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         atomic_t ref_count;
  92         struct list_head list;
  93
  94         struct block_device *bdev;
  95         struct dm_block_manager *bm;
  96         struct dm_space_map *metadata_sm;
  97         struct dm_transaction_manager *tm;
  98
  99         struct dm_array_info info;
 100         struct dm_array_info hint_info;
 101         struct dm_disk_bitset discard_info;
 102
 103         struct rw_semaphore root_lock;
 104         dm_block_t root;
 105         dm_block_t hint_root;
 106         dm_block_t discard_root;
 107
 108         sector_t discard_block_size;
 109         dm_dblock_t discard_nr_blocks;
 110
 111         sector_t data_block_size;
 112         dm_cblock_t cache_blocks;
 113         bool changed:1;
 114         bool clean_when_opened:1;
 115
 116         char policy_name[CACHE_POLICY_NAME_SIZE];
 117         unsigned policy_version[CACHE_POLICY_VERSION_SIZE];
 118         size_t policy_hint_size;
 119         struct dm_cache_statistics stats;
 120
 121         /*
 122          * Reading the space map root can fail, so we read it into this
 123          * buffer before the superblock is locked and updated.
 124          */
 125         __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
 126 };
 127
 128 /*-------------------------------------------------------------------
 129  * superblock validator
 130  *-----------------------------------------------------------------*/
 131
 132 #define SUPERBLOCK_CSUM_XOR 9031977
 133
 134 static void sb_prepare_for_write(struct dm_block_validator *v,
 135                                  struct dm_block *b,
 136                                  size_t sb_block_size)
 137 {
 138         struct cache_disk_superblock *disk_super = dm_block_data(b);
 139
 140         disk_super->blocknr = cpu_to_le64(dm_block_location(b));
 141         disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
 142                                                       sb_block_size - sizeof(__le32),
 143                                                       SUPERBLOCK_CSUM_XOR));
 144 }
 145
 146 static int sb_check(struct dm_block_validator *v,
 147                     struct dm_block *b,
 148                     size_t sb_block_size)
 149 {
 150         struct cache_disk_superblock *disk_super = dm_block_data(b);
 151         __le32 csum_le;
 152
 153         if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
 154                 DMERR("sb_check failed: blocknr %llu: wanted %llu",
 155                       le64_to_cpu(disk_super->blocknr),
 156                       (unsigned long long)dm_block_location(b));
 157                 return -ENOTBLK;
 158         }
 159
 160         if (le64_to_cpu(disk_super->magic) != CACHE_SUPERBLOCK_MAGIC) {
 161                 DMERR("sb_check failed: magic %llu: wanted %llu",
 162                       le64_to_cpu(disk_super->magic),
 163                       (unsigned long long)CACHE_SUPERBLOCK_MAGIC);
 164                 return -EILSEQ;
 165         }
 166
 167         csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
 168                                              sb_block_size - sizeof(__le32),
 169                                              SUPERBLOCK_CSUM_XOR));
 170         if (csum_le != disk_super->csum) {
 171                 DMERR("sb_check failed: csum %u: wanted %u",
 172                       le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
 173                 return -EILSEQ;
 174         }
 175
 176         return 0;
 177 }
 178
 179 static struct dm_block_validator sb_validator = {
 180         .name = "superblock",
 181         .prepare_for_write = sb_prepare_for_write,
 182         .check = sb_check
 183 };
 184
 185 /*----------------------------------------------------------------*/
 186
 187 static int superblock_read_lock(struct dm_cache_metadata *cmd,
 188                                 struct dm_block **sblock)
 189 {
 190         return dm_bm_read_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 191                                &sb_validator, sblock);
 192 }
 193
 194 static int superblock_lock_zero(struct dm_cache_metadata *cmd,
 195                                 struct dm_block **sblock)
 196 {
 197         return dm_bm_write_lock_zero(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 198                                      &sb_validator, sblock);
 199 }
 200
 201 static int superblock_lock(struct dm_cache_metadata *cmd,
 202                            struct dm_block **sblock)
 203 {
 204         return dm_bm_write_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 205                                 &sb_validator, sblock);
 206 }
 207
 208 /*----------------------------------------------------------------*/
 209
 210 static int __superblock_all_zeroes(struct dm_block_manager *bm, int *result)
 211 {
 212         int r;
 213         unsigned i;
 214         struct dm_block *b;
 215         __le64 *data_le, zero = cpu_to_le64(0);
 216         unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
 217
 218         /*
 219          * We can't use a validator here - it may be all zeroes.
 220          */
 221         r = dm_bm_read_lock(bm, CACHE_SUPERBLOCK_LOCATION, NULL, &b);
 222         if (r)
 223                 return r;
 224
 225         data_le = dm_block_data(b);
 226         *result = 1;
 227         for (i = 0; i < sb_block_size; i++) {
 228                 if (data_le[i] != zero) {
 229                         *result = 0;
 230                         break;
 231                 }
 232         }
 233
 234         return dm_bm_unlock(b);
 235 }
 236
 237 static void __setup_mapping_info(struct dm_cache_metadata *cmd)
 238 {
 239         struct dm_btree_value_type vt;
 240
 241         vt.context = NULL;
 242         vt.size = sizeof(__le64);
 243         vt.inc = NULL;
 244         vt.dec = NULL;
 245         vt.equal = NULL;
 246         dm_array_info_init(&cmd->info, cmd->tm, &vt);
 247
 248         if (cmd->policy_hint_size) {
 249                 vt.size = sizeof(__le32);
 250                 dm_array_info_init(&cmd->hint_info, cmd->tm, &vt);
 251         }
 252 }
 253
 254 static int __save_sm_root(struct dm_cache_metadata *cmd)
 255 {
 256         int r;
 257         size_t metadata_len;
 258
 259         r = dm_sm_root_size(cmd->metadata_sm, &metadata_len);
 260         if (r < 0)
 261                 return r;
 262
 263         return dm_sm_copy_root(cmd->metadata_sm, &cmd->metadata_space_map_root,
 264                                metadata_len);
 265 }
 266
 267 static void __copy_sm_root(struct dm_cache_metadata *cmd,
 268                            struct cache_disk_superblock *disk_super)
 269 {
 270         memcpy(&disk_super->metadata_space_map_root,
 271                &cmd->metadata_space_map_root,
 272                sizeof(cmd->metadata_space_map_root));
 273 }
 274
 275 static int __write_initial_superblock(struct dm_cache_metadata *cmd)
 276 {
 277         int r;
 278         struct dm_block *sblock;
 279         struct cache_disk_superblock *disk_super;
 280         sector_t bdev_size = i_size_read(cmd->bdev->bd_inode) >> SECTOR_SHIFT;
 281
 282         /* FIXME: see if we can lose the max sectors limit */
 283         if (bdev_size > DM_CACHE_METADATA_MAX_SECTORS)
 284                 bdev_size = DM_CACHE_METADATA_MAX_SECTORS;
 285
 286         r = dm_tm_pre_commit(cmd->tm);
 287         if (r < 0)
 288                 return r;
 289
 290         /*
 291          * dm_sm_copy_root() can fail.  So we need to do it before we start
 292          * updating the superblock.
 293          */
 294         r = __save_sm_root(cmd);
 295         if (r)
 296                 return r;
 297
 298         r = superblock_lock_zero(cmd, &sblock);
 299         if (r)
 300                 return r;
 301
 302         disk_super = dm_block_data(sblock);
 303         disk_super->flags = 0;
 304         memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
 305         disk_super->magic = cpu_to_le64(CACHE_SUPERBLOCK_MAGIC);
 306         disk_super->version = cpu_to_le32(CACHE_VERSION);
 307         memset(disk_super->policy_name, 0, sizeof(disk_super->policy_name));
 308         memset(disk_super->policy_version, 0, sizeof(disk_super->policy_version));
 309         disk_super->policy_hint_size = 0;
 310
 311         __copy_sm_root(cmd, disk_super);
 312
 313         disk_super->mapping_root = cpu_to_le64(cmd->root);
 314         disk_super->hint_root = cpu_to_le64(cmd->hint_root);
 315         disk_super->discard_root = cpu_to_le64(cmd->discard_root);
 316         disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
 317         disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
 318         disk_super->metadata_block_size = cpu_to_le32(DM_CACHE_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
 319         disk_super->data_block_size = cpu_to_le32(cmd->data_block_size);
 320         disk_super->cache_blocks = cpu_to_le32(0);
 321
 322         disk_super->read_hits = cpu_to_le32(0);
 323         disk_super->read_misses = cpu_to_le32(0);
 324         disk_super->write_hits = cpu_to_le32(0);
 325         disk_super->write_misses = cpu_to_le32(0);
 326
 327         return dm_tm_commit(cmd->tm, sblock);
 328 }
 329
 330 static int __format_metadata(struct dm_cache_metadata *cmd)
 331 {
 332         int r;
 333
 334         r = dm_tm_create_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 335                                  &cmd->tm, &cmd->metadata_sm);
 336         if (r < 0) {
 337                 DMERR("tm_create_with_sm failed");
 338                 return r;
 339         }
 340
 341         __setup_mapping_info(cmd);
 342
 343         r = dm_array_empty(&cmd->info, &cmd->root);
 344         if (r < 0)
 345                 goto bad;
 346
 347         dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
 348
 349         r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root);
 350         if (r < 0)
 351                 goto bad;
 352
 353         cmd->discard_block_size = 0;
 354         cmd->discard_nr_blocks = 0;
 355
 356         r = __write_initial_superblock(cmd);
 357         if (r)
 358                 goto bad;
 359
 360         cmd->clean_when_opened = true;
 361         return 0;
 362
 363 bad:
 364         dm_tm_destroy(cmd->tm);
 365         dm_sm_destroy(cmd->metadata_sm);
 366
 367         return r;
 368 }
 369
 370 static int __check_incompat_features(struct cache_disk_superblock *disk_super,
 371                                      struct dm_cache_metadata *cmd)
 372 {
 373         uint32_t features;
 374
 375         features = le32_to_cpu(disk_super->incompat_flags) & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
 376         if (features) {
 377                 DMERR("could not access metadata due to unsupported optional features (%lx).",
 378                       (unsigned long)features);
 379                 return -EINVAL;
 380         }
 381
 382         /*
 383          * Check for read-only metadata to skip the following RDWR checks.
 384          */
 385         if (get_disk_ro(cmd->bdev->bd_disk))
 386                 return 0;
 387
 388         features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP;
 389         if (features) {
 390                 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
 391                       (unsigned long)features);
 392                 return -EINVAL;
 393         }
 394
 395         return 0;
 396 }
 397
 398 static int __open_metadata(struct dm_cache_metadata *cmd)
 399 {
 400         int r;
 401         struct dm_block *sblock;
 402         struct cache_disk_superblock *disk_super;
 403         unsigned long sb_flags;
 404
 405         r = superblock_read_lock(cmd, &sblock);
 406         if (r < 0) {
 407                 DMERR("couldn't read lock superblock");
 408                 return r;
 409         }
 410
 411         disk_super = dm_block_data(sblock);
 412
 413         /* Verify the data block size hasn't changed */
 414         if (le32_to_cpu(disk_super->data_block_size) != cmd->data_block_size) {
 415                 DMERR("changing the data block size (from %u to %llu) is not supported",
 416                       le32_to_cpu(disk_super->data_block_size),
 417                       (unsigned long long)cmd->data_block_size);
 418                 r = -EINVAL;
 419                 goto bad;
 420         }
 421
 422         r = __check_incompat_features(disk_super, cmd);
 423         if (r < 0)
 424                 goto bad;
 425
 426         r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
 427                                disk_super->metadata_space_map_root,
 428                                sizeof(disk_super->metadata_space_map_root),
 429                                &cmd->tm, &cmd->metadata_sm);
 430         if (r < 0) {
 431                 DMERR("tm_open_with_sm failed");
 432                 goto bad;
 433         }
 434
 435         __setup_mapping_info(cmd);
 436         dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
 437         sb_flags = le32_to_cpu(disk_super->flags);
 438         cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags);
 439         return dm_bm_unlock(sblock);
 440
 441 bad:
 442         dm_bm_unlock(sblock);
 443         return r;
 444 }
 445
 446 static int __open_or_format_metadata(struct dm_cache_metadata *cmd,
 447                                      bool format_device)
 448 {
 449         int r, unformatted;
 450
 451         r = __superblock_all_zeroes(cmd->bm, &unformatted);
 452         if (r)
 453                 return r;
 454
 455         if (unformatted)
 456                 return format_device ? __format_metadata(cmd) : -EPERM;
 457
 458         return __open_metadata(cmd);
 459 }
 460
 461 static int __create_persistent_data_objects(struct dm_cache_metadata *cmd,
 462                                             bool may_format_device)
 463 {
 464         int r;
 465         cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE,
 466                                           CACHE_METADATA_CACHE_SIZE,
 467                                           CACHE_MAX_CONCURRENT_LOCKS);
 468         if (IS_ERR(cmd->bm)) {
 469                 DMERR("could not create block manager");
 470                 return PTR_ERR(cmd->bm);
 471         }
 472
 473         r = __open_or_format_metadata(cmd, may_format_device);
 474         if (r)
 475                 dm_block_manager_destroy(cmd->bm);
 476
 477         return r;
 478 }
 479
 480 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd)
 481 {
 482         dm_sm_destroy(cmd->metadata_sm);
 483         dm_tm_destroy(cmd->tm);
 484         dm_block_manager_destroy(cmd->bm);
 485 }
 486
 487 typedef unsigned long (*flags_mutator)(unsigned long);
 488
 489 static void update_flags(struct cache_disk_superblock *disk_super,
 490                          flags_mutator mutator)
 491 {
 492         uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
 493         disk_super->flags = cpu_to_le32(sb_flags);
 494 }
 495
 496 static unsigned long set_clean_shutdown(unsigned long flags)
 497 {
 498         set_bit(CLEAN_SHUTDOWN, &flags);
 499         return flags;
 500 }
 501
 502 static unsigned long clear_clean_shutdown(unsigned long flags)
 503 {
 504         clear_bit(CLEAN_SHUTDOWN, &flags);
 505         return flags;
 506 }
 507
 508 static void read_superblock_fields(struct dm_cache_metadata *cmd,
 509                                    struct cache_disk_superblock *disk_super)
 510 {
 511         cmd->root = le64_to_cpu(disk_super->mapping_root);
 512         cmd->hint_root = le64_to_cpu(disk_super->hint_root);
 513         cmd->discard_root = le64_to_cpu(disk_super->discard_root);
 514         cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
 515         cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
 516         cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
 517         cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
 518         strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
 519         cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
 520         cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
 521         cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
 522         cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
 523
 524         cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
 525         cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
 526         cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
 527         cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
 528
 529         cmd->changed = false;
 530 }
 531
 532 /*
 533  * The mutator updates the superblock flags.
 534  */
 535 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
 536                                      flags_mutator mutator)
 537 {
 538         int r;
 539         struct cache_disk_superblock *disk_super;
 540         struct dm_block *sblock;
 541
 542         r = superblock_lock(cmd, &sblock);
 543         if (r)
 544                 return r;
 545
 546         disk_super = dm_block_data(sblock);
 547         update_flags(disk_super, mutator);
 548         read_superblock_fields(cmd, disk_super);
 549         dm_bm_unlock(sblock);
 550
 551         return dm_bm_flush(cmd->bm);
 552 }
 553
 554 static int __begin_transaction(struct dm_cache_metadata *cmd)
 555 {
 556         int r;
 557         struct cache_disk_superblock *disk_super;
 558         struct dm_block *sblock;
 559
 560         /*
 561          * We re-read the superblock every time.  Shouldn't need to do this
 562          * really.
 563          */
 564         r = superblock_read_lock(cmd, &sblock);
 565         if (r)
 566                 return r;
 567
 568         disk_super = dm_block_data(sblock);
 569         read_superblock_fields(cmd, disk_super);
 570         dm_bm_unlock(sblock);
 571
 572         return 0;
 573 }
 574
 575 static int __commit_transaction(struct dm_cache_metadata *cmd,
 576                                 flags_mutator mutator)
 577 {
 578         int r;
 579         struct cache_disk_superblock *disk_super;
 580         struct dm_block *sblock;
 581
 582         /*
 583          * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
 584          */
 585         BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
 586
 587         r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
 588                             &cmd->discard_root);
 589         if (r)
 590                 return r;
 591
 592         r = dm_tm_pre_commit(cmd->tm);
 593         if (r < 0)
 594                 return r;
 595
 596         r = __save_sm_root(cmd);
 597         if (r)
 598                 return r;
 599
 600         r = superblock_lock(cmd, &sblock);
 601         if (r)
 602                 return r;
 603
 604         disk_super = dm_block_data(sblock);
 605
 606         if (mutator)
 607                 update_flags(disk_super, mutator);
 608
 609         disk_super->mapping_root = cpu_to_le64(cmd->root);
 610         disk_super->hint_root = cpu_to_le64(cmd->hint_root);
 611         disk_super->discard_root = cpu_to_le64(cmd->discard_root);
 612         disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
 613         disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
 614         disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
 615         strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
 616         disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
 617         disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
 618         disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
 619
 620         disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
 621         disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
 622         disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
 623         disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
 624         __copy_sm_root(cmd, disk_super);
 625
 626         return dm_tm_commit(cmd->tm, sblock);
 627 }
 628
 629 /*----------------------------------------------------------------*/
 630
 631 /*
 632  * The mappings are held in a dm-array that has 64-bit values stored in
 633  * little-endian format.  The index is the cblock, the high 48bits of the
 634  * value are the oblock and the low 16 bit the flags.
 635  */
 636 #define FLAGS_MASK ((1 << 16) - 1)
 637
 638 static __le64 pack_value(dm_oblock_t block, unsigned flags)
 639 {
 640         uint64_t value = from_oblock(block);
 641         value <<= 16;
 642         value = value | (flags & FLAGS_MASK);
 643         return cpu_to_le64(value);
 644 }
 645
 646 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
 647 {
 648         uint64_t value = le64_to_cpu(value_le);
 649         uint64_t b = value >> 16;
 650         *block = to_oblock(b);
 651         *flags = value & FLAGS_MASK;
 652 }
 653
 654 /*----------------------------------------------------------------*/
 655
 656 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
 657                                                sector_t data_block_size,
 658                                                bool may_format_device,
 659                                                size_t policy_hint_size)
 660 {
 661         int r;
 662         struct dm_cache_metadata *cmd;
 663
 664         cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
 665         if (!cmd) {
 666                 DMERR("could not allocate metadata struct");
 667                 return ERR_PTR(-ENOMEM);
 668         }
 669
 670         atomic_set(&cmd->ref_count, 1);
 671         init_rwsem(&cmd->root_lock);
 672         cmd->bdev = bdev;
 673         cmd->data_block_size = data_block_size;
 674         cmd->cache_blocks = 0;
 675         cmd->policy_hint_size = policy_hint_size;
 676         cmd->changed = true;
 677
 678         r = __create_persistent_data_objects(cmd, may_format_device);
 679         if (r) {
 680                 kfree(cmd);
 681                 return ERR_PTR(r);
 682         }
 683
 684         r = __begin_transaction_flags(cmd, clear_clean_shutdown);
 685         if (r < 0) {
 686                 dm_cache_metadata_close(cmd);
 687                 return ERR_PTR(r);
 688         }
 689
 690         return cmd;
 691 }
 692
 693 /*
 694  * We keep a little list of ref counted metadata objects to prevent two
 695  * different target instances creating separate bufio instances.  This is
 696  * an issue if a table is reloaded before the suspend.
 697  */
 698 static DEFINE_MUTEX(table_lock);
 699 static LIST_HEAD(table);
 700
 701 static struct dm_cache_metadata *lookup(struct block_device *bdev)
 702 {
 703         struct dm_cache_metadata *cmd;
 704
 705         list_for_each_entry(cmd, &table, list)
 706                 if (cmd->bdev == bdev) {
 707                         atomic_inc(&cmd->ref_count);
 708                         return cmd;
 709                 }
 710
 711         return NULL;
 712 }
 713
 714 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
 715                                                 sector_t data_block_size,
 716                                                 bool may_format_device,
 717                                                 size_t policy_hint_size)
 718 {
 719         struct dm_cache_metadata *cmd, *cmd2;
 720
 721         mutex_lock(&table_lock);
 722         cmd = lookup(bdev);
 723         mutex_unlock(&table_lock);
 724
 725         if (cmd)
 726                 return cmd;
 727
 728         cmd = metadata_open(bdev, data_block_size, may_format_device, policy_hint_size);
 729         if (!IS_ERR(cmd)) {
 730                 mutex_lock(&table_lock);
 731                 cmd2 = lookup(bdev);
 732                 if (cmd2) {
 733                         mutex_unlock(&table_lock);
 734                         __destroy_persistent_data_objects(cmd);
 735                         kfree(cmd);
 736                         return cmd2;
 737                 }
 738                 list_add(&cmd->list, &table);
 739                 mutex_unlock(&table_lock);
 740         }
 741
 742         return cmd;
 743 }
 744
 745 static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
 746 {
 747         if (cmd->data_block_size != data_block_size) {
 748                 DMERR("data_block_size (%llu) different from that in metadata (%llu)\n",
 749                       (unsigned long long) data_block_size,
 750                       (unsigned long long) cmd->data_block_size);
 751                 return false;
 752         }
 753
 754         return true;
 755 }
 756
 757 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
 758                                                  sector_t data_block_size,
 759                                                  bool may_format_device,
 760                                                  size_t policy_hint_size)
 761 {
 762         struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size,
 763                                                        may_format_device, policy_hint_size);
 764
 765         if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
 766                 dm_cache_metadata_close(cmd);
 767                 return ERR_PTR(-EINVAL);
 768         }
 769
 770         return cmd;
 771 }
 772
 773 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
 774 {
 775         if (atomic_dec_and_test(&cmd->ref_count)) {
 776                 mutex_lock(&table_lock);
 777                 list_del(&cmd->list);
 778                 mutex_unlock(&table_lock);
 779
 780                 __destroy_persistent_data_objects(cmd);
 781                 kfree(cmd);
 782         }
 783 }
 784
 785 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
 786 {
 787         int r;
 788         __le64 null_mapping = pack_value(0, 0);
 789
 790         down_write(&cmd->root_lock);
 791         __dm_bless_for_disk(&null_mapping);
 792         r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
 793                             from_cblock(new_cache_size),
 794                             &null_mapping, &cmd->root);
 795         if (!r)
 796                 cmd->cache_blocks = new_cache_size;
 797         cmd->changed = true;
 798         up_write(&cmd->root_lock);
 799
 800         return r;
 801 }
 802
 803 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
 804                                    sector_t discard_block_size,
 805                                    dm_dblock_t new_nr_entries)
 806 {
 807         int r;
 808
 809         down_write(&cmd->root_lock);
 810         r = dm_bitset_resize(&cmd->discard_info,
 811                              cmd->discard_root,
 812                              from_dblock(cmd->discard_nr_blocks),
 813                              from_dblock(new_nr_entries),
 814                              false, &cmd->discard_root);
 815         if (!r) {
 816                 cmd->discard_block_size = discard_block_size;
 817                 cmd->discard_nr_blocks = new_nr_entries;
 818         }
 819
 820         cmd->changed = true;
 821         up_write(&cmd->root_lock);
 822
 823         return r;
 824 }
 825
 826 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
 827 {
 828         return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
 829                                  from_dblock(b), &cmd->discard_root);
 830 }
 831
 832 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
 833 {
 834         return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
 835                                    from_dblock(b), &cmd->discard_root);
 836 }
 837
 838 static int __is_discarded(struct dm_cache_metadata *cmd, dm_dblock_t b,
 839                           bool *is_discarded)
 840 {
 841         return dm_bitset_test_bit(&cmd->discard_info, cmd->discard_root,
 842                                   from_dblock(b), &cmd->discard_root,
 843                                   is_discarded);
 844 }
 845
 846 static int __discard(struct dm_cache_metadata *cmd,
 847                      dm_dblock_t dblock, bool discard)
 848 {
 849         int r;
 850
 851         r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
 852         if (r)
 853                 return r;
 854
 855         cmd->changed = true;
 856         return 0;
 857 }
 858
 859 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
 860                          dm_dblock_t dblock, bool discard)
 861 {
 862         int r;
 863
 864         down_write(&cmd->root_lock);
 865         r = __discard(cmd, dblock, discard);
 866         up_write(&cmd->root_lock);
 867
 868         return r;
 869 }
 870
 871 static int __load_discards(struct dm_cache_metadata *cmd,
 872                            load_discard_fn fn, void *context)
 873 {
 874         int r = 0;
 875         dm_block_t b;
 876         bool discard;
 877
 878         for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
 879                 dm_dblock_t dblock = to_dblock(b);
 880
 881                 if (cmd->clean_when_opened) {
 882                         r = __is_discarded(cmd, dblock, &discard);
 883                         if (r)
 884                                 return r;
 885                 } else
 886                         discard = false;
 887
 888                 r = fn(context, cmd->discard_block_size, dblock, discard);
 889                 if (r)
 890                         break;
 891         }
 892
 893         return r;
 894 }
 895
 896 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
 897                            load_discard_fn fn, void *context)
 898 {
 899         int r;
 900
 901         down_read(&cmd->root_lock);
 902         r = __load_discards(cmd, fn, context);
 903         up_read(&cmd->root_lock);
 904
 905         return r;
 906 }
 907
 908 dm_cblock_t dm_cache_size(struct dm_cache_metadata *cmd)
 909 {
 910         dm_cblock_t r;
 911
 912         down_read(&cmd->root_lock);
 913         r = cmd->cache_blocks;
 914         up_read(&cmd->root_lock);
 915
 916         return r;
 917 }
 918
 919 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
 920 {
 921         int r;
 922         __le64 value = pack_value(0, 0);
 923
 924         __dm_bless_for_disk(&value);
 925         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
 926                                &value, &cmd->root);
 927         if (r)
 928                 return r;
 929
 930         cmd->changed = true;
 931         return 0;
 932 }
 933
 934 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
 935 {
 936         int r;
 937
 938         down_write(&cmd->root_lock);
 939         r = __remove(cmd, cblock);
 940         up_write(&cmd->root_lock);
 941
 942         return r;
 943 }
 944
 945 static int __insert(struct dm_cache_metadata *cmd,
 946                     dm_cblock_t cblock, dm_oblock_t oblock)
 947 {
 948         int r;
 949         __le64 value = pack_value(oblock, M_VALID);
 950         __dm_bless_for_disk(&value);
 951
 952         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
 953                                &value, &cmd->root);
 954         if (r)
 955                 return r;
 956
 957         cmd->changed = true;
 958         return 0;
 959 }
 960
 961 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
 962                             dm_cblock_t cblock, dm_oblock_t oblock)
 963 {
 964         int r;
 965
 966         down_write(&cmd->root_lock);
 967         r = __insert(cmd, cblock, oblock);
 968         up_write(&cmd->root_lock);
 969
 970         return r;
 971 }
 972
 973 struct thunk {
 974         load_mapping_fn fn;
 975         void *context;
 976
 977         struct dm_cache_metadata *cmd;
 978         bool respect_dirty_flags;
 979         bool hints_valid;
 980 };
 981
 982 static bool policy_unchanged(struct dm_cache_metadata *cmd,
 983                              struct dm_cache_policy *policy)
 984 {
 985         const char *policy_name = dm_cache_policy_get_name(policy);
 986         const unsigned *policy_version = dm_cache_policy_get_version(policy);
 987         size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
 988
 989         /*
 990          * Ensure policy names match.
 991          */
 992         if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
 993                 return false;
 994
 995         /*
 996          * Ensure policy major versions match.
 997          */
 998         if (cmd->policy_version[0] != policy_version[0])
 999                 return false;
1000
1001         /*
1002          * Ensure policy hint sizes match.
1003          */
1004         if (cmd->policy_hint_size != policy_hint_size)
1005                 return false;
1006
1007         return true;
1008 }
1009
1010 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1011 {
1012         return cmd->hint_root && cmd->policy_hint_size;
1013 }
1014
1015 static bool hints_array_available(struct dm_cache_metadata *cmd,
1016                                   struct dm_cache_policy *policy)
1017 {
1018         return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1019                 hints_array_initialized(cmd);
1020 }
1021
1022 static int __load_mapping(void *context, uint64_t cblock, void *leaf)
1023 {
1024         int r = 0;
1025         bool dirty;
1026         __le64 value;
1027         __le32 hint_value = 0;
1028         dm_oblock_t oblock;
1029         unsigned flags;
1030         struct thunk *thunk = context;
1031         struct dm_cache_metadata *cmd = thunk->cmd;
1032
1033         memcpy(&value, leaf, sizeof(value));
1034         unpack_value(value, &oblock, &flags);
1035
1036         if (flags & M_VALID) {
1037                 if (thunk->hints_valid) {
1038                         r = dm_array_get_value(&cmd->hint_info, cmd->hint_root,
1039                                                cblock, &hint_value);
1040                         if (r && r != -ENODATA)
1041                                 return r;
1042                 }
1043
1044                 dirty = thunk->respect_dirty_flags ? (flags & M_DIRTY) : true;
1045                 r = thunk->fn(thunk->context, oblock, to_cblock(cblock),
1046                               dirty, le32_to_cpu(hint_value), thunk->hints_valid);
1047         }
1048
1049         return r;
1050 }
1051
1052 static int __load_mappings(struct dm_cache_metadata *cmd,
1053                            struct dm_cache_policy *policy,
1054                            load_mapping_fn fn, void *context)
1055 {
1056         struct thunk thunk;
1057
1058         thunk.fn = fn;
1059         thunk.context = context;
1060
1061         thunk.cmd = cmd;
1062         thunk.respect_dirty_flags = cmd->clean_when_opened;
1063         thunk.hints_valid = hints_array_available(cmd, policy);
1064
1065         return dm_array_walk(&cmd->info, cmd->root, __load_mapping, &thunk);
1066 }
1067
1068 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1069                            struct dm_cache_policy *policy,
1070                            load_mapping_fn fn, void *context)
1071 {
1072         int r;
1073
1074         down_read(&cmd->root_lock);
1075         r = __load_mappings(cmd, policy, fn, context);
1076         up_read(&cmd->root_lock);
1077
1078         return r;
1079 }
1080
1081 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
1082 {
1083         int r = 0;
1084         __le64 value;
1085         dm_oblock_t oblock;
1086         unsigned flags;
1087
1088         memcpy(&value, leaf, sizeof(value));
1089         unpack_value(value, &oblock, &flags);
1090
1091         return r;
1092 }
1093
1094 static int __dump_mappings(struct dm_cache_metadata *cmd)
1095 {
1096         return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1097 }
1098
1099 void dm_cache_dump(struct dm_cache_metadata *cmd)
1100 {
1101         down_read(&cmd->root_lock);
1102         __dump_mappings(cmd);
1103         up_read(&cmd->root_lock);
1104 }
1105
1106 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1107 {
1108         int r;
1109
1110         down_read(&cmd->root_lock);
1111         r = cmd->changed;
1112         up_read(&cmd->root_lock);
1113
1114         return r;
1115 }
1116
1117 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1118 {
1119         int r;
1120         unsigned flags;
1121         dm_oblock_t oblock;
1122         __le64 value;
1123
1124         r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1125         if (r)
1126                 return r;
1127
1128         unpack_value(value, &oblock, &flags);
1129
1130         if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1131                 /* nothing to be done */
1132                 return 0;
1133
1134         value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1135         __dm_bless_for_disk(&value);
1136
1137         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1138                                &value, &cmd->root);
1139         if (r)
1140                 return r;
1141
1142         cmd->changed = true;
1143         return 0;
1144
1145 }
1146
1147 int dm_cache_set_dirty(struct dm_cache_metadata *cmd,
1148                        dm_cblock_t cblock, bool dirty)
1149 {
1150         int r;
1151
1152         down_write(&cmd->root_lock);
1153         r = __dirty(cmd, cblock, dirty);
1154         up_write(&cmd->root_lock);
1155
1156         return r;
1157 }
1158
1159 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1160                                  struct dm_cache_statistics *stats)
1161 {
1162         down_read(&cmd->root_lock);
1163         *stats = cmd->stats;
1164         up_read(&cmd->root_lock);
1165 }
1166
1167 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1168                                  struct dm_cache_statistics *stats)
1169 {
1170         down_write(&cmd->root_lock);
1171         cmd->stats = *stats;
1172         up_write(&cmd->root_lock);
1173 }
1174
1175 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1176 {
1177         int r;
1178         flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1179                                  clear_clean_shutdown);
1180
1181         down_write(&cmd->root_lock);
1182         r = __commit_transaction(cmd, mutator);
1183         if (r)
1184                 goto out;
1185
1186         r = __begin_transaction(cmd);
1187
1188 out:
1189         up_write(&cmd->root_lock);
1190         return r;
1191 }
1192
1193 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1194                                            dm_block_t *result)
1195 {
1196         int r = -EINVAL;
1197
1198         down_read(&cmd->root_lock);
1199         r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1200         up_read(&cmd->root_lock);
1201
1202         return r;
1203 }
1204
1205 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1206                                    dm_block_t *result)
1207 {
1208         int r = -EINVAL;
1209
1210         down_read(&cmd->root_lock);
1211         r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1212         up_read(&cmd->root_lock);
1213
1214         return r;
1215 }
1216
1217 /*----------------------------------------------------------------*/
1218
1219 static int begin_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1220 {
1221         int r;
1222         __le32 value;
1223         size_t hint_size;
1224         const char *policy_name = dm_cache_policy_get_name(policy);
1225         const unsigned *policy_version = dm_cache_policy_get_version(policy);
1226
1227         if (!policy_name[0] ||
1228             (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1229                 return -EINVAL;
1230
1231         if (!policy_unchanged(cmd, policy)) {
1232                 strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1233                 memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1234
1235                 hint_size = dm_cache_policy_get_hint_size(policy);
1236                 if (!hint_size)
1237                         return 0; /* short-circuit hints initialization */
1238                 cmd->policy_hint_size = hint_size;
1239
1240                 if (cmd->hint_root) {
1241                         r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1242                         if (r)
1243                                 return r;
1244                 }
1245
1246                 r = dm_array_empty(&cmd->hint_info, &cmd->hint_root);
1247                 if (r)
1248                         return r;
1249
1250                 value = cpu_to_le32(0);
1251                 __dm_bless_for_disk(&value);
1252                 r = dm_array_resize(&cmd->hint_info, cmd->hint_root, 0,
1253                                     from_cblock(cmd->cache_blocks),
1254                                     &value, &cmd->hint_root);
1255                 if (r)
1256                         return r;
1257         }
1258
1259         return 0;
1260 }
1261
1262 int dm_cache_begin_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1263 {
1264         int r;
1265
1266         down_write(&cmd->root_lock);
1267         r = begin_hints(cmd, policy);
1268         up_write(&cmd->root_lock);
1269
1270         return r;
1271 }
1272
1273 static int save_hint(struct dm_cache_metadata *cmd, dm_cblock_t cblock,
1274                      uint32_t hint)
1275 {
1276         int r;
1277         __le32 value = cpu_to_le32(hint);
1278         __dm_bless_for_disk(&value);
1279
1280         r = dm_array_set_value(&cmd->hint_info, cmd->hint_root,
1281                                from_cblock(cblock), &value, &cmd->hint_root);
1282         cmd->changed = true;
1283
1284         return r;
1285 }
1286
1287 int dm_cache_save_hint(struct dm_cache_metadata *cmd, dm_cblock_t cblock,
1288                        uint32_t hint)
1289 {
1290         int r;
1291
1292         if (!hints_array_initialized(cmd))
1293                 return 0;
1294
1295         down_write(&cmd->root_lock);
1296         r = save_hint(cmd, cblock, hint);
1297         up_write(&cmd->root_lock);
1298
1299         return r;
1300 }