drivers/md/dm-era-target.c

   1 #include "dm.h"
   2 #include "persistent-data/dm-transaction-manager.h"
   3 #include "persistent-data/dm-bitset.h"
   4 #include "persistent-data/dm-space-map.h"
   5
   6 #include <linux/dm-io.h>
   7 #include <linux/dm-kcopyd.h>
   8 #include <linux/init.h>
   9 #include <linux/mempool.h>
  10 #include <linux/module.h>
  11 #include <linux/slab.h>
  12 #include <linux/vmalloc.h>
  13
  14 #define DM_MSG_PREFIX "era"
  15
  16 #define SUPERBLOCK_LOCATION 0
  17 #define SUPERBLOCK_MAGIC 2126579579
  18 #define SUPERBLOCK_CSUM_XOR 146538381
  19 #define MIN_ERA_VERSION 1
  20 #define MAX_ERA_VERSION 1
  21 #define INVALID_WRITESET_ROOT SUPERBLOCK_LOCATION
  22 #define MIN_BLOCK_SIZE 8
  23
  24 /*----------------------------------------------------------------
  25  * Writeset
  26  *--------------------------------------------------------------*/
  27 struct writeset_metadata {
  28         uint32_t nr_bits;
  29         dm_block_t root;
  30 };
  31
  32 struct writeset {
  33         struct writeset_metadata md;
  34
  35         /*
  36          * An in core copy of the bits to save constantly doing look ups on
  37          * disk.
  38          */
  39         unsigned long *bits;
  40 };
  41
  42 /*
  43  * This does not free off the on disk bitset as this will normally be done
  44  * after digesting into the era array.
  45  */
  46 static void writeset_free(struct writeset *ws)
  47 {
  48         vfree(ws->bits);
  49 }
  50
  51 static int setup_on_disk_bitset(struct dm_disk_bitset *info,
  52                                 unsigned nr_bits, dm_block_t *root)
  53 {
  54         int r;
  55
  56         r = dm_bitset_empty(info, root);
  57         if (r)
  58                 return r;
  59
  60         return dm_bitset_resize(info, *root, 0, nr_bits, false, root);
  61 }
  62
  63 static size_t bitset_size(unsigned nr_bits)
  64 {
  65         return sizeof(unsigned long) * dm_div_up(nr_bits, BITS_PER_LONG);
  66 }
  67
  68 /*
  69  * Allocates memory for the in core bitset.
  70  */
  71 static int writeset_alloc(struct writeset *ws, dm_block_t nr_blocks)
  72 {
  73         ws->md.nr_bits = nr_blocks;
  74         ws->md.root = INVALID_WRITESET_ROOT;
  75         ws->bits = vzalloc(bitset_size(nr_blocks));
  76         if (!ws->bits) {
  77                 DMERR("%s: couldn't allocate in memory bitset", __func__);
  78                 return -ENOMEM;
  79         }
  80
  81         return 0;
  82 }
  83
  84 /*
  85  * Wipes the in-core bitset, and creates a new on disk bitset.
  86  */
  87 static int writeset_init(struct dm_disk_bitset *info, struct writeset *ws)
  88 {
  89         int r;
  90
  91         memset(ws->bits, 0, bitset_size(ws->md.nr_bits));
  92
  93         r = setup_on_disk_bitset(info, ws->md.nr_bits, &ws->md.root);
  94         if (r) {
  95                 DMERR("%s: setup_on_disk_bitset failed", __func__);
  96                 return r;
  97         }
  98
  99         return 0;
 100 }
 101
 102 static bool writeset_marked(struct writeset *ws, dm_block_t block)
 103 {
 104         return test_bit(block, ws->bits);
 105 }
 106
 107 static int writeset_marked_on_disk(struct dm_disk_bitset *info,
 108                                    struct writeset_metadata *m, dm_block_t block,
 109                                    bool *result)
 110 {
 111         dm_block_t old = m->root;
 112
 113         /*
 114          * The bitset was flushed when it was archived, so we know there'll
 115          * be no change to the root.
 116          */
 117         int r = dm_bitset_test_bit(info, m->root, block, &m->root, result);
 118         if (r) {
 119                 DMERR("%s: dm_bitset_test_bit failed", __func__);
 120                 return r;
 121         }
 122
 123         BUG_ON(m->root != old);
 124
 125         return r;
 126 }
 127
 128 /*
 129  * Returns < 0 on error, 0 if the bit wasn't previously set, 1 if it was.
 130  */
 131 static int writeset_test_and_set(struct dm_disk_bitset *info,
 132                                  struct writeset *ws, uint32_t block)
 133 {
 134         int r;
 135
 136         if (!test_and_set_bit(block, ws->bits)) {
 137                 r = dm_bitset_set_bit(info, ws->md.root, block, &ws->md.root);
 138                 if (r) {
 139                         /* FIXME: fail mode */
 140                         return r;
 141                 }
 142
 143                 return 0;
 144         }
 145
 146         return 1;
 147 }
 148
 149 /*----------------------------------------------------------------
 150  * On disk metadata layout
 151  *--------------------------------------------------------------*/
 152 #define SPACE_MAP_ROOT_SIZE 128
 153 #define UUID_LEN 16
 154
 155 struct writeset_disk {
 156         __le32 nr_bits;
 157         __le64 root;
 158 } __packed;
 159
 160 struct superblock_disk {
 161         __le32 csum;
 162         __le32 flags;
 163         __le64 blocknr;
 164
 165         __u8 uuid[UUID_LEN];
 166         __le64 magic;
 167         __le32 version;
 168
 169         __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
 170
 171         __le32 data_block_size;
 172         __le32 metadata_block_size;
 173         __le32 nr_blocks;
 174
 175         __le32 current_era;
 176         struct writeset_disk current_writeset;
 177
 178         /*
 179          * Only these two fields are valid within the metadata snapshot.
 180          */
 181         __le64 writeset_tree_root;
 182         __le64 era_array_root;
 183
 184         __le64 metadata_snap;
 185 } __packed;
 186
 187 /*----------------------------------------------------------------
 188  * Superblock validation
 189  *--------------------------------------------------------------*/
 190 static void sb_prepare_for_write(struct dm_block_validator *v,
 191                                  struct dm_block *b,
 192                                  size_t sb_block_size)
 193 {
 194         struct superblock_disk *disk = dm_block_data(b);
 195
 196         disk->blocknr = cpu_to_le64(dm_block_location(b));
 197         disk->csum = cpu_to_le32(dm_bm_checksum(&disk->flags,
 198                                                 sb_block_size - sizeof(__le32),
 199                                                 SUPERBLOCK_CSUM_XOR));
 200 }
 201
 202 static int check_metadata_version(struct superblock_disk *disk)
 203 {
 204         uint32_t metadata_version = le32_to_cpu(disk->version);
 205         if (metadata_version < MIN_ERA_VERSION || metadata_version > MAX_ERA_VERSION) {
 206                 DMERR("Era metadata version %u found, but only versions between %u and %u supported.",
 207                       metadata_version, MIN_ERA_VERSION, MAX_ERA_VERSION);
 208                 return -EINVAL;
 209         }
 210
 211         return 0;
 212 }
 213
 214 static int sb_check(struct dm_block_validator *v,
 215                     struct dm_block *b,
 216                     size_t sb_block_size)
 217 {
 218         struct superblock_disk *disk = dm_block_data(b);
 219         __le32 csum_le;
 220
 221         if (dm_block_location(b) != le64_to_cpu(disk->blocknr)) {
 222                 DMERR("sb_check failed: blocknr %llu: wanted %llu",
 223                       le64_to_cpu(disk->blocknr),
 224                       (unsigned long long)dm_block_location(b));
 225                 return -ENOTBLK;
 226         }
 227
 228         if (le64_to_cpu(disk->magic) != SUPERBLOCK_MAGIC) {
 229                 DMERR("sb_check failed: magic %llu: wanted %llu",
 230                       le64_to_cpu(disk->magic),
 231                       (unsigned long long) SUPERBLOCK_MAGIC);
 232                 return -EILSEQ;
 233         }
 234
 235         csum_le = cpu_to_le32(dm_bm_checksum(&disk->flags,
 236                                              sb_block_size - sizeof(__le32),
 237                                              SUPERBLOCK_CSUM_XOR));
 238         if (csum_le != disk->csum) {
 239                 DMERR("sb_check failed: csum %u: wanted %u",
 240                       le32_to_cpu(csum_le), le32_to_cpu(disk->csum));
 241                 return -EILSEQ;
 242         }
 243
 244         return check_metadata_version(disk);
 245 }
 246
 247 static struct dm_block_validator sb_validator = {
 248         .name = "superblock",
 249         .prepare_for_write = sb_prepare_for_write,
 250         .check = sb_check
 251 };
 252
 253 /*----------------------------------------------------------------
 254  * Low level metadata handling
 255  *--------------------------------------------------------------*/
 256 #define DM_ERA_METADATA_BLOCK_SIZE 4096
 257 #define DM_ERA_METADATA_CACHE_SIZE 64
 258 #define ERA_MAX_CONCURRENT_LOCKS 5
 259
 260 struct era_metadata {
 261         struct block_device *bdev;
 262         struct dm_block_manager *bm;
 263         struct dm_space_map *sm;
 264         struct dm_transaction_manager *tm;
 265
 266         dm_block_t block_size;
 267         uint32_t nr_blocks;
 268
 269         uint32_t current_era;
 270
 271         /*
 272          * We preallocate 2 writesets.  When an era rolls over we
 273          * switch between them. This means the allocation is done at
 274          * preresume time, rather than on the io path.
 275          */
 276         struct writeset writesets[2];
 277         struct writeset *current_writeset;
 278
 279         dm_block_t writeset_tree_root;
 280         dm_block_t era_array_root;
 281
 282         struct dm_disk_bitset bitset_info;
 283         struct dm_btree_info writeset_tree_info;
 284         struct dm_array_info era_array_info;
 285
 286         dm_block_t metadata_snap;
 287
 288         /*
 289          * A flag that is set whenever a writeset has been archived.
 290          */
 291         bool archived_writesets;
 292
 293         /*
 294          * Reading the space map root can fail, so we read it into this
 295          * buffer before the superblock is locked and updated.
 296          */
 297         __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
 298 };
 299
 300 static int superblock_read_lock(struct era_metadata *md,
 301                                 struct dm_block **sblock)
 302 {
 303         return dm_bm_read_lock(md->bm, SUPERBLOCK_LOCATION,
 304                                &sb_validator, sblock);
 305 }
 306
 307 static int superblock_lock_zero(struct era_metadata *md,
 308                                 struct dm_block **sblock)
 309 {
 310         return dm_bm_write_lock_zero(md->bm, SUPERBLOCK_LOCATION,
 311                                      &sb_validator, sblock);
 312 }
 313
 314 static int superblock_lock(struct era_metadata *md,
 315                            struct dm_block **sblock)
 316 {
 317         return dm_bm_write_lock(md->bm, SUPERBLOCK_LOCATION,
 318                                 &sb_validator, sblock);
 319 }
 320
 321 /* FIXME: duplication with cache and thin */
 322 static int superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
 323 {
 324         int r;
 325         unsigned i;
 326         struct dm_block *b;
 327         __le64 *data_le, zero = cpu_to_le64(0);
 328         unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
 329
 330         /*
 331          * We can't use a validator here - it may be all zeroes.
 332          */
 333         r = dm_bm_read_lock(bm, SUPERBLOCK_LOCATION, NULL, &b);
 334         if (r)
 335                 return r;
 336
 337         data_le = dm_block_data(b);
 338         *result = true;
 339         for (i = 0; i < sb_block_size; i++) {
 340                 if (data_le[i] != zero) {
 341                         *result = false;
 342                         break;
 343                 }
 344         }
 345
 346         dm_bm_unlock(b);
 347
 348         return 0;
 349 }
 350
 351 /*----------------------------------------------------------------*/
 352
 353 static void ws_pack(const struct writeset_metadata *core, struct writeset_disk *disk)
 354 {
 355         disk->nr_bits = cpu_to_le32(core->nr_bits);
 356         disk->root = cpu_to_le64(core->root);
 357 }
 358
 359 static void ws_unpack(const struct writeset_disk *disk, struct writeset_metadata *core)
 360 {
 361         core->nr_bits = le32_to_cpu(disk->nr_bits);
 362         core->root = le64_to_cpu(disk->root);
 363 }
 364
 365 static void ws_inc(void *context, const void *value)
 366 {
 367         struct era_metadata *md = context;
 368         struct writeset_disk ws_d;
 369         dm_block_t b;
 370
 371         memcpy(&ws_d, value, sizeof(ws_d));
 372         b = le64_to_cpu(ws_d.root);
 373
 374         dm_tm_inc(md->tm, b);
 375 }
 376
 377 static void ws_dec(void *context, const void *value)
 378 {
 379         struct era_metadata *md = context;
 380         struct writeset_disk ws_d;
 381         dm_block_t b;
 382
 383         memcpy(&ws_d, value, sizeof(ws_d));
 384         b = le64_to_cpu(ws_d.root);
 385
 386         dm_bitset_del(&md->bitset_info, b);
 387 }
 388
 389 static int ws_eq(void *context, const void *value1, const void *value2)
 390 {
 391         return !memcmp(value1, value2, sizeof(struct writeset_metadata));
 392 }
 393
 394 /*----------------------------------------------------------------*/
 395
 396 static void setup_writeset_tree_info(struct era_metadata *md)
 397 {
 398         struct dm_btree_value_type *vt = &md->writeset_tree_info.value_type;
 399         md->writeset_tree_info.tm = md->tm;
 400         md->writeset_tree_info.levels = 1;
 401         vt->context = md;
 402         vt->size = sizeof(struct writeset_disk);
 403         vt->inc = ws_inc;
 404         vt->dec = ws_dec;
 405         vt->equal = ws_eq;
 406 }
 407
 408 static void setup_era_array_info(struct era_metadata *md)
 409
 410 {
 411         struct dm_btree_value_type vt;
 412         vt.context = NULL;
 413         vt.size = sizeof(__le32);
 414         vt.inc = NULL;
 415         vt.dec = NULL;
 416         vt.equal = NULL;
 417
 418         dm_array_info_init(&md->era_array_info, md->tm, &vt);
 419 }
 420
 421 static void setup_infos(struct era_metadata *md)
 422 {
 423         dm_disk_bitset_init(md->tm, &md->bitset_info);
 424         setup_writeset_tree_info(md);
 425         setup_era_array_info(md);
 426 }
 427
 428 /*----------------------------------------------------------------*/
 429
 430 static int create_fresh_metadata(struct era_metadata *md)
 431 {
 432         int r;
 433
 434         r = dm_tm_create_with_sm(md->bm, SUPERBLOCK_LOCATION,
 435                                  &md->tm, &md->sm);
 436         if (r < 0) {
 437                 DMERR("dm_tm_create_with_sm failed");
 438                 return r;
 439         }
 440
 441         setup_infos(md);
 442
 443         r = dm_btree_empty(&md->writeset_tree_info, &md->writeset_tree_root);
 444         if (r) {
 445                 DMERR("couldn't create new writeset tree");
 446                 goto bad;
 447         }
 448
 449         r = dm_array_empty(&md->era_array_info, &md->era_array_root);
 450         if (r) {
 451                 DMERR("couldn't create era array");
 452                 goto bad;
 453         }
 454
 455         return 0;
 456
 457 bad:
 458         dm_sm_destroy(md->sm);
 459         dm_tm_destroy(md->tm);
 460
 461         return r;
 462 }
 463
 464 static int save_sm_root(struct era_metadata *md)
 465 {
 466         int r;
 467         size_t metadata_len;
 468
 469         r = dm_sm_root_size(md->sm, &metadata_len);
 470         if (r < 0)
 471                 return r;
 472
 473         return dm_sm_copy_root(md->sm, &md->metadata_space_map_root,
 474                                metadata_len);
 475 }
 476
 477 static void copy_sm_root(struct era_metadata *md, struct superblock_disk *disk)
 478 {
 479         memcpy(&disk->metadata_space_map_root,
 480                &md->metadata_space_map_root,
 481                sizeof(md->metadata_space_map_root));
 482 }
 483
 484 /*
 485  * Writes a superblock, including the static fields that don't get updated
 486  * with every commit (possible optimisation here).  'md' should be fully
 487  * constructed when this is called.
 488  */
 489 static void prepare_superblock(struct era_metadata *md, struct superblock_disk *disk)
 490 {
 491         disk->magic = cpu_to_le64(SUPERBLOCK_MAGIC);
 492         disk->flags = cpu_to_le32(0ul);
 493
 494         /* FIXME: can't keep blanking the uuid (uuid is currently unused though) */
 495         memset(disk->uuid, 0, sizeof(disk->uuid));
 496         disk->version = cpu_to_le32(MAX_ERA_VERSION);
 497
 498         copy_sm_root(md, disk);
 499
 500         disk->data_block_size = cpu_to_le32(md->block_size);
 501         disk->metadata_block_size = cpu_to_le32(DM_ERA_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
 502         disk->nr_blocks = cpu_to_le32(md->nr_blocks);
 503         disk->current_era = cpu_to_le32(md->current_era);
 504
 505         ws_pack(&md->current_writeset->md, &disk->current_writeset);
 506         disk->writeset_tree_root = cpu_to_le64(md->writeset_tree_root);
 507         disk->era_array_root = cpu_to_le64(md->era_array_root);
 508         disk->metadata_snap = cpu_to_le64(md->metadata_snap);
 509 }
 510
 511 static int write_superblock(struct era_metadata *md)
 512 {
 513         int r;
 514         struct dm_block *sblock;
 515         struct superblock_disk *disk;
 516
 517         r = save_sm_root(md);
 518         if (r) {
 519                 DMERR("%s: save_sm_root failed", __func__);
 520                 return r;
 521         }
 522
 523         r = superblock_lock_zero(md, &sblock);
 524         if (r)
 525                 return r;
 526
 527         disk = dm_block_data(sblock);
 528         prepare_superblock(md, disk);
 529
 530         return dm_tm_commit(md->tm, sblock);
 531 }
 532
 533 /*
 534  * Assumes block_size and the infos are set.
 535  */
 536 static int format_metadata(struct era_metadata *md)
 537 {
 538         int r;
 539
 540         r = create_fresh_metadata(md);
 541         if (r)
 542                 return r;
 543
 544         r = write_superblock(md);
 545         if (r) {
 546                 dm_sm_destroy(md->sm);
 547                 dm_tm_destroy(md->tm);
 548                 return r;
 549         }
 550
 551         return 0;
 552 }
 553
 554 static int open_metadata(struct era_metadata *md)
 555 {
 556         int r;
 557         struct dm_block *sblock;
 558         struct superblock_disk *disk;
 559
 560         r = superblock_read_lock(md, &sblock);
 561         if (r) {
 562                 DMERR("couldn't read_lock superblock");
 563                 return r;
 564         }
 565
 566         disk = dm_block_data(sblock);
 567         r = dm_tm_open_with_sm(md->bm, SUPERBLOCK_LOCATION,
 568                                disk->metadata_space_map_root,
 569                                sizeof(disk->metadata_space_map_root),
 570                                &md->tm, &md->sm);
 571         if (r) {
 572                 DMERR("dm_tm_open_with_sm failed");
 573                 goto bad;
 574         }
 575
 576         setup_infos(md);
 577
 578         md->block_size = le32_to_cpu(disk->data_block_size);
 579         md->nr_blocks = le32_to_cpu(disk->nr_blocks);
 580         md->current_era = le32_to_cpu(disk->current_era);
 581
 582         md->writeset_tree_root = le64_to_cpu(disk->writeset_tree_root);
 583         md->era_array_root = le64_to_cpu(disk->era_array_root);
 584         md->metadata_snap = le64_to_cpu(disk->metadata_snap);
 585         md->archived_writesets = true;
 586
 587         dm_bm_unlock(sblock);
 588
 589         return 0;
 590
 591 bad:
 592         dm_bm_unlock(sblock);
 593         return r;
 594 }
 595
 596 static int open_or_format_metadata(struct era_metadata *md,
 597                                    bool may_format)
 598 {
 599         int r;
 600         bool unformatted = false;
 601
 602         r = superblock_all_zeroes(md->bm, &unformatted);
 603         if (r)
 604                 return r;
 605
 606         if (unformatted)
 607                 return may_format ? format_metadata(md) : -EPERM;
 608
 609         return open_metadata(md);
 610 }
 611
 612 static int create_persistent_data_objects(struct era_metadata *md,
 613                                           bool may_format)
 614 {
 615         int r;
 616
 617         md->bm = dm_block_manager_create(md->bdev, DM_ERA_METADATA_BLOCK_SIZE,
 618                                          DM_ERA_METADATA_CACHE_SIZE,
 619                                          ERA_MAX_CONCURRENT_LOCKS);
 620         if (IS_ERR(md->bm)) {
 621                 DMERR("could not create block manager");
 622                 return PTR_ERR(md->bm);
 623         }
 624
 625         r = open_or_format_metadata(md, may_format);
 626         if (r)
 627                 dm_block_manager_destroy(md->bm);
 628
 629         return r;
 630 }
 631
 632 static void destroy_persistent_data_objects(struct era_metadata *md)
 633 {
 634         dm_sm_destroy(md->sm);
 635         dm_tm_destroy(md->tm);
 636         dm_block_manager_destroy(md->bm);
 637 }
 638
 639 /*
 640  * This waits until all era_map threads have picked up the new filter.
 641  */
 642 static void swap_writeset(struct era_metadata *md, struct writeset *new_writeset)
 643 {
 644         rcu_assign_pointer(md->current_writeset, new_writeset);
 645         synchronize_rcu();
 646 }
 647
 648 /*----------------------------------------------------------------
 649  * Writesets get 'digested' into the main era array.
 650  *
 651  * We're using a coroutine here so the worker thread can do the digestion,
 652  * thus avoiding synchronisation of the metadata.  Digesting a whole
 653  * writeset in one go would cause too much latency.
 654  *--------------------------------------------------------------*/
 655 struct digest {
 656         uint32_t era;
 657         unsigned nr_bits, current_bit;
 658         struct writeset_metadata writeset;
 659         __le32 value;
 660         struct dm_disk_bitset info;
 661
 662         int (*step)(struct era_metadata *, struct digest *);
 663 };
 664
 665 static int metadata_digest_lookup_writeset(struct era_metadata *md,
 666                                            struct digest *d);
 667
 668 static int metadata_digest_remove_writeset(struct era_metadata *md,
 669                                            struct digest *d)
 670 {
 671         int r;
 672         uint64_t key = d->era;
 673
 674         r = dm_btree_remove(&md->writeset_tree_info, md->writeset_tree_root,
 675                             &key, &md->writeset_tree_root);
 676         if (r) {
 677                 DMERR("%s: dm_btree_remove failed", __func__);
 678                 return r;
 679         }
 680
 681         d->step = metadata_digest_lookup_writeset;
 682         return 0;
 683 }
 684
 685 #define INSERTS_PER_STEP 100
 686
 687 static int metadata_digest_transcribe_writeset(struct era_metadata *md,
 688                                                struct digest *d)
 689 {
 690         int r;
 691         bool marked;
 692         unsigned b, e = min(d->current_bit + INSERTS_PER_STEP, d->nr_bits);
 693
 694         for (b = d->current_bit; b < e; b++) {
 695                 r = writeset_marked_on_disk(&d->info, &d->writeset, b, &marked);
 696                 if (r) {
 697                         DMERR("%s: writeset_marked_on_disk failed", __func__);
 698                         return r;
 699                 }
 700
 701                 if (!marked)
 702                         continue;
 703
 704                 __dm_bless_for_disk(&d->value);
 705                 r = dm_array_set_value(&md->era_array_info, md->era_array_root,
 706                                        b, &d->value, &md->era_array_root);
 707                 if (r) {
 708                         DMERR("%s: dm_array_set_value failed", __func__);
 709                         return r;
 710                 }
 711         }
 712
 713         if (b == d->nr_bits)
 714                 d->step = metadata_digest_remove_writeset;
 715         else
 716                 d->current_bit = b;
 717
 718         return 0;
 719 }
 720
 721 static int metadata_digest_lookup_writeset(struct era_metadata *md,
 722                                            struct digest *d)
 723 {
 724         int r;
 725         uint64_t key;
 726         struct writeset_disk disk;
 727
 728         r = dm_btree_find_lowest_key(&md->writeset_tree_info,
 729                                      md->writeset_tree_root, &key);
 730         if (r < 0)
 731                 return r;
 732
 733         d->era = key;
 734
 735         r = dm_btree_lookup(&md->writeset_tree_info,
 736                             md->writeset_tree_root, &key, &disk);
 737         if (r) {
 738                 if (r == -ENODATA) {
 739                         d->step = NULL;
 740                         return 0;
 741                 }
 742
 743                 DMERR("%s: dm_btree_lookup failed", __func__);
 744                 return r;
 745         }
 746
 747         ws_unpack(&disk, &d->writeset);
 748         d->value = cpu_to_le32(key);
 749
 750         d->nr_bits = min(d->writeset.nr_bits, md->nr_blocks);
 751         d->current_bit = 0;
 752         d->step = metadata_digest_transcribe_writeset;
 753
 754         return 0;
 755 }
 756
 757 static int metadata_digest_start(struct era_metadata *md, struct digest *d)
 758 {
 759         if (d->step)
 760                 return 0;
 761
 762         memset(d, 0, sizeof(*d));
 763
 764         /*
 765          * We initialise another bitset info to avoid any caching side
 766          * effects with the previous one.
 767          */
 768         dm_disk_bitset_init(md->tm, &d->info);
 769         d->step = metadata_digest_lookup_writeset;
 770
 771         return 0;
 772 }
 773
 774 /*----------------------------------------------------------------
 775  * High level metadata interface.  Target methods should use these, and not
 776  * the lower level ones.
 777  *--------------------------------------------------------------*/
 778 static struct era_metadata *metadata_open(struct block_device *bdev,
 779                                           sector_t block_size,
 780                                           bool may_format)
 781 {
 782         int r;
 783         struct era_metadata *md = kzalloc(sizeof(*md), GFP_KERNEL);
 784
 785         if (!md)
 786                 return NULL;
 787
 788         md->bdev = bdev;
 789         md->block_size = block_size;
 790
 791         md->writesets[0].md.root = INVALID_WRITESET_ROOT;
 792         md->writesets[1].md.root = INVALID_WRITESET_ROOT;
 793         md->current_writeset = &md->writesets[0];
 794
 795         r = create_persistent_data_objects(md, may_format);
 796         if (r) {
 797                 kfree(md);
 798                 return ERR_PTR(r);
 799         }
 800
 801         return md;
 802 }
 803
 804 static void metadata_close(struct era_metadata *md)
 805 {
 806         destroy_persistent_data_objects(md);
 807         kfree(md);
 808 }
 809
 810 static bool valid_nr_blocks(dm_block_t n)
 811 {
 812         /*
 813          * dm_bitset restricts us to 2^32.  test_bit & co. restrict us
 814          * further to 2^31 - 1
 815          */
 816         return n < (1ull << 31);
 817 }
 818
 819 static int metadata_resize(struct era_metadata *md, void *arg)
 820 {
 821         int r;
 822         dm_block_t *new_size = arg;
 823         __le32 value;
 824
 825         if (!valid_nr_blocks(*new_size)) {
 826                 DMERR("Invalid number of origin blocks %llu",
 827                       (unsigned long long) *new_size);
 828                 return -EINVAL;
 829         }
 830
 831         writeset_free(&md->writesets[0]);
 832         writeset_free(&md->writesets[1]);
 833
 834         r = writeset_alloc(&md->writesets[0], *new_size);
 835         if (r) {
 836                 DMERR("%s: writeset_alloc failed for writeset 0", __func__);
 837                 return r;
 838         }
 839
 840         r = writeset_alloc(&md->writesets[1], *new_size);
 841         if (r) {
 842                 DMERR("%s: writeset_alloc failed for writeset 1", __func__);
 843                 return r;
 844         }
 845
 846         value = cpu_to_le32(0u);
 847         __dm_bless_for_disk(&value);
 848         r = dm_array_resize(&md->era_array_info, md->era_array_root,
 849                             md->nr_blocks, *new_size,
 850                             &value, &md->era_array_root);
 851         if (r) {
 852                 DMERR("%s: dm_array_resize failed", __func__);
 853                 return r;
 854         }
 855
 856         md->nr_blocks = *new_size;
 857         return 0;
 858 }
 859
 860 static int metadata_era_archive(struct era_metadata *md)
 861 {
 862         int r;
 863         uint64_t keys[1];
 864         struct writeset_disk value;
 865
 866         r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root,
 867                             &md->current_writeset->md.root);
 868         if (r) {
 869                 DMERR("%s: dm_bitset_flush failed", __func__);
 870                 return r;
 871         }
 872
 873         ws_pack(&md->current_writeset->md, &value);
 874         md->current_writeset->md.root = INVALID_WRITESET_ROOT;
 875
 876         keys[0] = md->current_era;
 877         __dm_bless_for_disk(&value);
 878         r = dm_btree_insert(&md->writeset_tree_info, md->writeset_tree_root,
 879                             keys, &value, &md->writeset_tree_root);
 880         if (r) {
 881                 DMERR("%s: couldn't insert writeset into btree", __func__);
 882                 /* FIXME: fail mode */
 883                 return r;
 884         }
 885
 886         md->archived_writesets = true;
 887
 888         return 0;
 889 }
 890
 891 static struct writeset *next_writeset(struct era_metadata *md)
 892 {
 893         return (md->current_writeset == &md->writesets[0]) ?
 894                 &md->writesets[1] : &md->writesets[0];
 895 }
 896
 897 static int metadata_new_era(struct era_metadata *md)
 898 {
 899         int r;
 900         struct writeset *new_writeset = next_writeset(md);
 901
 902         r = writeset_init(&md->bitset_info, new_writeset);
 903         if (r) {
 904                 DMERR("%s: writeset_init failed", __func__);
 905                 return r;
 906         }
 907
 908         swap_writeset(md, new_writeset);
 909         md->current_era++;
 910
 911         return 0;
 912 }
 913
 914 static int metadata_era_rollover(struct era_metadata *md)
 915 {
 916         int r;
 917
 918         if (md->current_writeset->md.root != INVALID_WRITESET_ROOT) {
 919                 r = metadata_era_archive(md);
 920                 if (r) {
 921                         DMERR("%s: metadata_archive_era failed", __func__);
 922                         /* FIXME: fail mode? */
 923                         return r;
 924                 }
 925         }
 926
 927         r = metadata_new_era(md);
 928         if (r) {
 929                 DMERR("%s: new era failed", __func__);
 930                 /* FIXME: fail mode */
 931                 return r;
 932         }
 933
 934         return 0;
 935 }
 936
 937 static bool metadata_current_marked(struct era_metadata *md, dm_block_t block)
 938 {
 939         bool r;
 940         struct writeset *ws;
 941
 942         rcu_read_lock();
 943         ws = rcu_dereference(md->current_writeset);
 944         r = writeset_marked(ws, block);
 945         rcu_read_unlock();
 946
 947         return r;
 948 }
 949
 950 static int metadata_commit(struct era_metadata *md)
 951 {
 952         int r;
 953         struct dm_block *sblock;
 954
 955         if (md->current_writeset->md.root != SUPERBLOCK_LOCATION) {
 956                 r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root,
 957                                     &md->current_writeset->md.root);
 958                 if (r) {
 959                         DMERR("%s: bitset flush failed", __func__);
 960                         return r;
 961                 }
 962         }
 963
 964         r = save_sm_root(md);
 965         if (r) {
 966                 DMERR("%s: save_sm_root failed", __func__);
 967                 return r;
 968         }
 969
 970         r = dm_tm_pre_commit(md->tm);
 971         if (r) {
 972                 DMERR("%s: pre commit failed", __func__);
 973                 return r;
 974         }
 975
 976         r = superblock_lock(md, &sblock);
 977         if (r) {
 978                 DMERR("%s: superblock lock failed", __func__);
 979                 return r;
 980         }
 981
 982         prepare_superblock(md, dm_block_data(sblock));
 983
 984         return dm_tm_commit(md->tm, sblock);
 985 }
 986
 987 static int metadata_checkpoint(struct era_metadata *md)
 988 {
 989         /*
 990          * For now we just rollover, but later I want to put a check in to
 991          * avoid this if the filter is still pretty fresh.
 992          */
 993         return metadata_era_rollover(md);
 994 }
 995
 996 /*
 997  * Metadata snapshots allow userland to access era data.
 998  */
 999 static int metadata_take_snap(struct era_metadata *md)
1000 {
1001         int r, inc;
1002         struct dm_block *clone;
1003
1004         if (md->metadata_snap != SUPERBLOCK_LOCATION) {
1005                 DMERR("%s: metadata snapshot already exists", __func__);
1006                 return -EINVAL;
1007         }
1008
1009         r = metadata_era_rollover(md);
1010         if (r) {
1011                 DMERR("%s: era rollover failed", __func__);
1012                 return r;
1013         }
1014
1015         r = metadata_commit(md);
1016         if (r) {
1017                 DMERR("%s: pre commit failed", __func__);
1018                 return r;
1019         }
1020
1021         r = dm_sm_inc_block(md->sm, SUPERBLOCK_LOCATION);
1022         if (r) {
1023                 DMERR("%s: couldn't increment superblock", __func__);
1024                 return r;
1025         }
1026
1027         r = dm_tm_shadow_block(md->tm, SUPERBLOCK_LOCATION,
1028                                &sb_validator, &clone, &inc);
1029         if (r) {
1030                 DMERR("%s: couldn't shadow superblock", __func__);
1031                 dm_sm_dec_block(md->sm, SUPERBLOCK_LOCATION);
1032                 return r;
1033         }
1034         BUG_ON(!inc);
1035
1036         r = dm_sm_inc_block(md->sm, md->writeset_tree_root);
1037         if (r) {
1038                 DMERR("%s: couldn't inc writeset tree root", __func__);
1039                 dm_tm_unlock(md->tm, clone);
1040                 return r;
1041         }
1042
1043         r = dm_sm_inc_block(md->sm, md->era_array_root);
1044         if (r) {
1045                 DMERR("%s: couldn't inc era tree root", __func__);
1046                 dm_sm_dec_block(md->sm, md->writeset_tree_root);
1047                 dm_tm_unlock(md->tm, clone);
1048                 return r;
1049         }
1050
1051         md->metadata_snap = dm_block_location(clone);
1052
1053         dm_tm_unlock(md->tm, clone);
1054
1055         return 0;
1056 }
1057
1058 static int metadata_drop_snap(struct era_metadata *md)
1059 {
1060         int r;
1061         dm_block_t location;
1062         struct dm_block *clone;
1063         struct superblock_disk *disk;
1064
1065         if (md->metadata_snap == SUPERBLOCK_LOCATION) {
1066                 DMERR("%s: no snap to drop", __func__);
1067                 return -EINVAL;
1068         }
1069
1070         r = dm_tm_read_lock(md->tm, md->metadata_snap, &sb_validator, &clone);
1071         if (r) {
1072                 DMERR("%s: couldn't read lock superblock clone", __func__);
1073                 return r;
1074         }
1075
1076         /*
1077          * Whatever happens now we'll commit with no record of the metadata
1078          * snap.
1079          */
1080         md->metadata_snap = SUPERBLOCK_LOCATION;
1081
1082         disk = dm_block_data(clone);
1083         r = dm_btree_del(&md->writeset_tree_info,
1084                          le64_to_cpu(disk->writeset_tree_root));
1085         if (r) {
1086                 DMERR("%s: error deleting writeset tree clone", __func__);
1087                 dm_tm_unlock(md->tm, clone);
1088                 return r;
1089         }
1090
1091         r = dm_array_del(&md->era_array_info, le64_to_cpu(disk->era_array_root));
1092         if (r) {
1093                 DMERR("%s: error deleting era array clone", __func__);
1094                 dm_tm_unlock(md->tm, clone);
1095                 return r;
1096         }
1097
1098         location = dm_block_location(clone);
1099         dm_tm_unlock(md->tm, clone);
1100
1101         return dm_sm_dec_block(md->sm, location);
1102 }
1103
1104 struct metadata_stats {
1105         dm_block_t used;
1106         dm_block_t total;
1107         dm_block_t snap;
1108         uint32_t era;
1109 };
1110
1111 static int metadata_get_stats(struct era_metadata *md, void *ptr)
1112 {
1113         int r;
1114         struct metadata_stats *s = ptr;
1115         dm_block_t nr_free, nr_total;
1116
1117         r = dm_sm_get_nr_free(md->sm, &nr_free);
1118         if (r) {
1119                 DMERR("dm_sm_get_nr_free returned %d", r);
1120                 return r;
1121         }
1122
1123         r = dm_sm_get_nr_blocks(md->sm, &nr_total);
1124         if (r) {
1125                 DMERR("dm_pool_get_metadata_dev_size returned %d", r);
1126                 return r;
1127         }
1128
1129         s->used = nr_total - nr_free;
1130         s->total = nr_total;
1131         s->snap = md->metadata_snap;
1132         s->era = md->current_era;
1133
1134         return 0;
1135 }
1136
1137 /*----------------------------------------------------------------*/
1138
1139 struct era {
1140         struct dm_target *ti;
1141         struct dm_target_callbacks callbacks;
1142
1143         struct dm_dev *metadata_dev;
1144         struct dm_dev *origin_dev;
1145
1146         dm_block_t nr_blocks;
1147         uint32_t sectors_per_block;
1148         int sectors_per_block_shift;
1149         struct era_metadata *md;
1150
1151         struct workqueue_struct *wq;
1152         struct work_struct worker;
1153
1154         spinlock_t deferred_lock;
1155         struct bio_list deferred_bios;
1156
1157         spinlock_t rpc_lock;
1158         struct list_head rpc_calls;
1159
1160         struct digest digest;
1161         atomic_t suspended;
1162 };
1163
1164 struct rpc {
1165         struct list_head list;
1166
1167         int (*fn0)(struct era_metadata *);
1168         int (*fn1)(struct era_metadata *, void *);
1169         void *arg;
1170         int result;
1171
1172         struct completion complete;
1173 };
1174
1175 /*----------------------------------------------------------------
1176  * Remapping.
1177  *---------------------------------------------------------------*/
1178 static bool block_size_is_power_of_two(struct era *era)
1179 {
1180         return era->sectors_per_block_shift >= 0;
1181 }
1182
1183 static dm_block_t get_block(struct era *era, struct bio *bio)
1184 {
1185         sector_t block_nr = bio->bi_iter.bi_sector;
1186
1187         if (!block_size_is_power_of_two(era))
1188                 (void) sector_div(block_nr, era->sectors_per_block);
1189         else
1190                 block_nr >>= era->sectors_per_block_shift;
1191
1192         return block_nr;
1193 }
1194
1195 static void remap_to_origin(struct era *era, struct bio *bio)
1196 {
1197         bio->bi_bdev = era->origin_dev->bdev;
1198 }
1199
1200 /*----------------------------------------------------------------
1201  * Worker thread
1202  *--------------------------------------------------------------*/
1203 static void wake_worker(struct era *era)
1204 {
1205         if (!atomic_read(&era->suspended))
1206                 queue_work(era->wq, &era->worker);
1207 }
1208
1209 static void process_old_eras(struct era *era)
1210 {
1211         int r;
1212
1213         if (!era->digest.step)
1214                 return;
1215
1216         r = era->digest.step(era->md, &era->digest);
1217         if (r < 0) {
1218                 DMERR("%s: digest step failed, stopping digestion", __func__);
1219                 era->digest.step = NULL;
1220
1221         } else if (era->digest.step)
1222                 wake_worker(era);
1223 }
1224
1225 static void process_deferred_bios(struct era *era)
1226 {
1227         int r;
1228         struct bio_list deferred_bios, marked_bios;
1229         struct bio *bio;
1230         bool commit_needed = false;
1231         bool failed = false;
1232
1233         bio_list_init(&deferred_bios);
1234         bio_list_init(&marked_bios);
1235
1236         spin_lock(&era->deferred_lock);
1237         bio_list_merge(&deferred_bios, &era->deferred_bios);
1238         bio_list_init(&era->deferred_bios);
1239         spin_unlock(&era->deferred_lock);
1240
1241         while ((bio = bio_list_pop(&deferred_bios))) {
1242                 r = writeset_test_and_set(&era->md->bitset_info,
1243                                           era->md->current_writeset,
1244                                           get_block(era, bio));
1245                 if (r < 0) {
1246                         /*
1247                          * This is bad news, we need to rollback.
1248                          * FIXME: finish.
1249                          */
1250                         failed = true;
1251
1252                 } else if (r == 0)
1253                         commit_needed = true;
1254
1255                 bio_list_add(&marked_bios, bio);
1256         }
1257
1258         if (commit_needed) {
1259                 r = metadata_commit(era->md);
1260                 if (r)
1261                         failed = true;
1262         }
1263
1264         if (failed)
1265                 while ((bio = bio_list_pop(&marked_bios)))
1266                         bio_io_error(bio);
1267         else
1268                 while ((bio = bio_list_pop(&marked_bios)))
1269                         generic_make_request(bio);
1270 }
1271
1272 static void process_rpc_calls(struct era *era)
1273 {
1274         int r;
1275         bool need_commit = false;
1276         struct list_head calls;
1277         struct rpc *rpc, *tmp;
1278
1279         INIT_LIST_HEAD(&calls);
1280         spin_lock(&era->rpc_lock);
1281         list_splice_init(&era->rpc_calls, &calls);
1282         spin_unlock(&era->rpc_lock);
1283
1284         list_for_each_entry_safe(rpc, tmp, &calls, list) {
1285                 rpc->result = rpc->fn0 ? rpc->fn0(era->md) : rpc->fn1(era->md, rpc->arg);
1286                 need_commit = true;
1287         }
1288
1289         if (need_commit) {
1290                 r = metadata_commit(era->md);
1291                 if (r)
1292                         list_for_each_entry_safe(rpc, tmp, &calls, list)
1293                                 rpc->result = r;
1294         }
1295
1296         list_for_each_entry_safe(rpc, tmp, &calls, list)
1297                 complete(&rpc->complete);
1298 }
1299
1300 static void kick_off_digest(struct era *era)
1301 {
1302         if (era->md->archived_writesets) {
1303                 era->md->archived_writesets = false;
1304                 metadata_digest_start(era->md, &era->digest);
1305         }
1306 }
1307
1308 static void do_work(struct work_struct *ws)
1309 {
1310         struct era *era = container_of(ws, struct era, worker);
1311
1312         kick_off_digest(era);
1313         process_old_eras(era);
1314         process_deferred_bios(era);
1315         process_rpc_calls(era);
1316 }
1317
1318 static void defer_bio(struct era *era, struct bio *bio)
1319 {
1320         spin_lock(&era->deferred_lock);
1321         bio_list_add(&era->deferred_bios, bio);
1322         spin_unlock(&era->deferred_lock);
1323
1324         wake_worker(era);
1325 }
1326
1327 /*
1328  * Make an rpc call to the worker to change the metadata.
1329  */
1330 static int perform_rpc(struct era *era, struct rpc *rpc)
1331 {
1332         rpc->result = 0;
1333         init_completion(&rpc->complete);
1334
1335         spin_lock(&era->rpc_lock);
1336         list_add(&rpc->list, &era->rpc_calls);
1337         spin_unlock(&era->rpc_lock);
1338
1339         wake_worker(era);
1340         wait_for_completion(&rpc->complete);
1341
1342         return rpc->result;
1343 }
1344
1345 static int in_worker0(struct era *era, int (*fn)(struct era_metadata *))
1346 {
1347         struct rpc rpc;
1348         rpc.fn0 = fn;
1349         rpc.fn1 = NULL;
1350
1351         return perform_rpc(era, &rpc);
1352 }
1353
1354 static int in_worker1(struct era *era,
1355                       int (*fn)(struct era_metadata *, void *), void *arg)
1356 {
1357         struct rpc rpc;
1358         rpc.fn0 = NULL;
1359         rpc.fn1 = fn;
1360         rpc.arg = arg;
1361
1362         return perform_rpc(era, &rpc);
1363 }
1364
1365 static void start_worker(struct era *era)
1366 {
1367         atomic_set(&era->suspended, 0);
1368 }
1369
1370 static void stop_worker(struct era *era)
1371 {
1372         atomic_set(&era->suspended, 1);
1373         flush_workqueue(era->wq);
1374 }
1375
1376 /*----------------------------------------------------------------
1377  * Target methods
1378  *--------------------------------------------------------------*/
1379 static int dev_is_congested(struct dm_dev *dev, int bdi_bits)
1380 {
1381         struct request_queue *q = bdev_get_queue(dev->bdev);
1382         return bdi_congested(&q->backing_dev_info, bdi_bits);
1383 }
1384
1385 static int era_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
1386 {
1387         struct era *era = container_of(cb, struct era, callbacks);
1388         return dev_is_congested(era->origin_dev, bdi_bits);
1389 }
1390
1391 static void era_destroy(struct era *era)
1392 {
1393         if (era->md)
1394                 metadata_close(era->md);
1395
1396         if (era->wq)
1397                 destroy_workqueue(era->wq);
1398
1399         if (era->origin_dev)
1400                 dm_put_device(era->ti, era->origin_dev);
1401
1402         if (era->metadata_dev)
1403                 dm_put_device(era->ti, era->metadata_dev);
1404
1405         kfree(era);
1406 }
1407
1408 static dm_block_t calc_nr_blocks(struct era *era)
1409 {
1410         return dm_sector_div_up(era->ti->len, era->sectors_per_block);
1411 }
1412
1413 static bool valid_block_size(dm_block_t block_size)
1414 {
1415         bool greater_than_zero = block_size > 0;
1416         bool multiple_of_min_block_size = (block_size & (MIN_BLOCK_SIZE - 1)) == 0;
1417
1418         return greater_than_zero && multiple_of_min_block_size;
1419 }
1420
1421 /*
1422  * <metadata dev> <data dev> <data block size (sectors)>
1423  */
1424 static int era_ctr(struct dm_target *ti, unsigned argc, char **argv)
1425 {
1426         int r;
1427         char dummy;
1428         struct era *era;
1429         struct era_metadata *md;
1430
1431         if (argc != 3) {
1432                 ti->error = "Invalid argument count";
1433                 return -EINVAL;
1434         }
1435
1436         era = kzalloc(sizeof(*era), GFP_KERNEL);
1437         if (!era) {
1438                 ti->error = "Error allocating era structure";
1439                 return -ENOMEM;
1440         }
1441
1442         era->ti = ti;
1443
1444         r = dm_get_device(ti, argv[0], FMODE_READ | FMODE_WRITE, &era->metadata_dev);
1445         if (r) {
1446                 ti->error = "Error opening metadata device";
1447                 era_destroy(era);
1448                 return -EINVAL;
1449         }
1450
1451         r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &era->origin_dev);
1452         if (r) {
1453                 ti->error = "Error opening data device";
1454                 era_destroy(era);
1455                 return -EINVAL;
1456         }
1457
1458         r = sscanf(argv[2], "%u%c", &era->sectors_per_block, &dummy);
1459         if (r != 1) {
1460                 ti->error = "Error parsing block size";
1461                 era_destroy(era);
1462                 return -EINVAL;
1463         }
1464
1465         r = dm_set_target_max_io_len(ti, era->sectors_per_block);
1466         if (r) {
1467                 ti->error = "could not set max io len";
1468                 era_destroy(era);
1469                 return -EINVAL;
1470         }
1471
1472         if (!valid_block_size(era->sectors_per_block)) {
1473                 ti->error = "Invalid block size";
1474                 era_destroy(era);
1475                 return -EINVAL;
1476         }
1477         if (era->sectors_per_block & (era->sectors_per_block - 1))
1478                 era->sectors_per_block_shift = -1;
1479         else
1480                 era->sectors_per_block_shift = __ffs(era->sectors_per_block);
1481
1482         md = metadata_open(era->metadata_dev->bdev, era->sectors_per_block, true);
1483         if (IS_ERR(md)) {
1484                 ti->error = "Error reading metadata";
1485                 era_destroy(era);
1486                 return PTR_ERR(md);
1487         }
1488         era->md = md;
1489
1490         era->nr_blocks = calc_nr_blocks(era);
1491
1492         r = metadata_resize(era->md, &era->nr_blocks);
1493         if (r) {
1494                 ti->error = "couldn't resize metadata";
1495                 era_destroy(era);
1496                 return -ENOMEM;
1497         }
1498
1499         era->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
1500         if (!era->wq) {
1501                 ti->error = "could not create workqueue for metadata object";
1502                 era_destroy(era);
1503                 return -ENOMEM;
1504         }
1505         INIT_WORK(&era->worker, do_work);
1506
1507         spin_lock_init(&era->deferred_lock);
1508         bio_list_init(&era->deferred_bios);
1509
1510         spin_lock_init(&era->rpc_lock);
1511         INIT_LIST_HEAD(&era->rpc_calls);
1512
1513         ti->private = era;
1514         ti->num_flush_bios = 1;
1515         ti->flush_supported = true;
1516
1517         ti->num_discard_bios = 1;
1518         ti->discards_supported = true;
1519         era->callbacks.congested_fn = era_is_congested;
1520         dm_table_add_target_callbacks(ti->table, &era->callbacks);
1521
1522         return 0;
1523 }
1524
1525 static void era_dtr(struct dm_target *ti)
1526 {
1527         era_destroy(ti->private);
1528 }
1529
1530 static int era_map(struct dm_target *ti, struct bio *bio)
1531 {
1532         struct era *era = ti->private;
1533         dm_block_t block = get_block(era, bio);
1534
1535         /*
1536          * All bios get remapped to the origin device.  We do this now, but
1537          * it may not get issued until later.  Depending on whether the
1538          * block is marked in this era.
1539          */
1540         remap_to_origin(era, bio);
1541
1542         /*
1543          * REQ_PREFLUSH bios carry no data, so we're not interested in them.
1544          */
1545         if (!(bio->bi_rw & REQ_PREFLUSH) &&
1546             (bio_data_dir(bio) == WRITE) &&
1547             !metadata_current_marked(era->md, block)) {
1548                 defer_bio(era, bio);
1549                 return DM_MAPIO_SUBMITTED;
1550         }
1551
1552         return DM_MAPIO_REMAPPED;
1553 }
1554
1555 static void era_postsuspend(struct dm_target *ti)
1556 {
1557         int r;
1558         struct era *era = ti->private;
1559
1560         r = in_worker0(era, metadata_era_archive);
1561         if (r) {
1562                 DMERR("%s: couldn't archive current era", __func__);
1563                 /* FIXME: fail mode */
1564         }
1565
1566         stop_worker(era);
1567 }
1568
1569 static int era_preresume(struct dm_target *ti)
1570 {
1571         int r;
1572         struct era *era = ti->private;
1573         dm_block_t new_size = calc_nr_blocks(era);
1574
1575         if (era->nr_blocks != new_size) {
1576                 r = in_worker1(era, metadata_resize, &new_size);
1577                 if (r)
1578                         return r;
1579
1580                 era->nr_blocks = new_size;
1581         }
1582
1583         start_worker(era);
1584
1585         r = in_worker0(era, metadata_new_era);
1586         if (r) {
1587                 DMERR("%s: metadata_era_rollover failed", __func__);
1588                 return r;
1589         }
1590
1591         return 0;
1592 }
1593
1594 /*
1595  * Status format:
1596  *
1597  * <metadata block size> <#used metadata blocks>/<#total metadata blocks>
1598  * <current era> <held metadata root | '-'>
1599  */
1600 static void era_status(struct dm_target *ti, status_type_t type,
1601                        unsigned status_flags, char *result, unsigned maxlen)
1602 {
1603         int r;
1604         struct era *era = ti->private;
1605         ssize_t sz = 0;
1606         struct metadata_stats stats;
1607         char buf[BDEVNAME_SIZE];
1608
1609         switch (type) {
1610         case STATUSTYPE_INFO:
1611                 r = in_worker1(era, metadata_get_stats, &stats);
1612                 if (r)
1613                         goto err;
1614
1615                 DMEMIT("%u %llu/%llu %u",
1616                        (unsigned) (DM_ERA_METADATA_BLOCK_SIZE >> SECTOR_SHIFT),
1617                        (unsigned long long) stats.used,
1618                        (unsigned long long) stats.total,
1619                        (unsigned) stats.era);
1620
1621                 if (stats.snap != SUPERBLOCK_LOCATION)
1622                         DMEMIT(" %llu", stats.snap);
1623                 else
1624                         DMEMIT(" -");
1625                 break;
1626
1627         case STATUSTYPE_TABLE:
1628                 format_dev_t(buf, era->metadata_dev->bdev->bd_dev);
1629                 DMEMIT("%s ", buf);
1630                 format_dev_t(buf, era->origin_dev->bdev->bd_dev);
1631                 DMEMIT("%s %u", buf, era->sectors_per_block);
1632                 break;
1633         }
1634
1635         return;
1636
1637 err:
1638         DMEMIT("Error");
1639 }
1640
1641 static int era_message(struct dm_target *ti, unsigned argc, char **argv)
1642 {
1643         struct era *era = ti->private;
1644
1645         if (argc != 1) {
1646                 DMERR("incorrect number of message arguments");
1647                 return -EINVAL;
1648         }
1649
1650         if (!strcasecmp(argv[0], "checkpoint"))
1651                 return in_worker0(era, metadata_checkpoint);
1652
1653         if (!strcasecmp(argv[0], "take_metadata_snap"))
1654                 return in_worker0(era, metadata_take_snap);
1655
1656         if (!strcasecmp(argv[0], "drop_metadata_snap"))
1657                 return in_worker0(era, metadata_drop_snap);
1658
1659         DMERR("unsupported message '%s'", argv[0]);
1660         return -EINVAL;
1661 }
1662
1663 static sector_t get_dev_size(struct dm_dev *dev)
1664 {
1665         return i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT;
1666 }
1667
1668 static int era_iterate_devices(struct dm_target *ti,
1669                                iterate_devices_callout_fn fn, void *data)
1670 {
1671         struct era *era = ti->private;
1672         return fn(ti, era->origin_dev, 0, get_dev_size(era->origin_dev), data);
1673 }
1674
1675 static void era_io_hints(struct dm_target *ti, struct queue_limits *limits)
1676 {
1677         struct era *era = ti->private;
1678         uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
1679
1680         /*
1681          * If the system-determined stacked limits are compatible with the
1682          * era device's blocksize (io_opt is a factor) do not override them.
1683          */
1684         if (io_opt_sectors < era->sectors_per_block ||
1685             do_div(io_opt_sectors, era->sectors_per_block)) {
1686                 blk_limits_io_min(limits, 0);
1687                 blk_limits_io_opt(limits, era->sectors_per_block << SECTOR_SHIFT);
1688         }
1689 }
1690
1691 /*----------------------------------------------------------------*/
1692
1693 static struct target_type era_target = {
1694         .name = "era",
1695         .version = {1, 0, 0},
1696         .module = THIS_MODULE,
1697         .ctr = era_ctr,
1698         .dtr = era_dtr,
1699         .map = era_map,
1700         .postsuspend = era_postsuspend,
1701         .preresume = era_preresume,
1702         .status = era_status,
1703         .message = era_message,
1704         .iterate_devices = era_iterate_devices,
1705         .io_hints = era_io_hints
1706 };
1707
1708 static int __init dm_era_init(void)
1709 {
1710         int r;
1711
1712         r = dm_register_target(&era_target);
1713         if (r) {
1714                 DMERR("era target registration failed: %d", r);
1715                 return r;
1716         }
1717
1718         return 0;
1719 }
1720
1721 static void __exit dm_era_exit(void)
1722 {
1723         dm_unregister_target(&era_target);
1724 }
1725
1726 module_init(dm_era_init);
1727 module_exit(dm_era_exit);
1728
1729 MODULE_DESCRIPTION(DM_NAME " era target");
1730 MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>");
1731 MODULE_LICENSE("GPL");