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 1
  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 >> SECTOR_SHIFT);
 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         /* Verify the data block size hasn't changed */
 595         if (le32_to_cpu(disk_super->data_block_size) != pmd->data_block_size) {
 596                 DMERR("changing the data block size (from %u to %llu) is not supported",
 597                       le32_to_cpu(disk_super->data_block_size),
 598                       (unsigned long long)pmd->data_block_size);
 599                 r = -EINVAL;
 600                 goto bad_unlock_sblock;
 601         }
 602
 603         r = __check_incompat_features(disk_super, pmd);
 604         if (r < 0)
 605                 goto bad_unlock_sblock;
 606
 607         r = dm_tm_open_with_sm(pmd->bm, THIN_SUPERBLOCK_LOCATION,
 608                                disk_super->metadata_space_map_root,
 609                                sizeof(disk_super->metadata_space_map_root),
 610                                &pmd->tm, &pmd->metadata_sm);
 611         if (r < 0) {
 612                 DMERR("tm_open_with_sm failed");
 613                 goto bad_unlock_sblock;
 614         }
 615
 616         pmd->data_sm = dm_sm_disk_open(pmd->tm, disk_super->data_space_map_root,
 617                                        sizeof(disk_super->data_space_map_root));
 618         if (IS_ERR(pmd->data_sm)) {
 619                 DMERR("sm_disk_open failed");
 620                 r = PTR_ERR(pmd->data_sm);
 621                 goto bad_cleanup_tm;
 622         }
 623
 624         pmd->nb_tm = dm_tm_create_non_blocking_clone(pmd->tm);
 625         if (!pmd->nb_tm) {
 626                 DMERR("could not create non-blocking clone tm");
 627                 r = -ENOMEM;
 628                 goto bad_cleanup_data_sm;
 629         }
 630
 631         __setup_btree_details(pmd);
 632         return dm_bm_unlock(sblock);
 633
 634 bad_cleanup_data_sm:
 635         dm_sm_destroy(pmd->data_sm);
 636 bad_cleanup_tm:
 637         dm_tm_destroy(pmd->tm);
 638         dm_sm_destroy(pmd->metadata_sm);
 639 bad_unlock_sblock:
 640         dm_bm_unlock(sblock);
 641
 642         return r;
 643 }
 644
 645 static int __open_or_format_metadata(struct dm_pool_metadata *pmd, bool format_device)
 646 {
 647         int r, unformatted;
 648
 649         r = __superblock_all_zeroes(pmd->bm, &unformatted);
 650         if (r)
 651                 return r;
 652
 653         if (unformatted)
 654                 return format_device ? __format_metadata(pmd) : -EPERM;
 655
 656         return __open_metadata(pmd);
 657 }
 658
 659 static int __create_persistent_data_objects(struct dm_pool_metadata *pmd, bool format_device)
 660 {
 661         int r;
 662
 663         pmd->bm = dm_block_manager_create(pmd->bdev, THIN_METADATA_BLOCK_SIZE,
 664                                           THIN_METADATA_CACHE_SIZE,
 665                                           THIN_MAX_CONCURRENT_LOCKS);
 666         if (IS_ERR(pmd->bm)) {
 667                 DMERR("could not create block manager");
 668                 return PTR_ERR(pmd->bm);
 669         }
 670
 671         r = __open_or_format_metadata(pmd, format_device);
 672         if (r)
 673                 dm_block_manager_destroy(pmd->bm);
 674
 675         return r;
 676 }
 677
 678 static void __destroy_persistent_data_objects(struct dm_pool_metadata *pmd)
 679 {
 680         dm_sm_destroy(pmd->data_sm);
 681         dm_sm_destroy(pmd->metadata_sm);
 682         dm_tm_destroy(pmd->nb_tm);
 683         dm_tm_destroy(pmd->tm);
 684         dm_block_manager_destroy(pmd->bm);
 685 }
 686
 687 static int __begin_transaction(struct dm_pool_metadata *pmd)
 688 {
 689         int r;
 690         struct thin_disk_superblock *disk_super;
 691         struct dm_block *sblock;
 692
 693         /*
 694          * We re-read the superblock every time.  Shouldn't need to do this
 695          * really.
 696          */
 697         r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
 698                             &sb_validator, &sblock);
 699         if (r)
 700                 return r;
 701
 702         disk_super = dm_block_data(sblock);
 703         pmd->time = le32_to_cpu(disk_super->time);
 704         pmd->root = le64_to_cpu(disk_super->data_mapping_root);
 705         pmd->details_root = le64_to_cpu(disk_super->device_details_root);
 706         pmd->trans_id = le64_to_cpu(disk_super->trans_id);
 707         pmd->flags = le32_to_cpu(disk_super->flags);
 708         pmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
 709
 710         dm_bm_unlock(sblock);
 711         return 0;
 712 }
 713
 714 static int __write_changed_details(struct dm_pool_metadata *pmd)
 715 {
 716         int r;
 717         struct dm_thin_device *td, *tmp;
 718         struct disk_device_details details;
 719         uint64_t key;
 720
 721         list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
 722                 if (!td->changed)
 723                         continue;
 724
 725                 key = td->id;
 726
 727                 details.mapped_blocks = cpu_to_le64(td->mapped_blocks);
 728                 details.transaction_id = cpu_to_le64(td->transaction_id);
 729                 details.creation_time = cpu_to_le32(td->creation_time);
 730                 details.snapshotted_time = cpu_to_le32(td->snapshotted_time);
 731                 __dm_bless_for_disk(&details);
 732
 733                 r = dm_btree_insert(&pmd->details_info, pmd->details_root,
 734                                     &key, &details, &pmd->details_root);
 735                 if (r)
 736                         return r;
 737
 738                 if (td->open_count)
 739                         td->changed = 0;
 740                 else {
 741                         list_del(&td->list);
 742                         kfree(td);
 743                 }
 744         }
 745
 746         return 0;
 747 }
 748
 749 static int __commit_transaction(struct dm_pool_metadata *pmd)
 750 {
 751         int r;
 752         size_t metadata_len, data_len;
 753         struct thin_disk_superblock *disk_super;
 754         struct dm_block *sblock;
 755
 756         /*
 757          * We need to know if the thin_disk_superblock exceeds a 512-byte sector.
 758          */
 759         BUILD_BUG_ON(sizeof(struct thin_disk_superblock) > 512);
 760
 761         r = __write_changed_details(pmd);
 762         if (r < 0)
 763                 return r;
 764
 765         r = dm_sm_commit(pmd->data_sm);
 766         if (r < 0)
 767                 return r;
 768
 769         r = dm_tm_pre_commit(pmd->tm);
 770         if (r < 0)
 771                 return r;
 772
 773         r = dm_sm_root_size(pmd->metadata_sm, &metadata_len);
 774         if (r < 0)
 775                 return r;
 776
 777         r = dm_sm_root_size(pmd->data_sm, &data_len);
 778         if (r < 0)
 779                 return r;
 780
 781         r = superblock_lock(pmd, &sblock);
 782         if (r)
 783                 return r;
 784
 785         disk_super = dm_block_data(sblock);
 786         disk_super->time = cpu_to_le32(pmd->time);
 787         disk_super->data_mapping_root = cpu_to_le64(pmd->root);
 788         disk_super->device_details_root = cpu_to_le64(pmd->details_root);
 789         disk_super->trans_id = cpu_to_le64(pmd->trans_id);
 790         disk_super->flags = cpu_to_le32(pmd->flags);
 791
 792         r = dm_sm_copy_root(pmd->metadata_sm, &disk_super->metadata_space_map_root,
 793                             metadata_len);
 794         if (r < 0)
 795                 goto out_locked;
 796
 797         r = dm_sm_copy_root(pmd->data_sm, &disk_super->data_space_map_root,
 798                             data_len);
 799         if (r < 0)
 800                 goto out_locked;
 801
 802         return dm_tm_commit(pmd->tm, sblock);
 803
 804 out_locked:
 805         dm_bm_unlock(sblock);
 806         return r;
 807 }
 808
 809 struct dm_pool_metadata *dm_pool_metadata_open(struct block_device *bdev,
 810                                                sector_t data_block_size,
 811                                                bool format_device)
 812 {
 813         int r;
 814         struct dm_pool_metadata *pmd;
 815
 816         pmd = kmalloc(sizeof(*pmd), GFP_KERNEL);
 817         if (!pmd) {
 818                 DMERR("could not allocate metadata struct");
 819                 return ERR_PTR(-ENOMEM);
 820         }
 821
 822         init_rwsem(&pmd->root_lock);
 823         pmd->time = 0;
 824         INIT_LIST_HEAD(&pmd->thin_devices);
 825         pmd->read_only = false;
 826         pmd->fail_io = false;
 827         pmd->bdev = bdev;
 828         pmd->data_block_size = data_block_size;
 829
 830         r = __create_persistent_data_objects(pmd, format_device);
 831         if (r) {
 832                 kfree(pmd);
 833                 return ERR_PTR(r);
 834         }
 835
 836         r = __begin_transaction(pmd);
 837         if (r < 0) {
 838                 if (dm_pool_metadata_close(pmd) < 0)
 839                         DMWARN("%s: dm_pool_metadata_close() failed.", __func__);
 840                 return ERR_PTR(r);
 841         }
 842
 843         return pmd;
 844 }
 845
 846 int dm_pool_metadata_close(struct dm_pool_metadata *pmd)
 847 {
 848         int r;
 849         unsigned open_devices = 0;
 850         struct dm_thin_device *td, *tmp;
 851
 852         down_read(&pmd->root_lock);
 853         list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
 854                 if (td->open_count)
 855                         open_devices++;
 856                 else {
 857                         list_del(&td->list);
 858                         kfree(td);
 859                 }
 860         }
 861         up_read(&pmd->root_lock);
 862
 863         if (open_devices) {
 864                 DMERR("attempt to close pmd when %u device(s) are still open",
 865                        open_devices);
 866                 return -EBUSY;
 867         }
 868
 869         if (!pmd->read_only && !pmd->fail_io) {
 870                 r = __commit_transaction(pmd);
 871                 if (r < 0)
 872                         DMWARN("%s: __commit_transaction() failed, error = %d",
 873                                __func__, r);
 874         }
 875
 876         if (!pmd->fail_io)
 877                 __destroy_persistent_data_objects(pmd);
 878
 879         kfree(pmd);
 880         return 0;
 881 }
 882
 883 /*
 884  * __open_device: Returns @td corresponding to device with id @dev,
 885  * creating it if @create is set and incrementing @td->open_count.
 886  * On failure, @td is undefined.
 887  */
 888 static int __open_device(struct dm_pool_metadata *pmd,
 889                          dm_thin_id dev, int create,
 890                          struct dm_thin_device **td)
 891 {
 892         int r, changed = 0;
 893         struct dm_thin_device *td2;
 894         uint64_t key = dev;
 895         struct disk_device_details details_le;
 896
 897         /*
 898          * If the device is already open, return it.
 899          */
 900         list_for_each_entry(td2, &pmd->thin_devices, list)
 901                 if (td2->id == dev) {
 902                         /*
 903                          * May not create an already-open device.
 904                          */
 905                         if (create)
 906                                 return -EEXIST;
 907
 908                         td2->open_count++;
 909                         *td = td2;
 910                         return 0;
 911                 }
 912
 913         /*
 914          * Check the device exists.
 915          */
 916         r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
 917                             &key, &details_le);
 918         if (r) {
 919                 if (r != -ENODATA || !create)
 920                         return r;
 921
 922                 /*
 923                  * Create new device.
 924                  */
 925                 changed = 1;
 926                 details_le.mapped_blocks = 0;
 927                 details_le.transaction_id = cpu_to_le64(pmd->trans_id);
 928                 details_le.creation_time = cpu_to_le32(pmd->time);
 929                 details_le.snapshotted_time = cpu_to_le32(pmd->time);
 930         }
 931
 932         *td = kmalloc(sizeof(**td), GFP_NOIO);
 933         if (!*td)
 934                 return -ENOMEM;
 935
 936         (*td)->pmd = pmd;
 937         (*td)->id = dev;
 938         (*td)->open_count = 1;
 939         (*td)->changed = changed;
 940         (*td)->aborted_with_changes = false;
 941         (*td)->mapped_blocks = le64_to_cpu(details_le.mapped_blocks);
 942         (*td)->transaction_id = le64_to_cpu(details_le.transaction_id);
 943         (*td)->creation_time = le32_to_cpu(details_le.creation_time);
 944         (*td)->snapshotted_time = le32_to_cpu(details_le.snapshotted_time);
 945
 946         list_add(&(*td)->list, &pmd->thin_devices);
 947
 948         return 0;
 949 }
 950
 951 static void __close_device(struct dm_thin_device *td)
 952 {
 953         --td->open_count;
 954 }
 955
 956 static int __create_thin(struct dm_pool_metadata *pmd,
 957                          dm_thin_id dev)
 958 {
 959         int r;
 960         dm_block_t dev_root;
 961         uint64_t key = dev;
 962         struct disk_device_details details_le;
 963         struct dm_thin_device *td;
 964         __le64 value;
 965
 966         r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
 967                             &key, &details_le);
 968         if (!r)
 969                 return -EEXIST;
 970
 971         /*
 972          * Create an empty btree for the mappings.
 973          */
 974         r = dm_btree_empty(&pmd->bl_info, &dev_root);
 975         if (r)
 976                 return r;
 977
 978         /*
 979          * Insert it into the main mapping tree.
 980          */
 981         value = cpu_to_le64(dev_root);
 982         __dm_bless_for_disk(&value);
 983         r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
 984         if (r) {
 985                 dm_btree_del(&pmd->bl_info, dev_root);
 986                 return r;
 987         }
 988
 989         r = __open_device(pmd, dev, 1, &td);
 990         if (r) {
 991                 dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
 992                 dm_btree_del(&pmd->bl_info, dev_root);
 993                 return r;
 994         }
 995         __close_device(td);
 996
 997         return r;
 998 }
 999
1000 int dm_pool_create_thin(struct dm_pool_metadata *pmd, dm_thin_id dev)
1001 {
1002         int r = -EINVAL;
1003
1004         down_write(&pmd->root_lock);
1005         if (!pmd->fail_io)
1006                 r = __create_thin(pmd, dev);
1007         up_write(&pmd->root_lock);
1008
1009         return r;
1010 }
1011
1012 static int __set_snapshot_details(struct dm_pool_metadata *pmd,
1013                                   struct dm_thin_device *snap,
1014                                   dm_thin_id origin, uint32_t time)
1015 {
1016         int r;
1017         struct dm_thin_device *td;
1018
1019         r = __open_device(pmd, origin, 0, &td);
1020         if (r)
1021                 return r;
1022
1023         td->changed = 1;
1024         td->snapshotted_time = time;
1025
1026         snap->mapped_blocks = td->mapped_blocks;
1027         snap->snapshotted_time = time;
1028         __close_device(td);
1029
1030         return 0;
1031 }
1032
1033 static int __create_snap(struct dm_pool_metadata *pmd,
1034                          dm_thin_id dev, dm_thin_id origin)
1035 {
1036         int r;
1037         dm_block_t origin_root;
1038         uint64_t key = origin, dev_key = dev;
1039         struct dm_thin_device *td;
1040         struct disk_device_details details_le;
1041         __le64 value;
1042
1043         /* check this device is unused */
1044         r = dm_btree_lookup(&pmd->details_info, pmd->details_root,
1045                             &dev_key, &details_le);
1046         if (!r)
1047                 return -EEXIST;
1048
1049         /* find the mapping tree for the origin */
1050         r = dm_btree_lookup(&pmd->tl_info, pmd->root, &key, &value);
1051         if (r)
1052                 return r;
1053         origin_root = le64_to_cpu(value);
1054
1055         /* clone the origin, an inc will do */
1056         dm_tm_inc(pmd->tm, origin_root);
1057
1058         /* insert into the main mapping tree */
1059         value = cpu_to_le64(origin_root);
1060         __dm_bless_for_disk(&value);
1061         key = dev;
1062         r = dm_btree_insert(&pmd->tl_info, pmd->root, &key, &value, &pmd->root);
1063         if (r) {
1064                 dm_tm_dec(pmd->tm, origin_root);
1065                 return r;
1066         }
1067
1068         pmd->time++;
1069
1070         r = __open_device(pmd, dev, 1, &td);
1071         if (r)
1072                 goto bad;
1073
1074         r = __set_snapshot_details(pmd, td, origin, pmd->time);
1075         __close_device(td);
1076
1077         if (r)
1078                 goto bad;
1079
1080         return 0;
1081
1082 bad:
1083         dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1084         dm_btree_remove(&pmd->details_info, pmd->details_root,
1085                         &key, &pmd->details_root);
1086         return r;
1087 }
1088
1089 int dm_pool_create_snap(struct dm_pool_metadata *pmd,
1090                                  dm_thin_id dev,
1091                                  dm_thin_id origin)
1092 {
1093         int r = -EINVAL;
1094
1095         down_write(&pmd->root_lock);
1096         if (!pmd->fail_io)
1097                 r = __create_snap(pmd, dev, origin);
1098         up_write(&pmd->root_lock);
1099
1100         return r;
1101 }
1102
1103 static int __delete_device(struct dm_pool_metadata *pmd, dm_thin_id dev)
1104 {
1105         int r;
1106         uint64_t key = dev;
1107         struct dm_thin_device *td;
1108
1109         /* TODO: failure should mark the transaction invalid */
1110         r = __open_device(pmd, dev, 0, &td);
1111         if (r)
1112                 return r;
1113
1114         if (td->open_count > 1) {
1115                 __close_device(td);
1116                 return -EBUSY;
1117         }
1118
1119         list_del(&td->list);
1120         kfree(td);
1121         r = dm_btree_remove(&pmd->details_info, pmd->details_root,
1122                             &key, &pmd->details_root);
1123         if (r)
1124                 return r;
1125
1126         r = dm_btree_remove(&pmd->tl_info, pmd->root, &key, &pmd->root);
1127         if (r)
1128                 return r;
1129
1130         return 0;
1131 }
1132
1133 int dm_pool_delete_thin_device(struct dm_pool_metadata *pmd,
1134                                dm_thin_id dev)
1135 {
1136         int r = -EINVAL;
1137
1138         down_write(&pmd->root_lock);
1139         if (!pmd->fail_io)
1140                 r = __delete_device(pmd, dev);
1141         up_write(&pmd->root_lock);
1142
1143         return r;
1144 }
1145
1146 int dm_pool_set_metadata_transaction_id(struct dm_pool_metadata *pmd,
1147                                         uint64_t current_id,
1148                                         uint64_t new_id)
1149 {
1150         int r = -EINVAL;
1151
1152         down_write(&pmd->root_lock);
1153
1154         if (pmd->fail_io)
1155                 goto out;
1156
1157         if (pmd->trans_id != current_id) {
1158                 DMERR("mismatched transaction id");
1159                 goto out;
1160         }
1161
1162         pmd->trans_id = new_id;
1163         r = 0;
1164
1165 out:
1166         up_write(&pmd->root_lock);
1167
1168         return r;
1169 }
1170
1171 int dm_pool_get_metadata_transaction_id(struct dm_pool_metadata *pmd,
1172                                         uint64_t *result)
1173 {
1174         int r = -EINVAL;
1175
1176         down_read(&pmd->root_lock);
1177         if (!pmd->fail_io) {
1178                 *result = pmd->trans_id;
1179                 r = 0;
1180         }
1181         up_read(&pmd->root_lock);
1182
1183         return r;
1184 }
1185
1186 static int __reserve_metadata_snap(struct dm_pool_metadata *pmd)
1187 {
1188         int r, inc;
1189         struct thin_disk_superblock *disk_super;
1190         struct dm_block *copy, *sblock;
1191         dm_block_t held_root;
1192
1193         /*
1194          * Copy the superblock.
1195          */
1196         dm_sm_inc_block(pmd->metadata_sm, THIN_SUPERBLOCK_LOCATION);
1197         r = dm_tm_shadow_block(pmd->tm, THIN_SUPERBLOCK_LOCATION,
1198                                &sb_validator, &copy, &inc);
1199         if (r)
1200                 return r;
1201
1202         BUG_ON(!inc);
1203
1204         held_root = dm_block_location(copy);
1205         disk_super = dm_block_data(copy);
1206
1207         if (le64_to_cpu(disk_super->held_root)) {
1208                 DMWARN("Pool metadata snapshot already exists: release this before taking another.");
1209
1210                 dm_tm_dec(pmd->tm, held_root);
1211                 dm_tm_unlock(pmd->tm, copy);
1212                 return -EBUSY;
1213         }
1214
1215         /*
1216          * Wipe the spacemap since we're not publishing this.
1217          */
1218         memset(&disk_super->data_space_map_root, 0,
1219                sizeof(disk_super->data_space_map_root));
1220         memset(&disk_super->metadata_space_map_root, 0,
1221                sizeof(disk_super->metadata_space_map_root));
1222
1223         /*
1224          * Increment the data structures that need to be preserved.
1225          */
1226         dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->data_mapping_root));
1227         dm_tm_inc(pmd->tm, le64_to_cpu(disk_super->device_details_root));
1228         dm_tm_unlock(pmd->tm, copy);
1229
1230         /*
1231          * Write the held root into the superblock.
1232          */
1233         r = superblock_lock(pmd, &sblock);
1234         if (r) {
1235                 dm_tm_dec(pmd->tm, held_root);
1236                 return r;
1237         }
1238
1239         disk_super = dm_block_data(sblock);
1240         disk_super->held_root = cpu_to_le64(held_root);
1241         dm_bm_unlock(sblock);
1242         return 0;
1243 }
1244
1245 int dm_pool_reserve_metadata_snap(struct dm_pool_metadata *pmd)
1246 {
1247         int r = -EINVAL;
1248
1249         down_write(&pmd->root_lock);
1250         if (!pmd->fail_io)
1251                 r = __reserve_metadata_snap(pmd);
1252         up_write(&pmd->root_lock);
1253
1254         return r;
1255 }
1256
1257 static int __release_metadata_snap(struct dm_pool_metadata *pmd)
1258 {
1259         int r;
1260         struct thin_disk_superblock *disk_super;
1261         struct dm_block *sblock, *copy;
1262         dm_block_t held_root;
1263
1264         r = superblock_lock(pmd, &sblock);
1265         if (r)
1266                 return r;
1267
1268         disk_super = dm_block_data(sblock);
1269         held_root = le64_to_cpu(disk_super->held_root);
1270         disk_super->held_root = cpu_to_le64(0);
1271
1272         dm_bm_unlock(sblock);
1273
1274         if (!held_root) {
1275                 DMWARN("No pool metadata snapshot found: nothing to release.");
1276                 return -EINVAL;
1277         }
1278
1279         r = dm_tm_read_lock(pmd->tm, held_root, &sb_validator, &copy);
1280         if (r)
1281                 return r;
1282
1283         disk_super = dm_block_data(copy);
1284         dm_sm_dec_block(pmd->metadata_sm, le64_to_cpu(disk_super->data_mapping_root));
1285         dm_sm_dec_block(pmd->metadata_sm, le64_to_cpu(disk_super->device_details_root));
1286         dm_sm_dec_block(pmd->metadata_sm, held_root);
1287
1288         return dm_tm_unlock(pmd->tm, copy);
1289 }
1290
1291 int dm_pool_release_metadata_snap(struct dm_pool_metadata *pmd)
1292 {
1293         int r = -EINVAL;
1294
1295         down_write(&pmd->root_lock);
1296         if (!pmd->fail_io)
1297                 r = __release_metadata_snap(pmd);
1298         up_write(&pmd->root_lock);
1299
1300         return r;
1301 }
1302
1303 static int __get_metadata_snap(struct dm_pool_metadata *pmd,
1304                                dm_block_t *result)
1305 {
1306         int r;
1307         struct thin_disk_superblock *disk_super;
1308         struct dm_block *sblock;
1309
1310         r = dm_bm_read_lock(pmd->bm, THIN_SUPERBLOCK_LOCATION,
1311                             &sb_validator, &sblock);
1312         if (r)
1313                 return r;
1314
1315         disk_super = dm_block_data(sblock);
1316         *result = le64_to_cpu(disk_super->held_root);
1317
1318         return dm_bm_unlock(sblock);
1319 }
1320
1321 int dm_pool_get_metadata_snap(struct dm_pool_metadata *pmd,
1322                               dm_block_t *result)
1323 {
1324         int r = -EINVAL;
1325
1326         down_read(&pmd->root_lock);
1327         if (!pmd->fail_io)
1328                 r = __get_metadata_snap(pmd, result);
1329         up_read(&pmd->root_lock);
1330
1331         return r;
1332 }
1333
1334 int dm_pool_open_thin_device(struct dm_pool_metadata *pmd, dm_thin_id dev,
1335                              struct dm_thin_device **td)
1336 {
1337         int r = -EINVAL;
1338
1339         down_write(&pmd->root_lock);
1340         if (!pmd->fail_io)
1341                 r = __open_device(pmd, dev, 0, td);
1342         up_write(&pmd->root_lock);
1343
1344         return r;
1345 }
1346
1347 int dm_pool_close_thin_device(struct dm_thin_device *td)
1348 {
1349         down_write(&td->pmd->root_lock);
1350         __close_device(td);
1351         up_write(&td->pmd->root_lock);
1352
1353         return 0;
1354 }
1355
1356 dm_thin_id dm_thin_dev_id(struct dm_thin_device *td)
1357 {
1358         return td->id;
1359 }
1360
1361 /*
1362  * Check whether @time (of block creation) is older than @td's last snapshot.
1363  * If so then the associated block is shared with the last snapshot device.
1364  * Any block on a device created *after* the device last got snapshotted is
1365  * necessarily not shared.
1366  */
1367 static bool __snapshotted_since(struct dm_thin_device *td, uint32_t time)
1368 {
1369         return td->snapshotted_time > time;
1370 }
1371
1372 int dm_thin_find_block(struct dm_thin_device *td, dm_block_t block,
1373                        int can_block, struct dm_thin_lookup_result *result)
1374 {
1375         int r = -EINVAL;
1376         uint64_t block_time = 0;
1377         __le64 value;
1378         struct dm_pool_metadata *pmd = td->pmd;
1379         dm_block_t keys[2] = { td->id, block };
1380         struct dm_btree_info *info;
1381
1382         if (can_block) {
1383                 down_read(&pmd->root_lock);
1384                 info = &pmd->info;
1385         } else if (down_read_trylock(&pmd->root_lock))
1386                 info = &pmd->nb_info;
1387         else
1388                 return -EWOULDBLOCK;
1389
1390         if (pmd->fail_io)
1391                 goto out;
1392
1393         r = dm_btree_lookup(info, pmd->root, keys, &value);
1394         if (!r)
1395                 block_time = le64_to_cpu(value);
1396
1397 out:
1398         up_read(&pmd->root_lock);
1399
1400         if (!r) {
1401                 dm_block_t exception_block;
1402                 uint32_t exception_time;
1403                 unpack_block_time(block_time, &exception_block,
1404                                   &exception_time);
1405                 result->block = exception_block;
1406                 result->shared = __snapshotted_since(td, exception_time);
1407         }
1408
1409         return r;
1410 }
1411
1412 static int __insert(struct dm_thin_device *td, dm_block_t block,
1413                     dm_block_t data_block)
1414 {
1415         int r, inserted;
1416         __le64 value;
1417         struct dm_pool_metadata *pmd = td->pmd;
1418         dm_block_t keys[2] = { td->id, block };
1419
1420         value = cpu_to_le64(pack_block_time(data_block, pmd->time));
1421         __dm_bless_for_disk(&value);
1422
1423         r = dm_btree_insert_notify(&pmd->info, pmd->root, keys, &value,
1424                                    &pmd->root, &inserted);
1425         if (r)
1426                 return r;
1427
1428         td->changed = 1;
1429         if (inserted)
1430                 td->mapped_blocks++;
1431
1432         return 0;
1433 }
1434
1435 int dm_thin_insert_block(struct dm_thin_device *td, dm_block_t block,
1436                          dm_block_t data_block)
1437 {
1438         int r = -EINVAL;
1439
1440         down_write(&td->pmd->root_lock);
1441         if (!td->pmd->fail_io)
1442                 r = __insert(td, block, data_block);
1443         up_write(&td->pmd->root_lock);
1444
1445         return r;
1446 }
1447
1448 static int __remove(struct dm_thin_device *td, dm_block_t block)
1449 {
1450         int r;
1451         struct dm_pool_metadata *pmd = td->pmd;
1452         dm_block_t keys[2] = { td->id, block };
1453
1454         r = dm_btree_remove(&pmd->info, pmd->root, keys, &pmd->root);
1455         if (r)
1456                 return r;
1457
1458         td->mapped_blocks--;
1459         td->changed = 1;
1460
1461         return 0;
1462 }
1463
1464 int dm_thin_remove_block(struct dm_thin_device *td, dm_block_t block)
1465 {
1466         int r = -EINVAL;
1467
1468         down_write(&td->pmd->root_lock);
1469         if (!td->pmd->fail_io)
1470                 r = __remove(td, block);
1471         up_write(&td->pmd->root_lock);
1472
1473         return r;
1474 }
1475
1476 int dm_pool_block_is_used(struct dm_pool_metadata *pmd, dm_block_t b, bool *result)
1477 {
1478         int r;
1479         uint32_t ref_count;
1480
1481         down_read(&pmd->root_lock);
1482         r = dm_sm_get_count(pmd->data_sm, b, &ref_count);
1483         if (!r)
1484                 *result = (ref_count != 0);
1485         up_read(&pmd->root_lock);
1486
1487         return r;
1488 }
1489
1490 bool dm_thin_changed_this_transaction(struct dm_thin_device *td)
1491 {
1492         int r;
1493
1494         down_read(&td->pmd->root_lock);
1495         r = td->changed;
1496         up_read(&td->pmd->root_lock);
1497
1498         return r;
1499 }
1500
1501 bool dm_pool_changed_this_transaction(struct dm_pool_metadata *pmd)
1502 {
1503         bool r = false;
1504         struct dm_thin_device *td, *tmp;
1505
1506         down_read(&pmd->root_lock);
1507         list_for_each_entry_safe(td, tmp, &pmd->thin_devices, list) {
1508                 if (td->changed) {
1509                         r = td->changed;
1510                         break;
1511                 }
1512         }
1513         up_read(&pmd->root_lock);
1514
1515         return r;
1516 }
1517
1518 bool dm_thin_aborted_changes(struct dm_thin_device *td)
1519 {
1520         bool r;
1521
1522         down_read(&td->pmd->root_lock);
1523         r = td->aborted_with_changes;
1524         up_read(&td->pmd->root_lock);
1525
1526         return r;
1527 }
1528
1529 int dm_pool_alloc_data_block(struct dm_pool_metadata *pmd, dm_block_t *result)
1530 {
1531         int r = -EINVAL;
1532
1533         down_write(&pmd->root_lock);
1534         if (!pmd->fail_io)
1535                 r = dm_sm_new_block(pmd->data_sm, result);
1536         up_write(&pmd->root_lock);
1537
1538         return r;
1539 }
1540
1541 int dm_pool_commit_metadata(struct dm_pool_metadata *pmd)
1542 {
1543         int r = -EINVAL;
1544
1545         down_write(&pmd->root_lock);
1546         if (pmd->fail_io)
1547                 goto out;
1548
1549         r = __commit_transaction(pmd);
1550         if (r <= 0)
1551                 goto out;
1552
1553         /*
1554          * Open the next transaction.
1555          */
1556         r = __begin_transaction(pmd);
1557 out:
1558         up_write(&pmd->root_lock);
1559         return r;
1560 }
1561
1562 static void __set_abort_with_changes_flags(struct dm_pool_metadata *pmd)
1563 {
1564         struct dm_thin_device *td;
1565
1566         list_for_each_entry(td, &pmd->thin_devices, list)
1567                 td->aborted_with_changes = td->changed;
1568 }
1569
1570 int dm_pool_abort_metadata(struct dm_pool_metadata *pmd)
1571 {
1572         int r = -EINVAL;
1573
1574         down_write(&pmd->root_lock);
1575         if (pmd->fail_io)
1576                 goto out;
1577
1578         __set_abort_with_changes_flags(pmd);
1579         __destroy_persistent_data_objects(pmd);
1580         r = __create_persistent_data_objects(pmd, false);
1581         if (r)
1582                 pmd->fail_io = true;
1583
1584 out:
1585         up_write(&pmd->root_lock);
1586
1587         return r;
1588 }
1589
1590 int dm_pool_get_free_block_count(struct dm_pool_metadata *pmd, dm_block_t *result)
1591 {
1592         int r = -EINVAL;
1593
1594         down_read(&pmd->root_lock);
1595         if (!pmd->fail_io)
1596                 r = dm_sm_get_nr_free(pmd->data_sm, result);
1597         up_read(&pmd->root_lock);
1598
1599         return r;
1600 }
1601
1602 int dm_pool_get_free_metadata_block_count(struct dm_pool_metadata *pmd,
1603                                           dm_block_t *result)
1604 {
1605         int r = -EINVAL;
1606
1607         down_read(&pmd->root_lock);
1608         if (!pmd->fail_io)
1609                 r = dm_sm_get_nr_free(pmd->metadata_sm, result);
1610         up_read(&pmd->root_lock);
1611
1612         return r;
1613 }
1614
1615 int dm_pool_get_metadata_dev_size(struct dm_pool_metadata *pmd,
1616                                   dm_block_t *result)
1617 {
1618         int r = -EINVAL;
1619
1620         down_read(&pmd->root_lock);
1621         if (!pmd->fail_io)
1622                 r = dm_sm_get_nr_blocks(pmd->metadata_sm, result);
1623         up_read(&pmd->root_lock);
1624
1625         return r;
1626 }
1627
1628 int dm_pool_get_data_block_size(struct dm_pool_metadata *pmd, sector_t *result)
1629 {
1630         down_read(&pmd->root_lock);
1631         *result = pmd->data_block_size;
1632         up_read(&pmd->root_lock);
1633
1634         return 0;
1635 }
1636
1637 int dm_pool_get_data_dev_size(struct dm_pool_metadata *pmd, dm_block_t *result)
1638 {
1639         int r = -EINVAL;
1640
1641         down_read(&pmd->root_lock);
1642         if (!pmd->fail_io)
1643                 r = dm_sm_get_nr_blocks(pmd->data_sm, result);
1644         up_read(&pmd->root_lock);
1645
1646         return r;
1647 }
1648
1649 int dm_thin_get_mapped_count(struct dm_thin_device *td, dm_block_t *result)
1650 {
1651         int r = -EINVAL;
1652         struct dm_pool_metadata *pmd = td->pmd;
1653
1654         down_read(&pmd->root_lock);
1655         if (!pmd->fail_io) {
1656                 *result = td->mapped_blocks;
1657                 r = 0;
1658         }
1659         up_read(&pmd->root_lock);
1660
1661         return r;
1662 }
1663
1664 static int __highest_block(struct dm_thin_device *td, dm_block_t *result)
1665 {
1666         int r;
1667         __le64 value_le;
1668         dm_block_t thin_root;
1669         struct dm_pool_metadata *pmd = td->pmd;
1670
1671         r = dm_btree_lookup(&pmd->tl_info, pmd->root, &td->id, &value_le);
1672         if (r)
1673                 return r;
1674
1675         thin_root = le64_to_cpu(value_le);
1676
1677         return dm_btree_find_highest_key(&pmd->bl_info, thin_root, result);
1678 }
1679
1680 int dm_thin_get_highest_mapped_block(struct dm_thin_device *td,
1681                                      dm_block_t *result)
1682 {
1683         int r = -EINVAL;
1684         struct dm_pool_metadata *pmd = td->pmd;
1685
1686         down_read(&pmd->root_lock);
1687         if (!pmd->fail_io)
1688                 r = __highest_block(td, result);
1689         up_read(&pmd->root_lock);
1690
1691         return r;
1692 }
1693
1694 static int __resize_space_map(struct dm_space_map *sm, dm_block_t new_count)
1695 {
1696         int r;
1697         dm_block_t old_count;
1698
1699         r = dm_sm_get_nr_blocks(sm, &old_count);
1700         if (r)
1701                 return r;
1702
1703         if (new_count == old_count)
1704                 return 0;
1705
1706         if (new_count < old_count) {
1707                 DMERR("cannot reduce size of space map");
1708                 return -EINVAL;
1709         }
1710
1711         return dm_sm_extend(sm, new_count - old_count);
1712 }
1713
1714 int dm_pool_resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1715 {
1716         int r = -EINVAL;
1717
1718         down_write(&pmd->root_lock);
1719         if (!pmd->fail_io)
1720                 r = __resize_space_map(pmd->data_sm, new_count);
1721         up_write(&pmd->root_lock);
1722
1723         return r;
1724 }
1725
1726 int dm_pool_resize_metadata_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)
1727 {
1728         int r = -EINVAL;
1729
1730         down_write(&pmd->root_lock);
1731         if (!pmd->fail_io)
1732                 r = __resize_space_map(pmd->metadata_sm, new_count);
1733         up_write(&pmd->root_lock);
1734
1735         return r;
1736 }
1737
1738 void dm_pool_metadata_read_only(struct dm_pool_metadata *pmd)
1739 {
1740         down_write(&pmd->root_lock);
1741         pmd->read_only = true;
1742         dm_bm_set_read_only(pmd->bm);
1743         up_write(&pmd->root_lock);
1744 }
1745
1746 int dm_pool_register_metadata_threshold(struct dm_pool_metadata *pmd,
1747                                         dm_block_t threshold,
1748                                         dm_sm_threshold_fn fn,
1749                                         void *context)
1750 {
1751         int r;
1752
1753         down_write(&pmd->root_lock);
1754         r = dm_sm_register_threshold_callback(pmd->metadata_sm, threshold, fn, context);
1755         up_write(&pmd->root_lock);
1756
1757         return r;
1758 }