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                                               bool destroy_bm)
 527 {
 528         dm_sm_destroy(cmd->metadata_sm);
 529         dm_tm_destroy(cmd->tm);
 530         if (destroy_bm)
 531                 dm_block_manager_destroy(cmd->bm);
 532 }
 533
 534 typedef unsigned long (*flags_mutator)(unsigned long);
 535
 536 static void update_flags(struct cache_disk_superblock *disk_super,
 537                          flags_mutator mutator)
 538 {
 539         uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
 540         disk_super->flags = cpu_to_le32(sb_flags);
 541 }
 542
 543 static unsigned long set_clean_shutdown(unsigned long flags)
 544 {
 545         set_bit(CLEAN_SHUTDOWN, &flags);
 546         return flags;
 547 }
 548
 549 static unsigned long clear_clean_shutdown(unsigned long flags)
 550 {
 551         clear_bit(CLEAN_SHUTDOWN, &flags);
 552         return flags;
 553 }
 554
 555 static void read_superblock_fields(struct dm_cache_metadata *cmd,
 556                                    struct cache_disk_superblock *disk_super)
 557 {
 558         cmd->flags = le32_to_cpu(disk_super->flags);
 559         cmd->root = le64_to_cpu(disk_super->mapping_root);
 560         cmd->hint_root = le64_to_cpu(disk_super->hint_root);
 561         cmd->discard_root = le64_to_cpu(disk_super->discard_root);
 562         cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
 563         cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
 564         cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
 565         cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
 566         strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
 567         cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
 568         cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
 569         cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
 570         cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
 571
 572         cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
 573         cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
 574         cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
 575         cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
 576
 577         cmd->changed = false;
 578 }
 579
 580 /*
 581  * The mutator updates the superblock flags.
 582  */
 583 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
 584                                      flags_mutator mutator)
 585 {
 586         int r;
 587         struct cache_disk_superblock *disk_super;
 588         struct dm_block *sblock;
 589
 590         r = superblock_lock(cmd, &sblock);
 591         if (r)
 592                 return r;
 593
 594         disk_super = dm_block_data(sblock);
 595         update_flags(disk_super, mutator);
 596         read_superblock_fields(cmd, disk_super);
 597         dm_bm_unlock(sblock);
 598
 599         return dm_bm_flush(cmd->bm);
 600 }
 601
 602 static int __begin_transaction(struct dm_cache_metadata *cmd)
 603 {
 604         int r;
 605         struct cache_disk_superblock *disk_super;
 606         struct dm_block *sblock;
 607
 608         /*
 609          * We re-read the superblock every time.  Shouldn't need to do this
 610          * really.
 611          */
 612         r = superblock_read_lock(cmd, &sblock);
 613         if (r)
 614                 return r;
 615
 616         disk_super = dm_block_data(sblock);
 617         read_superblock_fields(cmd, disk_super);
 618         dm_bm_unlock(sblock);
 619
 620         return 0;
 621 }
 622
 623 static int __commit_transaction(struct dm_cache_metadata *cmd,
 624                                 flags_mutator mutator)
 625 {
 626         int r;
 627         struct cache_disk_superblock *disk_super;
 628         struct dm_block *sblock;
 629
 630         /*
 631          * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
 632          */
 633         BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
 634
 635         r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
 636                             &cmd->discard_root);
 637         if (r)
 638                 return r;
 639
 640         r = dm_tm_pre_commit(cmd->tm);
 641         if (r < 0)
 642                 return r;
 643
 644         r = __save_sm_root(cmd);
 645         if (r)
 646                 return r;
 647
 648         r = superblock_lock(cmd, &sblock);
 649         if (r)
 650                 return r;
 651
 652         disk_super = dm_block_data(sblock);
 653
 654         disk_super->flags = cpu_to_le32(cmd->flags);
 655         if (mutator)
 656                 update_flags(disk_super, mutator);
 657
 658         disk_super->mapping_root = cpu_to_le64(cmd->root);
 659         disk_super->hint_root = cpu_to_le64(cmd->hint_root);
 660         disk_super->discard_root = cpu_to_le64(cmd->discard_root);
 661         disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
 662         disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
 663         disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
 664         strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
 665         disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
 666         disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
 667         disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
 668         disk_super->policy_hint_size = cpu_to_le32(cmd->policy_hint_size);
 669
 670         disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
 671         disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
 672         disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
 673         disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
 674         __copy_sm_root(cmd, disk_super);
 675
 676         return dm_tm_commit(cmd->tm, sblock);
 677 }
 678
 679 /*----------------------------------------------------------------*/
 680
 681 /*
 682  * The mappings are held in a dm-array that has 64-bit values stored in
 683  * little-endian format.  The index is the cblock, the high 48bits of the
 684  * value are the oblock and the low 16 bit the flags.
 685  */
 686 #define FLAGS_MASK ((1 << 16) - 1)
 687
 688 static __le64 pack_value(dm_oblock_t block, unsigned flags)
 689 {
 690         uint64_t value = from_oblock(block);
 691         value <<= 16;
 692         value = value | (flags & FLAGS_MASK);
 693         return cpu_to_le64(value);
 694 }
 695
 696 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
 697 {
 698         uint64_t value = le64_to_cpu(value_le);
 699         uint64_t b = value >> 16;
 700         *block = to_oblock(b);
 701         *flags = value & FLAGS_MASK;
 702 }
 703
 704 /*----------------------------------------------------------------*/
 705
 706 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
 707                                                sector_t data_block_size,
 708                                                bool may_format_device,
 709                                                size_t policy_hint_size)
 710 {
 711         int r;
 712         struct dm_cache_metadata *cmd;
 713
 714         cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
 715         if (!cmd) {
 716                 DMERR("could not allocate metadata struct");
 717                 return ERR_PTR(-ENOMEM);
 718         }
 719
 720         atomic_set(&cmd->ref_count, 1);
 721         init_rwsem(&cmd->root_lock);
 722         cmd->bdev = bdev;
 723         cmd->data_block_size = data_block_size;
 724         cmd->cache_blocks = 0;
 725         cmd->policy_hint_size = policy_hint_size;
 726         cmd->changed = true;
 727         cmd->fail_io = false;
 728
 729         r = __create_persistent_data_objects(cmd, may_format_device);
 730         if (r) {
 731                 kfree(cmd);
 732                 return ERR_PTR(r);
 733         }
 734
 735         r = __begin_transaction_flags(cmd, clear_clean_shutdown);
 736         if (r < 0) {
 737                 dm_cache_metadata_close(cmd);
 738                 return ERR_PTR(r);
 739         }
 740
 741         return cmd;
 742 }
 743
 744 /*
 745  * We keep a little list of ref counted metadata objects to prevent two
 746  * different target instances creating separate bufio instances.  This is
 747  * an issue if a table is reloaded before the suspend.
 748  */
 749 static DEFINE_MUTEX(table_lock);
 750 static LIST_HEAD(table);
 751
 752 static struct dm_cache_metadata *lookup(struct block_device *bdev)
 753 {
 754         struct dm_cache_metadata *cmd;
 755
 756         list_for_each_entry(cmd, &table, list)
 757                 if (cmd->bdev == bdev) {
 758                         atomic_inc(&cmd->ref_count);
 759                         return cmd;
 760                 }
 761
 762         return NULL;
 763 }
 764
 765 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
 766                                                 sector_t data_block_size,
 767                                                 bool may_format_device,
 768                                                 size_t policy_hint_size)
 769 {
 770         struct dm_cache_metadata *cmd, *cmd2;
 771
 772         mutex_lock(&table_lock);
 773         cmd = lookup(bdev);
 774         mutex_unlock(&table_lock);
 775
 776         if (cmd)
 777                 return cmd;
 778
 779         cmd = metadata_open(bdev, data_block_size, may_format_device, policy_hint_size);
 780         if (!IS_ERR(cmd)) {
 781                 mutex_lock(&table_lock);
 782                 cmd2 = lookup(bdev);
 783                 if (cmd2) {
 784                         mutex_unlock(&table_lock);
 785                         __destroy_persistent_data_objects(cmd, true);
 786                         kfree(cmd);
 787                         return cmd2;
 788                 }
 789                 list_add(&cmd->list, &table);
 790                 mutex_unlock(&table_lock);
 791         }
 792
 793         return cmd;
 794 }
 795
 796 static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
 797 {
 798         if (cmd->data_block_size != data_block_size) {
 799                 DMERR("data_block_size (%llu) different from that in metadata (%llu)\n",
 800                       (unsigned long long) data_block_size,
 801                       (unsigned long long) cmd->data_block_size);
 802                 return false;
 803         }
 804
 805         return true;
 806 }
 807
 808 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
 809                                                  sector_t data_block_size,
 810                                                  bool may_format_device,
 811                                                  size_t policy_hint_size)
 812 {
 813         struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size,
 814                                                        may_format_device, policy_hint_size);
 815
 816         if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
 817                 dm_cache_metadata_close(cmd);
 818                 return ERR_PTR(-EINVAL);
 819         }
 820
 821         return cmd;
 822 }
 823
 824 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
 825 {
 826         if (atomic_dec_and_test(&cmd->ref_count)) {
 827                 mutex_lock(&table_lock);
 828                 list_del(&cmd->list);
 829                 mutex_unlock(&table_lock);
 830
 831                 if (!cmd->fail_io)
 832                         __destroy_persistent_data_objects(cmd, true);
 833                 kfree(cmd);
 834         }
 835 }
 836
 837 /*
 838  * Checks that the given cache block is either unmapped or clean.
 839  */
 840 static int block_unmapped_or_clean(struct dm_cache_metadata *cmd, dm_cblock_t b,
 841                                    bool *result)
 842 {
 843         int r;
 844         __le64 value;
 845         dm_oblock_t ob;
 846         unsigned flags;
 847
 848         r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
 849         if (r) {
 850                 DMERR("block_unmapped_or_clean failed");
 851                 return r;
 852         }
 853
 854         unpack_value(value, &ob, &flags);
 855         *result = !((flags & M_VALID) && (flags & M_DIRTY));
 856
 857         return 0;
 858 }
 859
 860 static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
 861                                         dm_cblock_t begin, dm_cblock_t end,
 862                                         bool *result)
 863 {
 864         int r;
 865         *result = true;
 866
 867         while (begin != end) {
 868                 r = block_unmapped_or_clean(cmd, begin, result);
 869                 if (r)
 870                         return r;
 871
 872                 if (!*result) {
 873                         DMERR("cache block %llu is dirty",
 874                               (unsigned long long) from_cblock(begin));
 875                         return 0;
 876                 }
 877
 878                 begin = to_cblock(from_cblock(begin) + 1);
 879         }
 880
 881         return 0;
 882 }
 883
 884 static bool cmd_write_lock(struct dm_cache_metadata *cmd)
 885 {
 886         down_write(&cmd->root_lock);
 887         if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) {
 888                 up_write(&cmd->root_lock);
 889                 return false;
 890         }
 891         return true;
 892 }
 893
 894 #define WRITE_LOCK(cmd)                         \
 895         do {                                    \
 896                 if (!cmd_write_lock((cmd)))     \
 897                         return -EINVAL;         \
 898         } while(0)
 899
 900 #define WRITE_LOCK_VOID(cmd)                    \
 901         do {                                    \
 902                 if (!cmd_write_lock((cmd)))     \
 903                         return;                 \
 904         } while(0)
 905
 906 #define WRITE_UNLOCK(cmd) \
 907         up_write(&(cmd)->root_lock)
 908
 909 static bool cmd_read_lock(struct dm_cache_metadata *cmd)
 910 {
 911         down_read(&cmd->root_lock);
 912         if (cmd->fail_io) {
 913                 up_read(&cmd->root_lock);
 914                 return false;
 915         }
 916         return true;
 917 }
 918
 919 #define READ_LOCK(cmd)                          \
 920         do {                                    \
 921                 if (!cmd_read_lock((cmd)))      \
 922                         return -EINVAL;         \
 923         } while(0)
 924
 925 #define READ_LOCK_VOID(cmd)                     \
 926         do {                                    \
 927                 if (!cmd_read_lock((cmd)))      \
 928                         return;                 \
 929         } while(0)
 930
 931 #define READ_UNLOCK(cmd) \
 932         up_read(&(cmd)->root_lock)
 933
 934 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
 935 {
 936         int r;
 937         bool clean;
 938         __le64 null_mapping = pack_value(0, 0);
 939
 940         WRITE_LOCK(cmd);
 941         __dm_bless_for_disk(&null_mapping);
 942
 943         if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) {
 944                 r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean);
 945                 if (r) {
 946                         __dm_unbless_for_disk(&null_mapping);
 947                         goto out;
 948                 }
 949
 950                 if (!clean) {
 951                         DMERR("unable to shrink cache due to dirty blocks");
 952                         r = -EINVAL;
 953                         __dm_unbless_for_disk(&null_mapping);
 954                         goto out;
 955                 }
 956         }
 957
 958         r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
 959                             from_cblock(new_cache_size),
 960                             &null_mapping, &cmd->root);
 961         if (!r)
 962                 cmd->cache_blocks = new_cache_size;
 963         cmd->changed = true;
 964
 965 out:
 966         WRITE_UNLOCK(cmd);
 967
 968         return r;
 969 }
 970
 971 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
 972                                    sector_t discard_block_size,
 973                                    dm_dblock_t new_nr_entries)
 974 {
 975         int r;
 976
 977         WRITE_LOCK(cmd);
 978         r = dm_bitset_resize(&cmd->discard_info,
 979                              cmd->discard_root,
 980                              from_dblock(cmd->discard_nr_blocks),
 981                              from_dblock(new_nr_entries),
 982                              false, &cmd->discard_root);
 983         if (!r) {
 984                 cmd->discard_block_size = discard_block_size;
 985                 cmd->discard_nr_blocks = new_nr_entries;
 986         }
 987
 988         cmd->changed = true;
 989         WRITE_UNLOCK(cmd);
 990
 991         return r;
 992 }
 993
 994 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
 995 {
 996         return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
 997                                  from_dblock(b), &cmd->discard_root);
 998 }
 999
1000 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
1001 {
1002         return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
1003                                    from_dblock(b), &cmd->discard_root);
1004 }
1005
1006 static int __is_discarded(struct dm_cache_metadata *cmd, dm_dblock_t b,
1007                           bool *is_discarded)
1008 {
1009         return dm_bitset_test_bit(&cmd->discard_info, cmd->discard_root,
1010                                   from_dblock(b), &cmd->discard_root,
1011                                   is_discarded);
1012 }
1013
1014 static int __discard(struct dm_cache_metadata *cmd,
1015                      dm_dblock_t dblock, bool discard)
1016 {
1017         int r;
1018
1019         r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
1020         if (r)
1021                 return r;
1022
1023         cmd->changed = true;
1024         return 0;
1025 }
1026
1027 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
1028                          dm_dblock_t dblock, bool discard)
1029 {
1030         int r;
1031
1032         WRITE_LOCK(cmd);
1033         r = __discard(cmd, dblock, discard);
1034         WRITE_UNLOCK(cmd);
1035
1036         return r;
1037 }
1038
1039 static int __load_discards(struct dm_cache_metadata *cmd,
1040                            load_discard_fn fn, void *context)
1041 {
1042         int r = 0;
1043         dm_block_t b;
1044         bool discard;
1045
1046         for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1047                 dm_dblock_t dblock = to_dblock(b);
1048
1049                 if (cmd->clean_when_opened) {
1050                         r = __is_discarded(cmd, dblock, &discard);
1051                         if (r)
1052                                 return r;
1053                 } else
1054                         discard = false;
1055
1056                 r = fn(context, cmd->discard_block_size, dblock, discard);
1057                 if (r)
1058                         break;
1059         }
1060
1061         return r;
1062 }
1063
1064 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
1065                            load_discard_fn fn, void *context)
1066 {
1067         int r;
1068
1069         READ_LOCK(cmd);
1070         r = __load_discards(cmd, fn, context);
1071         READ_UNLOCK(cmd);
1072
1073         return r;
1074 }
1075
1076 int dm_cache_size(struct dm_cache_metadata *cmd, dm_cblock_t *result)
1077 {
1078         READ_LOCK(cmd);
1079         *result = cmd->cache_blocks;
1080         READ_UNLOCK(cmd);
1081
1082         return 0;
1083 }
1084
1085 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1086 {
1087         int r;
1088         __le64 value = pack_value(0, 0);
1089
1090         __dm_bless_for_disk(&value);
1091         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1092                                &value, &cmd->root);
1093         if (r)
1094                 return r;
1095
1096         cmd->changed = true;
1097         return 0;
1098 }
1099
1100 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1101 {
1102         int r;
1103
1104         WRITE_LOCK(cmd);
1105         r = __remove(cmd, cblock);
1106         WRITE_UNLOCK(cmd);
1107
1108         return r;
1109 }
1110
1111 static int __insert(struct dm_cache_metadata *cmd,
1112                     dm_cblock_t cblock, dm_oblock_t oblock)
1113 {
1114         int r;
1115         __le64 value = pack_value(oblock, M_VALID);
1116         __dm_bless_for_disk(&value);
1117
1118         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1119                                &value, &cmd->root);
1120         if (r)
1121                 return r;
1122
1123         cmd->changed = true;
1124         return 0;
1125 }
1126
1127 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
1128                             dm_cblock_t cblock, dm_oblock_t oblock)
1129 {
1130         int r;
1131
1132         WRITE_LOCK(cmd);
1133         r = __insert(cmd, cblock, oblock);
1134         WRITE_UNLOCK(cmd);
1135
1136         return r;
1137 }
1138
1139 struct thunk {
1140         load_mapping_fn fn;
1141         void *context;
1142
1143         struct dm_cache_metadata *cmd;
1144         bool respect_dirty_flags;
1145         bool hints_valid;
1146 };
1147
1148 static bool policy_unchanged(struct dm_cache_metadata *cmd,
1149                              struct dm_cache_policy *policy)
1150 {
1151         const char *policy_name = dm_cache_policy_get_name(policy);
1152         const unsigned *policy_version = dm_cache_policy_get_version(policy);
1153         size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
1154
1155         /*
1156          * Ensure policy names match.
1157          */
1158         if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
1159                 return false;
1160
1161         /*
1162          * Ensure policy major versions match.
1163          */
1164         if (cmd->policy_version[0] != policy_version[0])
1165                 return false;
1166
1167         /*
1168          * Ensure policy hint sizes match.
1169          */
1170         if (cmd->policy_hint_size != policy_hint_size)
1171                 return false;
1172
1173         return true;
1174 }
1175
1176 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1177 {
1178         return cmd->hint_root && cmd->policy_hint_size;
1179 }
1180
1181 static bool hints_array_available(struct dm_cache_metadata *cmd,
1182                                   struct dm_cache_policy *policy)
1183 {
1184         return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1185                 hints_array_initialized(cmd);
1186 }
1187
1188 static int __load_mapping(struct dm_cache_metadata *cmd,
1189                           uint64_t cb, bool hints_valid,
1190                           struct dm_array_cursor *mapping_cursor,
1191                           struct dm_array_cursor *hint_cursor,
1192                           load_mapping_fn fn, void *context)
1193 {
1194         int r = 0;
1195
1196         __le64 mapping;
1197         __le32 hint = 0;
1198
1199         __le64 *mapping_value_le;
1200         __le32 *hint_value_le;
1201
1202         dm_oblock_t oblock;
1203         unsigned flags;
1204
1205         dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1206         memcpy(&mapping, mapping_value_le, sizeof(mapping));
1207         unpack_value(mapping, &oblock, &flags);
1208
1209         if (flags & M_VALID) {
1210                 if (hints_valid) {
1211                         dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1212                         memcpy(&hint, hint_value_le, sizeof(hint));
1213                 }
1214
1215                 r = fn(context, oblock, to_cblock(cb), flags & M_DIRTY,
1216                        le32_to_cpu(hint), hints_valid);
1217                 if (r)
1218                         DMERR("policy couldn't load cblock");
1219         }
1220
1221         return r;
1222 }
1223
1224 static int __load_mappings(struct dm_cache_metadata *cmd,
1225                            struct dm_cache_policy *policy,
1226                            load_mapping_fn fn, void *context)
1227 {
1228         int r;
1229         uint64_t cb;
1230
1231         bool hints_valid = hints_array_available(cmd, policy);
1232
1233         if (from_cblock(cmd->cache_blocks) == 0)
1234                 /* Nothing to do */
1235                 return 0;
1236
1237         r = dm_array_cursor_begin(&cmd->info, cmd->root, &cmd->mapping_cursor);
1238         if (r)
1239                 return r;
1240
1241         if (hints_valid) {
1242                 r = dm_array_cursor_begin(&cmd->hint_info, cmd->hint_root, &cmd->hint_cursor);
1243                 if (r) {
1244                         dm_array_cursor_end(&cmd->mapping_cursor);
1245                         return r;
1246                 }
1247         }
1248
1249         for (cb = 0; ; cb++) {
1250                 r = __load_mapping(cmd, cb, hints_valid,
1251                                    &cmd->mapping_cursor, &cmd->hint_cursor,
1252                                    fn, context);
1253                 if (r)
1254                         goto out;
1255
1256                 /*
1257                  * We need to break out before we move the cursors.
1258                  */
1259                 if (cb >= (from_cblock(cmd->cache_blocks) - 1))
1260                         break;
1261
1262                 r = dm_array_cursor_next(&cmd->mapping_cursor);
1263                 if (r) {
1264                         DMERR("dm_array_cursor_next for mapping failed");
1265                         goto out;
1266                 }
1267
1268                 if (hints_valid) {
1269                         r = dm_array_cursor_next(&cmd->hint_cursor);
1270                         if (r) {
1271                                 dm_array_cursor_end(&cmd->hint_cursor);
1272                                 hints_valid = false;
1273                         }
1274                 }
1275         }
1276 out:
1277         dm_array_cursor_end(&cmd->mapping_cursor);
1278         if (hints_valid)
1279                 dm_array_cursor_end(&cmd->hint_cursor);
1280
1281         return r;
1282 }
1283
1284 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1285                            struct dm_cache_policy *policy,
1286                            load_mapping_fn fn, void *context)
1287 {
1288         int r;
1289
1290         READ_LOCK(cmd);
1291         r = __load_mappings(cmd, policy, fn, context);
1292         READ_UNLOCK(cmd);
1293
1294         return r;
1295 }
1296
1297 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
1298 {
1299         int r = 0;
1300         __le64 value;
1301         dm_oblock_t oblock;
1302         unsigned flags;
1303
1304         memcpy(&value, leaf, sizeof(value));
1305         unpack_value(value, &oblock, &flags);
1306
1307         return r;
1308 }
1309
1310 static int __dump_mappings(struct dm_cache_metadata *cmd)
1311 {
1312         return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1313 }
1314
1315 void dm_cache_dump(struct dm_cache_metadata *cmd)
1316 {
1317         READ_LOCK_VOID(cmd);
1318         __dump_mappings(cmd);
1319         READ_UNLOCK(cmd);
1320 }
1321
1322 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1323 {
1324         int r;
1325
1326         READ_LOCK(cmd);
1327         r = cmd->changed;
1328         READ_UNLOCK(cmd);
1329
1330         return r;
1331 }
1332
1333 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1334 {
1335         int r;
1336         unsigned flags;
1337         dm_oblock_t oblock;
1338         __le64 value;
1339
1340         r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1341         if (r)
1342                 return r;
1343
1344         unpack_value(value, &oblock, &flags);
1345
1346         if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1347                 /* nothing to be done */
1348                 return 0;
1349
1350         value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1351         __dm_bless_for_disk(&value);
1352
1353         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1354                                &value, &cmd->root);
1355         if (r)
1356                 return r;
1357
1358         cmd->changed = true;
1359         return 0;
1360
1361 }
1362
1363 int dm_cache_set_dirty(struct dm_cache_metadata *cmd,
1364                        dm_cblock_t cblock, bool dirty)
1365 {
1366         int r;
1367
1368         WRITE_LOCK(cmd);
1369         r = __dirty(cmd, cblock, dirty);
1370         WRITE_UNLOCK(cmd);
1371
1372         return r;
1373 }
1374
1375 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1376                                  struct dm_cache_statistics *stats)
1377 {
1378         READ_LOCK_VOID(cmd);
1379         *stats = cmd->stats;
1380         READ_UNLOCK(cmd);
1381 }
1382
1383 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1384                                  struct dm_cache_statistics *stats)
1385 {
1386         WRITE_LOCK_VOID(cmd);
1387         cmd->stats = *stats;
1388         WRITE_UNLOCK(cmd);
1389 }
1390
1391 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1392 {
1393         int r = -EINVAL;
1394         flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1395                                  clear_clean_shutdown);
1396
1397         WRITE_LOCK(cmd);
1398         if (cmd->fail_io)
1399                 goto out;
1400
1401         r = __commit_transaction(cmd, mutator);
1402         if (r)
1403                 goto out;
1404
1405         r = __begin_transaction(cmd);
1406 out:
1407         WRITE_UNLOCK(cmd);
1408         return r;
1409 }
1410
1411 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1412                                            dm_block_t *result)
1413 {
1414         int r = -EINVAL;
1415
1416         READ_LOCK(cmd);
1417         if (!cmd->fail_io)
1418                 r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1419         READ_UNLOCK(cmd);
1420
1421         return r;
1422 }
1423
1424 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1425                                    dm_block_t *result)
1426 {
1427         int r = -EINVAL;
1428
1429         READ_LOCK(cmd);
1430         if (!cmd->fail_io)
1431                 r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1432         READ_UNLOCK(cmd);
1433
1434         return r;
1435 }
1436
1437 /*----------------------------------------------------------------*/
1438
1439 static int get_hint(uint32_t index, void *value_le, void *context)
1440 {
1441         uint32_t value;
1442         struct dm_cache_policy *policy = context;
1443
1444         value = policy_get_hint(policy, to_cblock(index));
1445         *((__le32 *) value_le) = cpu_to_le32(value);
1446
1447         return 0;
1448 }
1449
1450 /*
1451  * It's quicker to always delete the hint array, and recreate with
1452  * dm_array_new().
1453  */
1454 static int write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1455 {
1456         int r;
1457         size_t hint_size;
1458         const char *policy_name = dm_cache_policy_get_name(policy);
1459         const unsigned *policy_version = dm_cache_policy_get_version(policy);
1460
1461         if (!policy_name[0] ||
1462             (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1463                 return -EINVAL;
1464
1465         strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1466         memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1467
1468         hint_size = dm_cache_policy_get_hint_size(policy);
1469         if (!hint_size)
1470                 return 0; /* short-circuit hints initialization */
1471         cmd->policy_hint_size = hint_size;
1472
1473         if (cmd->hint_root) {
1474                 r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1475                 if (r)
1476                         return r;
1477         }
1478
1479         return dm_array_new(&cmd->hint_info, &cmd->hint_root,
1480                             from_cblock(cmd->cache_blocks),
1481                             get_hint, policy);
1482 }
1483
1484 int dm_cache_write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1485 {
1486         int r;
1487
1488         WRITE_LOCK(cmd);
1489         r = write_hints(cmd, policy);
1490         WRITE_UNLOCK(cmd);
1491
1492         return r;
1493 }
1494
1495 int dm_cache_metadata_all_clean(struct dm_cache_metadata *cmd, bool *result)
1496 {
1497         int r;
1498
1499         READ_LOCK(cmd);
1500         r = blocks_are_unmapped_or_clean(cmd, 0, cmd->cache_blocks, result);
1501         READ_UNLOCK(cmd);
1502
1503         return r;
1504 }
1505
1506 void dm_cache_metadata_set_read_only(struct dm_cache_metadata *cmd)
1507 {
1508         WRITE_LOCK_VOID(cmd);
1509         dm_bm_set_read_only(cmd->bm);
1510         WRITE_UNLOCK(cmd);
1511 }
1512
1513 void dm_cache_metadata_set_read_write(struct dm_cache_metadata *cmd)
1514 {
1515         WRITE_LOCK_VOID(cmd);
1516         dm_bm_set_read_write(cmd->bm);
1517         WRITE_UNLOCK(cmd);
1518 }
1519
1520 int dm_cache_metadata_set_needs_check(struct dm_cache_metadata *cmd)
1521 {
1522         int r;
1523         struct dm_block *sblock;
1524         struct cache_disk_superblock *disk_super;
1525
1526         WRITE_LOCK(cmd);
1527         set_bit(NEEDS_CHECK, &cmd->flags);
1528
1529         r = superblock_lock(cmd, &sblock);
1530         if (r) {
1531                 DMERR("couldn't read superblock");
1532                 goto out;
1533         }
1534
1535         disk_super = dm_block_data(sblock);
1536         disk_super->flags = cpu_to_le32(cmd->flags);
1537
1538         dm_bm_unlock(sblock);
1539
1540 out:
1541         WRITE_UNLOCK(cmd);
1542         return r;
1543 }
1544
1545 int dm_cache_metadata_needs_check(struct dm_cache_metadata *cmd, bool *result)
1546 {
1547         READ_LOCK(cmd);
1548         *result = !!test_bit(NEEDS_CHECK, &cmd->flags);
1549         READ_UNLOCK(cmd);
1550
1551         return 0;
1552 }
1553
1554 int dm_cache_metadata_abort(struct dm_cache_metadata *cmd)
1555 {
1556         int r = -EINVAL;
1557         struct dm_block_manager *old_bm = NULL, *new_bm = NULL;
1558
1559         /* fail_io is double-checked with cmd->root_lock held below */
1560         if (unlikely(cmd->fail_io))
1561                 return r;
1562
1563         /*
1564          * Replacement block manager (new_bm) is created and old_bm destroyed outside of
1565          * cmd root_lock to avoid ABBA deadlock that would result (due to life-cycle of
1566          * shrinker associated with the block manager's bufio client vs cmd root_lock).
1567          * - must take shrinker_rwsem without holding cmd->root_lock
1568          */
1569         new_bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
1570                                          CACHE_METADATA_CACHE_SIZE,
1571                                          CACHE_MAX_CONCURRENT_LOCKS);
1572
1573         WRITE_LOCK(cmd);
1574         if (cmd->fail_io) {
1575                 WRITE_UNLOCK(cmd);
1576                 goto out;
1577         }
1578
1579         __destroy_persistent_data_objects(cmd, false);
1580         old_bm = cmd->bm;
1581         if (IS_ERR(new_bm)) {
1582                 DMERR("could not create block manager during abort");
1583                 cmd->bm = NULL;
1584                 r = PTR_ERR(new_bm);
1585                 goto out_unlock;
1586         }
1587
1588         cmd->bm = new_bm;
1589         r = __open_or_format_metadata(cmd, false);
1590         if (r) {
1591                 cmd->bm = NULL;
1592                 goto out_unlock;
1593         }
1594         new_bm = NULL;
1595 out_unlock:
1596         if (r)
1597                 cmd->fail_io = true;
1598         WRITE_UNLOCK(cmd);
1599         dm_block_manager_destroy(old_bm);
1600 out:
1601         if (new_bm && !IS_ERR(new_bm))
1602                 dm_block_manager_destroy(new_bm);
1603
1604         return r;
1605 }