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         ws->bits = NULL;
  50 }
  51
  52 static int setup_on_disk_bitset(struct dm_disk_bitset *info,
  53                                 unsigned nr_bits, dm_block_t *root)
  54 {
  55         int r;
  56
  57         r = dm_bitset_empty(info, root);
  58         if (r)
  59                 return r;
  60
  61         return dm_bitset_resize(info, *root, 0, nr_bits, false, root);
  62 }
  63
  64 static size_t bitset_size(unsigned nr_bits)
  65 {
  66         return sizeof(unsigned long) * dm_div_up(nr_bits, BITS_PER_LONG);
  67 }
  68
  69 /*
  70  * Allocates memory for the in core bitset.
  71  */
  72 static int writeset_alloc(struct writeset *ws, dm_block_t nr_blocks)
  73 {
  74         ws->bits = vzalloc(bitset_size(nr_blocks));
  75         if (!ws->bits) {
  76                 DMERR("%s: couldn't allocate in memory bitset", __func__);
  77                 return -ENOMEM;
  78         }
  79
  80         return 0;
  81 }
  82
  83 /*
  84  * Wipes the in-core bitset, and creates a new on disk bitset.
  85  */
  86 static int writeset_init(struct dm_disk_bitset *info, struct writeset *ws,
  87                          dm_block_t nr_blocks)
  88 {
  89         int r;
  90
  91         memset(ws->bits, 0, bitset_size(nr_blocks));
  92
  93         ws->md.nr_bits = nr_blocks;
  94         r = setup_on_disk_bitset(info, ws->md.nr_bits, &ws->md.root);
  95         if (r) {
  96                 DMERR("%s: setup_on_disk_bitset failed", __func__);
  97                 return r;
  98         }
  99
 100         return 0;
 101 }
 102
 103 static bool writeset_marked(struct writeset *ws, dm_block_t block)
 104 {
 105         return test_bit(block, ws->bits);
 106 }
 107
 108 static int writeset_marked_on_disk(struct dm_disk_bitset *info,
 109                                    struct writeset_metadata *m, dm_block_t block,
 110                                    bool *result)
 111 {
 112         dm_block_t old = m->root;
 113
 114         /*
 115          * The bitset was flushed when it was archived, so we know there'll
 116          * be no change to the root.
 117          */
 118         int r = dm_bitset_test_bit(info, m->root, block, &m->root, result);
 119         if (r) {
 120                 DMERR("%s: dm_bitset_test_bit failed", __func__);
 121                 return r;
 122         }
 123
 124         BUG_ON(m->root != old);
 125
 126         return r;
 127 }
 128
 129 /*
 130  * Returns < 0 on error, 0 if the bit wasn't previously set, 1 if it was.
 131  */
 132 static int writeset_test_and_set(struct dm_disk_bitset *info,
 133                                  struct writeset *ws, uint32_t block)
 134 {
 135         int r;
 136
 137         if (!test_bit(block, ws->bits)) {
 138                 r = dm_bitset_set_bit(info, ws->md.root, block, &ws->md.root);
 139                 if (r) {
 140                         /* FIXME: fail mode */
 141                         return r;
 142                 }
 143
 144                 return 0;
 145         }
 146
 147         return 1;
 148 }
 149
 150 /*----------------------------------------------------------------
 151  * On disk metadata layout
 152  *--------------------------------------------------------------*/
 153 #define SPACE_MAP_ROOT_SIZE 128
 154 #define UUID_LEN 16
 155
 156 struct writeset_disk {
 157         __le32 nr_bits;
 158         __le64 root;
 159 } __packed;
 160
 161 struct superblock_disk {
 162         __le32 csum;
 163         __le32 flags;
 164         __le64 blocknr;
 165
 166         __u8 uuid[UUID_LEN];
 167         __le64 magic;
 168         __le32 version;
 169
 170         __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
 171
 172         __le32 data_block_size;
 173         __le32 metadata_block_size;
 174         __le32 nr_blocks;
 175
 176         __le32 current_era;
 177         struct writeset_disk current_writeset;
 178
 179         /*
 180          * Only these two fields are valid within the metadata snapshot.
 181          */
 182         __le64 writeset_tree_root;
 183         __le64 era_array_root;
 184
 185         __le64 metadata_snap;
 186 } __packed;
 187
 188 /*----------------------------------------------------------------
 189  * Superblock validation
 190  *--------------------------------------------------------------*/
 191 static void sb_prepare_for_write(struct dm_block_validator *v,
 192                                  struct dm_block *b,
 193                                  size_t sb_block_size)
 194 {
 195         struct superblock_disk *disk = dm_block_data(b);
 196
 197         disk->blocknr = cpu_to_le64(dm_block_location(b));
 198         disk->csum = cpu_to_le32(dm_bm_checksum(&disk->flags,
 199                                                 sb_block_size - sizeof(__le32),
 200                                                 SUPERBLOCK_CSUM_XOR));
 201 }
 202
 203 static int check_metadata_version(struct superblock_disk *disk)
 204 {
 205         uint32_t metadata_version = le32_to_cpu(disk->version);
 206         if (metadata_version < MIN_ERA_VERSION || metadata_version > MAX_ERA_VERSION) {
 207                 DMERR("Era metadata version %u found, but only versions between %u and %u supported.",
 208                       metadata_version, MIN_ERA_VERSION, MAX_ERA_VERSION);
 209                 return -EINVAL;
 210         }
 211
 212         return 0;
 213 }
 214
 215 static int sb_check(struct dm_block_validator *v,
 216                     struct dm_block *b,
 217                     size_t sb_block_size)
 218 {
 219         struct superblock_disk *disk = dm_block_data(b);
 220         __le32 csum_le;
 221
 222         if (dm_block_location(b) != le64_to_cpu(disk->blocknr)) {
 223                 DMERR("sb_check failed: blocknr %llu: wanted %llu",
 224                       le64_to_cpu(disk->blocknr),
 225                       (unsigned long long)dm_block_location(b));
 226                 return -ENOTBLK;
 227         }
 228
 229         if (le64_to_cpu(disk->magic) != SUPERBLOCK_MAGIC) {
 230                 DMERR("sb_check failed: magic %llu: wanted %llu",
 231                       le64_to_cpu(disk->magic),
 232                       (unsigned long long) SUPERBLOCK_MAGIC);
 233                 return -EILSEQ;
 234         }
 235
 236         csum_le = cpu_to_le32(dm_bm_checksum(&disk->flags,
 237                                              sb_block_size - sizeof(__le32),
 238                                              SUPERBLOCK_CSUM_XOR));
 239         if (csum_le != disk->csum) {
 240                 DMERR("sb_check failed: csum %u: wanted %u",
 241                       le32_to_cpu(csum_le), le32_to_cpu(disk->csum));
 242                 return -EILSEQ;
 243         }
 244
 245         return check_metadata_version(disk);
 246 }
 247
 248 static struct dm_block_validator sb_validator = {
 249         .name = "superblock",
 250         .prepare_for_write = sb_prepare_for_write,
 251         .check = sb_check
 252 };
 253
 254 /*----------------------------------------------------------------
 255  * Low level metadata handling
 256  *--------------------------------------------------------------*/
 257 #define DM_ERA_METADATA_BLOCK_SIZE 4096
 258 #define DM_ERA_METADATA_CACHE_SIZE 64
 259 #define ERA_MAX_CONCURRENT_LOCKS 5
 260
 261 struct era_metadata {
 262         struct block_device *bdev;
 263         struct dm_block_manager *bm;
 264         struct dm_space_map *sm;
 265         struct dm_transaction_manager *tm;
 266
 267         dm_block_t block_size;
 268         uint32_t nr_blocks;
 269
 270         uint32_t current_era;
 271
 272         /*
 273          * We preallocate 2 writesets.  When an era rolls over we
 274          * switch between them. This means the allocation is done at
 275          * preresume time, rather than on the io path.
 276          */
 277         struct writeset writesets[2];
 278         struct writeset *current_writeset;
 279
 280         dm_block_t writeset_tree_root;
 281         dm_block_t era_array_root;
 282
 283         struct dm_disk_bitset bitset_info;
 284         struct dm_btree_info writeset_tree_info;
 285         struct dm_array_info era_array_info;
 286
 287         dm_block_t metadata_snap;
 288
 289         /*
 290          * A flag that is set whenever a writeset has been archived.
 291          */
 292         bool archived_writesets;
 293
 294         /*
 295          * Reading the space map root can fail, so we read it into this
 296          * buffer before the superblock is locked and updated.
 297          */
 298         __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
 299 };
 300
 301 static int superblock_read_lock(struct era_metadata *md,
 302                                 struct dm_block **sblock)
 303 {
 304         return dm_bm_read_lock(md->bm, SUPERBLOCK_LOCATION,
 305                                &sb_validator, sblock);
 306 }
 307
 308 static int superblock_lock_zero(struct era_metadata *md,
 309                                 struct dm_block **sblock)
 310 {
 311         return dm_bm_write_lock_zero(md->bm, SUPERBLOCK_LOCATION,
 312                                      &sb_validator, sblock);
 313 }
 314
 315 static int superblock_lock(struct era_metadata *md,
 316                            struct dm_block **sblock)
 317 {
 318         return dm_bm_write_lock(md->bm, SUPERBLOCK_LOCATION,
 319                                 &sb_validator, sblock);
 320 }
 321
 322 /* FIXME: duplication with cache and thin */
 323 static int superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
 324 {
 325         int r;
 326         unsigned i;
 327         struct dm_block *b;
 328         __le64 *data_le, zero = cpu_to_le64(0);
 329         unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
 330
 331         /*
 332          * We can't use a validator here - it may be all zeroes.
 333          */
 334         r = dm_bm_read_lock(bm, SUPERBLOCK_LOCATION, NULL, &b);
 335         if (r)
 336                 return r;
 337
 338         data_le = dm_block_data(b);
 339         *result = true;
 340         for (i = 0; i < sb_block_size; i++) {
 341                 if (data_le[i] != zero) {
 342                         *result = false;
 343                         break;
 344                 }
 345         }
 346
 347         dm_bm_unlock(b);
 348
 349         return 0;
 350 }
 351
 352 /*----------------------------------------------------------------*/
 353
 354 static void ws_pack(const struct writeset_metadata *core, struct writeset_disk *disk)
 355 {
 356         disk->nr_bits = cpu_to_le32(core->nr_bits);
 357         disk->root = cpu_to_le64(core->root);
 358 }
 359
 360 static void ws_unpack(const struct writeset_disk *disk, struct writeset_metadata *core)
 361 {
 362         core->nr_bits = le32_to_cpu(disk->nr_bits);
 363         core->root = le64_to_cpu(disk->root);
 364 }
 365
 366 static void ws_inc(void *context, const void *value)
 367 {
 368         struct era_metadata *md = context;
 369         struct writeset_disk ws_d;
 370         dm_block_t b;
 371
 372         memcpy(&ws_d, value, sizeof(ws_d));
 373         b = le64_to_cpu(ws_d.root);
 374
 375         dm_tm_inc(md->tm, b);
 376 }
 377
 378 static void ws_dec(void *context, const void *value)
 379 {
 380         struct era_metadata *md = context;
 381         struct writeset_disk ws_d;
 382         dm_block_t b;
 383
 384         memcpy(&ws_d, value, sizeof(ws_d));
 385         b = le64_to_cpu(ws_d.root);
 386
 387         dm_bitset_del(&md->bitset_info, b);
 388 }
 389
 390 static int ws_eq(void *context, const void *value1, const void *value2)
 391 {
 392         return !memcmp(value1, value2, sizeof(struct writeset_disk));
 393 }
 394
 395 /*----------------------------------------------------------------*/
 396
 397 static void setup_writeset_tree_info(struct era_metadata *md)
 398 {
 399         struct dm_btree_value_type *vt = &md->writeset_tree_info.value_type;
 400         md->writeset_tree_info.tm = md->tm;
 401         md->writeset_tree_info.levels = 1;
 402         vt->context = md;
 403         vt->size = sizeof(struct writeset_disk);
 404         vt->inc = ws_inc;
 405         vt->dec = ws_dec;
 406         vt->equal = ws_eq;
 407 }
 408
 409 static void setup_era_array_info(struct era_metadata *md)
 410
 411 {
 412         struct dm_btree_value_type vt;
 413         vt.context = NULL;
 414         vt.size = sizeof(__le32);
 415         vt.inc = NULL;
 416         vt.dec = NULL;
 417         vt.equal = NULL;
 418
 419         dm_array_info_init(&md->era_array_info, md->tm, &vt);
 420 }
 421
 422 static void setup_infos(struct era_metadata *md)
 423 {
 424         dm_disk_bitset_init(md->tm, &md->bitset_info);
 425         setup_writeset_tree_info(md);
 426         setup_era_array_info(md);
 427 }
 428
 429 /*----------------------------------------------------------------*/
 430
 431 static int create_fresh_metadata(struct era_metadata *md)
 432 {
 433         int r;
 434
 435         r = dm_tm_create_with_sm(md->bm, SUPERBLOCK_LOCATION,
 436                                  &md->tm, &md->sm);
 437         if (r < 0) {
 438                 DMERR("dm_tm_create_with_sm failed");
 439                 return r;
 440         }
 441
 442         setup_infos(md);
 443
 444         r = dm_btree_empty(&md->writeset_tree_info, &md->writeset_tree_root);
 445         if (r) {
 446                 DMERR("couldn't create new writeset tree");
 447                 goto bad;
 448         }
 449
 450         r = dm_array_empty(&md->era_array_info, &md->era_array_root);
 451         if (r) {
 452                 DMERR("couldn't create era array");
 453                 goto bad;
 454         }
 455
 456         return 0;
 457
 458 bad:
 459         dm_sm_destroy(md->sm);
 460         dm_tm_destroy(md->tm);
 461
 462         return r;
 463 }
 464
 465 static int save_sm_root(struct era_metadata *md)
 466 {
 467         int r;
 468         size_t metadata_len;
 469
 470         r = dm_sm_root_size(md->sm, &metadata_len);
 471         if (r < 0)
 472                 return r;
 473
 474         return dm_sm_copy_root(md->sm, &md->metadata_space_map_root,
 475                                metadata_len);
 476 }
 477
 478 static void copy_sm_root(struct era_metadata *md, struct superblock_disk *disk)
 479 {
 480         memcpy(&disk->metadata_space_map_root,
 481                &md->metadata_space_map_root,
 482                sizeof(md->metadata_space_map_root));
 483 }
 484
 485 /*
 486  * Writes a superblock, including the static fields that don't get updated
 487  * with every commit (possible optimisation here).  'md' should be fully
 488  * constructed when this is called.
 489  */
 490 static void prepare_superblock(struct era_metadata *md, struct superblock_disk *disk)
 491 {
 492         disk->magic = cpu_to_le64(SUPERBLOCK_MAGIC);
 493         disk->flags = cpu_to_le32(0ul);
 494
 495         /* FIXME: can't keep blanking the uuid (uuid is currently unused though) */
 496         memset(disk->uuid, 0, sizeof(disk->uuid));
 497         disk->version = cpu_to_le32(MAX_ERA_VERSION);
 498
 499         copy_sm_root(md, disk);
 500
 501         disk->data_block_size = cpu_to_le32(md->block_size);
 502         disk->metadata_block_size = cpu_to_le32(DM_ERA_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
 503         disk->nr_blocks = cpu_to_le32(md->nr_blocks);
 504         disk->current_era = cpu_to_le32(md->current_era);
 505
 506         ws_pack(&md->current_writeset->md, &disk->current_writeset);
 507         disk->writeset_tree_root = cpu_to_le64(md->writeset_tree_root);
 508         disk->era_array_root = cpu_to_le64(md->era_array_root);
 509         disk->metadata_snap = cpu_to_le64(md->metadata_snap);
 510 }
 511
 512 static int write_superblock(struct era_metadata *md)
 513 {
 514         int r;
 515         struct dm_block *sblock;
 516         struct superblock_disk *disk;
 517
 518         r = save_sm_root(md);
 519         if (r) {
 520                 DMERR("%s: save_sm_root failed", __func__);
 521                 return r;
 522         }
 523
 524         r = superblock_lock_zero(md, &sblock);
 525         if (r)
 526                 return r;
 527
 528         disk = dm_block_data(sblock);
 529         prepare_superblock(md, disk);
 530
 531         return dm_tm_commit(md->tm, sblock);
 532 }
 533
 534 /*
 535  * Assumes block_size and the infos are set.
 536  */
 537 static int format_metadata(struct era_metadata *md)
 538 {
 539         int r;
 540
 541         r = create_fresh_metadata(md);
 542         if (r)
 543                 return r;
 544
 545         r = write_superblock(md);
 546         if (r) {
 547                 dm_sm_destroy(md->sm);
 548                 dm_tm_destroy(md->tm);
 549                 return r;
 550         }
 551
 552         return 0;
 553 }
 554
 555 static int open_metadata(struct era_metadata *md)
 556 {
 557         int r;
 558         struct dm_block *sblock;
 559         struct superblock_disk *disk;
 560
 561         r = superblock_read_lock(md, &sblock);
 562         if (r) {
 563                 DMERR("couldn't read_lock superblock");
 564                 return r;
 565         }
 566
 567         disk = dm_block_data(sblock);
 568
 569         /* Verify the data block size hasn't changed */
 570         if (le32_to_cpu(disk->data_block_size) != md->block_size) {
 571                 DMERR("changing the data block size (from %u to %llu) is not supported",
 572                       le32_to_cpu(disk->data_block_size), md->block_size);
 573                 r = -EINVAL;
 574                 goto bad;
 575         }
 576
 577         r = dm_tm_open_with_sm(md->bm, SUPERBLOCK_LOCATION,
 578                                disk->metadata_space_map_root,
 579                                sizeof(disk->metadata_space_map_root),
 580                                &md->tm, &md->sm);
 581         if (r) {
 582                 DMERR("dm_tm_open_with_sm failed");
 583                 goto bad;
 584         }
 585
 586         setup_infos(md);
 587
 588         md->nr_blocks = le32_to_cpu(disk->nr_blocks);
 589         md->current_era = le32_to_cpu(disk->current_era);
 590
 591         ws_unpack(&disk->current_writeset, &md->current_writeset->md);
 592         md->writeset_tree_root = le64_to_cpu(disk->writeset_tree_root);
 593         md->era_array_root = le64_to_cpu(disk->era_array_root);
 594         md->metadata_snap = le64_to_cpu(disk->metadata_snap);
 595         md->archived_writesets = true;
 596
 597         dm_bm_unlock(sblock);
 598
 599         return 0;
 600
 601 bad:
 602         dm_bm_unlock(sblock);
 603         return r;
 604 }
 605
 606 static int open_or_format_metadata(struct era_metadata *md,
 607                                    bool may_format)
 608 {
 609         int r;
 610         bool unformatted = false;
 611
 612         r = superblock_all_zeroes(md->bm, &unformatted);
 613         if (r)
 614                 return r;
 615
 616         if (unformatted)
 617                 return may_format ? format_metadata(md) : -EPERM;
 618
 619         return open_metadata(md);
 620 }
 621
 622 static int create_persistent_data_objects(struct era_metadata *md,
 623                                           bool may_format)
 624 {
 625         int r;
 626
 627         md->bm = dm_block_manager_create(md->bdev, DM_ERA_METADATA_BLOCK_SIZE,
 628                                          DM_ERA_METADATA_CACHE_SIZE,
 629                                          ERA_MAX_CONCURRENT_LOCKS);
 630         if (IS_ERR(md->bm)) {
 631                 DMERR("could not create block manager");
 632                 return PTR_ERR(md->bm);
 633         }
 634
 635         r = open_or_format_metadata(md, may_format);
 636         if (r)
 637                 dm_block_manager_destroy(md->bm);
 638
 639         return r;
 640 }
 641
 642 static void destroy_persistent_data_objects(struct era_metadata *md)
 643 {
 644         dm_sm_destroy(md->sm);
 645         dm_tm_destroy(md->tm);
 646         dm_block_manager_destroy(md->bm);
 647 }
 648
 649 /*
 650  * This waits until all era_map threads have picked up the new filter.
 651  */
 652 static void swap_writeset(struct era_metadata *md, struct writeset *new_writeset)
 653 {
 654         rcu_assign_pointer(md->current_writeset, new_writeset);
 655         synchronize_rcu();
 656 }
 657
 658 /*----------------------------------------------------------------
 659  * Writesets get 'digested' into the main era array.
 660  *
 661  * We're using a coroutine here so the worker thread can do the digestion,
 662  * thus avoiding synchronisation of the metadata.  Digesting a whole
 663  * writeset in one go would cause too much latency.
 664  *--------------------------------------------------------------*/
 665 struct digest {
 666         uint32_t era;
 667         unsigned nr_bits, current_bit;
 668         struct writeset_metadata writeset;
 669         __le32 value;
 670         struct dm_disk_bitset info;
 671
 672         int (*step)(struct era_metadata *, struct digest *);
 673 };
 674
 675 static int metadata_digest_lookup_writeset(struct era_metadata *md,
 676                                            struct digest *d);
 677
 678 static int metadata_digest_remove_writeset(struct era_metadata *md,
 679                                            struct digest *d)
 680 {
 681         int r;
 682         uint64_t key = d->era;
 683
 684         r = dm_btree_remove(&md->writeset_tree_info, md->writeset_tree_root,
 685                             &key, &md->writeset_tree_root);
 686         if (r) {
 687                 DMERR("%s: dm_btree_remove failed", __func__);
 688                 return r;
 689         }
 690
 691         d->step = metadata_digest_lookup_writeset;
 692         return 0;
 693 }
 694
 695 #define INSERTS_PER_STEP 100
 696
 697 static int metadata_digest_transcribe_writeset(struct era_metadata *md,
 698                                                struct digest *d)
 699 {
 700         int r;
 701         bool marked;
 702         unsigned b, e = min(d->current_bit + INSERTS_PER_STEP, d->nr_bits);
 703
 704         for (b = d->current_bit; b < e; b++) {
 705                 r = writeset_marked_on_disk(&d->info, &d->writeset, b, &marked);
 706                 if (r) {
 707                         DMERR("%s: writeset_marked_on_disk failed", __func__);
 708                         return r;
 709                 }
 710
 711                 if (!marked)
 712                         continue;
 713
 714                 __dm_bless_for_disk(&d->value);
 715                 r = dm_array_set_value(&md->era_array_info, md->era_array_root,
 716                                        b, &d->value, &md->era_array_root);
 717                 if (r) {
 718                         DMERR("%s: dm_array_set_value failed", __func__);
 719                         return r;
 720                 }
 721         }
 722
 723         if (b == d->nr_bits)
 724                 d->step = metadata_digest_remove_writeset;
 725         else
 726                 d->current_bit = b;
 727
 728         return 0;
 729 }
 730
 731 static int metadata_digest_lookup_writeset(struct era_metadata *md,
 732                                            struct digest *d)
 733 {
 734         int r;
 735         uint64_t key;
 736         struct writeset_disk disk;
 737
 738         r = dm_btree_find_lowest_key(&md->writeset_tree_info,
 739                                      md->writeset_tree_root, &key);
 740         if (r < 0)
 741                 return r;
 742
 743         d->era = key;
 744
 745         r = dm_btree_lookup(&md->writeset_tree_info,
 746                             md->writeset_tree_root, &key, &disk);
 747         if (r) {
 748                 if (r == -ENODATA) {
 749                         d->step = NULL;
 750                         return 0;
 751                 }
 752
 753                 DMERR("%s: dm_btree_lookup failed", __func__);
 754                 return r;
 755         }
 756
 757         ws_unpack(&disk, &d->writeset);
 758         d->value = cpu_to_le32(key);
 759
 760         /*
 761          * We initialise another bitset info to avoid any caching side effects
 762          * with the previous one.
 763          */
 764         dm_disk_bitset_init(md->tm, &d->info);
 765
 766         d->nr_bits = min(d->writeset.nr_bits, md->nr_blocks);
 767         d->current_bit = 0;
 768         d->step = metadata_digest_transcribe_writeset;
 769
 770         return 0;
 771 }
 772
 773 static int metadata_digest_start(struct era_metadata *md, struct digest *d)
 774 {
 775         if (d->step)
 776                 return 0;
 777
 778         memset(d, 0, sizeof(*d));
 779         d->step = metadata_digest_lookup_writeset;
 780
 781         return 0;
 782 }
 783
 784 /*----------------------------------------------------------------
 785  * High level metadata interface.  Target methods should use these, and not
 786  * the lower level ones.
 787  *--------------------------------------------------------------*/
 788 static struct era_metadata *metadata_open(struct block_device *bdev,
 789                                           sector_t block_size,
 790                                           bool may_format)
 791 {
 792         int r;
 793         struct era_metadata *md = kzalloc(sizeof(*md), GFP_KERNEL);
 794
 795         if (!md)
 796                 return NULL;
 797
 798         md->bdev = bdev;
 799         md->block_size = block_size;
 800
 801         md->writesets[0].md.root = INVALID_WRITESET_ROOT;
 802         md->writesets[1].md.root = INVALID_WRITESET_ROOT;
 803         md->current_writeset = &md->writesets[0];
 804
 805         r = create_persistent_data_objects(md, may_format);
 806         if (r) {
 807                 kfree(md);
 808                 return ERR_PTR(r);
 809         }
 810
 811         return md;
 812 }
 813
 814 static void metadata_close(struct era_metadata *md)
 815 {
 816         writeset_free(&md->writesets[0]);
 817         writeset_free(&md->writesets[1]);
 818         destroy_persistent_data_objects(md);
 819         kfree(md);
 820 }
 821
 822 static bool valid_nr_blocks(dm_block_t n)
 823 {
 824         /*
 825          * dm_bitset restricts us to 2^32.  test_bit & co. restrict us
 826          * further to 2^31 - 1
 827          */
 828         return n < (1ull << 31);
 829 }
 830
 831 static int metadata_resize(struct era_metadata *md, void *arg)
 832 {
 833         int r;
 834         dm_block_t *new_size = arg;
 835         __le32 value;
 836
 837         if (!valid_nr_blocks(*new_size)) {
 838                 DMERR("Invalid number of origin blocks %llu",
 839                       (unsigned long long) *new_size);
 840                 return -EINVAL;
 841         }
 842
 843         writeset_free(&md->writesets[0]);
 844         writeset_free(&md->writesets[1]);
 845
 846         r = writeset_alloc(&md->writesets[0], *new_size);
 847         if (r) {
 848                 DMERR("%s: writeset_alloc failed for writeset 0", __func__);
 849                 return r;
 850         }
 851
 852         r = writeset_alloc(&md->writesets[1], *new_size);
 853         if (r) {
 854                 DMERR("%s: writeset_alloc failed for writeset 1", __func__);
 855                 writeset_free(&md->writesets[0]);
 856                 return r;
 857         }
 858
 859         value = cpu_to_le32(0u);
 860         __dm_bless_for_disk(&value);
 861         r = dm_array_resize(&md->era_array_info, md->era_array_root,
 862                             md->nr_blocks, *new_size,
 863                             &value, &md->era_array_root);
 864         if (r) {
 865                 DMERR("%s: dm_array_resize failed", __func__);
 866                 writeset_free(&md->writesets[0]);
 867                 writeset_free(&md->writesets[1]);
 868                 return r;
 869         }
 870
 871         md->nr_blocks = *new_size;
 872         return 0;
 873 }
 874
 875 static int metadata_era_archive(struct era_metadata *md)
 876 {
 877         int r;
 878         uint64_t keys[1];
 879         struct writeset_disk value;
 880
 881         r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root,
 882                             &md->current_writeset->md.root);
 883         if (r) {
 884                 DMERR("%s: dm_bitset_flush failed", __func__);
 885                 return r;
 886         }
 887
 888         ws_pack(&md->current_writeset->md, &value);
 889
 890         keys[0] = md->current_era;
 891         __dm_bless_for_disk(&value);
 892         r = dm_btree_insert(&md->writeset_tree_info, md->writeset_tree_root,
 893                             keys, &value, &md->writeset_tree_root);
 894         if (r) {
 895                 DMERR("%s: couldn't insert writeset into btree", __func__);
 896                 /* FIXME: fail mode */
 897                 return r;
 898         }
 899
 900         md->current_writeset->md.root = INVALID_WRITESET_ROOT;
 901         md->archived_writesets = true;
 902
 903         return 0;
 904 }
 905
 906 static struct writeset *next_writeset(struct era_metadata *md)
 907 {
 908         return (md->current_writeset == &md->writesets[0]) ?
 909                 &md->writesets[1] : &md->writesets[0];
 910 }
 911
 912 static int metadata_new_era(struct era_metadata *md)
 913 {
 914         int r;
 915         struct writeset *new_writeset = next_writeset(md);
 916
 917         r = writeset_init(&md->bitset_info, new_writeset, md->nr_blocks);
 918         if (r) {
 919                 DMERR("%s: writeset_init failed", __func__);
 920                 return r;
 921         }
 922
 923         swap_writeset(md, new_writeset);
 924         md->current_era++;
 925
 926         return 0;
 927 }
 928
 929 static int metadata_era_rollover(struct era_metadata *md)
 930 {
 931         int r;
 932
 933         if (md->current_writeset->md.root != INVALID_WRITESET_ROOT) {
 934                 r = metadata_era_archive(md);
 935                 if (r) {
 936                         DMERR("%s: metadata_archive_era failed", __func__);
 937                         /* FIXME: fail mode? */
 938                         return r;
 939                 }
 940         }
 941
 942         r = metadata_new_era(md);
 943         if (r) {
 944                 DMERR("%s: new era failed", __func__);
 945                 /* FIXME: fail mode */
 946                 return r;
 947         }
 948
 949         return 0;
 950 }
 951
 952 static bool metadata_current_marked(struct era_metadata *md, dm_block_t block)
 953 {
 954         bool r;
 955         struct writeset *ws;
 956
 957         rcu_read_lock();
 958         ws = rcu_dereference(md->current_writeset);
 959         r = writeset_marked(ws, block);
 960         rcu_read_unlock();
 961
 962         return r;
 963 }
 964
 965 static int metadata_commit(struct era_metadata *md)
 966 {
 967         int r;
 968         struct dm_block *sblock;
 969
 970         if (md->current_writeset->md.root != INVALID_WRITESET_ROOT) {
 971                 r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root,
 972                                     &md->current_writeset->md.root);
 973                 if (r) {
 974                         DMERR("%s: bitset flush failed", __func__);
 975                         return r;
 976                 }
 977         }
 978
 979         r = dm_tm_pre_commit(md->tm);
 980         if (r) {
 981                 DMERR("%s: pre commit failed", __func__);
 982                 return r;
 983         }
 984
 985         r = save_sm_root(md);
 986         if (r) {
 987                 DMERR("%s: save_sm_root failed", __func__);
 988                 return r;
 989         }
 990
 991         r = superblock_lock(md, &sblock);
 992         if (r) {
 993                 DMERR("%s: superblock lock failed", __func__);
 994                 return r;
 995         }
 996
 997         prepare_superblock(md, dm_block_data(sblock));
 998
 999         return dm_tm_commit(md->tm, sblock);
1000 }
1001
1002 static int metadata_checkpoint(struct era_metadata *md)
1003 {
1004         /*
1005          * For now we just rollover, but later I want to put a check in to
1006          * avoid this if the filter is still pretty fresh.
1007          */
1008         return metadata_era_rollover(md);
1009 }
1010
1011 /*
1012  * Metadata snapshots allow userland to access era data.
1013  */
1014 static int metadata_take_snap(struct era_metadata *md)
1015 {
1016         int r, inc;
1017         struct dm_block *clone;
1018
1019         if (md->metadata_snap != SUPERBLOCK_LOCATION) {
1020                 DMERR("%s: metadata snapshot already exists", __func__);
1021                 return -EINVAL;
1022         }
1023
1024         r = metadata_era_rollover(md);
1025         if (r) {
1026                 DMERR("%s: era rollover failed", __func__);
1027                 return r;
1028         }
1029
1030         r = metadata_commit(md);
1031         if (r) {
1032                 DMERR("%s: pre commit failed", __func__);
1033                 return r;
1034         }
1035
1036         r = dm_sm_inc_block(md->sm, SUPERBLOCK_LOCATION);
1037         if (r) {
1038                 DMERR("%s: couldn't increment superblock", __func__);
1039                 return r;
1040         }
1041
1042         r = dm_tm_shadow_block(md->tm, SUPERBLOCK_LOCATION,
1043                                &sb_validator, &clone, &inc);
1044         if (r) {
1045                 DMERR("%s: couldn't shadow superblock", __func__);
1046                 dm_sm_dec_block(md->sm, SUPERBLOCK_LOCATION);
1047                 return r;
1048         }
1049         BUG_ON(!inc);
1050
1051         r = dm_sm_inc_block(md->sm, md->writeset_tree_root);
1052         if (r) {
1053                 DMERR("%s: couldn't inc writeset tree root", __func__);
1054                 dm_tm_unlock(md->tm, clone);
1055                 return r;
1056         }
1057
1058         r = dm_sm_inc_block(md->sm, md->era_array_root);
1059         if (r) {
1060                 DMERR("%s: couldn't inc era tree root", __func__);
1061                 dm_sm_dec_block(md->sm, md->writeset_tree_root);
1062                 dm_tm_unlock(md->tm, clone);
1063                 return r;
1064         }
1065
1066         md->metadata_snap = dm_block_location(clone);
1067
1068         dm_tm_unlock(md->tm, clone);
1069
1070         return 0;
1071 }
1072
1073 static int metadata_drop_snap(struct era_metadata *md)
1074 {
1075         int r;
1076         dm_block_t location;
1077         struct dm_block *clone;
1078         struct superblock_disk *disk;
1079
1080         if (md->metadata_snap == SUPERBLOCK_LOCATION) {
1081                 DMERR("%s: no snap to drop", __func__);
1082                 return -EINVAL;
1083         }
1084
1085         r = dm_tm_read_lock(md->tm, md->metadata_snap, &sb_validator, &clone);
1086         if (r) {
1087                 DMERR("%s: couldn't read lock superblock clone", __func__);
1088                 return r;
1089         }
1090
1091         /*
1092          * Whatever happens now we'll commit with no record of the metadata
1093          * snap.
1094          */
1095         md->metadata_snap = SUPERBLOCK_LOCATION;
1096
1097         disk = dm_block_data(clone);
1098         r = dm_btree_del(&md->writeset_tree_info,
1099                          le64_to_cpu(disk->writeset_tree_root));
1100         if (r) {
1101                 DMERR("%s: error deleting writeset tree clone", __func__);
1102                 dm_tm_unlock(md->tm, clone);
1103                 return r;
1104         }
1105
1106         r = dm_array_del(&md->era_array_info, le64_to_cpu(disk->era_array_root));
1107         if (r) {
1108                 DMERR("%s: error deleting era array clone", __func__);
1109                 dm_tm_unlock(md->tm, clone);
1110                 return r;
1111         }
1112
1113         location = dm_block_location(clone);
1114         dm_tm_unlock(md->tm, clone);
1115
1116         return dm_sm_dec_block(md->sm, location);
1117 }
1118
1119 struct metadata_stats {
1120         dm_block_t used;
1121         dm_block_t total;
1122         dm_block_t snap;
1123         uint32_t era;
1124 };
1125
1126 static int metadata_get_stats(struct era_metadata *md, void *ptr)
1127 {
1128         int r;
1129         struct metadata_stats *s = ptr;
1130         dm_block_t nr_free, nr_total;
1131
1132         r = dm_sm_get_nr_free(md->sm, &nr_free);
1133         if (r) {
1134                 DMERR("dm_sm_get_nr_free returned %d", r);
1135                 return r;
1136         }
1137
1138         r = dm_sm_get_nr_blocks(md->sm, &nr_total);
1139         if (r) {
1140                 DMERR("dm_pool_get_metadata_dev_size returned %d", r);
1141                 return r;
1142         }
1143
1144         s->used = nr_total - nr_free;
1145         s->total = nr_total;
1146         s->snap = md->metadata_snap;
1147         s->era = md->current_era;
1148
1149         return 0;
1150 }
1151
1152 /*----------------------------------------------------------------*/
1153
1154 struct era {
1155         struct dm_target *ti;
1156         struct dm_target_callbacks callbacks;
1157
1158         struct dm_dev *metadata_dev;
1159         struct dm_dev *origin_dev;
1160
1161         dm_block_t nr_blocks;
1162         uint32_t sectors_per_block;
1163         int sectors_per_block_shift;
1164         struct era_metadata *md;
1165
1166         struct workqueue_struct *wq;
1167         struct work_struct worker;
1168
1169         spinlock_t deferred_lock;
1170         struct bio_list deferred_bios;
1171
1172         spinlock_t rpc_lock;
1173         struct list_head rpc_calls;
1174
1175         struct digest digest;
1176         atomic_t suspended;
1177 };
1178
1179 struct rpc {
1180         struct list_head list;
1181
1182         int (*fn0)(struct era_metadata *);
1183         int (*fn1)(struct era_metadata *, void *);
1184         void *arg;
1185         int result;
1186
1187         struct completion complete;
1188 };
1189
1190 /*----------------------------------------------------------------
1191  * Remapping.
1192  *---------------------------------------------------------------*/
1193 static bool block_size_is_power_of_two(struct era *era)
1194 {
1195         return era->sectors_per_block_shift >= 0;
1196 }
1197
1198 static dm_block_t get_block(struct era *era, struct bio *bio)
1199 {
1200         sector_t block_nr = bio->bi_iter.bi_sector;
1201
1202         if (!block_size_is_power_of_two(era))
1203                 (void) sector_div(block_nr, era->sectors_per_block);
1204         else
1205                 block_nr >>= era->sectors_per_block_shift;
1206
1207         return block_nr;
1208 }
1209
1210 static void remap_to_origin(struct era *era, struct bio *bio)
1211 {
1212         bio->bi_bdev = era->origin_dev->bdev;
1213 }
1214
1215 /*----------------------------------------------------------------
1216  * Worker thread
1217  *--------------------------------------------------------------*/
1218 static void wake_worker(struct era *era)
1219 {
1220         if (!atomic_read(&era->suspended))
1221                 queue_work(era->wq, &era->worker);
1222 }
1223
1224 static void process_old_eras(struct era *era)
1225 {
1226         int r;
1227
1228         if (!era->digest.step)
1229                 return;
1230
1231         r = era->digest.step(era->md, &era->digest);
1232         if (r < 0) {
1233                 DMERR("%s: digest step failed, stopping digestion", __func__);
1234                 era->digest.step = NULL;
1235
1236         } else if (era->digest.step)
1237                 wake_worker(era);
1238 }
1239
1240 static void process_deferred_bios(struct era *era)
1241 {
1242         int r;
1243         struct bio_list deferred_bios, marked_bios;
1244         struct bio *bio;
1245         struct blk_plug plug;
1246         bool commit_needed = false;
1247         bool failed = false;
1248         struct writeset *ws = era->md->current_writeset;
1249
1250         bio_list_init(&deferred_bios);
1251         bio_list_init(&marked_bios);
1252
1253         spin_lock(&era->deferred_lock);
1254         bio_list_merge(&deferred_bios, &era->deferred_bios);
1255         bio_list_init(&era->deferred_bios);
1256         spin_unlock(&era->deferred_lock);
1257
1258         if (bio_list_empty(&deferred_bios))
1259                 return;
1260
1261         while ((bio = bio_list_pop(&deferred_bios))) {
1262                 r = writeset_test_and_set(&era->md->bitset_info, ws,
1263                                           get_block(era, bio));
1264                 if (r < 0) {
1265                         /*
1266                          * This is bad news, we need to rollback.
1267                          * FIXME: finish.
1268                          */
1269                         failed = true;
1270                 } else if (r == 0)
1271                         commit_needed = true;
1272
1273                 bio_list_add(&marked_bios, bio);
1274         }
1275
1276         if (commit_needed) {
1277                 r = metadata_commit(era->md);
1278                 if (r)
1279                         failed = true;
1280         }
1281
1282         if (failed)
1283                 while ((bio = bio_list_pop(&marked_bios)))
1284                         bio_io_error(bio);
1285         else {
1286                 blk_start_plug(&plug);
1287                 while ((bio = bio_list_pop(&marked_bios))) {
1288                         /*
1289                          * Only update the in-core writeset if the on-disk one
1290                          * was updated too.
1291                          */
1292                         if (commit_needed)
1293                                 set_bit(get_block(era, bio), ws->bits);
1294                         generic_make_request(bio);
1295                 }
1296                 blk_finish_plug(&plug);
1297         }
1298 }
1299
1300 static void process_rpc_calls(struct era *era)
1301 {
1302         int r;
1303         bool need_commit = false;
1304         struct list_head calls;
1305         struct rpc *rpc, *tmp;
1306
1307         INIT_LIST_HEAD(&calls);
1308         spin_lock(&era->rpc_lock);
1309         list_splice_init(&era->rpc_calls, &calls);
1310         spin_unlock(&era->rpc_lock);
1311
1312         list_for_each_entry_safe(rpc, tmp, &calls, list) {
1313                 rpc->result = rpc->fn0 ? rpc->fn0(era->md) : rpc->fn1(era->md, rpc->arg);
1314                 need_commit = true;
1315         }
1316
1317         if (need_commit) {
1318                 r = metadata_commit(era->md);
1319                 if (r)
1320                         list_for_each_entry_safe(rpc, tmp, &calls, list)
1321                                 rpc->result = r;
1322         }
1323
1324         list_for_each_entry_safe(rpc, tmp, &calls, list)
1325                 complete(&rpc->complete);
1326 }
1327
1328 static void kick_off_digest(struct era *era)
1329 {
1330         if (era->md->archived_writesets) {
1331                 era->md->archived_writesets = false;
1332                 metadata_digest_start(era->md, &era->digest);
1333         }
1334 }
1335
1336 static void do_work(struct work_struct *ws)
1337 {
1338         struct era *era = container_of(ws, struct era, worker);
1339
1340         kick_off_digest(era);
1341         process_old_eras(era);
1342         process_deferred_bios(era);
1343         process_rpc_calls(era);
1344 }
1345
1346 static void defer_bio(struct era *era, struct bio *bio)
1347 {
1348         spin_lock(&era->deferred_lock);
1349         bio_list_add(&era->deferred_bios, bio);
1350         spin_unlock(&era->deferred_lock);
1351
1352         wake_worker(era);
1353 }
1354
1355 /*
1356  * Make an rpc call to the worker to change the metadata.
1357  */
1358 static int perform_rpc(struct era *era, struct rpc *rpc)
1359 {
1360         rpc->result = 0;
1361         init_completion(&rpc->complete);
1362
1363         spin_lock(&era->rpc_lock);
1364         list_add(&rpc->list, &era->rpc_calls);
1365         spin_unlock(&era->rpc_lock);
1366
1367         wake_worker(era);
1368         wait_for_completion(&rpc->complete);
1369
1370         return rpc->result;
1371 }
1372
1373 static int in_worker0(struct era *era, int (*fn)(struct era_metadata *))
1374 {
1375         struct rpc rpc;
1376         rpc.fn0 = fn;
1377         rpc.fn1 = NULL;
1378
1379         return perform_rpc(era, &rpc);
1380 }
1381
1382 static int in_worker1(struct era *era,
1383                       int (*fn)(struct era_metadata *, void *), void *arg)
1384 {
1385         struct rpc rpc;
1386         rpc.fn0 = NULL;
1387         rpc.fn1 = fn;
1388         rpc.arg = arg;
1389
1390         return perform_rpc(era, &rpc);
1391 }
1392
1393 static void start_worker(struct era *era)
1394 {
1395         atomic_set(&era->suspended, 0);
1396 }
1397
1398 static void stop_worker(struct era *era)
1399 {
1400         atomic_set(&era->suspended, 1);
1401         flush_workqueue(era->wq);
1402 }
1403
1404 /*----------------------------------------------------------------
1405  * Target methods
1406  *--------------------------------------------------------------*/
1407 static int dev_is_congested(struct dm_dev *dev, int bdi_bits)
1408 {
1409         struct request_queue *q = bdev_get_queue(dev->bdev);
1410         return bdi_congested(&q->backing_dev_info, bdi_bits);
1411 }
1412
1413 static int era_is_congested(struct dm_target_callbacks *cb, int bdi_bits)
1414 {
1415         struct era *era = container_of(cb, struct era, callbacks);
1416         return dev_is_congested(era->origin_dev, bdi_bits);
1417 }
1418
1419 static void era_destroy(struct era *era)
1420 {
1421         if (era->md)
1422                 metadata_close(era->md);
1423
1424         if (era->wq)
1425                 destroy_workqueue(era->wq);
1426
1427         if (era->origin_dev)
1428                 dm_put_device(era->ti, era->origin_dev);
1429
1430         if (era->metadata_dev)
1431                 dm_put_device(era->ti, era->metadata_dev);
1432
1433         kfree(era);
1434 }
1435
1436 static dm_block_t calc_nr_blocks(struct era *era)
1437 {
1438         return dm_sector_div_up(era->ti->len, era->sectors_per_block);
1439 }
1440
1441 static bool valid_block_size(dm_block_t block_size)
1442 {
1443         bool greater_than_zero = block_size > 0;
1444         bool multiple_of_min_block_size = (block_size & (MIN_BLOCK_SIZE - 1)) == 0;
1445
1446         return greater_than_zero && multiple_of_min_block_size;
1447 }
1448
1449 /*
1450  * <metadata dev> <data dev> <data block size (sectors)>
1451  */
1452 static int era_ctr(struct dm_target *ti, unsigned argc, char **argv)
1453 {
1454         int r;
1455         char dummy;
1456         struct era *era;
1457         struct era_metadata *md;
1458
1459         if (argc != 3) {
1460                 ti->error = "Invalid argument count";
1461                 return -EINVAL;
1462         }
1463
1464         era = kzalloc(sizeof(*era), GFP_KERNEL);
1465         if (!era) {
1466                 ti->error = "Error allocating era structure";
1467                 return -ENOMEM;
1468         }
1469
1470         era->ti = ti;
1471
1472         r = dm_get_device(ti, argv[0], FMODE_READ | FMODE_WRITE, &era->metadata_dev);
1473         if (r) {
1474                 ti->error = "Error opening metadata device";
1475                 era_destroy(era);
1476                 return -EINVAL;
1477         }
1478
1479         r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &era->origin_dev);
1480         if (r) {
1481                 ti->error = "Error opening data device";
1482                 era_destroy(era);
1483                 return -EINVAL;
1484         }
1485
1486         r = sscanf(argv[2], "%u%c", &era->sectors_per_block, &dummy);
1487         if (r != 1) {
1488                 ti->error = "Error parsing block size";
1489                 era_destroy(era);
1490                 return -EINVAL;
1491         }
1492
1493         r = dm_set_target_max_io_len(ti, era->sectors_per_block);
1494         if (r) {
1495                 ti->error = "could not set max io len";
1496                 era_destroy(era);
1497                 return -EINVAL;
1498         }
1499
1500         if (!valid_block_size(era->sectors_per_block)) {
1501                 ti->error = "Invalid block size";
1502                 era_destroy(era);
1503                 return -EINVAL;
1504         }
1505         if (era->sectors_per_block & (era->sectors_per_block - 1))
1506                 era->sectors_per_block_shift = -1;
1507         else
1508                 era->sectors_per_block_shift = __ffs(era->sectors_per_block);
1509
1510         md = metadata_open(era->metadata_dev->bdev, era->sectors_per_block, true);
1511         if (IS_ERR(md)) {
1512                 ti->error = "Error reading metadata";
1513                 era_destroy(era);
1514                 return PTR_ERR(md);
1515         }
1516         era->md = md;
1517
1518         era->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
1519         if (!era->wq) {
1520                 ti->error = "could not create workqueue for metadata object";
1521                 era_destroy(era);
1522                 return -ENOMEM;
1523         }
1524         INIT_WORK(&era->worker, do_work);
1525
1526         spin_lock_init(&era->deferred_lock);
1527         bio_list_init(&era->deferred_bios);
1528
1529         spin_lock_init(&era->rpc_lock);
1530         INIT_LIST_HEAD(&era->rpc_calls);
1531
1532         ti->private = era;
1533         ti->num_flush_bios = 1;
1534         ti->flush_supported = true;
1535
1536         ti->num_discard_bios = 1;
1537         ti->discards_supported = true;
1538         era->callbacks.congested_fn = era_is_congested;
1539         dm_table_add_target_callbacks(ti->table, &era->callbacks);
1540
1541         return 0;
1542 }
1543
1544 static void era_dtr(struct dm_target *ti)
1545 {
1546         era_destroy(ti->private);
1547 }
1548
1549 static int era_map(struct dm_target *ti, struct bio *bio)
1550 {
1551         struct era *era = ti->private;
1552         dm_block_t block = get_block(era, bio);
1553
1554         /*
1555          * All bios get remapped to the origin device.  We do this now, but
1556          * it may not get issued until later.  Depending on whether the
1557          * block is marked in this era.
1558          */
1559         remap_to_origin(era, bio);
1560
1561         /*
1562          * REQ_FLUSH bios carry no data, so we're not interested in them.
1563          */
1564         if (!(bio->bi_rw & REQ_FLUSH) &&
1565             (bio_data_dir(bio) == WRITE) &&
1566             !metadata_current_marked(era->md, block)) {
1567                 defer_bio(era, bio);
1568                 return DM_MAPIO_SUBMITTED;
1569         }
1570
1571         return DM_MAPIO_REMAPPED;
1572 }
1573
1574 static void era_postsuspend(struct dm_target *ti)
1575 {
1576         int r;
1577         struct era *era = ti->private;
1578
1579         r = in_worker0(era, metadata_era_archive);
1580         if (r) {
1581                 DMERR("%s: couldn't archive current era", __func__);
1582                 /* FIXME: fail mode */
1583         }
1584
1585         stop_worker(era);
1586 }
1587
1588 static int era_preresume(struct dm_target *ti)
1589 {
1590         int r;
1591         struct era *era = ti->private;
1592         dm_block_t new_size = calc_nr_blocks(era);
1593
1594         if (era->nr_blocks != new_size) {
1595                 r = metadata_resize(era->md, &new_size);
1596                 if (r) {
1597                         DMERR("%s: metadata_resize failed", __func__);
1598                         return r;
1599                 }
1600
1601                 r = metadata_commit(era->md);
1602                 if (r) {
1603                         DMERR("%s: metadata_commit failed", __func__);
1604                         return r;
1605                 }
1606
1607                 era->nr_blocks = new_size;
1608         }
1609
1610         start_worker(era);
1611
1612         r = in_worker0(era, metadata_era_rollover);
1613         if (r) {
1614                 DMERR("%s: metadata_era_rollover failed", __func__);
1615                 return r;
1616         }
1617
1618         return 0;
1619 }
1620
1621 /*
1622  * Status format:
1623  *
1624  * <metadata block size> <#used metadata blocks>/<#total metadata blocks>
1625  * <current era> <held metadata root | '-'>
1626  */
1627 static void era_status(struct dm_target *ti, status_type_t type,
1628                        unsigned status_flags, char *result, unsigned maxlen)
1629 {
1630         int r;
1631         struct era *era = ti->private;
1632         ssize_t sz = 0;
1633         struct metadata_stats stats;
1634         char buf[BDEVNAME_SIZE];
1635
1636         switch (type) {
1637         case STATUSTYPE_INFO:
1638                 r = in_worker1(era, metadata_get_stats, &stats);
1639                 if (r)
1640                         goto err;
1641
1642                 DMEMIT("%u %llu/%llu %u",
1643                        (unsigned) (DM_ERA_METADATA_BLOCK_SIZE >> SECTOR_SHIFT),
1644                        (unsigned long long) stats.used,
1645                        (unsigned long long) stats.total,
1646                        (unsigned) stats.era);
1647
1648                 if (stats.snap != SUPERBLOCK_LOCATION)
1649                         DMEMIT(" %llu", stats.snap);
1650                 else
1651                         DMEMIT(" -");
1652                 break;
1653
1654         case STATUSTYPE_TABLE:
1655                 format_dev_t(buf, era->metadata_dev->bdev->bd_dev);
1656                 DMEMIT("%s ", buf);
1657                 format_dev_t(buf, era->origin_dev->bdev->bd_dev);
1658                 DMEMIT("%s %u", buf, era->sectors_per_block);
1659                 break;
1660         }
1661
1662         return;
1663
1664 err:
1665         DMEMIT("Error");
1666 }
1667
1668 static int era_message(struct dm_target *ti, unsigned argc, char **argv)
1669 {
1670         struct era *era = ti->private;
1671
1672         if (argc != 1) {
1673                 DMERR("incorrect number of message arguments");
1674                 return -EINVAL;
1675         }
1676
1677         if (!strcasecmp(argv[0], "checkpoint"))
1678                 return in_worker0(era, metadata_checkpoint);
1679
1680         if (!strcasecmp(argv[0], "take_metadata_snap"))
1681                 return in_worker0(era, metadata_take_snap);
1682
1683         if (!strcasecmp(argv[0], "drop_metadata_snap"))
1684                 return in_worker0(era, metadata_drop_snap);
1685
1686         DMERR("unsupported message '%s'", argv[0]);
1687         return -EINVAL;
1688 }
1689
1690 static sector_t get_dev_size(struct dm_dev *dev)
1691 {
1692         return i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT;
1693 }
1694
1695 static int era_iterate_devices(struct dm_target *ti,
1696                                iterate_devices_callout_fn fn, void *data)
1697 {
1698         struct era *era = ti->private;
1699         return fn(ti, era->origin_dev, 0, get_dev_size(era->origin_dev), data);
1700 }
1701
1702 static void era_io_hints(struct dm_target *ti, struct queue_limits *limits)
1703 {
1704         struct era *era = ti->private;
1705         uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
1706
1707         /*
1708          * If the system-determined stacked limits are compatible with the
1709          * era device's blocksize (io_opt is a factor) do not override them.
1710          */
1711         if (io_opt_sectors < era->sectors_per_block ||
1712             do_div(io_opt_sectors, era->sectors_per_block)) {
1713                 blk_limits_io_min(limits, 0);
1714                 blk_limits_io_opt(limits, era->sectors_per_block << SECTOR_SHIFT);
1715         }
1716 }
1717
1718 /*----------------------------------------------------------------*/
1719
1720 static struct target_type era_target = {
1721         .name = "era",
1722         .version = {1, 0, 0},
1723         .module = THIS_MODULE,
1724         .ctr = era_ctr,
1725         .dtr = era_dtr,
1726         .map = era_map,
1727         .postsuspend = era_postsuspend,
1728         .preresume = era_preresume,
1729         .status = era_status,
1730         .message = era_message,
1731         .iterate_devices = era_iterate_devices,
1732         .io_hints = era_io_hints
1733 };
1734
1735 static int __init dm_era_init(void)
1736 {
1737         int r;
1738
1739         r = dm_register_target(&era_target);
1740         if (r) {
1741                 DMERR("era target registration failed: %d", r);
1742                 return r;
1743         }
1744
1745         return 0;
1746 }
1747
1748 static void __exit dm_era_exit(void)
1749 {
1750         dm_unregister_target(&era_target);
1751 }
1752
1753 module_init(dm_era_init);
1754 module_exit(dm_era_exit);
1755
1756 MODULE_DESCRIPTION(DM_NAME " era target");
1757 MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>");
1758 MODULE_LICENSE("GPL");