drivers/md/dm-thin-metadata.c

   1 /*
   2  * Copyright (C) 2011-2012 Red Hat, Inc.
   3  *
   4  * This file is released under the GPL.
   5  */
   6
   7 #include "dm-thin-metadata.h"
   8 #include "persistent-data/dm-btree.h"
   9 #include "persistent-data/dm-space-map.h"
  10 #include "persistent-data/dm-space-map-disk.h"
  11 #include "persistent-data/dm-transaction-manager.h"
  12
  13 #include <linux/list.h>
  14 #include <linux/device-mapper.h>
  15 #include <linux/workqueue.h>
  16
  17 /*--------------------------------------------------------------------------
  18  * As far as the metadata goes, there is:
  19  *
  20  * - A superblock in block zero, taking up fewer than 512 bytes for
  21  *   atomic writes.
  22  *
  23  * - A space map managing the metadata blocks.
  24  *
  25  * - A space map managing the data blocks.
  26  *
  27  * - A btree mapping our internal thin dev ids onto struct disk_device_details.
  28  *
  29  * - A hierarchical btree, with 2 levels which effectively maps (thin
  30  *   dev id, virtual block) -> block_time.  Block time is a 64-bit
  31  *   field holding the time in the low 24 bits, and block in the top 48
  32  *   bits.
  33  *
  34  * BTrees consist solely of btree_nodes, that fill a block.  Some are
  35  * internal nodes, as such their values are a __le64 pointing to other
  36  * nodes.  Leaf nodes can store data of any reasonable size (ie. much
  37  * smaller than the block size).  The nodes consist of the header,
  38  * followed by an array of keys, followed by an array of values.  We have
  39  * to binary search on the keys so they're all held together to help the
  40  * cpu cache.
  41  *
  42  * Space maps have 2 btrees:
  43  *
  44  * - One maps a uint64_t onto a struct index_entry.  Which points to a
  45  *   bitmap block, and has some details about how many free entries there
  46  *   are etc.
  47  *
  48  * - The bitmap blocks have a header (for the checksum).  Then the rest
  49  *   of the block is pairs of bits.  With the meaning being:
  50  *
  51  *   0 - ref count is 0
  52  *   1 - ref count is 1
  53  *   2 - ref count is 2
  54  *   3 - ref count is higher than 2
  55  *
  56  * - If the count is higher than 2 then the ref count is entered in a
  57  *   second btree that directly maps the block_address to a uint32_t ref
  58  *   count.
  59  *
  60  * The space map metadata variant doesn't have a bitmaps btree.  Instead
  61  * it has one single blocks worth of index_entries.  This avoids
  62  * recursive issues with the bitmap btree needing to allocate space in
  63  * order to insert.  With a small data block size such as 64k the
  64  * metadata support data devices that are hundreds of terrabytes.
  65  *
  66  * The space maps allocate space linearly from front to back.  Space that
  67  * is freed in a transaction is never recycled within that transaction.
  68  * To try and avoid fragmenting _free_ space the allocator always goes
  69  * back and fills in gaps.
  70  *
  71  * All metadata io is in THIN_METADATA_BLOCK_SIZE sized/aligned chunks
  72  * from the block manager.
  73  *--------------------------------------------------------------------------*/
  74
  75 #define DM_MSG_PREFIX   "thin metadata"
  76
  77 #define THIN_SUPERBLOCK_MAGIC 27022010
  78 #define THIN_SUPERBLOCK_LOCATION 0
  79 #define THIN_VERSION 2
  80 #define THIN_METADATA_CACHE_SIZE 64
  81 #define SECTOR_TO_BLOCK_SHIFT 3
  82
  83 /*
  84  *  3 for btree insert +
  85  *  2 for btree lookup used within space map
  86  */
  87 #define THIN_MAX_CONCURRENT_LOCKS 5
  88
  89 /* This should be plenty */
  90 #define SPACE_MAP_ROOT_SIZE 128
  91
  92 /*
  93  * Little endian on-disk superblock and device details.
  94  */
  95 struct thin_disk_superblock {
  96         __le32 csum;    /* Checksum of superblock except for this field. */
  97         __le32 flags;
  98         __le64 blocknr; /* This block number, dm_block_t. */
  99
 100         __u8 uuid[16];
 101         __le64 magic;
 102         __le32 version;
 103         __le32 time;
 104
 105         __le64 trans_id;
 106
 107         /*
 108          * Root held by userspace transactions.
 109          */
 110         __le64 held_root;
 111
 112         __u8 data_space_map_root[SPACE_MAP_ROOT_SIZE];
 113         __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
 114
 115         /*
 116          * 2-level btree mapping (dev_id, (dev block, time)) -> data block
 117          */
 118         __le64 data_mapping_root;
 119
 120         /*
 121          * Device detail root mapping dev_id -> device_details
 122          */
 123         __le64 device_details_root;
 124
 125         __le32 data_block_size;         /* In 512-byte sectors. */
 126
 127         __le32 metadata_block_size;     /* In 512-byte sectors. */
 128         __le64 metadata_nr_blocks;
 129
 130         __le32 compat_flags;
 131         __le32 compat_ro_flags;
 132         __le32 incompat_flags;
 133 } __packed;
 134
 135 struct disk_device_details {
 136         __le64 mapped_blocks;
 137         __le64 transaction_id;          /* When created. */
 138         __le32 creation_time;
 139         __le32 snapshotted_time;
 140 } __packed;
 141
 142 struct dm_pool_metadata {
 143         struct hlist_node hash;
 144
 145         struct block_device *bdev;
 146         struct dm_block_manager *bm;
 147         struct dm_space_map *metadata_sm;
 148         struct dm_space_map *data_sm;
 149         struct dm_transaction_manager *tm;
 150         struct dm_transaction_manager *nb_tm;
 151
 152         /*
 153          * Two-level btree.
 154          * First level holds thin_dev_t.
 155          * Second level holds mappings.
 156          */
 157         struct dm_btree_info info;
 158
 159         /*
 160          * Non-blocking version of the above.
 161          */
 162         struct dm_btree_info nb_info;
 163
 164         /*
 165          * Just the top level for deleting whole devices.
 166          */
 167         struct dm_btree_info tl_info;
 168
 169         /*
 170          * Just the bottom level for creating new devices.
 171          */
 172         struct dm_btree_info bl_info;
 173
 174         /*
 175          * Describes the device details btree.
 176          */
 177         struct dm_btree_info details_info;
 178
 179         struct rw_semaphore root_lock;
 180         uint32_t time;
 181         dm_block_t root;
 182         dm_block_t details_root;
 183         struct list_head thin_devices;
 184         uint64_t trans_id;
 185         unsigned long flags;
 186         sector_t data_block_size;
 187         bool read_only:1;
 188
 189         /*
 190          * Set if a transaction has to be aborted but the attempt to roll back
 191          * to the previous (good) transaction failed.  The only pool metadata
 192          * operation possible in this state is the closing of the device.
 193          */
 194         bool fail_io:1;
 195 };
 196
 197 struct dm_thin_device {
 198         struct list_head list;
 199         struct dm_pool_metadata *pmd;
 200         dm_thin_id id;
 201
 202         int open_count;
 203         bool changed:1;
 204         bool aborted_with_changes:1;
 205         uint64_t mapped_blocks;
 206         uint64_t transaction_id;
 207         uint32_t creation_time;
 208         uint32_t snapshotted_time;
 209 };
 210
 211 /*----------------------------------------------------------------
 212  * superblock validator
 213  *--------------------------------------------------------------*/
 214
 215 #define SUPERBLOCK_CSUM_XOR 160774
 216
 217 static void sb_prepare_for_write(struct dm_block_validator *v,
 218                                  struct dm_block *b,
 219                                  size_t block_size)
 220 {
 221         struct thin_disk_superblock *disk_super = dm_block_data(b);
 222
 223         disk_super->blocknr = cpu_to_le64(dm_block_location(b));
 224         disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
 225                                                       block_size - sizeof(__le32),
 226                                                       SUPERBLOCK_CSUM_XOR));
 227 }
 228
 229 static int sb_check(struct dm_block_validator *v,
 230                     struct dm_block *b,
 231                     size_t block_size)
 232 {
 233         struct thin_disk_superblock *disk_super = dm_block_data(b);
 234         __le32 csum_le;
 235
 236         if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
 237                 DMERR("sb_check failed: blocknr %llu: "
 238                       "wanted %llu", le64_to_cpu(disk_super->blocknr),
 239                       (unsigned long long)dm_block_location(b));
 240                 return -ENOTBLK;
 241         }
 242
 243         if (le64_to_cpu(disk_super->magic) != THIN_SUPERBLOCK_MAGIC) {
 244                 DMERR("sb_check failed: magic %llu: "
 245                       "wanted %llu", le64_to_cpu(disk_super->magic),
 246                       (unsigned long long)THIN_SUPERBLOCK_MAGIC);
 247                 return -EILSEQ;
 248         }
 249
 250         csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
 251                                              block_size - sizeof(__le32),
 252                                              SUPERBLOCK_CSUM_XOR));
 253         if (csum_le != disk_super->csum) {
 254                 DMERR("sb_check failed: csum %u: wanted %u",
 255                       le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
 256                 return -EILSEQ;
 257         }
 258
 259         return 0;
 260 }
 261
 262 static struct dm_block_validator sb_validator = {
 263         .name = "superblock",
 264         .prepare_for_write = sb_prepare_for_write,
 265         .check = sb_check
 266 };
 267
 268 /*----------------------------------------------------------------
 269  * Methods for the btree value types
 270  *--------------------------------------------------------------*/
 271
 272 static uint64_t pack_block_time(dm_block_t b, uint32_t t)
 273 {
 274         return (b << 24) | t;
 275 }
 276
 277 static void unpack_block_time(uint64_t v, dm_block_t *b, uint32_t *t)
 278 {
 279         *b = v >> 24;
 280         *t = v & ((1 << 24) - 1);
 281 }
 282
 283 static void data_block_inc(void *context, const void *value_le)
 284 {
 285         struct dm_space_map *sm = context;
 286         __le64 v_le;
 287         uint64_t b;
 288         uint32_t t;
 289
 290         memcpy(&v_le, value_le, sizeof(v_le));
 291         unpack_block_time(le64_to_cpu(v_le), &b, &t);
 292         dm_sm_inc_block(sm, b);
 293 }
 294
 295 static void data_block_dec(void *context, const void *value_le)
 296 {
 297         struct dm_space_map *sm = context;
 298         __le64 v_le;
 299         uint64_t b;
 300         uint32_t t;
 301
 302         memcpy(&v_le, value_le, sizeof(v_le));
 303         unpack_block_time(le64_to_cpu(v_le), &b, &t);
 304         dm_sm_dec_block(sm, b);
 305 }
 306
 307 static int data_block_equal(void *context, const void *value1_le, const void *value2_le)
 308 {
 309         __le64 v1_le, v2_le;
 310         uint64_t b1, b2;
 311         uint32_t t;
 312
 313         memcpy(&v1_le, value1_le, sizeof(v1_le));
 314         memcpy(&v2_le, value2_le, sizeof(v2_le));
 315         unpack_block_time(le64_to_cpu(v1_le), &b1, &t);
 316         unpack_block_time(le64_to_cpu(v2_le), &b2, &t);
 317
 318         return b1 == b2;
 319 }
 320
 321 static void subtree_inc(void *context, const void *value)
 322 {
 323         struct dm_btree_info *info = context;
 324         __le64 root_le;
 325         uint64_t root;
 326
 327         memcpy(&root_le, value, sizeof(root_le));
 328         root = le64_to_cpu(root_le);
 329         dm_tm_inc(info->tm, root);
 330 }
 331
 332 static void subtree_dec(void *context, const void *value)
 333 {
 334         struct dm_btree_info *info = context;
 335         __le64 root_le;
 336         uint64_t root;
 337
 338         memcpy(&root_le, value, sizeof(root_le));
 339         root = le64_to_cpu(root_le);
 340         if (dm_btree_del(info, root))
 341                 DMERR("btree delete failed\n");
 342 }
 343
 344 static int subtree_equal(void *context, const void *value1_le, const void *value2_le)
 345 {
 346         __le64 v1_le, v2_le;
 347         memcpy(&v1_le, value1_le, sizeof(v1_le));
 348         memcpy(&v2_le, value2_le, sizeof(v2_le));
 349
 350         return v1_le == v2_le;
 351 }
 352
 353 /*----------------------------------------------------------------*/
 354
 355 static int superblock_lock_zero(struct dm_pool_metadata *pmd,
 356                                 struct dm_block **sblock)
 357 {
 358         return dm_bm_write_lock_zero(pmd->bm, THIN_SUPERBLOCK_LOCATION,
 359                                      &sb_validator, sblock);
 360 }
 361
 362 static int superblock_lock(struct dm_pool_metadata *pmd,
 363                            struct dm_block **sblock)
 364 {
 365         return dm_bm_write_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
 366                                 &sb_validator, sblock);
 367 }
 368
 369 static int __superblock_all_zeroes(struct dm_block_manager *bm, int *result)
 370 {
 371         int r;
 372         unsigned i;
 373         struct dm_block *b;
 374         __le64 *data_le, zero = cpu_to_le64(0);
 375         unsigned block_size = dm_bm_block_size(bm) / sizeof(__le64);
 376
 377         /*
 378          * We can't use a validator here - it may be all zeroes.
 379          */
 380         r = dm_bm_read_lock(bm, THIN_SUPERBLOCK_LOCATION, NULL, &b);
 381         if (r)
 382                 return r;
 383
 384         data_le = dm_block_data(b);
 385         *result = 1;
 386         for (i = 0; i < block_size; i++) {
 387                 if (data_le[i] != zero) {
 388                         *result = 0;
 389                         break;
 390                 }
 391         }
 392
 393         return dm_bm_unlock(b);
 394 }
 395
 396 static void __setup_btree_details(struct dm_pool_metadata *pmd)
 397 {
 398         pmd->info.tm = pmd->tm;
 399         pmd->info.levels = 2;
 400         pmd->info.value_type.context = pmd->data_sm;
 401         pmd->info.value_type.size = sizeof(__le64);
 402         pmd->info.value_type.inc = data_block_inc;
 403         pmd->info.value_type.dec = data_block_dec;
 404         pmd->info.value_type.equal = data_block_equal;
 405
 406         memcpy(&pmd->nb_info, &pmd->info, sizeof(pmd->nb_info));
 407         pmd->nb_info.tm = pmd->nb_tm;
 408
 409         pmd->tl_info.tm = pmd->tm;
 410         pmd->tl_info.levels = 1;
 411         pmd->tl_info.value_type.context = &pmd->bl_info;
 412         pmd->tl_info.value_type.size = sizeof(__le64);
 413         pmd->tl_info.value_type.inc = subtree_inc;
 414         pmd->tl_info.value_type.dec = subtree_dec;
 415         pmd->tl_info.value_type.equal = subtree_equal;
 416
 417         pmd->bl_info.tm = pmd->tm;
 418         pmd->bl_info.levels = 1;
 419         pmd->bl_info.value_type.context = pmd->data_sm;
 420         pmd->bl_info.value_type.size = sizeof(__le64);
 421         pmd->bl_info.value_type.inc = data_block_inc;
 422         pmd->bl_info.value_type.dec = data_block_dec;
 423         pmd->bl_info.value_type.equal = data_block_equal;
 424
 425         pmd->details_info.tm = pmd->tm;
 426         pmd->details_info.levels = 1;
 427         pmd->details_info.value_type.context = NULL;
 428         pmd->details_info.value_type.size = sizeof(struct disk_device_details);
 429         pmd->details_info.value_type.inc = NULL;
 430         pmd->details_info.value_type.dec = NULL;
 431         pmd->details_info.value_type.equal = NULL;
 432 }
 433
 434 static int __write_initial_superblock(struct dm_pool_metadata *pmd)
 435 {
 436         int r;
 437         struct dm_block *sblock;
 438         size_t metadata_len, data_len;
 439         struct thin_disk_superblock *disk_super;
 440         sector_t bdev_size = i_size_read(pmd->bdev->bd_inode) >> SECTOR_SHIFT;
 441
 442         if (bdev_size > THIN_METADATA_MAX_SECTORS)
 443                 bdev_size = THIN_METADATA_MAX_SECTORS;
 444
 445         r = dm_sm_root_size(pmd->metadata_sm, &metadata_len);
 446         if (r < 0)
 447                 return r;
 448
 449         r = dm_sm_root_size(pmd->data_sm, &data_len);
 450         if (r < 0)
 451                 return r;
 452
 453         r = dm_sm_commit(pmd->data_sm);
 454         if (r < 0)
 455                 return r;
 456
 457         r = dm_tm_pre_commit(pmd->tm);
 458         if (r < 0)
 459                 return r;
 460
 461         r = superblock_lock_zero(pmd, &sblock);
 462         if (r)
 463                 return r;
 464
 465         disk_super = dm_block_data(sblock);
 466         disk_super->flags = 0;
 467         memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
 468         disk_super->magic = cpu_to_le64(THIN_SUPERBLOCK_MAGIC);
 469         disk_super->version = cpu_to_le32(THIN_VERSION);
 470         disk_super->time = 0;
 471         disk_super->trans_id = 0;
 472         disk_super->held_root = 0;
 473
 474         r = dm_sm_copy_root(pmd->metadata_sm, &disk_super->metadata_space_map_root,
 475                             metadata_len);
 476         if (r < 0)
 477                 goto bad_locked;
 478
 479         r = dm_sm_copy_root(pmd->data_sm, &disk_super->data_space_map_root,
 480                             data_len);
 481         if (r < 0)
 482                 goto bad_locked;
 483
 484         disk_super->data_mapping_root = cpu_to_le64(pmd->root);
 485         disk_super->device_details_root = cpu_to_le64(pmd->details_root);
 486         disk_super->metadata_block_size = cpu_to_le32(THIN_METADATA_BLOCK_SIZE);
 487         disk_super->metadata_nr_blocks = cpu_to_le64(bdev_size >> SECTOR_TO_BLOCK_SHIFT);
 488         disk_super->data_block_size = cpu_to_le32(pmd->data_block_size);
 489
 490         return dm_tm_commit(pmd->tm, sblock);
 491
 492 bad_locked:
 493         dm_bm_unlock(sblock);
 494         return r;
 495 }
 496
 497 static int __format_metadata(struct dm_pool_metadata *pmd)
 498 {
 499         int r;
 500
 501         r = dm_tm_create_with_sm(pmd->bm, THIN_SUPERBLOCK_LOCATION,
 502                                  &pmd->tm, &pmd->metadata_sm);
 503         if (r < 0) {
 504                 DMERR("tm_create_with_sm failed");
 505                 return r;
 506         }
 507
 508         pmd->data_sm = dm_sm_disk_create(pmd->tm, 0);
 509         if (IS_ERR(pmd->data_sm)) {
 510                 DMERR("sm_disk_create failed");
 511                 r = PTR_ERR(pmd->data_sm);
 512                 goto bad_cleanup_tm;
 513         }
 514
 515         pmd->nb_tm = dm_tm_create_non_blocking_clone(pmd->tm);
 516         if (!pmd->nb_tm) {
 517                 DMERR("could not create non-blocking clone tm");
 518                 r = -ENOMEM;
 519                 goto bad_cleanup_data_sm;
 520         }
 521
 522         __setup_btree_details(pmd);
 523
 524         r = dm_btree_empty(&pmd->info, &pmd->root);
 525         if (r < 0)
 526                 goto bad_cleanup_nb_tm;
 527
 528         r = dm_btree_empty(&pmd->details_info, &pmd->details_root);
 529         if (r < 0) {
 530                 DMERR("couldn't create devices root");
 531                 goto bad_cleanup_nb_tm;
 532         }
 533
 534         r = __write_initial_superblock(pmd);
 535         if (r)
 536                 goto bad_cleanup_nb_tm;
 537
 538         return 0;
 539
 540 bad_cleanup_nb_tm:
 541         dm_tm_destroy(pmd->nb_tm);
 542 bad_cleanup_data_sm:
 543         dm_sm_destroy(pmd->data_sm);
 544 bad_cleanup_tm:
 545         dm_tm_destroy(pmd->tm);
 546         dm_sm_destroy(pmd->metadata_sm);
 547
 548         return r;
 549 }
 550
 551 static int __check_incompat_features(struct thin_disk_superblock *disk_super,
 552                                      struct dm_pool_metadata *pmd)
 553 {
 554         uint32_t features;
 555
 556         features = le32_to_cpu(disk_super->incompat_flags) & ~THIN_FEATURE_INCOMPAT_SUPP;
 557         if (features) {
 558                 DMERR("could not access metadata due to unsupported optional features (%lx).",
 559                       (unsigned long)features);
 560                 return -EINVAL;
 561         }
 562
 563         /*
 564          * Check for read-only metadata to skip the following RDWR checks.
 565          */
 566         if (get_disk_ro(pmd->bdev->bd_disk))
 567                 return 0;
 568
 569         features = le32_to_cpu(disk_super->compat_ro_flags) & ~THIN_FEATURE_COMPAT_RO_SUPP;
 570         if (features) {
 571                 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
 572                       (unsigned long)features);
 573                 return -EINVAL;
 574         }
 575
 576         return 0;
 577 }
 578
 579 static int __open_metadata(struct dm_pool_metadata *pmd)
 580 {
 581         int r;
 582         struct dm_block *sblock;
 583         struct thin_disk_superblock *disk_super;
 584
 585         r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
 586                             &sb_validator, &sblock);
 587         if (r < 0) {
 588                 DMERR("couldn't read superblock");
 589                 return r;
 590         }
 591
 592         disk_super = dm_block_data(sblock);
 593
 594         r = __check_incompat_features(disk_super, pmd);
 595         if (r < 0)
 596                 goto bad_unlock_sblock;
 597
 598         r = dm_tm_open_with_sm(pmd->bm, THIN_SUPERBLOCK_LOCATION,
 599                                disk_super->metadata_space_map_root,
 600                                sizeof(disk_super->metadata_space_map_root),
 601                                &pmd->tm, &pmd->metadata_sm);
 602         if (r < 0) {
 603                 DMERR("tm_open_with_sm failed");
 604                 goto bad_unlock_sblock;
 605         }
 606
 607         pmd->data_sm = dm_sm_disk_open(pmd->tm, disk_super->data_space_map_root,
 608                                        sizeof(disk_super->data_space_map_root));
 609         if (IS_ERR(pmd->data_sm)) {
 610                 DMERR("sm_disk_open failed");
 611                 r = PTR_ERR(pmd->data_sm);
 612                 goto bad_cleanup_tm;
 613         }
 614
 615         pmd->nb_tm = dm_tm_create_non_blocking_clone(pmd->tm);
 616         if (!pmd->nb_tm) {
 617                 DMERR("could not create non-blocking clone tm");
 618                 r = -ENOMEM;
 619                 goto bad_cleanup_data_sm;
 620         }
 621
 622         __setup_btree_details(pmd);
 623         return dm_bm_unlock(sblock);
 624
 625 bad_cleanup_data_sm:
 626         dm_sm_destroy(pmd->data_sm);
 627 bad_cleanup_tm:
 628         dm_tm_destroy(pmd->tm);
 629         dm_sm_destroy(pmd->metadata_sm);
 630 bad_unlock_sblock:
 631         dm_bm_unlock(sblock);
 632
 633         return r;
 634 }
 635
 636 static int __open_or_format_metadata(struct dm_pool_metadata *pmd, bool format_device)
 637 {
 638         int r, unformatted;
 639
 640         r = __superblock_all_zeroes(pmd->bm, &unformatted);
 641         if (r)
 642                 return r;
 643
 644         if (unformatted)
 645                 return format_device ? __format_metadata(pmd) : -EPERM;
 646
 647         return __open_metadata(pmd);
 648 }
 649
 650 static int __create_persistent_data_objects(struct dm_pool_metadata *pmd, bool format_device)
 651 {
 652         int r;
 653
 654         pmd->bm = dm_block_manager_create(pmd->bdev, THIN_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
 655                                           THIN_METADATA_CACHE_SIZE,
 656                                           THIN_MAX_CONCURRENT_LOCKS);
 657         if (IS_ERR(pmd->bm)) {
 658                 DMERR("could not create block manager");
 659                 return PTR_ERR(pmd->bm);
 660         }
 661
 662         r = __open_or_format_metadata(pmd, format_device);
 663         if (r)
 664                 dm_block_manager_destroy(pmd->bm);
 665
 666         return r;
 667 }
 668
 669 static void __destroy_persistent_data_objects(struct dm_pool_metadata *pmd)
 670 {
 671         dm_sm_destroy(pmd->data_sm);
 672         dm_sm_destroy(pmd->metadata_sm);
 673         dm_tm_destroy(pmd->nb_tm);
 674         dm_tm_destroy(pmd->tm);
 675         dm_block_manager_destroy(pmd->bm);
 676 }
 677
 678 static int __begin_transaction(struct dm_pool_metadata *pmd)
 679 {
 680         int r;
 681         struct thin_disk_superblock *disk_super;
 682         struct dm_block *sblock;
 683
 684         /*
 685          * We re-read the superblock every time.  Shouldn't need to do this
 686          * really.
 687          */
 688         r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
 689                             &sb_validator, &sblock);
 690         if (r)
 691                 return r;
 692
 693         disk_super = dm_block_data(sblock);
 694         pmd->time = le32_to_cpu(disk_super->time);
 695         pmd->root = le64_to_cpu(disk_super->data_mapping_root);
 696         pmd->details_root = le64_to_cpu(disk_super->device_details_root);
 697         pmd->trans_id = le64_to_cpu(disk_super->trans_id);
 698         pmd->flags = le32_to_cpu(disk_super->flags);
 699         pmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
 700
 701         dm_bm_unlock(sblock);
 702         return 0;
 703 }
 704
 705 static int __write_changed_details(struct dm_pool_metadata *pmd)
 706 {
 707         int r;
 708         struct dm_thin_device *td, *tmp;
 709         struct disk_device_details details;
 710         uint64_t key;
 711
 712         list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
 713                 if (!td->changed)
 714                         continue;
 715
 716                 key = td->id;
 717
 718                 details.mapped_blocks = cpu_to_le64(td->mapped_blocks);
 719                 details.transaction_id = cpu_to_le64(td->transaction_id);
 720                 details.creation_time = cpu_to_le32(td->creation_time);
 721                 details.snapshotted_time = cpu_to_le32(td->snapshotted_time);
 722                 __dm_bless_for_disk(&details);
 723
 724                 r = dm_btree_insert(&pmd->details_info, pmd->details_root,
 725                                     &key, &details, &pmd->details_root);
 726                 if (r)
 727                         return r;
 728
 729                 if (td->open_count)
 730                         td->changed = 0;
 731                 else {
 732                         list_del(&td->list);
 733                         kfree(td);
 734                 }
 735         }
 736
 737         return 0;
 738 }
 739
 740 static int __commit_transaction(struct dm_pool_metadata *pmd)
 741 {
 742         int r;
 743         size_t metadata_len, data_len;
 744         struct thin_disk_superblock *disk_super;
 745         struct dm_block *sblock;
 746
 747         /*
 748          * We need to know if the thin_disk_superblock exceeds a 512-byte sector.
 749          */
 750         BUILD_BUG_ON(sizeof(struct thin_disk_superblock) > 512);
 751
 752         r = __write_changed_details(pmd);
 753         if (r < 0)
 754                 return r;
 755
 756         r = dm_sm_commit(pmd->data_sm);
 757         if (r < 0)
 758                 return r;
 759
 760         r = dm_tm_pre_commit(pmd->tm);
 761         if (r < 0)
 762                 return r;
 763
 764         r = dm_sm_root_size(pmd->metadata_sm, &metadata_len);
 765         if (r < 0)
 766                 return r;
 767
 768         r = dm_sm_root_size(pmd->data_sm, &data_len);
 769         if (r < 0)
 770                 return r;
 771
 772         r = superblock_lock(pmd, &sblock);
 773         if (r)
 774                 return r;
 775
 776         disk_super = dm_block_data(sblock);
 777         disk_super->time = cpu_to_le32(pmd->time);
 778         disk_super->data_mapping_root = cpu_to_le64(pmd->root);
 779         disk_super->device_details_root = cpu_to_le64(pmd->details_root);
 780         disk_super->trans_id = cpu_to_le64(pmd->trans_id);
 781         disk_super->flags = cpu_to_le32(pmd->flags);
 782
 783         r = dm_sm_copy_root(pmd->metadata_sm, &disk_super->metadata_space_map_root,
 784                             metadata_len);
 785         if (r < 0)
 786                 goto out_locked;
 787
 788         r = dm_sm_copy_root(pmd->data_sm, &disk_super->data_space_map_root,
 789                             data_len);
 790         if (r < 0)
 791                 goto out_locked;
 792
 793         return dm_tm_commit(pmd->tm, sblock);
 794
 795 out_locked:
 796         dm_bm_unlock(sblock);
 797         return r;
 798 }
 799
 800 struct dm_pool_metadata *dm_pool_metadata_open(struct block_device *bdev,
 801                                                sector_t data_block_size,
 802                                                bool format_device)
 803 {
 804         int r;
 805         struct dm_pool_metadata *pmd;
 806
 807         pmd = kmalloc(sizeof(*pmd), GFP_KERNEL);
 808         if (!pmd) {
 809                 DMERR("could not allocate metadata struct");
 810                 return ERR_PTR(-ENOMEM);
 811         }
 812
 813         init_rwsem(&pmd->root_lock);
 814         pmd->time = 0;
 815         INIT_LIST_HEAD(&pmd->thin_devices);
 816         pmd->read_only = false;
 817         pmd->fail_io = false;
 818         pmd->bdev = bdev;
 819         pmd->data_block_size = data_block_size;
 820
 821         r = __create_persistent_data_objects(pmd, format_device);
 822         if (r) {
 823                 kfree(pmd);
 824                 return ERR_PTR(r);
 825         }
 826
 827         r = __begin_transaction(pmd);
 828         if (r < 0) {
 829                 if (dm_pool_metadata_close(pmd) < 0)
 830                         DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
 831                 return ERR_PTR(r);
 832         }
 833
 834         return pmd;
 835 }
 836
 837 int dm_pool_metadata_close(struct dm_pool_metadata *pmd)
 838 {
 839         int r;
 840         unsigned open_devices = 0;
 841         struct dm_thin_device *td, *tmp;
 842
 843         down_read(&pmd->root_lock);
 844         list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
 845                 if (td->open_count)
 846                         open_devices++;
 847                 else {
 848                         list_del(&td->list);
 849                         kfree(td);
 850                 }
 851         }
 852         up_read(&pmd->root_lock);
 853
 854         if (open_devices) {
 855                 DMERR("attempt to close pmd when %u device(s) are still open",
 856                        open_devices);
 857                 return -EBUSY;
 858         }
 859
 860         if (!pmd->read_only && !pmd->fail_io) {
 861                 r = __commit_transaction(pmd);
 862                 if (r < 0)
 863                         DMWARN("%s: __commit_transaction() failed, error = %d",
 864                                __func__, r);
 865         }
 866
 867         if (!pmd->fail_io)
 868                 __destroy_persistent_data_objects(pmd);
 869
 870         kfree(pmd);
 871         return 0;
 872 }
 873
 874 /*
 875  * __open_device: Returns @td corresponding to device with id @dev,
 876  * creating it if @create is set and incrementing @td->open_count.
 877  * On failure, @td is undefined.
 878  */
 879 static int __open_device(struct dm_pool_metadata *pmd,
 880                          dm_thin_id dev, int create,
 881                          struct dm_thin_device **td)
 882 {
 883         int r, changed = 0;
 884         struct dm_thin_device *td2;
 885         uint64_t key = dev;
 886         struct disk_device_details details_le;
 887
 888         /*
 889          * If the device is already open, return it.
 890          */
 891         list_for_each_entry(td2, &pmd->thin_devices, list)
 892                 if (td2->id == dev) {
 893                         /*
 894                          * May not create an already-open device.
 895                          */
 896                         if (create)
 897                                 return -EEXIST;
 898
 899                         td2->open_count++;
 900                         *td = td2;
 901                         return 0;
 902                 }
 903
 904         /*
 905          * Check the device exists.
 906          */
 907         r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
 908                             &key, &details_le);
 909         if (r) {
 910                 if (r != -ENODATA || !create)
 911                         return r;
 912
 913                 /*
 914                  * Create new device.
 915                  */
 916                 changed = 1;
 917                 details_le.mapped_blocks = 0;
 918                 details_le.transaction_id = cpu_to_le64(pmd->trans_id);
 919                 details_le.creation_time = cpu_to_le32(pmd->time);
 920                 details_le.snapshotted_time = cpu_to_le32(pmd->time);
 921         }
 922
 923         *td = kmalloc(sizeof(**td), GFP_NOIO);
 924         if (!*td)
 925                 return -ENOMEM;
 926
 927         (*td)->pmd = pmd;
 928         (*td)->id = dev;
 929         (*td)->open_count = 1;
 930         (*td)->changed = changed;
 931         (*td)->aborted_with_changes = false;
 932         (*td)->mapped_blocks = le64_to_cpu(details_le.mapped_blocks);
 933         (*td)->transaction_id = le64_to_cpu(details_le.transaction_id);
 934         (*td)->creation_time = le32_to_cpu(details_le.creation_time);
 935         (*td)->snapshotted_time = le32_to_cpu(details_le.snapshotted_time);
 936
 937         list_add(&(*td)->list, &pmd->thin_devices);
 938
 939         return 0;
 940 }
 941
 942 static void __close_device(struct dm_thin_device *td)
 943 {
 944         --td->open_count;
 945 }
 946
 947 static int __create_thin(struct dm_pool_metadata *pmd,
 948                          dm_thin_id dev)
 949 {
 950         int r;
 951         dm_block_t dev_root;
 952         uint64_t key = dev;
 953         struct disk_device_details details_le;
 954         struct dm_thin_device *td;
 955         __le64 value;
 956
 957         r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
 958                             &key, &details_le);
 959         if (!r)
 960                 return -EEXIST;
 961
 962         /*
 963          * Create an empty btree for the mappings.
 964          */
 965         r = dm_btree_empty(&pmd->bl_info, &dev_root);
 966         if (r)
 967                 return r;
 968
 969         /*
 970          * Insert it into the main mapping tree.
 971          */
 972         value = cpu_to_le64(dev_root);
 973         __dm_bless_for_disk(&value);
 974         r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
 975         if (r) {
 976                 dm_btree_del(&pmd->bl_info, dev_root);
 977                 return r;
 978         }
 979
 980         r = __open_device(pmd, dev, 1, &td);
 981         if (r) {
 982                 dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
 983                 dm_btree_del(&pmd->bl_info, dev_root);
 984                 return r;
 985         }
 986         __close_device(td);
 987
 988         return r;
 989 }
 990
 991 int dm_pool_create_thin(struct dm_pool_metadata *pmd, dm_thin_id dev)
 992 {
 993         int r = -EINVAL;
 994
 995         down_write(&pmd->root_lock);
 996         if (!pmd->fail_io)
 997                 r = __create_thin(pmd, dev);
 998         up_write(&pmd->root_lock);
 999
1000         return r;
1001 }
1002
1003 static int __set_snapshot_details(struct dm_pool_metadata *pmd,
1004                                   struct dm_thin_device *snap,
1005                                   dm_thin_id origin, uint32_t time)
1006 {
1007         int r;
1008         struct dm_thin_device *td;
1009
1010         r = __open_device(pmd, origin, 0, &td);
1011         if (r)
1012                 return r;
1013
1014         td->changed = 1;
1015         td->snapshotted_time = time;
1016
1017         snap->mapped_blocks = td->mapped_blocks;
1018         snap->snapshotted_time = time;
1019         __close_device(td);
1020
1021         return 0;
1022 }
1023
1024 static int __create_snap(struct dm_pool_metadata *pmd,
1025                          dm_thin_id dev, dm_thin_id origin)
1026 {
1027         int r;
1028         dm_block_t origin_root;
1029         uint64_t key = origin, dev_key = dev;
1030         struct dm_thin_device *td;
1031         struct disk_device_details details_le;
1032         __le64 value;
1033
1034         /* check this device is unused */
1035         r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
1036                             &dev_key, &details_le);
1037         if (!r)
1038                 return -EEXIST;
1039
1040         /* find the mapping tree for the origin */
1041         r = dm_btree_lookup(&pmd->tl_info, pmd->root, &key, &value);
1042         if (r)
1043                 return r;
1044         origin_root = le64_to_cpu(value);
1045
1046         /* clone the origin, an inc will do */
1047         dm_tm_inc(pmd->tm, origin_root);
1048
1049         /* insert into the main mapping tree */
1050         value = cpu_to_le64(origin_root);
1051         __dm_bless_for_disk(&value);
1052         key = dev;
1053         r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
1054         if (r) {
1055                 dm_tm_dec(pmd->tm, origin_root);
1056                 return r;
1057         }
1058
1059         pmd->time++;
1060
1061         r = __open_device(pmd, dev, 1, &td);
1062         if (r)
1063                 goto bad;
1064
1065         r = __set_snapshot_details(pmd, td, origin, pmd->time);
1066         __close_device(td);
1067
1068         if (r)
1069                 goto bad;
1070
1071         return 0;
1072
1073 bad:
1074         dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1075         dm_btree_remove(&pmd->details_info, pmd->details_root,
1076                         &key, &pmd->details_root);
1077         return r;
1078 }
1079
1080 int dm_pool_create_snap(struct dm_pool_metadata *pmd,
1081                                  dm_thin_id dev,
1082                                  dm_thin_id origin)
1083 {
1084         int r = -EINVAL;
1085
1086         down_write(&pmd->root_lock);
1087         if (!pmd->fail_io)
1088                 r = __create_snap(pmd, dev, origin);
1089         up_write(&pmd->root_lock);
1090
1091         return r;
1092 }
1093
1094 static int __delete_device(struct dm_pool_metadata *pmd, dm_thin_id dev)
1095 {
1096         int r;
1097         uint64_t key = dev;
1098         struct dm_thin_device *td;
1099
1100         /* TODO: failure should mark the transaction invalid */
1101         r = __open_device(pmd, dev, 0, &td);
1102         if (r)
1103                 return r;
1104
1105         if (td->open_count > 1) {
1106                 __close_device(td);
1107                 return -EBUSY;
1108         }
1109
1110         list_del(&td->list);
1111         kfree(td);
1112         r = dm_btree_remove(&pmd->details_info, pmd->details_root,
1113                             &key, &pmd->details_root);
1114         if (r)
1115                 return r;
1116
1117         r = dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1118         if (r)
1119                 return r;
1120
1121         return 0;
1122 }
1123
1124 int dm_pool_delete_thin_device(struct dm_pool_metadata *pmd,
1125                                dm_thin_id dev)
1126 {
1127         int r = -EINVAL;
1128
1129         down_write(&pmd->root_lock);
1130         if (!pmd->fail_io)
1131                 r = __delete_device(pmd, dev);
1132         up_write(&pmd->root_lock);
1133
1134         return r;
1135 }
1136
1137 int dm_pool_set_metadata_transaction_id(struct dm_pool_metadata *pmd,
1138                                         uint64_t current_id,
1139                                         uint64_t new_id)
1140 {
1141         int r = -EINVAL;
1142
1143         down_write(&pmd->root_lock);
1144
1145         if (pmd->fail_io)
1146                 goto out;
1147
1148         if (pmd->trans_id != current_id) {
1149                 DMERR("mismatched transaction id");
1150                 goto out;
1151         }
1152
1153         pmd->trans_id = new_id;
1154         r = 0;
1155
1156 out:
1157         up_write(&pmd->root_lock);
1158
1159         return r;
1160 }
1161
1162 int dm_pool_get_metadata_transaction_id(struct dm_pool_metadata *pmd,
1163                                         uint64_t *result)
1164 {
1165         int r = -EINVAL;
1166
1167         down_read(&pmd->root_lock);
1168         if (!pmd->fail_io) {
1169                 *result = pmd->trans_id;
1170                 r = 0;
1171         }
1172         up_read(&pmd->root_lock);
1173
1174         return r;
1175 }
1176
1177 static int __reserve_metadata_snap(struct dm_pool_metadata *pmd)
1178 {
1179         int r, inc;
1180         struct thin_disk_superblock *disk_super;
1181         struct dm_block *copy, *sblock;
1182         dm_block_t held_root;
1183
1184         /*
1185          * Copy the superblock.
1186          */
1187         dm_sm_inc_block(pmd->metadata_sm, THIN_SUPERBLOCK_LOCATION);
1188         r = dm_tm_shadow_block(pmd->tm, THIN_SUPERBLOCK_LOCATION,
1189                                &sb_validator, &copy, &inc);
1190         if (r)
1191                 return r;
1192
1193         BUG_ON(!inc);
1194
1195         held_root = dm_block_location(copy);
1196         disk_super = dm_block_data(copy);
1197
1198         if (le64_to_cpu(disk_super->held_root)) {
1199                 DMWARN("Pool metadata snapshot already exists: release this before taking another.");
1200
1201                 dm_tm_dec(pmd->tm, held_root);
1202                 dm_tm_unlock(pmd->tm, copy);
1203                 return -EBUSY;
1204         }
1205
1206         /*
1207          * Wipe the spacemap since we're not publishing this.
1208          */
1209         memset(&disk_super->data_space_map_root, 0,
1210                sizeof(disk_super->data_space_map_root));
1211         memset(&disk_super->metadata_space_map_root, 0,
1212                sizeof(disk_super->metadata_space_map_root));
1213
1214         /*
1215          * Increment the data structures that need to be preserved.
1216          */
1217         dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->data_mapping_root));
1218         dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->device_details_root));
1219         dm_tm_unlock(pmd->tm, copy);
1220
1221         /*
1222          * Write the held root into the superblock.
1223          */
1224         r = superblock_lock(pmd, &sblock);
1225         if (r) {
1226                 dm_tm_dec(pmd->tm, held_root);
1227                 return r;
1228         }
1229
1230         disk_super = dm_block_data(sblock);
1231         disk_super->held_root = cpu_to_le64(held_root);
1232         dm_bm_unlock(sblock);
1233         return 0;
1234 }
1235
1236 int dm_pool_reserve_metadata_snap(struct dm_pool_metadata *pmd)
1237 {
1238         int r = -EINVAL;
1239
1240         down_write(&pmd->root_lock);
1241         if (!pmd->fail_io)
1242                 r = __reserve_metadata_snap(pmd);
1243         up_write(&pmd->root_lock);
1244
1245         return r;
1246 }
1247
1248 static int __release_metadata_snap(struct dm_pool_metadata *pmd)
1249 {
1250         int r;
1251         struct thin_disk_superblock *disk_super;
1252         struct dm_block *sblock, *copy;
1253         dm_block_t held_root;
1254
1255         r = superblock_lock(pmd, &sblock);
1256         if (r)
1257                 return r;
1258
1259         disk_super = dm_block_data(sblock);
1260         held_root = le64_to_cpu(disk_super->held_root);
1261         disk_super->held_root = cpu_to_le64(0);
1262
1263         dm_bm_unlock(sblock);
1264
1265         if (!held_root) {
1266                 DMWARN("No pool metadata snapshot found: nothing to release.");
1267                 return -EINVAL;
1268         }
1269
1270         r = dm_tm_read_lock(pmd->tm, held_root, &sb_validator, &copy);
1271         if (r)
1272                 return r;
1273
1274         disk_super = dm_block_data(copy);
1275         dm_sm_dec_block(pmd->metadata_sm, le64_to_cpu(disk_super->data_mapping_root));
1276         dm_sm_dec_block(pmd->metadata_sm, le64_to_cpu(disk_super->device_details_root));
1277         dm_sm_dec_block(pmd->metadata_sm, held_root);
1278
1279         return dm_tm_unlock(pmd->tm, copy);
1280 }
1281
1282 int dm_pool_release_metadata_snap(struct dm_pool_metadata *pmd)
1283 {
1284         int r = -EINVAL;
1285
1286         down_write(&pmd->root_lock);
1287         if (!pmd->fail_io)
1288                 r = __release_metadata_snap(pmd);
1289         up_write(&pmd->root_lock);
1290
1291         return r;
1292 }
1293
1294 static int __get_metadata_snap(struct dm_pool_metadata *pmd,
1295                                dm_block_t *result)
1296 {
1297         int r;
1298         struct thin_disk_superblock *disk_super;
1299         struct dm_block *sblock;
1300
1301         r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
1302                             &sb_validator, &sblock);
1303         if (r)
1304                 return r;
1305
1306         disk_super = dm_block_data(sblock);
1307         *result = le64_to_cpu(disk_super->held_root);
1308
1309         return dm_bm_unlock(sblock);
1310 }
1311
1312 int dm_pool_get_metadata_snap(struct dm_pool_metadata *pmd,
1313                               dm_block_t *result)
1314 {
1315         int r = -EINVAL;
1316
1317         down_read(&pmd->root_lock);
1318         if (!pmd->fail_io)
1319                 r = __get_metadata_snap(pmd, result);
1320         up_read(&pmd->root_lock);
1321
1322         return r;
1323 }
1324
1325 int dm_pool_open_thin_device(struct dm_pool_metadata *pmd, dm_thin_id dev,
1326                              struct dm_thin_device **td)
1327 {
1328         int r = -EINVAL;
1329
1330         down_write(&pmd->root_lock);
1331         if (!pmd->fail_io)
1332                 r = __open_device(pmd, dev, 0, td);
1333         up_write(&pmd->root_lock);
1334
1335         return r;
1336 }
1337
1338 int dm_pool_close_thin_device(struct dm_thin_device *td)
1339 {
1340         down_write(&td->pmd->root_lock);
1341         __close_device(td);
1342         up_write(&td->pmd->root_lock);
1343
1344         return 0;
1345 }
1346
1347 dm_thin_id dm_thin_dev_id(struct dm_thin_device *td)
1348 {
1349         return td->id;
1350 }
1351
1352 /*
1353  * Check whether @time (of block creation) is older than @td's last snapshot.
1354  * If so then the associated block is shared with the last snapshot device.
1355  * Any block on a device created *after* the device last got snapshotted is
1356  * necessarily not shared.
1357  */
1358 static bool __snapshotted_since(struct dm_thin_device *td, uint32_t time)
1359 {
1360         return td->snapshotted_time > time;
1361 }
1362
1363 int dm_thin_find_block(struct dm_thin_device *td, dm_block_t block,
1364                        int can_block, struct dm_thin_lookup_result *result)
1365 {
1366         int r = -EINVAL;
1367         uint64_t block_time = 0;
1368         __le64 value;
1369         struct dm_pool_metadata *pmd = td->pmd;
1370         dm_block_t keys[2] = { td->id, block };
1371         struct dm_btree_info *info;
1372
1373         if (can_block) {
1374                 down_read(&pmd->root_lock);
1375                 info = &pmd->info;
1376         } else if (down_read_trylock(&pmd->root_lock))
1377                 info = &pmd->nb_info;
1378         else
1379                 return -EWOULDBLOCK;
1380
1381         if (pmd->fail_io)
1382                 goto out;
1383
1384         r = dm_btree_lookup(info, pmd->root, keys, &value);
1385         if (!r)
1386                 block_time = le64_to_cpu(value);
1387
1388 out:
1389         up_read(&pmd->root_lock);
1390
1391         if (!r) {
1392                 dm_block_t exception_block;
1393                 uint32_t exception_time;
1394                 unpack_block_time(block_time, &exception_block,
1395                                   &exception_time);
1396                 result->block = exception_block;
1397                 result->shared = __snapshotted_since(td, exception_time);
1398         }
1399
1400         return r;
1401 }
1402
1403 static int __insert(struct dm_thin_device *td, dm_block_t block,
1404                     dm_block_t data_block)
1405 {
1406         int r, inserted;
1407         __le64 value;
1408         struct dm_pool_metadata *pmd = td->pmd;
1409         dm_block_t keys[2] = { td->id, block };
1410
1411         value = cpu_to_le64(pack_block_time(data_block, pmd->time));
1412         __dm_bless_for_disk(&value);
1413
1414         r = dm_btree_insert_notify(&pmd->info, pmd->root, keys, &value,
1415                                    &pmd->root, &inserted);
1416         if (r)
1417                 return r;
1418
1419         td->changed = 1;
1420         if (inserted)
1421                 td->mapped_blocks++;
1422
1423         return 0;
1424 }
1425
1426 int dm_thin_insert_block(struct dm_thin_device *td, dm_block_t block,
1427                          dm_block_t data_block)
1428 {
1429         int r = -EINVAL;
1430
1431         down_write(&td->pmd->root_lock);
1432         if (!td->pmd->fail_io)
1433                 r = __insert(td, block, data_block);
1434         up_write(&td->pmd->root_lock);
1435
1436         return r;
1437 }
1438
1439 static int __remove(struct dm_thin_device *td, dm_block_t block)
1440 {
1441         int r;
1442         struct dm_pool_metadata *pmd = td->pmd;
1443         dm_block_t keys[2] = { td->id, block };
1444
1445         r = dm_btree_remove(&pmd->info, pmd->root, keys, &pmd->root);
1446         if (r)
1447                 return r;
1448
1449         td->mapped_blocks--;
1450         td->changed = 1;
1451
1452         return 0;
1453 }
1454
1455 int dm_thin_remove_block(struct dm_thin_device *td, dm_block_t block)
1456 {
1457         int r = -EINVAL;
1458
1459         down_write(&td->pmd->root_lock);
1460         if (!td->pmd->fail_io)
1461                 r = __remove(td, block);
1462         up_write(&td->pmd->root_lock);
1463
1464         return r;
1465 }
1466
1467 int dm_pool_block_is_used(struct dm_pool_metadata *pmd, dm_block_t b, bool *result)
1468 {
1469         int r;
1470         uint32_t ref_count;
1471
1472         down_read(&pmd->root_lock);
1473         r = dm_sm_get_count(pmd->data_sm, b, &ref_count);
1474         if (!r)
1475                 *result = (ref_count != 0);
1476         up_read(&pmd->root_lock);
1477
1478         return r;
1479 }
1480
1481 bool dm_thin_changed_this_transaction(struct dm_thin_device *td)
1482 {
1483         int r;
1484
1485         down_read(&td->pmd->root_lock);
1486         r = td->changed;
1487         up_read(&td->pmd->root_lock);
1488
1489         return r;
1490 }
1491
1492 bool dm_pool_changed_this_transaction(struct dm_pool_metadata *pmd)
1493 {
1494         bool r = false;
1495         struct dm_thin_device *td, *tmp;
1496
1497         down_read(&pmd->root_lock);
1498         list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
1499                 if (td->changed) {
1500                         r = td->changed;
1501                         break;
1502                 }
1503         }
1504         up_read(&pmd->root_lock);
1505
1506         return r;
1507 }
1508
1509 bool dm_thin_aborted_changes(struct dm_thin_device *td)
1510 {
1511         bool r;
1512
1513         down_read(&td->pmd->root_lock);
1514         r = td->aborted_with_changes;
1515         up_read(&td->pmd->root_lock);
1516
1517         return r;
1518 }
1519
1520 int dm_pool_alloc_data_block(struct dm_pool_metadata *pmd, dm_block_t *result)
1521 {
1522         int r = -EINVAL;
1523
1524         down_write(&pmd->root_lock);
1525         if (!pmd->fail_io)
1526                 r = dm_sm_new_block(pmd->data_sm, result);
1527         up_write(&pmd->root_lock);
1528
1529         return r;
1530 }
1531
1532 int dm_pool_commit_metadata(struct dm_pool_metadata *pmd)
1533 {
1534         int r = -EINVAL;
1535
1536         down_write(&pmd->root_lock);
1537         if (pmd->fail_io)
1538                 goto out;
1539
1540         r = __commit_transaction(pmd);
1541         if (r <= 0)
1542                 goto out;
1543
1544         /*
1545          * Open the next transaction.
1546          */
1547         r = __begin_transaction(pmd);
1548 out:
1549         up_write(&pmd->root_lock);
1550         return r;
1551 }
1552
1553 static void __set_abort_with_changes_flags(struct dm_pool_metadata *pmd)
1554 {
1555         struct dm_thin_device *td;
1556
1557         list_for_each_entry(td, &pmd->thin_devices, list)
1558                 td->aborted_with_changes = td->changed;
1559 }
1560
1561 int dm_pool_abort_metadata(struct dm_pool_metadata *pmd)
1562 {
1563         int r = -EINVAL;
1564
1565         down_write(&pmd->root_lock);
1566         if (pmd->fail_io)
1567                 goto out;
1568
1569         __set_abort_with_changes_flags(pmd);
1570         __destroy_persistent_data_objects(pmd);
1571         r = __create_persistent_data_objects(pmd, false);
1572         if (r)
1573                 pmd->fail_io = true;
1574
1575 out:
1576         up_write(&pmd->root_lock);
1577
1578         return r;
1579 }
1580
1581 int dm_pool_get_free_block_count(struct dm_pool_metadata *pmd, dm_block_t *result)
1582 {
1583         int r = -EINVAL;
1584
1585         down_read(&pmd->root_lock);
1586         if (!pmd->fail_io)
1587                 r = dm_sm_get_nr_free(pmd->data_sm, result);
1588         up_read(&pmd->root_lock);
1589
1590         return r;
1591 }
1592
1593 int dm_pool_get_free_metadata_block_count(struct dm_pool_metadata *pmd,
1594                                           dm_block_t *result)
1595 {
1596         int r = -EINVAL;
1597
1598         down_read(&pmd->root_lock);
1599         if (!pmd->fail_io)
1600                 r = dm_sm_get_nr_free(pmd->metadata_sm, result);
1601         up_read(&pmd->root_lock);
1602
1603         return r;
1604 }
1605
1606 int dm_pool_get_metadata_dev_size(struct dm_pool_metadata *pmd,
1607                                   dm_block_t *result)
1608 {
1609         int r = -EINVAL;
1610
1611         down_read(&pmd->root_lock);
1612         if (!pmd->fail_io)
1613                 r = dm_sm_get_nr_blocks(pmd->metadata_sm, result);
1614         up_read(&pmd->root_lock);
1615
1616         return r;
1617 }
1618
1619 int dm_pool_get_data_block_size(struct dm_pool_metadata *pmd, sector_t *result)
1620 {
1621         down_read(&pmd->root_lock);
1622         *result = pmd->data_block_size;
1623         up_read(&pmd->root_lock);
1624
1625         return 0;
1626 }
1627
1628 int dm_pool_get_data_dev_size(struct dm_pool_metadata *pmd, dm_block_t *result)
1629 {
1630         int r = -EINVAL;
1631
1632         down_read(&pmd->root_lock);
1633         if (!pmd->fail_io)
1634                 r = dm_sm_get_nr_blocks(pmd->data_sm, result);
1635         up_read(&pmd->root_lock);
1636
1637         return r;
1638 }
1639
1640 int dm_thin_get_mapped_count(struct dm_thin_device *td, dm_block_t *result)
1641 {
1642         int r = -EINVAL;
1643         struct dm_pool_metadata *pmd = td->pmd;
1644
1645         down_read(&pmd->root_lock);
1646         if (!pmd->fail_io) {
1647                 *result = td->mapped_blocks;
1648                 r = 0;
1649         }
1650         up_read(&pmd->root_lock);
1651
1652         return r;
1653 }
1654
1655 static int __highest_block(struct dm_thin_device *td, dm_block_t *result)
1656 {
1657         int r;
1658         __le64 value_le;
1659         dm_block_t thin_root;
1660         struct dm_pool_metadata *pmd = td->pmd;
1661
1662         r = dm_btree_lookup(&pmd->tl_info, pmd->root, &td->id, &value_le);
1663         if (r)
1664                 return r;
1665
1666         thin_root = le64_to_cpu(value_le);
1667
1668         return dm_btree_find_highest_key(&pmd->bl_info, thin_root, result);
1669 }
1670
1671 int dm_thin_get_highest_mapped_block(struct dm_thin_device *td,
1672                                      dm_block_t *result)
1673 {
1674         int r = -EINVAL;
1675         struct dm_pool_metadata *pmd = td->pmd;
1676
1677         down_read(&pmd->root_lock);
1678         if (!pmd->fail_io)
1679                 r = __highest_block(td, result);
1680         up_read(&pmd->root_lock);
1681
1682         return r;
1683 }
1684
1685 static int __resize_space_map(struct dm_space_map *sm, dm_block_t new_count)
1686 {
1687         int r;
1688         dm_block_t old_count;
1689
1690         r = dm_sm_get_nr_blocks(sm, &old_count);
1691         if (r)
1692                 return r;
1693
1694         if (new_count == old_count)
1695                 return 0;
1696
1697         if (new_count < old_count) {
1698                 DMERR("cannot reduce size of space map");
1699                 return -EINVAL;
1700         }
1701
1702         return dm_sm_extend(sm, new_count - old_count);
1703 }
1704
1705 int dm_pool_resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1706 {
1707         int r = -EINVAL;
1708
1709         down_write(&pmd->root_lock);
1710         if (!pmd->fail_io)
1711                 r = __resize_space_map(pmd->data_sm, new_count);
1712         up_write(&pmd->root_lock);
1713
1714         return r;
1715 }
1716
1717 int dm_pool_resize_metadata_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1718 {
1719         int r = -EINVAL;
1720
1721         down_write(&pmd->root_lock);
1722         if (!pmd->fail_io)
1723                 r = __resize_space_map(pmd->metadata_sm, new_count);
1724         up_write(&pmd->root_lock);
1725
1726         return r;
1727 }
1728
1729 void dm_pool_metadata_read_only(struct dm_pool_metadata *pmd)
1730 {
1731         down_write(&pmd->root_lock);
1732         pmd->read_only = true;
1733         dm_bm_set_read_only(pmd->bm);
1734         up_write(&pmd->root_lock);
1735 }
1736
1737 void dm_pool_metadata_read_write(struct dm_pool_metadata *pmd)
1738 {
1739         down_write(&pmd->root_lock);
1740         pmd->read_only = false;
1741         dm_bm_set_read_write(pmd->bm);
1742         up_write(&pmd->root_lock);
1743 }
1744
1745 int dm_pool_register_metadata_threshold(struct dm_pool_metadata *pmd,
1746                                         dm_block_t threshold,
1747                                         dm_sm_threshold_fn fn,
1748                                         void *context)
1749 {
1750         int r;
1751
1752         down_write(&pmd->root_lock);
1753         r = dm_sm_register_threshold_callback(pmd->metadata_sm, threshold, fn, context);
1754         up_write(&pmd->root_lock);
1755
1756         return r;
1757 }
1758
1759 int dm_pool_metadata_set_needs_check(struct dm_pool_metadata *pmd)
1760 {
1761         int r;
1762         struct dm_block *sblock;
1763         struct thin_disk_superblock *disk_super;
1764
1765         down_write(&pmd->root_lock);
1766         pmd->flags |= THIN_METADATA_NEEDS_CHECK_FLAG;
1767
1768         r = superblock_lock(pmd, &sblock);
1769         if (r) {
1770                 DMERR("couldn't read superblock");
1771                 goto out;
1772         }
1773
1774         disk_super = dm_block_data(sblock);
1775         disk_super->flags = cpu_to_le32(pmd->flags);
1776
1777         dm_bm_unlock(sblock);
1778 out:
1779         up_write(&pmd->root_lock);
1780         return r;
1781 }
1782
1783 bool dm_pool_metadata_needs_check(struct dm_pool_metadata *pmd)
1784 {
1785         bool needs_check;
1786
1787         down_read(&pmd->root_lock);
1788         needs_check = pmd->flags & THIN_METADATA_NEEDS_CHECK_FLAG;
1789         up_read(&pmd->root_lock);
1790
1791         return needs_check;
1792 }