clean_shutdown(unsigned long flags)
 542 {
 543         set_bit(CLEAN_SHUTDOWN, &flags);
 544         return flags;
 545 }
 546
 547 static unsigned long clear_clean_shutdown(unsigned long flags)
 548 {
 549         clear_bit(CLEAN_SHUTDOWN, &flags);
 550         return flags;
 551 }
 552
 553 static void read_superblock_fields(struct dm_cache_metadata *cmd,
 554                                    struct cache_disk_superblock *disk_super)
 555 {
 556         cmd->flags = le32_to_cpu(disk_super->flags);
 557         cmd->root = le64_to_cpu(disk_super->mapping_root);
 558         cmd->hint_root = le64_to_cpu(disk_super->hint_root);
 559         cmd->discard_root = le64_to_cpu(disk_super->discard_root);
 560         cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
 561         cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
 562         cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
 563         cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
 564         strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
 565         cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
 566         cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
 567         cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
 568         cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
 569
 570         cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
 571         cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
 572         cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
 573         cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
 574
 575         cmd->changed = false;
 576 }
 577
 578 /*
 579  * The mutator updates the superblock flags.
 580  */
 581 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
 582                                      flags_mutator mutator)
 583 {
 584         int r;
 585         struct cache_disk_superblock *disk_super;
 586         struct dm_block *sblock;
 587
 588         r = superblock_lock(cmd, &sblock);
 589         if (r)
 590                 return r;
 591
 592         disk_super = dm_block_data(sblock);
 593         update_flags(disk_super, mutator);
 594         read_superblock_fields(cmd, disk_super);
 595         dm_bm_unlock(sblock);
 596
 597         return dm_bm_flush(cmd->bm);
 598 }
 599
 600 static int __begin_transaction(struct dm_cache_metadata *cmd)
 601 {
 602         int r;
 603         struct cache_disk_superblock *disk_super;
 604         struct dm_block *sblock;
 605
 606         /*
 607          * We re-read the superblock every time.  Shouldn't need to do this
 608          * really.
 609          */
 610         r = superblock_read_lock(cmd, &sblock);
 611         if (r)
 612                 return r;
 613
 614         disk_super = dm_block_data(sblock);
 615         read_superblock_fields(cmd, disk_super);
 616         dm_bm_unlock(sblock);
 617
 618         return 0;
 619 }
 620
 621 static int __commit_transaction(struct dm_cache_metadata *cmd,
 622                                 flags_mutator mutator)
 623 {
 624         int r;
 625         struct cache_disk_superblock *disk_super;
 626         struct dm_block *sblock;
 627
 628         /*
 629          * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
 630          */
 631         BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
 632
 633         r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
 634                             &cmd->discard_root);
 635         if (r)
 636                 return r;
 637
 638         r = dm_tm_pre_commit(cmd->tm);
 639         if (r < 0)
 640                 return r;
 641
 642         r = __save_sm_root(cmd);
 643         if (r)
 644                 return r;
 645
 646         r = superblock_lock(cmd, &sblock);
 647         if (r)
 648                 return r;
 649
 650         disk_super = dm_block_data(sblock);
 651
 652         disk_super->flags = cpu_to_le32(cmd->flags);
 653         if (mutator)
 654                 update_flags(disk_super, mutator);
 655
 656         disk_super->mapping_root = cpu_to_le64(cmd->root);
 657         disk_super->hint_root = cpu_to_le64(cmd->hint_root);
 658         disk_super->discard_root = cpu_to_le64(cmd->discard_root);
 659         disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
 660         disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
 661         disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
 662         strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
 663         disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
 664         disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
 665         disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
 666         disk_super->policy_hint_size = cpu_to_le32(cmd->policy_hint_size);
 667
 668         disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
 669         disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
 670         disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
 671         disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
 672         __copy_sm_root(cmd, disk_super);
 673
 674         return dm_tm_commit(cmd->tm, sblock);
 675 }
 676
 677 /*----------------------------------------------------------------*/
 678
 679 /*
 680  * The mappings are held in a dm-array that has 64-bit values stored in
 681  * little-endian format.  The index is the cblock, the high 48bits of the
 682  * value are the oblock and the low 16 bit the flags.
 683  */
 684 #define FLAGS_MASK ((1 << 16) - 1)
 685
 686 static __le64 pack_value(dm_oblock_t block, unsigned flags)
 687 {
 688         uint64_t value = from_oblock(block);
 689         value <<= 16;
 690         value = value | (flags & FLAGS_MASK);
 691         return cpu_to_le64(value);
 692 }
 693
 694 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
 695 {
 696         uint64_t value = le64_to_cpu(value_le);
 697         uint64_t b = value >> 16;
 698         *block = to_oblock(b);
 699         *flags = value & FLAGS_MASK;
 700 }
 701
 702 /*----------------------------------------------------------------*/
 703
 704 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
 705                                                sector_t data_block_size,
 706                                                bool may_format_device,
 707                                                size_t policy_hint_size)
 708 {
 709         int r;
 710         struct dm_cache_metadata *cmd;
 711
 712         cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
 713         if (!cmd) {
 714                 DMERR("could not allocate metadata struct");
 715                 return ERR_PTR(-ENOMEM);
 716         }
 717
 718         atomic_set(&cmd->ref_count, 1);
 719         init_rwsem(&cmd->root_lock);
 720         cmd->bdev = bdev;
 721         cmd->data_block_size = data_block_size;
 722         cmd->cache_blocks = 0;
 723         cmd->policy_hint_size = policy_hint_size;
 724         cmd->changed = true;
 725         cmd->fail_io = false;
 726
 727         r = __create_persistent_data_objects(cmd, may_format_device);
 728         if (r) {
 729                 kfree(cmd);
 730                 return ERR_PTR(r);
 731         }
 732
 733         r = __begin_transaction_flags(cmd, clear_clean_shutdown);
 734         if (r < 0) {
 735                 dm_cache_metadata_close(cmd);
 736                 return ERR_PTR(r);
 737         }
 738
 739         return cmd;
 740 }
 741
 742 /*
 743  * We keep a little list of ref counted metadata objects to prevent two
 744  * different target instances creating separate bufio instances.  This is
 745  * an issue if a table is reloaded before the suspend.
 746  */
 747 static DEFINE_MUTEX(table_lock);
 748 static LIST_HEAD(table);
 749
 750 static struct dm_cache_metadata *lookup(struct block_device *bdev)
 751 {
 752         struct dm_cache_metadata *cmd;
 753
 754         list_for_each_entry(cmd, &table, list)
 755                 if (cmd->bdev == bdev) {
 756                         atomic_inc(&cmd->ref_count);
 757                         return cmd;
 758                 }
 759
 760         return NULL;
 761 }
 762
 763 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
 764                                                 sector_t data_block_size,
 765                                                 bool may_format_device,
 766                                                 size_t policy_hint_size)
 767 {
 768         struct dm_cache_metadata *cmd, *cmd2;
 769
 770         mutex_lock(&table_lock);
 771         cmd = lookup(bdev);
 772         mutex_unlock(&table_lock);
 773
 774         if (cmd)
 775                 return cmd;
 776
 777         cmd = metadata_open(bdev, data_block_size, may_format_device, policy_hint_size);
 778         if (!IS_ERR(cmd)) {
 779                 mutex_lock(&table_lock);
 780                 cmd2 = lookup(bdev);
 781                 if (cmd2) {
 782                         mutex_unlock(&table_lock);
 783                         __destroy_persistent_data_objects(cmd);
 784                         kfree(cmd);
 785                         return cmd2;
 786                 }
 787                 list_add(&cmd->list, &table);
 788                 mutex_unlock(&table_lock);
 789         }
 790
 791         return cmd;
 792 }
 793
 794 static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
 795 {
 796         if (cmd->data_block_size != data_block_size) {
 797                 DMERR("data_block_size (%llu) different from that in metadata (%llu)\n",
 798                       (unsigned long long) data_block_size,
 799                       (unsigned long long) cmd->data_block_size);
 800                 return false;
 801         }
 802
 803         return true;
 804 }
 805
 806 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
 807                                                  sector_t data_block_size,
 808                                                  bool may_format_device,
 809                                                  size_t policy_hint_size)
 810 {
 811         struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size,
 812                                                        may_format_device, policy_hint_size);
 813
 814         if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
 815                 dm_cache_metadata_close(cmd);
 816                 return ERR_PTR(-EINVAL);
 817         }
 818
 819         return cmd;
 820 }
 821
 822 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
 823 {
 824         if (atomic_dec_and_test(&cmd->ref_count)) {
 825                 mutex_lock(&table_lock);
 826                 list_del(&cmd->list);
 827                 mutex_unlock(&table_lock);
 828
 829                 if (!cmd->fail_io)
 830                         __destroy_persistent_data_objects(cmd);
 831                 kfree(cmd);
 832         }
 833 }
 834
 835 /*
 836  * Checks that the given cache block is either unmapped or clean.
 837  */
 838 static int block_unmapped_or_clean(struct dm_cache_metadata *cmd, dm_cblock_t b,
 839                                    bool *result)
 840 {
 841         int r;
 842         __le64 value;
 843         dm_oblock_t ob;
 844         unsigned flags;
 845
 846         r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
 847         if (r) {
 848                 DMERR("block_unmapped_or_clean failed");
 849                 return r;
 850         }
 851
 852         unpack_value(value, &ob, &flags);
 853         *result = !((flags & M_VALID) && (flags & M_DIRTY));
 854
 855         return 0;
 856 }
 857
 858 static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
 859                                         dm_cblock_t begin, dm_cblock_t end,
 860                                         bool *result)
 861 {
 862         int r;
 863         *result = true;
 864
 865         while (begin != end) {
 866                 r = block_unmapped_or_clean(cmd, begin, result);
 867                 if (r)
 868                         return r;
 869
 870                 if (!*result) {
 871                         DMERR("cache block %llu is dirty",
 872                               (unsigned long long) from_cblock(begin));
 873                         return 0;
 874                 }
 875
 876                 begin = to_cblock(from_cblock(begin) + 1);
 877         }
 878
 879         return 0;
 880 }
 881
 882 static bool cmd_write_lock(struct dm_cache_metadata *cmd)
 883 {
 884         down_write(&cmd->root_lock);
 885         if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) {
 886                 up_write(&cmd->root_lock);
 887                 return false;
 888         }
 889         return true;
 890 }
 891
 892 #define WRITE_LOCK(cmd)                         \
 893         do {                                    \
 894                 if (!cmd_write_lock((cmd)))     \
 895                         return -EINVAL;         \
 896         } while(0)
 897
 898 #define WRITE_LOCK_VOID(cmd)                    \
 899         do {                                    \
 900                 if (!cmd_write_lock((cmd)))     \
 901                         return;                 \
 902         } while(0)
 903
 904 #define WRITE_UNLOCK(cmd) \
 905         up_write(&(cmd)->root_lock)
 906
 907 static bool cmd_read_lock(struct dm_cache_metadata *cmd)
 908 {
 909         down_read(&cmd->root_lock);
 910         if (cmd->fail_io) {
 911                 up_read(&cmd->root_lock);
 912                 return false;
 913         }
 914         return true;
 915 }
 916
 917 #define READ_LOCK(cmd)                          \
 918         do {                                    \
 919                 if (!cmd_read_lock((cmd)))      \
 920                         return -EINVAL;         \
 921         } while(0)
 922
 923 #define READ_LOCK_VOID(cmd)                     \
 924         do {                                    \
 925                 if (!cmd_read_lock((cmd)))      \
 926                         return;                 \
 927         } while(0)
 928
 929 #define READ_UNLOCK(cmd) \
 930         up_read(&(cmd)->root_lock)
 931
 932 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
 933 {
 934         int r;
 935         bool clean;
 936         __le64 null_mapping = pack_value(0, 0);
 937
 938         WRITE_LOCK(cmd);
 939         __dm_bless_for_disk(&null_mapping);
 940
 941         if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) {
 942                 r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean);
 943                 if (r) {
 944                         __dm_unbless_for_disk(&null_mapping);
 945                         goto out;
 946                 }
 947
 948                 if (!clean) {
 949                         DMERR("unable to shrink cache due to dirty blocks");
 950                         r = -EINVAL;
 951                         __dm_unbless_for_disk(&null_mapping);
 952                         goto out;
 953                 }
 954         }
 955
 956         r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
 957                             from_cblock(new_cache_size),
 958                             &null_mapping, &cmd->root);
 959         if (!r)
 960                 cmd->cache_blocks = new_cache_size;
 961         cmd->changed = true;
 962
 963 out:
 964         WRITE_UNLOCK(cmd);
 965
 966         return r;
 967 }
 968
 969 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
 970                                    sector_t discard_block_size,
 971                                    dm_dblock_t new_nr_entries)
 972 {
 973         int r;
 974
 975         WRITE_LOCK(cmd);
 976         r = dm_bitset_resize(&cmd->discard_info,
 977                              cmd->discard_root,
 978                              from_dblock(cmd->discard_nr_blocks),
 979                              from_dblock(new_nr_entries),
 980                              false, &cmd->discard_root);
 981         if (!r) {
 982                 cmd->discard_block_size = discard_block_size;
 983                 cmd->discard_nr_blocks = new_nr_entries;
 984         }
 985
 986         cmd->changed = true;
 987         WRITE_UNLOCK(cmd);
 988
 989         return r;
 990 }
 991
 992 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
 993 {
 994         return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
 995                                  from_dblock(b), &cmd->discard_root);
 996 }
 997
 998 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
 999 {
1000         return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
1001                                    from_dblock(b), &cmd->discard_root);
1002 }
1003
1004 static int __is_discarded(struct dm_cache_metadata *cmd, dm_dblock_t b,
1005                           bool *is_discarded)
1006 {
1007         return dm_bitset_test_bit(&cmd->discard_info, cmd->discard_root,
1008                                   from_dblock(b), &cmd->discard_root,
1009                                   is_discarded);
1010 }
1011
1012 static int __discard(struct dm_cache_metadata *cmd,
1013                      dm_dblock_t dblock, bool discard)
1014 {
1015         int r;
1016
1017         r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
1018         if (r)
1019                 return r;
1020
1021         cmd->changed = true;
1022         return 0;
1023 }
1024
1025 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
1026                          dm_dblock_t dblock, bool discard)
1027 {
1028         int r;
1029
1030         WRITE_LOCK(cmd);
1031         r = __discard(cmd, dblock, discard);
1032         WRITE_UNLOCK(cmd);
1033
1034         return r;
1035 }
1036
1037 static int __load_discards(struct dm_cache_metadata *cmd,
1038                            load_discard_fn fn, void *context)
1039 {
1040         int r = 0;
1041         dm_block_t b;
1042         bool discard;
1043
1044         for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1045                 dm_dblock_t dblock = to_dblock(b);
1046
1047                 if (cmd->clean_when_opened) {
1048                         r = __is_discarded(cmd, dblock, &discard);
1049                         if (r)
1050                                 return r;
1051                 } else
1052                         discard = false;
1053
1054                 r = fn(context, cmd->discard_block_size, dblock, discard);
1055                 if (r)
1056                         break;
1057         }
1058
1059         return r;
1060 }
1061
1062 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
1063                            load_discard_fn fn, void *context)
1064 {
1065         int r;
1066
1067         READ_LOCK(cmd);
1068         r = __load_discards(cmd, fn, context);
1069         READ_UNLOCK(cmd);
1070
1071         return r;
1072 }
1073
1074 int dm_cache_size(struct dm_cache_metadata *cmd, dm_cblock_t *result)
1075 {
1076         READ_LOCK(cmd);
1077         *result = cmd->cache_blocks;
1078         READ_UNLOCK(cmd);
1079
1080         return 0;
1081 }
1082
1083 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1084 {
1085         int r;
1086         __le64 value = pack_value(0, 0);
1087
1088         __dm_bless_for_disk(&value);
1089         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1090                                &value, &cmd->root);
1091         if (r)
1092                 return r;
1093
1094         cmd->changed = true;
1095         return 0;
1096 }
1097
1098 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1099 {
1100         int r;
1101
1102         WRITE_LOCK(cmd);
1103         r = __remove(cmd, cblock);
1104         WRITE_UNLOCK(cmd);
1105
1106         return r;
1107 }
1108
1109 static int __insert(struct dm_cache_metadata *cmd,
1110                     dm_cblock_t cblock, dm_oblock_t oblock)
1111 {
1112         int r;
1113         __le64 value = pack_value(oblock, M_VALID);
1114         __dm_bless_for_disk(&value);
1115
1116         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1117                                &value, &cmd->root);
1118         if (r)
1119                 return r;
1120
1121         cmd->changed = true;
1122         return 0;
1123 }
1124
1125 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
1126                             dm_cblock_t cblock, dm_oblock_t oblock)
1127 {
1128         int r;
1129
1130         WRITE_LOCK(cmd);
1131         r = __insert(cmd, cblock, oblock);
1132         WRITE_UNLOCK(cmd);
1133
1134         return r;
1135 }
1136
1137 struct thunk {
1138         load_mapping_fn fn;
1139         void *context;
1140
1141         struct dm_cache_metadata *cmd;
1142         bool respect_dirty_flags;
1143         bool hints_valid;
1144 };
1145
1146 static bool policy_unchanged(struct dm_cache_metadata *cmd,
1147                              struct dm_cache_policy *policy)
1148 {
1149         const char *policy_name = dm_cache_policy_get_name(policy);
1150         const unsigned *policy_version = dm_cache_policy_get_version(policy);
1151         size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
1152
1153         /*
1154          * Ensure policy names match.
1155          */
1156         if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
1157                 return false;
1158
1159         /*
1160          * Ensure policy major versions match.
1161          */
1162         if (cmd->policy_version[0] != policy_version[0])
1163                 return false;
1164
1165         /*
1166          * Ensure policy hint sizes match.
1167          */
1168         if (cmd->policy_hint_size != policy_hint_size)
1169                 return false;
1170
1171         return true;
1172 }
1173
1174 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1175 {
1176         return cmd->hint_root && cmd->policy_hint_size;
1177 }
1178
1179 static bool hints_array_available(struct dm_cache_metadata *cmd,
1180                                   struct dm_cache_policy *policy)
1181 {
1182         return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1183                 hints_array_initialized(cmd);
1184 }
1185
1186 static int __load_mapping(struct dm_cache_metadata *cmd,
1187                           uint64_t cb, bool hints_valid,
1188                           struct dm_array_cursor *mapping_cursor,
1189                           struct dm_array_cursor *hint_cursor,
1190                           load_mapping_fn fn, void *context)
1191 {
1192         int r = 0;
1193
1194         __le64 mapping;
1195         __le32 hint = 0;
1196
1197         __le64 *mapping_value_le;
1198         __le32 *hint_value_le;
1199
1200         dm_oblock_t oblock;
1201         unsigned flags;
1202
1203         dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1204         memcpy(&mapping, mapping_value_le, sizeof(mapping));
1205         unpack_value(mapping, &oblock, &flags);
1206
1207         if (flags & M_VALID) {
1208                 if (hints_valid) {
1209                         dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1210                         memcpy(&hint, hint_value_le, sizeof(hint));
1211                 }
1212
1213                 r = fn(context, oblock, to_cblock(cb), flags & M_DIRTY,
1214                        le32_to_cpu(hint), hints_valid);
1215                 if (r)
1216                         DMERR("policy couldn't load cblock");
1217         }
1218
1219         return r;
1220 }
1221
1222 static int __load_mappings(struct dm_cache_metadata *cmd,
1223                            struct dm_cache_policy *policy,
1224                            load_mapping_fn fn, void *context)
1225 {
1226         int r;
1227         uint64_t cb;
1228
1229         bool hints_valid = hints_array_available(cmd, policy);
1230
1231         if (from_cblock(cmd->cache_blocks) == 0)
1232                 /* Nothing to do */
1233                 return 0;
1234
1235         r = dm_array_cursor_begin(&cmd->info, cmd->root, &cmd->mapping_cursor);
1236         if (r)
1237                 return r;
1238
1239         if (hints_valid) {
1240                 r = dm_array_cursor_begin(&cmd->hint_info, cmd->hint_root, &cmd->hint_cursor);
1241                 if (r) {
1242                         dm_array_cursor_end(&cmd->mapping_cursor);
1243                         return r;
1244                 }
1245         }
1246
1247         for (cb = 0; ; cb++) {
1248                 r = __load_mapping(cmd, cb, hints_valid,
1249                                    &cmd->mapping_cursor, &cmd->hint_cursor,
1250                                    fn, context);
1251                 if (r)
1252                         goto out;
1253
1254                 /*
1255                  * We need to break out before we move the cursors.
1256                  */
1257                 if (cb >= (from_cblock(cmd->cache_blocks) - 1))
1258                         break;
1259
1260                 r = dm_array_cursor_next(&cmd->mapping_cursor);
1261                 if (r) {
1262                         DMERR("dm_array_cursor_next for mapping failed");
1263                         goto out;
1264                 }
1265
1266                 if (hints_valid) {
1267                         r = dm_array_cursor_next(&cmd->hint_cursor);
1268                         if (r) {
1269                                 dm_array_cursor_end(&cmd->hint_cursor);
1270                                 hints_valid = false;
1271                         }
1272                 }
1273         }
1274 out:
1275         dm_array_cursor_end(&cmd->mapping_cursor);
1276         if (hints_valid)
1277                 dm_array_cursor_end(&cmd->hint_cursor);
1278
1279         return r;
1280 }
1281
1282 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1283                            struct dm_cache_policy *policy,
1284                            load_mapping_fn fn, void *context)
1285 {
1286         int r;
1287
1288         READ_LOCK(cmd);
1289         r = __load_mappings(cmd, policy, fn, context);
1290         READ_UNLOCK(cmd);
1291
1292         return r;
1293 }
1294
1295 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
1296 {
1297         int r = 0;
1298         __le64 value;
1299         dm_oblock_t oblock;
1300         unsigned flags;
1301
1302         memcpy(&value, leaf, sizeof(value));
1303         unpack_value(value, &oblock, &flags);
1304
1305         return r;
1306 }
1307
1308 static int __dump_mappings(struct dm_cache_metadata *cmd)
1309 {
1310         return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1311 }
1312
1313 void dm_cache_dump(struct dm_cache_metadata *cmd)
1314 {
1315         READ_LOCK_VOID(cmd);
1316         __dump_mappings(cmd);
1317         READ_UNLOCK(cmd);
1318 }
1319
1320 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1321 {
1322         int r;
1323
1324         READ_LOCK(cmd);
1325         r = cmd->changed;
1326         READ_UNLOCK(cmd);
1327
1328         return r;
1329 }
1330
1331 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1332 {
1333         int r;
1334         unsigned flags;
1335         dm_oblock_t oblock;
1336         __le64 value;
1337
1338         r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1339         if (r)
1340                 return r;
1341
1342         unpack_value(value, &oblock, &flags);
1343
1344         if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1345                 /* nothing to be done */
1346                 return 0;
1347
1348         value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1349         __dm_bless_for_disk(&value);
1350
1351         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1352                                &value, &cmd->root);
1353         if (r)
1354                 return r;
1355
1356         cmd->changed = true;
1357         return 0;
1358
1359 }
1360
1361 int dm_cache_set_dirty(struct dm_cache_metadata *cmd,
1362                        dm_cblock_t cblock, bool dirty)
1363 {
1364         int r;
1365
1366         WRITE_LOCK(cmd);
1367         r = __dirty(cmd, cblock, dirty);
1368         WRITE_UNLOCK(cmd);
1369
1370         return r;
1371 }
1372
1373 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1374                                  struct dm_cache_statistics *stats)
1375 {
1376         READ_LOCK_VOID(cmd);
1377         *stats = cmd->stats;
1378         READ_UNLOCK(cmd);
1379 }
1380
1381 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1382                                  struct dm_cache_statistics *stats)
1383 {
1384         WRITE_LOCK_VOID(cmd);
1385         cmd->stats = *stats;
1386         WRITE_UNLOCK(cmd);
1387 }
1388
1389 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1390 {
1391         int r = -EINVAL;
1392         flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1393                                  clear_clean_shutdown);
1394
1395         WRITE_LOCK(cmd);
1396         if (cmd->fail_io)
1397                 goto out;
1398
1399         r = __commit_transaction(cmd, mutator);
1400         if (r)
1401                 goto out;
1402
1403         r = __begin_transaction(cmd);
1404 out:
1405         WRITE_UNLOCK(cmd);
1406         return r;
1407 }
1408
1409 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1410                                            dm_block_t *result)
1411 {
1412         int r = -EINVAL;
1413
1414         READ_LOCK(cmd);
1415         if (!cmd->fail_io)
1416                 r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1417         READ_UNLOCK(cmd);
1418
1419         return r;
1420 }
1421
1422 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1423                                    dm_block_t *result)
1424 {
1425         int r = -EINVAL;
1426
1427         READ_LOCK(cmd);
1428         if (!cmd->fail_io)
1429                 r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1430         READ_UNLOCK(cmd);
1431
1432         return r;
1433 }
1434
1435 /*----------------------------------------------------------------*/
1436
1437 static int get_hint(uint32_t index, void *value_le, void *context)
1438 {
1439         uint32_t value;
1440         struct dm_cache_policy *policy = context;
1441
1442         value = policy_get_hint(policy, to_cblock(index));
1443         *((__le32 *) value_le) = cpu_to_le32(value);
1444
1445         return 0;
1446 }
1447
1448 /*
1449  * It's quicker to always delete the hint array, and recreate with
1450  * dm_array_new().
1451  */
1452 static int write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1453 {
1454         int r;
1455         size_t hint_size;
1456         const char *policy_name = dm_cache_policy_get_name(policy);
1457         const unsigned *policy_version = dm_cache_policy_get_version(policy);
1458
1459         if (!policy_name[0] ||
1460             (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1461                 return -EINVAL;
1462
1463         strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1464         memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1465
1466         hint_size = dm_cache_policy_get_hint_size(policy);
1467         if (!hint_size)
1468                 return 0; /* short-circuit hints initialization */
1469         cmd->policy_hint_size = hint_size;
1470
1471         if (cmd->hint_root) {
1472                 r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1473                 if (r)
1474                         return r;
1475         }
1476
1477         return dm_array_new(&cmd->hint_info, &cmd->hint_root,
1478                             from_cblock(cmd->cache_blocks),
1479                             get_hint, policy);
1480 }
1481
1482 int dm_cache_write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1483 {
1484         int r;
1485
1486         WRITE_LOCK(cmd);
1487         r = write_hints(cmd, policy);
1488         WRITE_UNLOCK(cmd);
1489
1490         return r;
1491 }
1492
1493 int dm_cache_metadata_all_clean(struct dm_cache_metadata *cmd, bool *result)
1494 {
1495         int r;
1496
1497         READ_LOCK(cmd);
1498         r = blocks_are_unmapped_or_clean(cmd, 0, cmd->cache_blocks, result);
1499         READ_UNLOCK(cmd);
1500
1501         return r;
1502 }
1503
1504 void dm_cache_metadata_set_read_only(struct dm_cache_metadata *cmd)
1505 {
1506         WRITE_LOCK_VOID(cmd);
1507         dm_bm_set_read_only(cmd->bm);
1508         WRITE_UNLOCK(cmd);
1509 }
1510
1511 void dm_cache_metadata_set_read_write(struct dm_cache_metadata *cmd)
1512 {
1513         WRITE_LOCK_VOID(cmd);
1514         dm_bm_set_read_write(cmd->bm);
1515         WRITE_UNLOCK(cmd);
1516 }
1517
1518 int dm_cache_metadata_set_needs_check(struct dm_cache_metadata *cmd)
1519 {
1520         int r;
1521         struct dm_block *sblock;
1522         struct cache_disk_superblock *disk_super;
1523
1524         WRITE_LOCK(cmd);
1525         set_bit(NEEDS_CHECK, &cmd->flags);
1526
1527         r = superblock_lock(cmd, &sblock);
1528         if (r) {
1529                 DMERR("couldn't read superblock");
1530                 goto out;
1531         }
1532
1533         disk_super = dm_block_data(sblock);
1534         disk_super->flags = cpu_to_le32(cmd->flags);
1535
1536         dm_bm_unlock(sblock);
1537
1538 out:
1539         WRITE_UNLOCK(cmd);
1540         return r;
1541 }
1542
1543 int dm_cache_metadata_needs_check(struct dm_cache_metadata *cmd, bool *result)
1544 {
1545         READ_LOCK(cmd);
1546         *result = !!test_bit(NEEDS_CHECK, &cmd->flags);
1547         READ_UNLOCK(cmd);
1548
1549         return 0;
1550 }
1551
1552 int dm_cache_metadata_abort(struct dm_cache_metadata *cmd)
1553 {
1554         int r;
1555
1556         WRITE_LOCK(cmd);
1557         __destroy_persistent_data_objects(cmd);
1558         r = __create_persistent_data_objects(cmd, false);
1559         if (r)
1560                 cmd->fail_io = true;
1561         WRITE_UNLOCK(cmd);
1562
1563         return r;
1564 }