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 2
  29
  30 /*
  31  *  3 for btree insert +
  32  *  2 for btree lookup used within space map
  33  */
  34 #define CACHE_MAX_CONCURRENT_LOCKS 5
  35 #define SPACE_MAP_ROOT_SIZE 128
  36
  37 enum superblock_flag_bits {
  38         /* for spotting crashes that would invalidate the dirty bitset */
  39         CLEAN_SHUTDOWN,
  40         /* metadata must be checked using the tools */
  41         NEEDS_CHECK,
  42 };
  43
  44 /*
  45  * Each mapping from cache block -> origin block carries a set of flags.
  46  */
  47 enum mapping_bits {
  48         /*
  49          * A valid mapping.  Because we're using an array we clear this
  50          * flag for an non existant mapping.
  51          */
  52         M_VALID = 1,
  53
  54         /*
  55          * The data on the cache is different from that on the origin.
  56          * This flag is only used by metadata format 1.
  57          */
  58         M_DIRTY = 2
  59 };
  60
  61 struct cache_disk_superblock {
  62         __le32 csum;
  63         __le32 flags;
  64         __le64 blocknr;
  65
  66         __u8 uuid[16];
  67         __le64 magic;
  68         __le32 version;
  69
  70         __u8 policy_name[CACHE_POLICY_NAME_SIZE];
  71         __le32 policy_hint_size;
  72
  73         __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
  74         __le64 mapping_root;
  75         __le64 hint_root;
  76
  77         __le64 discard_root;
  78         __le64 discard_block_size;
  79         __le64 discard_nr_blocks;
  80
  81         __le32 data_block_size;
  82         __le32 metadata_block_size;
  83         __le32 cache_blocks;
  84
  85         __le32 compat_flags;
  86         __le32 compat_ro_flags;
  87         __le32 incompat_flags;
  88
  89         __le32 read_hits;
  90         __le32 read_misses;
  91         __le32 write_hits;
  92         __le32 write_misses;
  93
  94         __le32 policy_version[CACHE_POLICY_VERSION_SIZE];
  95
  96         /*
  97          * Metadata format 2 fields.
  98          */
  99         __le64 dirty_root;
 100 } __packed;
 101
 102 struct dm_cache_metadata {
 103         atomic_t ref_count;
 104         struct list_head list;
 105
 106         unsigned version;
 107         struct block_device *bdev;
 108         struct dm_block_manager *bm;
 109         struct dm_space_map *metadata_sm;
 110         struct dm_transaction_manager *tm;
 111
 112         struct dm_array_info info;
 113         struct dm_array_info hint_info;
 114         struct dm_disk_bitset discard_info;
 115
 116         struct rw_semaphore root_lock;
 117         unsigned long flags;
 118         dm_block_t root;
 119         dm_block_t hint_root;
 120         dm_block_t discard_root;
 121
 122         sector_t discard_block_size;
 123         dm_dblock_t discard_nr_blocks;
 124
 125         sector_t data_block_size;
 126         dm_cblock_t cache_blocks;
 127         bool changed:1;
 128         bool clean_when_opened:1;
 129
 130         char policy_name[CACHE_POLICY_NAME_SIZE];
 131         unsigned policy_version[CACHE_POLICY_VERSION_SIZE];
 132         size_t policy_hint_size;
 133         struct dm_cache_statistics stats;
 134
 135         /*
 136          * Reading the space map root can fail, so we read it into this
 137          * buffer before the superblock is locked and updated.
 138          */
 139         __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
 140
 141         /*
 142          * Set if a transaction has to be aborted but the attempt to roll
 143          * back to the previous (good) transaction failed.  The only
 144          * metadata operation permissible in this state is the closing of
 145          * the device.
 146          */
 147         bool fail_io:1;
 148
 149         /*
 150          * Metadata format 2 fields.
 151          */
 152         dm_block_t dirty_root;
 153         struct dm_disk_bitset dirty_info;
 154
 155         /*
 156          * These structures are used when loading metadata.  They're too
 157          * big to put on the stack.
 158          */
 159         struct dm_array_cursor mapping_cursor;
 160         struct dm_array_cursor hint_cursor;
 161         struct dm_bitset_cursor dirty_cursor;
 162 };
 163
 164 /*-------------------------------------------------------------------
 165  * superblock validator
 166  *-----------------------------------------------------------------*/
 167
 168 #define SUPERBLOCK_CSUM_XOR 9031977
 169
 170 static void sb_prepare_for_write(struct dm_block_validator *v,
 171                                  struct dm_block *b,
 172                                  size_t sb_block_size)
 173 {
 174         struct cache_disk_superblock *disk_super = dm_block_data(b);
 175
 176         disk_super->blocknr = cpu_to_le64(dm_block_location(b));
 177         disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
 178                                                       sb_block_size - sizeof(__le32),
 179                                                       SUPERBLOCK_CSUM_XOR));
 180 }
 181
 182 static int check_metadata_version(struct cache_disk_superblock *disk_super)
 183 {
 184         uint32_t metadata_version = le32_to_cpu(disk_super->version);
 185
 186         if (metadata_version < MIN_CACHE_VERSION || metadata_version > MAX_CACHE_VERSION) {
 187                 DMERR("Cache metadata version %u found, but only versions between %u and %u supported.",
 188                       metadata_version, MIN_CACHE_VERSION, MAX_CACHE_VERSION);
 189                 return -EINVAL;
 190         }
 191
 192         return 0;
 193 }
 194
 195 static int sb_check(struct dm_block_validator *v,
 196                     struct dm_block *b,
 197                     size_t sb_block_size)
 198 {
 199         struct cache_disk_superblock *disk_super = dm_block_data(b);
 200         __le32 csum_le;
 201
 202         if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
 203                 DMERR("sb_check failed: blocknr %llu: wanted %llu",
 204                       le64_to_cpu(disk_super->blocknr),
 205                       (unsigned long long)dm_block_location(b));
 206                 return -ENOTBLK;
 207         }
 208
 209         if (le64_to_cpu(disk_super->magic) != CACHE_SUPERBLOCK_MAGIC) {
 210                 DMERR("sb_check failed: magic %llu: wanted %llu",
 211                       le64_to_cpu(disk_super->magic),
 212                       (unsigned long long)CACHE_SUPERBLOCK_MAGIC);
 213                 return -EILSEQ;
 214         }
 215
 216         csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
 217                                              sb_block_size - sizeof(__le32),
 218                                              SUPERBLOCK_CSUM_XOR));
 219         if (csum_le != disk_super->csum) {
 220                 DMERR("sb_check failed: csum %u: wanted %u",
 221                       le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
 222                 return -EILSEQ;
 223         }
 224
 225         return check_metadata_version(disk_super);
 226 }
 227
 228 static struct dm_block_validator sb_validator = {
 229         .name = "superblock",
 230         .prepare_for_write = sb_prepare_for_write,
 231         .check = sb_check
 232 };
 233
 234 /*----------------------------------------------------------------*/
 235
 236 static int superblock_read_lock(struct dm_cache_metadata *cmd,
 237                                 struct dm_block **sblock)
 238 {
 239         return dm_bm_read_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 240                                &sb_validator, sblock);
 241 }
 242
 243 static int superblock_lock_zero(struct dm_cache_metadata *cmd,
 244                                 struct dm_block **sblock)
 245 {
 246         return dm_bm_write_lock_zero(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 247                                      &sb_validator, sblock);
 248 }
 249
 250 static int superblock_lock(struct dm_cache_metadata *cmd,
 251                            struct dm_block **sblock)
 252 {
 253         return dm_bm_write_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 254                                 &sb_validator, sblock);
 255 }
 256
 257 /*----------------------------------------------------------------*/
 258
 259 static int __superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
 260 {
 261         int r;
 262         unsigned i;
 263         struct dm_block *b;
 264         __le64 *data_le, zero = cpu_to_le64(0);
 265         unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
 266
 267         /*
 268          * We can't use a validator here - it may be all zeroes.
 269          */
 270         r = dm_bm_read_lock(bm, CACHE_SUPERBLOCK_LOCATION, NULL, &b);
 271         if (r)
 272                 return r;
 273
 274         data_le = dm_block_data(b);
 275         *result = true;
 276         for (i = 0; i < sb_block_size; i++) {
 277                 if (data_le[i] != zero) {
 278                         *result = false;
 279                         break;
 280                 }
 281         }
 282
 283         dm_bm_unlock(b);
 284
 285         return 0;
 286 }
 287
 288 static void __setup_mapping_info(struct dm_cache_metadata *cmd)
 289 {
 290         struct dm_btree_value_type vt;
 291
 292         vt.context = NULL;
 293         vt.size = sizeof(__le64);
 294         vt.inc = NULL;
 295         vt.dec = NULL;
 296         vt.equal = NULL;
 297         dm_array_info_init(&cmd->info, cmd->tm, &vt);
 298
 299         if (cmd->policy_hint_size) {
 300                 vt.size = sizeof(__le32);
 301                 dm_array_info_init(&cmd->hint_info, cmd->tm, &vt);
 302         }
 303 }
 304
 305 static int __save_sm_root(struct dm_cache_metadata *cmd)
 306 {
 307         int r;
 308         size_t metadata_len;
 309
 310         r = dm_sm_root_size(cmd->metadata_sm, &metadata_len);
 311         if (r < 0)
 312                 return r;
 313
 314         return dm_sm_copy_root(cmd->metadata_sm, &cmd->metadata_space_map_root,
 315                                metadata_len);
 316 }
 317
 318 static void __copy_sm_root(struct dm_cache_metadata *cmd,
 319                            struct cache_disk_superblock *disk_super)
 320 {
 321         memcpy(&disk_super->metadata_space_map_root,
 322                &cmd->metadata_space_map_root,
 323                sizeof(cmd->metadata_space_map_root));
 324 }
 325
 326 static bool separate_dirty_bits(struct dm_cache_metadata *cmd)
 327 {
 328         return cmd->version >= 2;
 329 }
 330
 331 static int __write_initial_superblock(struct dm_cache_metadata *cmd)
 332 {
 333         int r;
 334         struct dm_block *sblock;
 335         struct cache_disk_superblock *disk_super;
 336         sector_t bdev_size = i_size_read(cmd->bdev->bd_inode) >> SECTOR_SHIFT;
 337
 338         /* FIXME: see if we can lose the max sectors limit */
 339         if (bdev_size > DM_CACHE_METADATA_MAX_SECTORS)
 340                 bdev_size = DM_CACHE_METADATA_MAX_SECTORS;
 341
 342         r = dm_tm_pre_commit(cmd->tm);
 343         if (r < 0)
 344                 return r;
 345
 346         /*
 347          * dm_sm_copy_root() can fail.  So we need to do it before we start
 348          * updating the superblock.
 349          */
 350         r = __save_sm_root(cmd);
 351         if (r)
 352                 return r;
 353
 354         r = superblock_lock_zero(cmd, &sblock);
 355         if (r)
 356                 return r;
 357
 358         disk_super = dm_block_data(sblock);
 359         disk_super->flags = 0;
 360         memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
 361         disk_super->magic = cpu_to_le64(CACHE_SUPERBLOCK_MAGIC);
 362         disk_super->version = cpu_to_le32(cmd->version);
 363         memset(disk_super->policy_name, 0, sizeof(disk_super->policy_name));
 364         memset(disk_super->policy_version, 0, sizeof(disk_super->policy_version));
 365         disk_super->policy_hint_size = 0;
 366
 367         __copy_sm_root(cmd, disk_super);
 368
 369         disk_super->mapping_root = cpu_to_le64(cmd->root);
 370         disk_super->hint_root = cpu_to_le64(cmd->hint_root);
 371         disk_super->discard_root = cpu_to_le64(cmd->discard_root);
 372         disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
 373         disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
 374         disk_super->metadata_block_size = cpu_to_le32(DM_CACHE_METADATA_BLOCK_SIZE);
 375         disk_super->data_block_size = cpu_to_le32(cmd->data_block_size);
 376         disk_super->cache_blocks = cpu_to_le32(0);
 377
 378         disk_super->read_hits = cpu_to_le32(0);
 379         disk_super->read_misses = cpu_to_le32(0);
 380         disk_super->write_hits = cpu_to_le32(0);
 381         disk_super->write_misses = cpu_to_le32(0);
 382
 383         if (separate_dirty_bits(cmd))
 384                 disk_super->dirty_root = cpu_to_le64(cmd->dirty_root);
 385
 386         return dm_tm_commit(cmd->tm, sblock);
 387 }
 388
 389 static int __format_metadata(struct dm_cache_metadata *cmd)
 390 {
 391         int r;
 392
 393         r = dm_tm_create_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 394                                  &cmd->tm, &cmd->metadata_sm);
 395         if (r < 0) {
 396                 DMERR("tm_create_with_sm failed");
 397                 return r;
 398         }
 399
 400         __setup_mapping_info(cmd);
 401
 402         r = dm_array_empty(&cmd->info, &cmd->root);
 403         if (r < 0)
 404                 goto bad;
 405
 406         if (separate_dirty_bits(cmd)) {
 407                 dm_disk_bitset_init(cmd->tm, &cmd->dirty_info);
 408                 r = dm_bitset_empty(&cmd->dirty_info, &cmd->dirty_root);
 409                 if (r < 0)
 410                         goto bad;
 411         }
 412
 413         dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
 414         r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root);
 415         if (r < 0)
 416                 goto bad;
 417
 418         cmd->discard_block_size = 0;
 419         cmd->discard_nr_blocks = 0;
 420
 421         r = __write_initial_superblock(cmd);
 422         if (r)
 423                 goto bad;
 424
 425         cmd->clean_when_opened = true;
 426         return 0;
 427
 428 bad:
 429         dm_tm_destroy(cmd->tm);
 430         dm_sm_destroy(cmd->metadata_sm);
 431
 432         return r;
 433 }
 434
 435 static int __check_incompat_features(struct cache_disk_superblock *disk_super,
 436                                      struct dm_cache_metadata *cmd)
 437 {
 438         uint32_t incompat_flags, features;
 439
 440         incompat_flags = le32_to_cpu(disk_super->incompat_flags);
 441         features = incompat_flags & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
 442         if (features) {
 443                 DMERR("could not access metadata due to unsupported optional features (%lx).",
 444                       (unsigned long)features);
 445                 return -EINVAL;
 446         }
 447
 448         /*
 449          * Check for read-only metadata to skip the following RDWR checks.
 450          */
 451         if (get_disk_ro(cmd->bdev->bd_disk))
 452                 return 0;
 453
 454         features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP;
 455         if (features) {
 456                 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
 457                       (unsigned long)features);
 458                 return -EINVAL;
 459         }
 460
 461         return 0;
 462 }
 463
 464 static int __open_metadata(struct dm_cache_metadata *cmd)
 465 {
 466         int r;
 467         struct dm_block *sblock;
 468         struct cache_disk_superblock *disk_super;
 469         unsigned long sb_flags;
 470
 471         r = superblock_read_lock(cmd, &sblock);
 472         if (r < 0) {
 473                 DMERR("couldn't read lock superblock");
 474                 return r;
 475         }
 476
 477         disk_super = dm_block_data(sblock);
 478
 479         /* Verify the data block size hasn't changed */
 480         if (le32_to_cpu(disk_super->data_block_size) != cmd->data_block_size) {
 481                 DMERR("changing the data block size (from %u to %llu) is not supported",
 482                       le32_to_cpu(disk_super->data_block_size),
 483                       (unsigned long long)cmd->data_block_size);
 484                 r = -EINVAL;
 485                 goto bad;
 486         }
 487
 488         r = __check_incompat_features(disk_super, cmd);
 489         if (r < 0)
 490                 goto bad;
 491
 492         r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 493                                disk_super->metadata_space_map_root,
 494                                sizeof(disk_super->metadata_space_map_root),
 495                                &cmd->tm, &cmd->metadata_sm);
 496         if (r < 0) {
 497                 DMERR("tm_open_with_sm failed");
 498                 goto bad;
 499         }
 500
 501         __setup_mapping_info(cmd);
 502         dm_disk_bitset_init(cmd->tm, &cmd->dirty_info);
 503         dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
 504         sb_flags = le32_to_cpu(disk_super->flags);
 505         cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags);
 506         dm_bm_unlock(sblock);
 507
 508         return 0;
 509
 510 bad:
 511         dm_bm_unlock(sblock);
 512         return r;
 513 }
 514
 515 static int __open_or_format_metadata(struct dm_cache_metadata *cmd,
 516                                      bool format_device)
 517 {
 518         int r;
 519         bool unformatted = false;
 520
 521         r = __superblock_all_zeroes(cmd->bm, &unformatted);
 522         if (r)
 523                 return r;
 524
 525         if (unformatted)
 526                 return format_device ? __format_metadata(cmd) : -EPERM;
 527
 528         return __open_metadata(cmd);
 529 }
 530
 531 static int __create_persistent_data_objects(struct dm_cache_metadata *cmd,
 532                                             bool may_format_device)
 533 {
 534         int r;
 535         cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
 536                                           CACHE_MAX_CONCURRENT_LOCKS);
 537         if (IS_ERR(cmd->bm)) {
 538                 DMERR("could not create block manager");
 539                 return PTR_ERR(cmd->bm);
 540         }
 541
 542         r = __open_or_format_metadata(cmd, may_format_device);
 543         if (r)
 544                 dm_block_manager_destroy(cmd->bm);
 545
 546         return r;
 547 }
 548
 549 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd)
 550 {
 551         dm_sm_destroy(cmd->metadata_sm);
 552         dm_tm_destroy(cmd->tm);
 553         dm_block_manager_destroy(cmd->bm);
 554 }
 555
 556 typedef unsigned long (*flags_mutator)(unsigned long);
 557
 558 static void update_flags(struct cache_disk_superblock *disk_super,
 559                          flags_mutator mutator)
 560 {
 561         uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
 562         disk_super->flags = cpu_to_le32(sb_flags);
 563 }
 564
 565 static unsigned long set_clean_shutdown(unsigned long flags)
 566 {
 567         set_bit(CLEAN_SHUTDOWN, &flags);
 568         return flags;
 569 }
 570
 571 static unsigned long clear_clean_shutdown(unsigned long flags)
 572 {
 573         clear_bit(CLEAN_SHUTDOWN, &flags);
 574         return flags;
 575 }
 576
 577 static void read_superblock_fields(struct dm_cache_metadata *cmd,
 578                                    struct cache_disk_superblock *disk_super)
 579 {
 580         cmd->version = le32_to_cpu(disk_super->version);
 581         cmd->flags = le32_to_cpu(disk_super->flags);
 582         cmd->root = le64_to_cpu(disk_super->mapping_root);
 583         cmd->hint_root = le64_to_cpu(disk_super->hint_root);
 584         cmd->discard_root = le64_to_cpu(disk_super->discard_root);
 585         cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
 586         cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
 587         cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
 588         cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
 589         strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
 590         cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
 591         cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
 592         cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
 593         cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
 594
 595         cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
 596         cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
 597         cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
 598         cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
 599
 600         if (separate_dirty_bits(cmd))
 601                 cmd->dirty_root = le64_to_cpu(disk_super->dirty_root);
 602
 603         cmd->changed = false;
 604 }
 605
 606 /*
 607  * The mutator updates the superblock flags.
 608  */
 609 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
 610                                      flags_mutator mutator)
 611 {
 612         int r;
 613         struct cache_disk_superblock *disk_super;
 614         struct dm_block *sblock;
 615
 616         r = superblock_lock(cmd, &sblock);
 617         if (r)
 618                 return r;
 619
 620         disk_super = dm_block_data(sblock);
 621         update_flags(disk_super, mutator);
 622         read_superblock_fields(cmd, disk_super);
 623         dm_bm_unlock(sblock);
 624
 625         return dm_bm_flush(cmd->bm);
 626 }
 627
 628 static int __begin_transaction(struct dm_cache_metadata *cmd)
 629 {
 630         int r;
 631         struct cache_disk_superblock *disk_super;
 632         struct dm_block *sblock;
 633
 634         /*
 635          * We re-read the superblock every time.  Shouldn't need to do this
 636          * really.
 637          */
 638         r = superblock_read_lock(cmd, &sblock);
 639         if (r)
 640                 return r;
 641
 642         disk_super = dm_block_data(sblock);
 643         read_superblock_fields(cmd, disk_super);
 644         dm_bm_unlock(sblock);
 645
 646         return 0;
 647 }
 648
 649 static int __commit_transaction(struct dm_cache_metadata *cmd,
 650                                 flags_mutator mutator)
 651 {
 652         int r;
 653         struct cache_disk_superblock *disk_super;
 654         struct dm_block *sblock;
 655
 656         /*
 657          * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
 658          */
 659         BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
 660
 661         if (separate_dirty_bits(cmd)) {
 662                 r = dm_bitset_flush(&cmd->dirty_info, cmd->dirty_root,
 663                                     &cmd->dirty_root);
 664                 if (r)
 665                         return r;
 666         }
 667
 668         r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
 669                             &cmd->discard_root);
 670         if (r)
 671                 return r;
 672
 673         r = dm_tm_pre_commit(cmd->tm);
 674         if (r < 0)
 675                 return r;
 676
 677         r = __save_sm_root(cmd);
 678         if (r)
 679                 return r;
 680
 681         r = superblock_lock(cmd, &sblock);
 682         if (r)
 683                 return r;
 684
 685         disk_super = dm_block_data(sblock);
 686
 687         disk_super->flags = cpu_to_le32(cmd->flags);
 688         if (mutator)
 689                 update_flags(disk_super, mutator);
 690
 691         disk_super->mapping_root = cpu_to_le64(cmd->root);
 692         if (separate_dirty_bits(cmd))
 693                 disk_super->dirty_root = cpu_to_le64(cmd->dirty_root);
 694         disk_super->hint_root = cpu_to_le64(cmd->hint_root);
 695         disk_super->discard_root = cpu_to_le64(cmd->discard_root);
 696         disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
 697         disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
 698         disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
 699         strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
 700         disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
 701         disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
 702         disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
 703
 704         disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
 705         disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
 706         disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
 707         disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
 708         __copy_sm_root(cmd, disk_super);
 709
 710         return dm_tm_commit(cmd->tm, sblock);
 711 }
 712
 713 /*----------------------------------------------------------------*/
 714
 715 /*
 716  * The mappings are held in a dm-array that has 64-bit values stored in
 717  * little-endian format.  The index is the cblock, the high 48bits of the
 718  * value are the oblock and the low 16 bit the flags.
 719  */
 720 #define FLAGS_MASK ((1 << 16) - 1)
 721
 722 static __le64 pack_value(dm_oblock_t block, unsigned flags)
 723 {
 724         uint64_t value = from_oblock(block);
 725         value <<= 16;
 726         value = value | (flags & FLAGS_MASK);
 727         return cpu_to_le64(value);
 728 }
 729
 730 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
 731 {
 732         uint64_t value = le64_to_cpu(value_le);
 733         uint64_t b = value >> 16;
 734         *block = to_oblock(b);
 735         *flags = value & FLAGS_MASK;
 736 }
 737
 738 /*----------------------------------------------------------------*/
 739
 740 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
 741                                                sector_t data_block_size,
 742                                                bool may_format_device,
 743                                                size_t policy_hint_size,
 744                                                unsigned metadata_version)
 745 {
 746         int r;
 747         struct dm_cache_metadata *cmd;
 748
 749         cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
 750         if (!cmd) {
 751                 DMERR("could not allocate metadata struct");
 752                 return ERR_PTR(-ENOMEM);
 753         }
 754
 755         cmd->version = metadata_version;
 756         atomic_set(&cmd->ref_count, 1);
 757         init_rwsem(&cmd->root_lock);
 758         cmd->bdev = bdev;
 759         cmd->data_block_size = data_block_size;
 760         cmd->cache_blocks = 0;
 761         cmd->policy_hint_size = policy_hint_size;
 762         cmd->changed = true;
 763         cmd->fail_io = false;
 764
 765         r = __create_persistent_data_objects(cmd, may_format_device);
 766         if (r) {
 767                 kfree(cmd);
 768                 return ERR_PTR(r);
 769         }
 770
 771         r = __begin_transaction_flags(cmd, clear_clean_shutdown);
 772         if (r < 0) {
 773                 dm_cache_metadata_close(cmd);
 774                 return ERR_PTR(r);
 775         }
 776
 777         return cmd;
 778 }
 779
 780 /*
 781  * We keep a little list of ref counted metadata objects to prevent two
 782  * different target instances creating separate bufio instances.  This is
 783  * an issue if a table is reloaded before the suspend.
 784  */
 785 static DEFINE_MUTEX(table_lock);
 786 static LIST_HEAD(table);
 787
 788 static struct dm_cache_metadata *lookup(struct block_device *bdev)
 789 {
 790         struct dm_cache_metadata *cmd;
 791
 792         list_for_each_entry(cmd, &table, list)
 793                 if (cmd->bdev == bdev) {
 794                         atomic_inc(&cmd->ref_count);
 795                         return cmd;
 796                 }
 797
 798         return NULL;
 799 }
 800
 801 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
 802                                                 sector_t data_block_size,
 803                                                 bool may_format_device,
 804                                                 size_t policy_hint_size,
 805                                                 unsigned metadata_version)
 806 {
 807         struct dm_cache_metadata *cmd, *cmd2;
 808
 809         mutex_lock(&table_lock);
 810         cmd = lookup(bdev);
 811         mutex_unlock(&table_lock);
 812
 813         if (cmd)
 814                 return cmd;
 815
 816         cmd = metadata_open(bdev, data_block_size, may_format_device,
 817                             policy_hint_size, metadata_version);
 818         if (!IS_ERR(cmd)) {
 819                 mutex_lock(&table_lock);
 820                 cmd2 = lookup(bdev);
 821                 if (cmd2) {
 822                         mutex_unlock(&table_lock);
 823                         __destroy_persistent_data_objects(cmd);
 824                         kfree(cmd);
 825                         return cmd2;
 826                 }
 827                 list_add(&cmd->list, &table);
 828                 mutex_unlock(&table_lock);
 829         }
 830
 831         return cmd;
 832 }
 833
 834 static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
 835 {
 836         if (cmd->data_block_size != data_block_size) {
 837                 DMERR("data_block_size (%llu) different from that in metadata (%llu)",
 838                       (unsigned long long) data_block_size,
 839                       (unsigned long long) cmd->data_block_size);
 840                 return false;
 841         }
 842
 843         return true;
 844 }
 845
 846 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
 847                                                  sector_t data_block_size,
 848                                                  bool may_format_device,
 849                                                  size_t policy_hint_size,
 850                                                  unsigned metadata_version)
 851 {
 852         struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size, may_format_device,
 853                                                        policy_hint_size, metadata_version);
 854
 855         if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
 856                 dm_cache_metadata_close(cmd);
 857                 return ERR_PTR(-EINVAL);
 858         }
 859
 860         return cmd;
 861 }
 862
 863 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
 864 {
 865         if (atomic_dec_and_test(&cmd->ref_count)) {
 866                 mutex_lock(&table_lock);
 867                 list_del(&cmd->list);
 868                 mutex_unlock(&table_lock);
 869
 870                 if (!cmd->fail_io)
 871                         __destroy_persistent_data_objects(cmd);
 872                 kfree(cmd);
 873         }
 874 }
 875
 876 /*
 877  * Checks that the given cache block is either unmapped or clean.
 878  */
 879 static int block_clean_combined_dirty(struct dm_cache_metadata *cmd, dm_cblock_t b,
 880                                       bool *result)
 881 {
 882         int r;
 883         __le64 value;
 884         dm_oblock_t ob;
 885         unsigned flags;
 886
 887         r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
 888         if (r)
 889                 return r;
 890
 891         unpack_value(value, &ob, &flags);
 892         *result = !((flags & M_VALID) && (flags & M_DIRTY));
 893
 894         return 0;
 895 }
 896
 897 static int blocks_are_clean_combined_dirty(struct dm_cache_metadata *cmd,
 898                                            dm_cblock_t begin, dm_cblock_t end,
 899                                            bool *result)
 900 {
 901         int r;
 902         *result = true;
 903
 904         while (begin != end) {
 905                 r = block_clean_combined_dirty(cmd, begin, result);
 906                 if (r) {
 907                         DMERR("block_clean_combined_dirty failed");
 908                         return r;
 909                 }
 910
 911                 if (!*result) {
 912                         DMERR("cache block %llu is dirty",
 913                               (unsigned long long) from_cblock(begin));
 914                         return 0;
 915                 }
 916
 917                 begin = to_cblock(from_cblock(begin) + 1);
 918         }
 919
 920         return 0;
 921 }
 922
 923 static int blocks_are_clean_separate_dirty(struct dm_cache_metadata *cmd,
 924                                            dm_cblock_t begin, dm_cblock_t end,
 925                                            bool *result)
 926 {
 927         int r;
 928         bool dirty_flag;
 929         *result = true;
 930
 931         r = dm_bitset_cursor_begin(&cmd->dirty_info, cmd->dirty_root,
 932                                    from_cblock(begin), &cmd->dirty_cursor);
 933         if (r) {
 934                 DMERR("%s: dm_bitset_cursor_begin for dirty failed", __func__);
 935                 return r;
 936         }
 937
 938         r = dm_bitset_cursor_skip(&cmd->dirty_cursor, from_cblock(begin));
 939         if (r) {
 940                 DMERR("%s: dm_bitset_cursor_skip for dirty failed", __func__);
 941                 dm_bitset_cursor_end(&cmd->dirty_cursor);
 942                 return r;
 943         }
 944
 945         while (begin != end) {
 946                 /*
 947                  * We assume that unmapped blocks have their dirty bit
 948                  * cleared.
 949                  */
 950                 dirty_flag = dm_bitset_cursor_get_value(&cmd->dirty_cursor);
 951                 if (dirty_flag) {
 952                         DMERR("%s: cache block %llu is dirty", __func__,
 953                               (unsigned long long) from_cblock(begin));
 954                         dm_bitset_cursor_end(&cmd->dirty_cursor);
 955                         *result = false;
 956                         return 0;
 957                 }
 958
 959                 r = dm_bitset_cursor_next(&cmd->dirty_cursor);
 960                 if (r) {
 961                         DMERR("%s: dm_bitset_cursor_next for dirty failed", __func__);
 962                         dm_bitset_cursor_end(&cmd->dirty_cursor);
 963                         return r;
 964                 }
 965
 966                 begin = to_cblock(from_cblock(begin) + 1);
 967         }
 968
 969         dm_bitset_cursor_end(&cmd->dirty_cursor);
 970
 971         return 0;
 972 }
 973
 974 static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
 975                                         dm_cblock_t begin, dm_cblock_t end,
 976                                         bool *result)
 977 {
 978         if (separate_dirty_bits(cmd))
 979                 return blocks_are_clean_separate_dirty(cmd, begin, end, result);
 980         else
 981                 return blocks_are_clean_combined_dirty(cmd, begin, end, result);
 982 }
 983
 984 static bool cmd_write_lock(struct dm_cache_metadata *cmd)
 985 {
 986         down_write(&cmd->root_lock);
 987         if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) {
 988                 up_write(&cmd->root_lock);
 989                 return false;
 990         }
 991         return true;
 992 }
 993
 994 #define WRITE_LOCK(cmd)                         \
 995         do {                                    \
 996                 if (!cmd_write_lock((cmd)))     \
 997                         return -EINVAL;         \
 998         } while(0)
 999
1000 #define WRITE_LOCK_VOID(cmd)                    \
1001         do {                                    \
1002                 if (!cmd_write_lock((cmd)))     \
1003                         return;                 \
1004         } while(0)
1005
1006 #define WRITE_UNLOCK(cmd) \
1007         up_write(&(cmd)->root_lock)
1008
1009 static bool cmd_read_lock(struct dm_cache_metadata *cmd)
1010 {
1011         down_read(&cmd->root_lock);
1012         if (cmd->fail_io) {
1013                 up_read(&cmd->root_lock);
1014                 return false;
1015         }
1016         return true;
1017 }
1018
1019 #define READ_LOCK(cmd)                          \
1020         do {                                    \
1021                 if (!cmd_read_lock((cmd)))      \
1022                         return -EINVAL;         \
1023         } while(0)
1024
1025 #define READ_LOCK_VOID(cmd)                     \
1026         do {                                    \
1027                 if (!cmd_read_lock((cmd)))      \
1028                         return;                 \
1029         } while(0)
1030
1031 #define READ_UNLOCK(cmd) \
1032         up_read(&(cmd)->root_lock)
1033
1034 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
1035 {
1036         int r;
1037         bool clean;
1038         __le64 null_mapping = pack_value(0, 0);
1039
1040         WRITE_LOCK(cmd);
1041         __dm_bless_for_disk(&null_mapping);
1042
1043         if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) {
1044                 r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean);
1045                 if (r) {
1046                         __dm_unbless_for_disk(&null_mapping);
1047                         goto out;
1048                 }
1049
1050                 if (!clean) {
1051                         DMERR("unable to shrink cache due to dirty blocks");
1052                         r = -EINVAL;
1053                         __dm_unbless_for_disk(&null_mapping);
1054                         goto out;
1055                 }
1056         }
1057
1058         r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
1059                             from_cblock(new_cache_size),
1060                             &null_mapping, &cmd->root);
1061         if (r)
1062                 goto out;
1063
1064         if (separate_dirty_bits(cmd)) {
1065                 r = dm_bitset_resize(&cmd->dirty_info, cmd->dirty_root,
1066                                      from_cblock(cmd->cache_blocks), from_cblock(new_cache_size),
1067                                      false, &cmd->dirty_root);
1068                 if (r)
1069                         goto out;
1070         }
1071
1072         cmd->cache_blocks = new_cache_size;
1073         cmd->changed = true;
1074
1075 out:
1076         WRITE_UNLOCK(cmd);
1077
1078         return r;
1079 }
1080
1081 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
1082                                    sector_t discard_block_size,
1083                                    dm_dblock_t new_nr_entries)
1084 {
1085         int r;
1086
1087         WRITE_LOCK(cmd);
1088         r = dm_bitset_resize(&cmd->discard_info,
1089                              cmd->discard_root,
1090                              from_dblock(cmd->discard_nr_blocks),
1091                              from_dblock(new_nr_entries),
1092                              false, &cmd->discard_root);
1093         if (!r) {
1094                 cmd->discard_block_size = discard_block_size;
1095                 cmd->discard_nr_blocks = new_nr_entries;
1096         }
1097
1098         cmd->changed = true;
1099         WRITE_UNLOCK(cmd);
1100
1101         return r;
1102 }
1103
1104 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
1105 {
1106         return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
1107                                  from_dblock(b), &cmd->discard_root);
1108 }
1109
1110 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
1111 {
1112         return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
1113                                    from_dblock(b), &cmd->discard_root);
1114 }
1115
1116 static int __discard(struct dm_cache_metadata *cmd,
1117                      dm_dblock_t dblock, bool discard)
1118 {
1119         int r;
1120
1121         r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
1122         if (r)
1123                 return r;
1124
1125         cmd->changed = true;
1126         return 0;
1127 }
1128
1129 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
1130                          dm_dblock_t dblock, bool discard)
1131 {
1132         int r;
1133
1134         WRITE_LOCK(cmd);
1135         r = __discard(cmd, dblock, discard);
1136         WRITE_UNLOCK(cmd);
1137
1138         return r;
1139 }
1140
1141 static int __load_discards(struct dm_cache_metadata *cmd,
1142                            load_discard_fn fn, void *context)
1143 {
1144         int r = 0;
1145         uint32_t b;
1146         struct dm_bitset_cursor c;
1147
1148         if (from_dblock(cmd->discard_nr_blocks) == 0)
1149                 /* nothing to do */
1150                 return 0;
1151
1152         if (cmd->clean_when_opened) {
1153                 r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root, &cmd->discard_root);
1154                 if (r)
1155                         return r;
1156
1157                 r = dm_bitset_cursor_begin(&cmd->discard_info, cmd->discard_root,
1158                                            from_dblock(cmd->discard_nr_blocks), &c);
1159                 if (r)
1160                         return r;
1161
1162                 for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1163                         r = fn(context, cmd->discard_block_size, to_dblock(b),
1164                                dm_bitset_cursor_get_value(&c));
1165                         if (r)
1166                                 break;
1167                 }
1168
1169                 dm_bitset_cursor_end(&c);
1170
1171         } else {
1172                 for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1173                         r = fn(context, cmd->discard_block_size, to_dblock(b), false);
1174                         if (r)
1175                                 return r;
1176                 }
1177         }
1178
1179         return r;
1180 }
1181
1182 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
1183                            load_discard_fn fn, void *context)
1184 {
1185         int r;
1186
1187         READ_LOCK(cmd);
1188         r = __load_discards(cmd, fn, context);
1189         READ_UNLOCK(cmd);
1190
1191         return r;
1192 }
1193
1194 int dm_cache_size(struct dm_cache_metadata *cmd, dm_cblock_t *result)
1195 {
1196         READ_LOCK(cmd);
1197         *result = cmd->cache_blocks;
1198         READ_UNLOCK(cmd);
1199
1200         return 0;
1201 }
1202
1203 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1204 {
1205         int r;
1206         __le64 value = pack_value(0, 0);
1207
1208         __dm_bless_for_disk(&value);
1209         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1210                                &value, &cmd->root);
1211         if (r)
1212                 return r;
1213
1214         cmd->changed = true;
1215         return 0;
1216 }
1217
1218 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1219 {
1220         int r;
1221
1222         WRITE_LOCK(cmd);
1223         r = __remove(cmd, cblock);
1224         WRITE_UNLOCK(cmd);
1225
1226         return r;
1227 }
1228
1229 static int __insert(struct dm_cache_metadata *cmd,
1230                     dm_cblock_t cblock, dm_oblock_t oblock)
1231 {
1232         int r;
1233         __le64 value = pack_value(oblock, M_VALID);
1234         __dm_bless_for_disk(&value);
1235
1236         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1237                                &value, &cmd->root);
1238         if (r)
1239                 return r;
1240
1241         cmd->changed = true;
1242         return 0;
1243 }
1244
1245 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
1246                             dm_cblock_t cblock, dm_oblock_t oblock)
1247 {
1248         int r;
1249
1250         WRITE_LOCK(cmd);
1251         r = __insert(cmd, cblock, oblock);
1252         WRITE_UNLOCK(cmd);
1253
1254         return r;
1255 }
1256
1257 struct thunk {
1258         load_mapping_fn fn;
1259         void *context;
1260
1261         struct dm_cache_metadata *cmd;
1262         bool respect_dirty_flags;
1263         bool hints_valid;
1264 };
1265
1266 static bool policy_unchanged(struct dm_cache_metadata *cmd,
1267                              struct dm_cache_policy *policy)
1268 {
1269         const char *policy_name = dm_cache_policy_get_name(policy);
1270         const unsigned *policy_version = dm_cache_policy_get_version(policy);
1271         size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
1272
1273         /*
1274          * Ensure policy names match.
1275          */
1276         if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
1277                 return false;
1278
1279         /*
1280          * Ensure policy major versions match.
1281          */
1282         if (cmd->policy_version[0] != policy_version[0])
1283                 return false;
1284
1285         /*
1286          * Ensure policy hint sizes match.
1287          */
1288         if (cmd->policy_hint_size != policy_hint_size)
1289                 return false;
1290
1291         return true;
1292 }
1293
1294 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1295 {
1296         return cmd->hint_root && cmd->policy_hint_size;
1297 }
1298
1299 static bool hints_array_available(struct dm_cache_metadata *cmd,
1300                                   struct dm_cache_policy *policy)
1301 {
1302         return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1303                 hints_array_initialized(cmd);
1304 }
1305
1306 static int __load_mapping_v1(struct dm_cache_metadata *cmd,
1307                              uint64_t cb, bool hints_valid,
1308                              struct dm_array_cursor *mapping_cursor,
1309                              struct dm_array_cursor *hint_cursor,
1310                              load_mapping_fn fn, void *context)
1311 {
1312         int r = 0;
1313
1314         __le64 mapping;
1315         __le32 hint = 0;
1316
1317         __le64 *mapping_value_le;
1318         __le32 *hint_value_le;
1319
1320         dm_oblock_t oblock;
1321         unsigned flags;
1322
1323         dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1324         memcpy(&mapping, mapping_value_le, sizeof(mapping));
1325         unpack_value(mapping, &oblock, &flags);
1326
1327         if (flags & M_VALID) {
1328                 if (hints_valid) {
1329                         dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1330                         memcpy(&hint, hint_value_le, sizeof(hint));
1331                 }
1332
1333                 r = fn(context, oblock, to_cblock(cb), flags & M_DIRTY,
1334                        le32_to_cpu(hint), hints_valid);
1335                 if (r) {
1336                         DMERR("policy couldn't load cache block %llu",
1337                               (unsigned long long) from_cblock(to_cblock(cb)));
1338                 }
1339         }
1340
1341         return r;
1342 }
1343
1344 static int __load_mapping_v2(struct dm_cache_metadata *cmd,
1345                              uint64_t cb, bool hints_valid,
1346                              struct dm_array_cursor *mapping_cursor,
1347                              struct dm_array_cursor *hint_cursor,
1348                              struct dm_bitset_cursor *dirty_cursor,
1349                              load_mapping_fn fn, void *context)
1350 {
1351         int r = 0;
1352
1353         __le64 mapping;
1354         __le32 hint = 0;
1355
1356         __le64 *mapping_value_le;
1357         __le32 *hint_value_le;
1358
1359         dm_oblock_t oblock;
1360         unsigned flags;
1361         bool dirty;
1362
1363         dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1364         memcpy(&mapping, mapping_value_le, sizeof(mapping));
1365         unpack_value(mapping, &oblock, &flags);
1366
1367         if (flags & M_VALID) {
1368                 if (hints_valid) {
1369                         dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1370                         memcpy(&hint, hint_value_le, sizeof(hint));
1371                 }
1372
1373                 dirty = dm_bitset_cursor_get_value(dirty_cursor);
1374                 r = fn(context, oblock, to_cblock(cb), dirty,
1375                        le32_to_cpu(hint), hints_valid);
1376                 if (r) {
1377                         DMERR("policy couldn't load cache block %llu",
1378                               (unsigned long long) from_cblock(to_cblock(cb)));
1379                 }
1380         }
1381
1382         return r;
1383 }
1384
1385 static int __load_mappings(struct dm_cache_metadata *cmd,
1386                            struct dm_cache_policy *policy,
1387                            load_mapping_fn fn, void *context)
1388 {
1389         int r;
1390         uint64_t cb;
1391
1392         bool hints_valid = hints_array_available(cmd, policy);
1393
1394         if (from_cblock(cmd->cache_blocks) == 0)
1395                 /* Nothing to do */
1396                 return 0;
1397
1398         r = dm_array_cursor_begin(&cmd->info, cmd->root, &cmd->mapping_cursor);
1399         if (r)
1400                 return r;
1401
1402         if (hints_valid) {
1403                 r = dm_array_cursor_begin(&cmd->hint_info, cmd->hint_root, &cmd->hint_cursor);
1404                 if (r) {
1405                         dm_array_cursor_end(&cmd->mapping_cursor);
1406                         return r;
1407                 }
1408         }
1409
1410         if (separate_dirty_bits(cmd)) {
1411                 r = dm_bitset_cursor_begin(&cmd->dirty_info, cmd->dirty_root,
1412                                            from_cblock(cmd->cache_blocks),
1413                                            &cmd->dirty_cursor);
1414                 if (r) {
1415                         dm_array_cursor_end(&cmd->hint_cursor);
1416                         dm_array_cursor_end(&cmd->mapping_cursor);
1417                         return r;
1418                 }
1419         }
1420
1421         for (cb = 0; ; cb++) {
1422                 if (separate_dirty_bits(cmd))
1423                         r = __load_mapping_v2(cmd, cb, hints_valid,
1424                                               &cmd->mapping_cursor,
1425                                               &cmd->hint_cursor,
1426                                               &cmd->dirty_cursor,
1427                                               fn, context);
1428                 else
1429                         r = __load_mapping_v1(cmd, cb, hints_valid,
1430                                               &cmd->mapping_cursor, &cmd->hint_cursor,
1431                                               fn, context);
1432                 if (r)
1433                         goto out;
1434
1435                 /*
1436                  * We need to break out before we move the cursors.
1437                  */
1438                 if (cb >= (from_cblock(cmd->cache_blocks) - 1))
1439                         break;
1440
1441                 r = dm_array_cursor_next(&cmd->mapping_cursor);
1442                 if (r) {
1443                         DMERR("dm_array_cursor_next for mapping failed");
1444                         goto out;
1445                 }
1446
1447                 if (hints_valid) {
1448                         r = dm_array_cursor_next(&cmd->hint_cursor);
1449                         if (r) {
1450                                 DMERR("dm_array_cursor_next for hint failed");
1451                                 goto out;
1452                         }
1453                 }
1454
1455                 if (separate_dirty_bits(cmd)) {
1456                         r = dm_bitset_cursor_next(&cmd->dirty_cursor);
1457                         if (r) {
1458                                 DMERR("dm_bitset_cursor_next for dirty failed");
1459                                 goto out;
1460                         }
1461                 }
1462         }
1463 out:
1464         dm_array_cursor_end(&cmd->mapping_cursor);
1465         if (hints_valid)
1466                 dm_array_cursor_end(&cmd->hint_cursor);
1467
1468         if (separate_dirty_bits(cmd))
1469                 dm_bitset_cursor_end(&cmd->dirty_cursor);
1470
1471         return r;
1472 }
1473
1474 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1475                            struct dm_cache_policy *policy,
1476                            load_mapping_fn fn, void *context)
1477 {
1478         int r;
1479
1480         READ_LOCK(cmd);
1481         r = __load_mappings(cmd, policy, fn, context);
1482         READ_UNLOCK(cmd);
1483
1484         return r;
1485 }
1486
1487 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
1488 {
1489         int r = 0;
1490         __le64 value;
1491         dm_oblock_t oblock;
1492         unsigned flags;
1493
1494         memcpy(&value, leaf, sizeof(value));
1495         unpack_value(value, &oblock, &flags);
1496
1497         return r;
1498 }
1499
1500 static int __dump_mappings(struct dm_cache_metadata *cmd)
1501 {
1502         return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1503 }
1504
1505 void dm_cache_dump(struct dm_cache_metadata *cmd)
1506 {
1507         READ_LOCK_VOID(cmd);
1508         __dump_mappings(cmd);
1509         READ_UNLOCK(cmd);
1510 }
1511
1512 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1513 {
1514         int r;
1515
1516         READ_LOCK(cmd);
1517         r = cmd->changed;
1518         READ_UNLOCK(cmd);
1519
1520         return r;
1521 }
1522
1523 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1524 {
1525         int r;
1526         unsigned flags;
1527         dm_oblock_t oblock;
1528         __le64 value;
1529
1530         r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1531         if (r)
1532                 return r;
1533
1534         unpack_value(value, &oblock, &flags);
1535
1536         if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1537                 /* nothing to be done */
1538                 return 0;
1539
1540         value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1541         __dm_bless_for_disk(&value);
1542
1543         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1544                                &value, &cmd->root);
1545         if (r)
1546                 return r;
1547
1548         cmd->changed = true;
1549         return 0;
1550
1551 }
1552
1553 static int __set_dirty_bits_v1(struct dm_cache_metadata *cmd, unsigned nr_bits, unsigned long *bits)
1554 {
1555         int r;
1556         unsigned i;
1557         for (i = 0; i < nr_bits; i++) {
1558                 r = __dirty(cmd, to_cblock(i), test_bit(i, bits));
1559                 if (r)
1560                         return r;
1561         }
1562
1563         return 0;
1564 }
1565
1566 static int is_dirty_callback(uint32_t index, bool *value, void *context)
1567 {
1568         unsigned long *bits = context;
1569         *value = test_bit(index, bits);
1570         return 0;
1571 }
1572
1573 static int __set_dirty_bits_v2(struct dm_cache_metadata *cmd, unsigned nr_bits, unsigned long *bits)
1574 {
1575         int r = 0;
1576
1577         /* nr_bits is really just a sanity check */
1578         if (nr_bits != from_cblock(cmd->cache_blocks)) {
1579                 DMERR("dirty bitset is wrong size");
1580                 return -EINVAL;
1581         }
1582
1583         r = dm_bitset_del(&cmd->dirty_info, cmd->dirty_root);
1584         if (r)
1585                 return r;
1586
1587         cmd->changed = true;
1588         return dm_bitset_new(&cmd->dirty_info, &cmd->dirty_root, nr_bits, is_dirty_callback, bits);
1589 }
1590
1591 int dm_cache_set_dirty_bits(struct dm_cache_metadata *cmd,
1592                             unsigned nr_bits,
1593                             unsigned long *bits)
1594 {
1595         int r;
1596
1597         WRITE_LOCK(cmd);
1598         if (separate_dirty_bits(cmd))
1599                 r = __set_dirty_bits_v2(cmd, nr_bits, bits);
1600         else
1601                 r = __set_dirty_bits_v1(cmd, nr_bits, bits);
1602         WRITE_UNLOCK(cmd);
1603
1604         return r;
1605 }
1606
1607 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1608                                  struct dm_cache_statistics *stats)
1609 {
1610         READ_LOCK_VOID(cmd);
1611         *stats = cmd->stats;
1612         READ_UNLOCK(cmd);
1613 }
1614
1615 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1616                                  struct dm_cache_statistics *stats)
1617 {
1618         WRITE_LOCK_VOID(cmd);
1619         cmd->stats = *stats;
1620         WRITE_UNLOCK(cmd);
1621 }
1622
1623 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1624 {
1625         int r;
1626         flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1627                                  clear_clean_shutdown);
1628
1629         WRITE_LOCK(cmd);
1630         r = __commit_transaction(cmd, mutator);
1631         if (r)
1632                 goto out;
1633
1634         r = __begin_transaction(cmd);
1635
1636 out:
1637         WRITE_UNLOCK(cmd);
1638         return r;
1639 }
1640
1641 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1642                                            dm_block_t *result)
1643 {
1644         int r = -EINVAL;
1645
1646         READ_LOCK(cmd);
1647         r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1648         READ_UNLOCK(cmd);
1649
1650         return r;
1651 }
1652
1653 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1654                                    dm_block_t *result)
1655 {
1656         int r = -EINVAL;
1657
1658         READ_LOCK(cmd);
1659         r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1660         READ_UNLOCK(cmd);
1661
1662         return r;
1663 }
1664
1665 /*----------------------------------------------------------------*/
1666
1667 static int get_hint(uint32_t index, void *value_le, void *context)
1668 {
1669         uint32_t value;
1670         struct dm_cache_policy *policy = context;
1671
1672         value = policy_get_hint(policy, to_cblock(index));
1673         *((__le32 *) value_le) = cpu_to_le32(value);
1674
1675         return 0;
1676 }
1677
1678 /*
1679  * It's quicker to always delete the hint array, and recreate with
1680  * dm_array_new().
1681  */
1682 static int write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1683 {
1684         int r;
1685         size_t hint_size;
1686         const char *policy_name = dm_cache_policy_get_name(policy);
1687         const unsigned *policy_version = dm_cache_policy_get_version(policy);
1688
1689         if (!policy_name[0] ||
1690             (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1691                 return -EINVAL;
1692
1693         strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1694         memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1695
1696         hint_size = dm_cache_policy_get_hint_size(policy);
1697         if (!hint_size)
1698                 return 0; /* short-circuit hints initialization */
1699         cmd->policy_hint_size = hint_size;
1700
1701         if (cmd->hint_root) {
1702                 r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1703                 if (r)
1704                         return r;
1705         }
1706
1707         return dm_array_new(&cmd->hint_info, &cmd->hint_root,
1708                             from_cblock(cmd->cache_blocks),
1709                             get_hint, policy);
1710 }
1711
1712 int dm_cache_write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1713 {
1714         int r;
1715
1716         WRITE_LOCK(cmd);
1717         r = write_hints(cmd, policy);
1718         WRITE_UNLOCK(cmd);
1719
1720         return r;
1721 }
1722
1723 int dm_cache_metadata_all_clean(struct dm_cache_metadata *cmd, bool *result)
1724 {
1725         int r;
1726
1727         READ_LOCK(cmd);
1728         r = blocks_are_unmapped_or_clean(cmd, 0, cmd->cache_blocks, result);
1729         READ_UNLOCK(cmd);
1730
1731         return r;
1732 }
1733
1734 void dm_cache_metadata_set_read_only(struct dm_cache_metadata *cmd)
1735 {
1736         WRITE_LOCK_VOID(cmd);
1737         dm_bm_set_read_only(cmd->bm);
1738         WRITE_UNLOCK(cmd);
1739 }
1740
1741 void dm_cache_metadata_set_read_write(struct dm_cache_metadata *cmd)
1742 {
1743         WRITE_LOCK_VOID(cmd);
1744         dm_bm_set_read_write(cmd->bm);
1745         WRITE_UNLOCK(cmd);
1746 }
1747
1748 int dm_cache_metadata_set_needs_check(struct dm_cache_metadata *cmd)
1749 {
1750         int r;
1751         struct dm_block *sblock;
1752         struct cache_disk_superblock *disk_super;
1753
1754         WRITE_LOCK(cmd);
1755         set_bit(NEEDS_CHECK, &cmd->flags);
1756
1757         r = superblock_lock(cmd, &sblock);
1758         if (r) {
1759                 DMERR("couldn't read superblock");
1760                 goto out;
1761         }
1762
1763         disk_super = dm_block_data(sblock);
1764         disk_super->flags = cpu_to_le32(cmd->flags);
1765
1766         dm_bm_unlock(sblock);
1767
1768 out:
1769         WRITE_UNLOCK(cmd);
1770         return r;
1771 }
1772
1773 int dm_cache_metadata_needs_check(struct dm_cache_metadata *cmd, bool *result)
1774 {
1775         READ_LOCK(cmd);
1776         *result = !!test_bit(NEEDS_CHECK, &cmd->flags);
1777         READ_UNLOCK(cmd);
1778
1779         return 0;
1780 }
1781
1782 int dm_cache_metadata_abort(struct dm_cache_metadata *cmd)
1783 {
1784         int r;
1785
1786         WRITE_LOCK(cmd);
1787         __destroy_persistent_data_objects(cmd);
1788         r = __create_persistent_data_objects(cmd, false);
1789         if (r)
1790                 cmd->fail_io = true;
1791         WRITE_UNLOCK(cmd);
1792
1793         return r;
1794 }