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 #define CACHE_VERSION 1
  24 #define CACHE_METADATA_CACHE_SIZE 64
  25
  26 /*
  27  *  3 for btree insert +
  28  *  2 for btree lookup used within space map
  29  */
  30 #define CACHE_MAX_CONCURRENT_LOCKS 5
  31 #define SPACE_MAP_ROOT_SIZE 128
  32
  33 enum superblock_flag_bits {
  34         /* for spotting crashes that would invalidate the dirty bitset */
  35         CLEAN_SHUTDOWN,
  36 };
  37
  38 /*
  39  * Each mapping from cache block -> origin block carries a set of flags.
  40  */
  41 enum mapping_bits {
  42         /*
  43          * A valid mapping.  Because we're using an array we clear this
  44          * flag for an non existant mapping.
  45          */
  46         M_VALID = 1,
  47
  48         /*
  49          * The data on the cache is different from that on the origin.
  50          */
  51         M_DIRTY = 2
  52 };
  53
  54 struct cache_disk_superblock {
  55         __le32 csum;
  56         __le32 flags;
  57         __le64 blocknr;
  58
  59         __u8 uuid[16];
  60         __le64 magic;
  61         __le32 version;
  62
  63         __u8 policy_name[CACHE_POLICY_NAME_SIZE];
  64         __le32 policy_hint_size;
  65
  66         __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
  67         __le64 mapping_root;
  68         __le64 hint_root;
  69
  70         __le64 discard_root;
  71         __le64 discard_block_size;
  72         __le64 discard_nr_blocks;
  73
  74         __le32 data_block_size;
  75         __le32 metadata_block_size;
  76         __le32 cache_blocks;
  77
  78         __le32 compat_flags;
  79         __le32 compat_ro_flags;
  80         __le32 incompat_flags;
  81
  82         __le32 read_hits;
  83         __le32 read_misses;
  84         __le32 write_hits;
  85         __le32 write_misses;
  86
  87         __le32 policy_version[CACHE_POLICY_VERSION_SIZE];
  88 } __packed;
  89
  90 struct dm_cache_metadata {
  91         struct block_device *bdev;
  92         struct dm_block_manager *bm;
  93         struct dm_space_map *metadata_sm;
  94         struct dm_transaction_manager *tm;
  95
  96         struct dm_array_info info;
  97         struct dm_array_info hint_info;
  98         struct dm_disk_bitset discard_info;
  99
 100         struct rw_semaphore root_lock;
 101         dm_block_t root;
 102         dm_block_t hint_root;
 103         dm_block_t discard_root;
 104
 105         sector_t discard_block_size;
 106         dm_dblock_t discard_nr_blocks;
 107
 108         sector_t data_block_size;
 109         dm_cblock_t cache_blocks;
 110         bool changed:1;
 111         bool clean_when_opened:1;
 112
 113         char policy_name[CACHE_POLICY_NAME_SIZE];
 114         unsigned policy_version[CACHE_POLICY_VERSION_SIZE];
 115         size_t policy_hint_size;
 116         struct dm_cache_statistics stats;
 117
 118         /*
 119          * Reading the space map root can fail, so we read it into this
 120          * buffer before the superblock is locked and updated.
 121          */
 122         __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
 123 };
 124
 125 /*-------------------------------------------------------------------
 126  * superblock validator
 127  *-----------------------------------------------------------------*/
 128
 129 #define SUPERBLOCK_CSUM_XOR 9031977
 130
 131 static void sb_prepare_for_write(struct dm_block_validator *v,
 132                                  struct dm_block *b,
 133                                  size_t sb_block_size)
 134 {
 135         struct cache_disk_superblock *disk_super = dm_block_data(b);
 136
 137         disk_super->blocknr = cpu_to_le64(dm_block_location(b));
 138         disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
 139                                                       sb_block_size - sizeof(__le32),
 140                                                       SUPERBLOCK_CSUM_XOR));
 141 }
 142
 143 static int sb_check(struct dm_block_validator *v,
 144                     struct dm_block *b,
 145                     size_t sb_block_size)
 146 {
 147         struct cache_disk_superblock *disk_super = dm_block_data(b);
 148         __le32 csum_le;
 149
 150         if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
 151                 DMERR("sb_check failed: blocknr %llu: wanted %llu",
 152                       le64_to_cpu(disk_super->blocknr),
 153                       (unsigned long long)dm_block_location(b));
 154                 return -ENOTBLK;
 155         }
 156
 157         if (le64_to_cpu(disk_super->magic) != CACHE_SUPERBLOCK_MAGIC) {
 158                 DMERR("sb_check failed: magic %llu: wanted %llu",
 159                       le64_to_cpu(disk_super->magic),
 160                       (unsigned long long)CACHE_SUPERBLOCK_MAGIC);
 161                 return -EILSEQ;
 162         }
 163
 164         csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
 165                                              sb_block_size - sizeof(__le32),
 166                                              SUPERBLOCK_CSUM_XOR));
 167         if (csum_le != disk_super->csum) {
 168                 DMERR("sb_check failed: csum %u: wanted %u",
 169                       le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
 170                 return -EILSEQ;
 171         }
 172
 173         return 0;
 174 }
 175
 176 static struct dm_block_validator sb_validator = {
 177         .name = "superblock",
 178         .prepare_for_write = sb_prepare_for_write,
 179         .check = sb_check
 180 };
 181
 182 /*----------------------------------------------------------------*/
 183
 184 static int superblock_read_lock(struct dm_cache_metadata *cmd,
 185                                 struct dm_block **sblock)
 186 {
 187         return dm_bm_read_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 188                                &sb_validator, sblock);
 189 }
 190
 191 static int superblock_lock_zero(struct dm_cache_metadata *cmd,
 192                                 struct dm_block **sblock)
 193 {
 194         return dm_bm_write_lock_zero(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 195                                      &sb_validator, sblock);
 196 }
 197
 198 static int superblock_lock(struct dm_cache_metadata *cmd,
 199                            struct dm_block **sblock)
 200 {
 201         return dm_bm_write_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 202                                 &sb_validator, sblock);
 203 }
 204
 205 /*----------------------------------------------------------------*/
 206
 207 static int __superblock_all_zeroes(struct dm_block_manager *bm, int *result)
 208 {
 209         int r;
 210         unsigned i;
 211         struct dm_block *b;
 212         __le64 *data_le, zero = cpu_to_le64(0);
 213         unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
 214
 215         /*
 216          * We can't use a validator here - it may be all zeroes.
 217          */
 218         r = dm_bm_read_lock(bm, CACHE_SUPERBLOCK_LOCATION, NULL, &b);
 219         if (r)
 220                 return r;
 221
 222         data_le = dm_block_data(b);
 223         *result = 1;
 224         for (i = 0; i < sb_block_size; i++) {
 225                 if (data_le[i] != zero) {
 226                         *result = 0;
 227                         break;
 228                 }
 229         }
 230
 231         return dm_bm_unlock(b);
 232 }
 233
 234 static void __setup_mapping_info(struct dm_cache_metadata *cmd)
 235 {
 236         struct dm_btree_value_type vt;
 237
 238         vt.context = NULL;
 239         vt.size = sizeof(__le64);
 240         vt.inc = NULL;
 241         vt.dec = NULL;
 242         vt.equal = NULL;
 243         dm_array_info_init(&cmd->info, cmd->tm, &vt);
 244
 245         if (cmd->policy_hint_size) {
 246                 vt.size = sizeof(__le32);
 247                 dm_array_info_init(&cmd->hint_info, cmd->tm, &vt);
 248         }
 249 }
 250
 251 static int __save_sm_root(struct dm_cache_metadata *cmd)
 252 {
 253         int r;
 254         size_t metadata_len;
 255
 256         r = dm_sm_root_size(cmd->metadata_sm, &metadata_len);
 257         if (r < 0)
 258                 return r;
 259
 260         return dm_sm_copy_root(cmd->metadata_sm, &cmd->metadata_space_map_root,
 261                                metadata_len);
 262 }
 263
 264 static void __copy_sm_root(struct dm_cache_metadata *cmd,
 265                            struct cache_disk_superblock *disk_super)
 266 {
 267         memcpy(&disk_super->metadata_space_map_root,
 268                &cmd->metadata_space_map_root,
 269                sizeof(cmd->metadata_space_map_root));
 270 }
 271
 272 static int __write_initial_superblock(struct dm_cache_metadata *cmd)
 273 {
 274         int r;
 275         struct dm_block *sblock;
 276         struct cache_disk_superblock *disk_super;
 277         sector_t bdev_size = i_size_read(cmd->bdev->bd_inode) >> SECTOR_SHIFT;
 278
 279         /* FIXME: see if we can lose the max sectors limit */
 280         if (bdev_size > DM_CACHE_METADATA_MAX_SECTORS)
 281                 bdev_size = DM_CACHE_METADATA_MAX_SECTORS;
 282
 283         r = dm_tm_pre_commit(cmd->tm);
 284         if (r < 0)
 285                 return r;
 286
 287         /*
 288          * dm_sm_copy_root() can fail.  So we need to do it before we start
 289          * updating the superblock.
 290          */
 291         r = __save_sm_root(cmd);
 292         if (r)
 293                 return r;
 294
 295         r = superblock_lock_zero(cmd, &sblock);
 296         if (r)
 297                 return r;
 298
 299         disk_super = dm_block_data(sblock);
 300         disk_super->flags = 0;
 301         memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
 302         disk_super->magic = cpu_to_le64(CACHE_SUPERBLOCK_MAGIC);
 303         disk_super->version = cpu_to_le32(CACHE_VERSION);
 304         memset(disk_super->policy_name, 0, sizeof(disk_super->policy_name));
 305         memset(disk_super->policy_version, 0, sizeof(disk_super->policy_version));
 306         disk_super->policy_hint_size = 0;
 307
 308         __copy_sm_root(cmd, disk_super);
 309
 310         disk_super->mapping_root = cpu_to_le64(cmd->root);
 311         disk_super->hint_root = cpu_to_le64(cmd->hint_root);
 312         disk_super->discard_root = cpu_to_le64(cmd->discard_root);
 313         disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
 314         disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
 315         disk_super->metadata_block_size = cpu_to_le32(DM_CACHE_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
 316         disk_super->data_block_size = cpu_to_le32(cmd->data_block_size);
 317         disk_super->cache_blocks = cpu_to_le32(0);
 318
 319         disk_super->read_hits = cpu_to_le32(0);
 320         disk_super->read_misses = cpu_to_le32(0);
 321         disk_super->write_hits = cpu_to_le32(0);
 322         disk_super->write_misses = cpu_to_le32(0);
 323
 324         return dm_tm_commit(cmd->tm, sblock);
 325 }
 326
 327 static int __format_metadata(struct dm_cache_metadata *cmd)
 328 {
 329         int r;
 330
 331         r = dm_tm_create_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 332                                  &cmd->tm, &cmd->metadata_sm);
 333         if (r < 0) {
 334                 DMERR("tm_create_with_sm failed");
 335                 return r;
 336         }
 337
 338         __setup_mapping_info(cmd);
 339
 340         r = dm_array_empty(&cmd->info, &cmd->root);
 341         if (r < 0)
 342                 goto bad;
 343
 344         dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
 345
 346         r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root);
 347         if (r < 0)
 348                 goto bad;
 349
 350         cmd->discard_block_size = 0;
 351         cmd->discard_nr_blocks = 0;
 352
 353         r = __write_initial_superblock(cmd);
 354         if (r)
 355                 goto bad;
 356
 357         cmd->clean_when_opened = true;
 358         return 0;
 359
 360 bad:
 361         dm_tm_destroy(cmd->tm);
 362         dm_sm_destroy(cmd->metadata_sm);
 363
 364         return r;
 365 }
 366
 367 static int __check_incompat_features(struct cache_disk_superblock *disk_super,
 368                                      struct dm_cache_metadata *cmd)
 369 {
 370         uint32_t features;
 371
 372         features = le32_to_cpu(disk_super->incompat_flags) & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
 373         if (features) {
 374                 DMERR("could not access metadata due to unsupported optional features (%lx).",
 375                       (unsigned long)features);
 376                 return -EINVAL;
 377         }
 378
 379         /*
 380          * Check for read-only metadata to skip the following RDWR checks.
 381          */
 382         if (get_disk_ro(cmd->bdev->bd_disk))
 383                 return 0;
 384
 385         features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP;
 386         if (features) {
 387                 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
 388                       (unsigned long)features);
 389                 return -EINVAL;
 390         }
 391
 392         return 0;
 393 }
 394
 395 static int __open_metadata(struct dm_cache_metadata *cmd)
 396 {
 397         int r;
 398         struct dm_block *sblock;
 399         struct cache_disk_superblock *disk_super;
 400         unsigned long sb_flags;
 401
 402         r = superblock_read_lock(cmd, &sblock);
 403         if (r < 0) {
 404                 DMERR("couldn't read lock superblock");
 405                 return r;
 406         }
 407
 408         disk_super = dm_block_data(sblock);
 409
 410         /* Verify the data block size hasn't changed */
 411         if (le32_to_cpu(disk_super->data_block_size) != cmd->data_block_size) {
 412                 DMERR("changing the data block size (from %u to %llu) is not supported",
 413                       le32_to_cpu(disk_super->data_block_size),
 414                       (unsigned long long)cmd->data_block_size);
 415                 r = -EINVAL;
 416                 goto bad;
 417         }
 418
 419         r = __check_incompat_features(disk_super, cmd);
 420         if (r < 0)
 421                 goto bad;
 422
 423         r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 424                                disk_super->metadata_space_map_root,
 425                                sizeof(disk_super->metadata_space_map_root),
 426                                &cmd->tm, &cmd->metadata_sm);
 427         if (r < 0) {
 428                 DMERR("tm_open_with_sm failed");
 429                 goto bad;
 430         }
 431
 432         __setup_mapping_info(cmd);
 433         dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
 434         sb_flags = le32_to_cpu(disk_super->flags);
 435         cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags);
 436         return dm_bm_unlock(sblock);
 437
 438 bad:
 439         dm_bm_unlock(sblock);
 440         return r;
 441 }
 442
 443 static int __open_or_format_metadata(struct dm_cache_metadata *cmd,
 444                                      bool format_device)
 445 {
 446         int r, unformatted;
 447
 448         r = __superblock_all_zeroes(cmd->bm, &unformatted);
 449         if (r)
 450                 return r;
 451
 452         if (unformatted)
 453                 return format_device ? __format_metadata(cmd) : -EPERM;
 454
 455         return __open_metadata(cmd);
 456 }
 457
 458 static int __create_persistent_data_objects(struct dm_cache_metadata *cmd,
 459                                             bool may_format_device)
 460 {
 461         int r;
 462         cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE,
 463                                           CACHE_METADATA_CACHE_SIZE,
 464                                           CACHE_MAX_CONCURRENT_LOCKS);
 465         if (IS_ERR(cmd->bm)) {
 466                 DMERR("could not create block manager");
 467                 return PTR_ERR(cmd->bm);
 468         }
 469
 470         r = __open_or_format_metadata(cmd, may_format_device);
 471         if (r)
 472                 dm_block_manager_destroy(cmd->bm);
 473
 474         return r;
 475 }
 476
 477 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd)
 478 {
 479         dm_sm_destroy(cmd->metadata_sm);
 480         dm_tm_destroy(cmd->tm);
 481         dm_block_manager_destroy(cmd->bm);
 482 }
 483
 484 typedef unsigned long (*flags_mutator)(unsigned long);
 485
 486 static void update_flags(struct cache_disk_superblock *disk_super,
 487                          flags_mutator mutator)
 488 {
 489         uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
 490         disk_super->flags = cpu_to_le32(sb_flags);
 491 }
 492
 493 static unsigned long set_clean_shutdown(unsigned long flags)
 494 {
 495         set_bit(CLEAN_SHUTDOWN, &flags);
 496         return flags;
 497 }
 498
 499 static unsigned long clear_clean_shutdown(unsigned long flags)
 500 {
 501         clear_bit(CLEAN_SHUTDOWN, &flags);
 502         return flags;
 503 }
 504
 505 static void read_superblock_fields(struct dm_cache_metadata *cmd,
 506                                    struct cache_disk_superblock *disk_super)
 507 {
 508         cmd->root = le64_to_cpu(disk_super->mapping_root);
 509         cmd->hint_root = le64_to_cpu(disk_super->hint_root);
 510         cmd->discard_root = le64_to_cpu(disk_super->discard_root);
 511         cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
 512         cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
 513         cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
 514         cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
 515         strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
 516         cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
 517         cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
 518         cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
 519         cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
 520
 521         cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
 522         cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
 523         cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
 524         cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
 525
 526         cmd->changed = false;
 527 }
 528
 529 /*
 530  * The mutator updates the superblock flags.
 531  */
 532 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
 533                                      flags_mutator mutator)
 534 {
 535         int r;
 536         struct cache_disk_superblock *disk_super;
 537         struct dm_block *sblock;
 538
 539         r = superblock_lock(cmd, &sblock);
 540         if (r)
 541                 return r;
 542
 543         disk_super = dm_block_data(sblock);
 544         update_flags(disk_super, mutator);
 545         read_superblock_fields(cmd, disk_super);
 546         dm_bm_unlock(sblock);
 547
 548         return dm_bm_flush(cmd->bm);
 549 }
 550
 551 static int __begin_transaction(struct dm_cache_metadata *cmd)
 552 {
 553         int r;
 554         struct cache_disk_superblock *disk_super;
 555         struct dm_block *sblock;
 556
 557         /*
 558          * We re-read the superblock every time.  Shouldn't need to do this
 559          * really.
 560          */
 561         r = superblock_read_lock(cmd, &sblock);
 562         if (r)
 563                 return r;
 564
 565         disk_super = dm_block_data(sblock);
 566         read_superblock_fields(cmd, disk_super);
 567         dm_bm_unlock(sblock);
 568
 569         return 0;
 570 }
 571
 572 static int __commit_transaction(struct dm_cache_metadata *cmd,
 573                                 flags_mutator mutator)
 574 {
 575         int r;
 576         struct cache_disk_superblock *disk_super;
 577         struct dm_block *sblock;
 578
 579         /*
 580          * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
 581          */
 582         BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
 583
 584         r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
 585                             &cmd->discard_root);
 586         if (r)
 587                 return r;
 588
 589         r = dm_tm_pre_commit(cmd->tm);
 590         if (r < 0)
 591                 return r;
 592
 593         r = __save_sm_root(cmd);
 594         if (r)
 595                 return r;
 596
 597         r = superblock_lock(cmd, &sblock);
 598         if (r)
 599                 return r;
 600
 601         disk_super = dm_block_data(sblock);
 602
 603         if (mutator)
 604                 update_flags(disk_super, mutator);
 605
 606         disk_super->mapping_root = cpu_to_le64(cmd->root);
 607         disk_super->hint_root = cpu_to_le64(cmd->hint_root);
 608         disk_super->discard_root = cpu_to_le64(cmd->discard_root);
 609         disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
 610         disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
 611         disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
 612         strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
 613         disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
 614         disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
 615         disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
 616
 617         disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
 618         disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
 619         disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
 620         disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
 621         __copy_sm_root(cmd, disk_super);
 622
 623         return dm_tm_commit(cmd->tm, sblock);
 624 }
 625
 626 /*----------------------------------------------------------------*/
 627
 628 /*
 629  * The mappings are held in a dm-array that has 64-bit values stored in
 630  * little-endian format.  The index is the cblock, the high 48bits of the
 631  * value are the oblock and the low 16 bit the flags.
 632  */
 633 #define FLAGS_MASK ((1 << 16) - 1)
 634
 635 static __le64 pack_value(dm_oblock_t block, unsigned flags)
 636 {
 637         uint64_t value = from_oblock(block);
 638         value <<= 16;
 639         value = value | (flags & FLAGS_MASK);
 640         return cpu_to_le64(value);
 641 }
 642
 643 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
 644 {
 645         uint64_t value = le64_to_cpu(value_le);
 646         uint64_t b = value >> 16;
 647         *block = to_oblock(b);
 648         *flags = value & FLAGS_MASK;
 649 }
 650
 651 /*----------------------------------------------------------------*/
 652
 653 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
 654                                                  sector_t data_block_size,
 655                                                  bool may_format_device,
 656                                                  size_t policy_hint_size)
 657 {
 658         int r;
 659         struct dm_cache_metadata *cmd;
 660
 661         cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
 662         if (!cmd) {
 663                 DMERR("could not allocate metadata struct");
 664                 return NULL;
 665         }
 666
 667         init_rwsem(&cmd->root_lock);
 668         cmd->bdev = bdev;
 669         cmd->data_block_size = data_block_size;
 670         cmd->cache_blocks = 0;
 671         cmd->policy_hint_size = policy_hint_size;
 672         cmd->changed = true;
 673
 674         r = __create_persistent_data_objects(cmd, may_format_device);
 675         if (r) {
 676                 kfree(cmd);
 677                 return ERR_PTR(r);
 678         }
 679
 680         r = __begin_transaction_flags(cmd, clear_clean_shutdown);
 681         if (r < 0) {
 682                 dm_cache_metadata_close(cmd);
 683                 return ERR_PTR(r);
 684         }
 685
 686         return cmd;
 687 }
 688
 689 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
 690 {
 691         __destroy_persistent_data_objects(cmd);
 692         kfree(cmd);
 693 }
 694
 695 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
 696 {
 697         int r;
 698         __le64 null_mapping = pack_value(0, 0);
 699
 700         down_write(&cmd->root_lock);
 701         __dm_bless_for_disk(&null_mapping);
 702         r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
 703                             from_cblock(new_cache_size),
 704                             &null_mapping, &cmd->root);
 705         if (!r)
 706                 cmd->cache_blocks = new_cache_size;
 707         cmd->changed = true;
 708         up_write(&cmd->root_lock);
 709
 710         return r;
 711 }
 712
 713 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
 714                                    sector_t discard_block_size,
 715                                    dm_dblock_t new_nr_entries)
 716 {
 717         int r;
 718
 719         down_write(&cmd->root_lock);
 720         r = dm_bitset_resize(&cmd->discard_info,
 721                              cmd->discard_root,
 722                              from_dblock(cmd->discard_nr_blocks),
 723                              from_dblock(new_nr_entries),
 724                              false, &cmd->discard_root);
 725         if (!r) {
 726                 cmd->discard_block_size = discard_block_size;
 727                 cmd->discard_nr_blocks = new_nr_entries;
 728         }
 729
 730         cmd->changed = true;
 731         up_write(&cmd->root_lock);
 732
 733         return r;
 734 }
 735
 736 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
 737 {
 738         return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
 739                                  from_dblock(b), &cmd->discard_root);
 740 }
 741
 742 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
 743 {
 744         return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
 745                                    from_dblock(b), &cmd->discard_root);
 746 }
 747
 748 static int __is_discarded(struct dm_cache_metadata *cmd, dm_dblock_t b,
 749                           bool *is_discarded)
 750 {
 751         return dm_bitset_test_bit(&cmd->discard_info, cmd->discard_root,
 752                                   from_dblock(b), &cmd->discard_root,
 753                                   is_discarded);
 754 }
 755
 756 static int __discard(struct dm_cache_metadata *cmd,
 757                      dm_dblock_t dblock, bool discard)
 758 {
 759         int r;
 760
 761         r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
 762         if (r)
 763                 return r;
 764
 765         cmd->changed = true;
 766         return 0;
 767 }
 768
 769 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
 770                          dm_dblock_t dblock, bool discard)
 771 {
 772         int r;
 773
 774         down_write(&cmd->root_lock);
 775         r = __discard(cmd, dblock, discard);
 776         up_write(&cmd->root_lock);
 777
 778         return r;
 779 }
 780
 781 static int __load_discards(struct dm_cache_metadata *cmd,
 782                            load_discard_fn fn, void *context)
 783 {
 784         int r = 0;
 785         dm_block_t b;
 786         bool discard;
 787
 788         for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
 789                 dm_dblock_t dblock = to_dblock(b);
 790
 791                 if (cmd->clean_when_opened) {
 792                         r = __is_discarded(cmd, dblock, &discard);
 793                         if (r)
 794                                 return r;
 795                 } else
 796                         discard = false;
 797
 798                 r = fn(context, cmd->discard_block_size, dblock, discard);
 799                 if (r)
 800                         break;
 801         }
 802
 803         return r;
 804 }
 805
 806 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
 807                            load_discard_fn fn, void *context)
 808 {
 809         int r;
 810
 811         down_read(&cmd->root_lock);
 812         r = __load_discards(cmd, fn, context);
 813         up_read(&cmd->root_lock);
 814
 815         return r;
 816 }
 817
 818 dm_cblock_t dm_cache_size(struct dm_cache_metadata *cmd)
 819 {
 820         dm_cblock_t r;
 821
 822         down_read(&cmd->root_lock);
 823         r = cmd->cache_blocks;
 824         up_read(&cmd->root_lock);
 825
 826         return r;
 827 }
 828
 829 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
 830 {
 831         int r;
 832         __le64 value = pack_value(0, 0);
 833
 834         __dm_bless_for_disk(&value);
 835         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
 836                                &value, &cmd->root);
 837         if (r)
 838                 return r;
 839
 840         cmd->changed = true;
 841         return 0;
 842 }
 843
 844 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
 845 {
 846         int r;
 847
 848         down_write(&cmd->root_lock);
 849         r = __remove(cmd, cblock);
 850         up_write(&cmd->root_lock);
 851
 852         return r;
 853 }
 854
 855 static int __insert(struct dm_cache_metadata *cmd,
 856                     dm_cblock_t cblock, dm_oblock_t oblock)
 857 {
 858         int r;
 859         __le64 value = pack_value(oblock, M_VALID);
 860         __dm_bless_for_disk(&value);
 861
 862         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
 863                                &value, &cmd->root);
 864         if (r)
 865                 return r;
 866
 867         cmd->changed = true;
 868         return 0;
 869 }
 870
 871 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
 872                             dm_cblock_t cblock, dm_oblock_t oblock)
 873 {
 874         int r;
 875
 876         down_write(&cmd->root_lock);
 877         r = __insert(cmd, cblock, oblock);
 878         up_write(&cmd->root_lock);
 879
 880         return r;
 881 }
 882
 883 struct thunk {
 884         load_mapping_fn fn;
 885         void *context;
 886
 887         struct dm_cache_metadata *cmd;
 888         bool respect_dirty_flags;
 889         bool hints_valid;
 890 };
 891
 892 static bool policy_unchanged(struct dm_cache_metadata *cmd,
 893                              struct dm_cache_policy *policy)
 894 {
 895         const char *policy_name = dm_cache_policy_get_name(policy);
 896         const unsigned *policy_version = dm_cache_policy_get_version(policy);
 897         size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
 898
 899         /*
 900          * Ensure policy names match.
 901          */
 902         if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
 903                 return false;
 904
 905         /*
 906          * Ensure policy major versions match.
 907          */
 908         if (cmd->policy_version[0] != policy_version[0])
 909                 return false;
 910
 911         /*
 912          * Ensure policy hint sizes match.
 913          */
 914         if (cmd->policy_hint_size != policy_hint_size)
 915                 return false;
 916
 917         return true;
 918 }
 919
 920 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
 921 {
 922         return cmd->hint_root && cmd->policy_hint_size;
 923 }
 924
 925 static bool hints_array_available(struct dm_cache_metadata *cmd,
 926                                   struct dm_cache_policy *policy)
 927 {
 928         return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
 929                 hints_array_initialized(cmd);
 930 }
 931
 932 static int __load_mapping(void *context, uint64_t cblock, void *leaf)
 933 {
 934         int r = 0;
 935         bool dirty;
 936         __le64 value;
 937         __le32 hint_value = 0;
 938         dm_oblock_t oblock;
 939         unsigned flags;
 940         struct thunk *thunk = context;
 941         struct dm_cache_metadata *cmd = thunk->cmd;
 942
 943         memcpy(&value, leaf, sizeof(value));
 944         unpack_value(value, &oblock, &flags);
 945
 946         if (flags & M_VALID) {
 947                 if (thunk->hints_valid) {
 948                         r = dm_array_get_value(&cmd->hint_info, cmd->hint_root,
 949                                                cblock, &hint_value);
 950                         if (r && r != -ENODATA)
 951                                 return r;
 952                 }
 953
 954                 dirty = thunk->respect_dirty_flags ? (flags & M_DIRTY) : true;
 955                 r = thunk->fn(thunk->context, oblock, to_cblock(cblock),
 956                               dirty, le32_to_cpu(hint_value), thunk->hints_valid);
 957         }
 958
 959         return r;
 960 }
 961
 962 static int __load_mappings(struct dm_cache_metadata *cmd,
 963                            struct dm_cache_policy *policy,
 964                            load_mapping_fn fn, void *context)
 965 {
 966         struct thunk thunk;
 967
 968         thunk.fn = fn;
 969         thunk.context = context;
 970
 971         thunk.cmd = cmd;
 972         thunk.respect_dirty_flags = cmd->clean_when_opened;
 973         thunk.hints_valid = hints_array_available(cmd, policy);
 974
 975         return dm_array_walk(&cmd->info, cmd->root, __load_mapping, &thunk);
 976 }
 977
 978 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
 979                            struct dm_cache_policy *policy,
 980                            load_mapping_fn fn, void *context)
 981 {
 982         int r;
 983
 984         down_read(&cmd->root_lock);
 985         r = __load_mappings(cmd, policy, fn, context);
 986         up_read(&cmd->root_lock);
 987
 988         return r;
 989 }
 990
 991 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
 992 {
 993         int r = 0;
 994         __le64 value;
 995         dm_oblock_t oblock;
 996         unsigned flags;
 997
 998         memcpy(&value, leaf, sizeof(value));
 999         unpack_value(value, &oblock, &flags);
1000
1001         return r;
1002 }
1003
1004 static int __dump_mappings(struct dm_cache_metadata *cmd)
1005 {
1006         return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1007 }
1008
1009 void dm_cache_dump(struct dm_cache_metadata *cmd)
1010 {
1011         down_read(&cmd->root_lock);
1012         __dump_mappings(cmd);
1013         up_read(&cmd->root_lock);
1014 }
1015
1016 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1017 {
1018         int r;
1019
1020         down_read(&cmd->root_lock);
1021         r = cmd->changed;
1022         up_read(&cmd->root_lock);
1023
1024         return r;
1025 }
1026
1027 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1028 {
1029         int r;
1030         unsigned flags;
1031         dm_oblock_t oblock;
1032         __le64 value;
1033
1034         r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1035         if (r)
1036                 return r;
1037
1038         unpack_value(value, &oblock, &flags);
1039
1040         if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1041                 /* nothing to be done */
1042                 return 0;
1043
1044         value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1045         __dm_bless_for_disk(&value);
1046
1047         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1048                                &value, &cmd->root);
1049         if (r)
1050                 return r;
1051
1052         cmd->changed = true;
1053         return 0;
1054
1055 }
1056
1057 int dm_cache_set_dirty(struct dm_cache_metadata *cmd,
1058                        dm_cblock_t cblock, bool dirty)
1059 {
1060         int r;
1061
1062         down_write(&cmd->root_lock);
1063         r = __dirty(cmd, cblock, dirty);
1064         up_write(&cmd->root_lock);
1065
1066         return r;
1067 }
1068
1069 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1070                                  struct dm_cache_statistics *stats)
1071 {
1072         down_read(&cmd->root_lock);
1073         *stats = cmd->stats;
1074         up_read(&cmd->root_lock);
1075 }
1076
1077 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1078                                  struct dm_cache_statistics *stats)
1079 {
1080         down_write(&cmd->root_lock);
1081         cmd->stats = *stats;
1082         up_write(&cmd->root_lock);
1083 }
1084
1085 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1086 {
1087         int r;
1088         flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1089                                  clear_clean_shutdown);
1090
1091         down_write(&cmd->root_lock);
1092         r = __commit_transaction(cmd, mutator);
1093         if (r)
1094                 goto out;
1095
1096         r = __begin_transaction(cmd);
1097
1098 out:
1099         up_write(&cmd->root_lock);
1100         return r;
1101 }
1102
1103 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1104                                            dm_block_t *result)
1105 {
1106         int r = -EINVAL;
1107
1108         down_read(&cmd->root_lock);
1109         r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1110         up_read(&cmd->root_lock);
1111
1112         return r;
1113 }
1114
1115 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1116                                    dm_block_t *result)
1117 {
1118         int r = -EINVAL;
1119
1120         down_read(&cmd->root_lock);
1121         r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1122         up_read(&cmd->root_lock);
1123
1124         return r;
1125 }
1126
1127 /*----------------------------------------------------------------*/
1128
1129 static int begin_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1130 {
1131         int r;
1132         __le32 value;
1133         size_t hint_size;
1134         const char *policy_name = dm_cache_policy_get_name(policy);
1135         const unsigned *policy_version = dm_cache_policy_get_version(policy);
1136
1137         if (!policy_name[0] ||
1138             (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1139                 return -EINVAL;
1140
1141         if (!policy_unchanged(cmd, policy)) {
1142                 strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1143                 memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1144
1145                 hint_size = dm_cache_policy_get_hint_size(policy);
1146                 if (!hint_size)
1147                         return 0; /* short-circuit hints initialization */
1148                 cmd->policy_hint_size = hint_size;
1149
1150                 if (cmd->hint_root) {
1151                         r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1152                         if (r)
1153                                 return r;
1154                 }
1155
1156                 r = dm_array_empty(&cmd->hint_info, &cmd->hint_root);
1157                 if (r)
1158                         return r;
1159
1160                 value = cpu_to_le32(0);
1161                 __dm_bless_for_disk(&value);
1162                 r = dm_array_resize(&cmd->hint_info, cmd->hint_root, 0,
1163                                     from_cblock(cmd->cache_blocks),
1164                                     &value, &cmd->hint_root);
1165                 if (r)
1166                         return r;
1167         }
1168
1169         return 0;
1170 }
1171
1172 int dm_cache_begin_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1173 {
1174         int r;
1175
1176         down_write(&cmd->root_lock);
1177         r = begin_hints(cmd, policy);
1178         up_write(&cmd->root_lock);
1179
1180         return r;
1181 }
1182
1183 static int save_hint(struct dm_cache_metadata *cmd, dm_cblock_t cblock,
1184                      uint32_t hint)
1185 {
1186         int r;
1187         __le32 value = cpu_to_le32(hint);
1188         __dm_bless_for_disk(&value);
1189
1190         r = dm_array_set_value(&cmd->hint_info, cmd->hint_root,
1191                                from_cblock(cblock), &value, &cmd->hint_root);
1192         cmd->changed = true;
1193
1194         return r;
1195 }
1196
1197 int dm_cache_save_hint(struct dm_cache_metadata *cmd, dm_cblock_t cblock,
1198                        uint32_t hint)
1199 {
1200         int r;
1201
1202         if (!hints_array_initialized(cmd))
1203                 return 0;
1204
1205         down_write(&cmd->root_lock);
1206         r = save_hint(cmd, cblock, hint);
1207         up_write(&cmd->root_lock);
1208
1209         return r;
1210 }