fs/bcachefs/recovery.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "backpointers.h"
   5 #include "bkey_buf.h"
   6 #include "alloc_background.h"
   7 #include "btree_gc.h"
   8 #include "btree_journal_iter.h"
   9 #include "btree_update.h"
  10 #include "btree_update_interior.h"
  11 #include "btree_io.h"
  12 #include "buckets.h"
  13 #include "dirent.h"
  14 #include "ec.h"
  15 #include "errcode.h"
  16 #include "error.h"
  17 #include "fs-common.h"
  18 #include "fsck.h"
  19 #include "journal_io.h"
  20 #include "journal_reclaim.h"
  21 #include "journal_seq_blacklist.h"
  22 #include "lru.h"
  23 #include "logged_ops.h"
  24 #include "move.h"
  25 #include "quota.h"
  26 #include "recovery.h"
  27 #include "replicas.h"
  28 #include "sb-clean.h"
  29 #include "snapshot.h"
  30 #include "subvolume.h"
  31 #include "super-io.h"
  32
  33 #include <linux/sort.h>
  34 #include <linux/stat.h>
  35
  36 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  37
  38 /* for -o reconstruct_alloc: */
  39 static void drop_alloc_keys(struct journal_keys *keys)
  40 {
  41         size_t src, dst;
  42
  43         for (src = 0, dst = 0; src < keys->nr; src++)
  44                 if (keys->d[src].btree_id != BTREE_ID_alloc)
  45                         keys->d[dst++] = keys->d[src];
  46
  47         keys->nr = dst;
  48 }
  49
  50 /*
  51  * Btree node pointers have a field to stack a pointer to the in memory btree
  52  * node; we need to zero out this field when reading in btree nodes, or when
  53  * reading in keys from the journal:
  54  */
  55 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
  56 {
  57         struct journal_key *i;
  58
  59         for (i = keys->d; i < keys->d + keys->nr; i++)
  60                 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
  61                         bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
  62 }
  63
  64 /* journal replay: */
  65
  66 static void replay_now_at(struct journal *j, u64 seq)
  67 {
  68         BUG_ON(seq < j->replay_journal_seq);
  69
  70         seq = min(seq, j->replay_journal_seq_end);
  71
  72         while (j->replay_journal_seq < seq)
  73                 bch2_journal_pin_put(j, j->replay_journal_seq++);
  74 }
  75
  76 static int bch2_journal_replay_key(struct btree_trans *trans,
  77                                    struct journal_key *k)
  78 {
  79         struct btree_iter iter;
  80         unsigned iter_flags =
  81                 BTREE_ITER_INTENT|
  82                 BTREE_ITER_NOT_EXTENTS;
  83         unsigned update_flags = BTREE_TRIGGER_NORUN;
  84         int ret;
  85
  86         /*
  87          * BTREE_UPDATE_KEY_CACHE_RECLAIM disables key cache lookup/update to
  88          * keep the key cache coherent with the underlying btree. Nothing
  89          * besides the allocator is doing updates yet so we don't need key cache
  90          * coherency for non-alloc btrees, and key cache fills for snapshots
  91          * btrees use BTREE_ITER_FILTER_SNAPSHOTS, which isn't available until
  92          * the snapshots recovery pass runs.
  93          */
  94         if (!k->level && k->btree_id == BTREE_ID_alloc)
  95                 iter_flags |= BTREE_ITER_CACHED;
  96         else
  97                 update_flags |= BTREE_UPDATE_KEY_CACHE_RECLAIM;
  98
  99         bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 100                                   BTREE_MAX_DEPTH, k->level,
 101                                   iter_flags);
 102         ret = bch2_btree_iter_traverse(&iter);
 103         if (ret)
 104                 goto out;
 105
 106         /* Must be checked with btree locked: */
 107         if (k->overwritten)
 108                 goto out;
 109
 110         ret = bch2_trans_update(trans, &iter, k->k, update_flags);
 111 out:
 112         bch2_trans_iter_exit(trans, &iter);
 113         return ret;
 114 }
 115
 116 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 117 {
 118         const struct journal_key *l = *((const struct journal_key **)_l);
 119         const struct journal_key *r = *((const struct journal_key **)_r);
 120
 121         return cmp_int(l->journal_seq, r->journal_seq);
 122 }
 123
 124 static int bch2_journal_replay(struct bch_fs *c)
 125 {
 126         struct journal_keys *keys = &c->journal_keys;
 127         struct journal_key **keys_sorted, *k;
 128         struct journal *j = &c->journal;
 129         u64 start_seq   = c->journal_replay_seq_start;
 130         u64 end_seq     = c->journal_replay_seq_start;
 131         size_t i;
 132         int ret;
 133
 134         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 135         keys->gap = keys->nr;
 136
 137         keys_sorted = kvmalloc_array(keys->nr, sizeof(*keys_sorted), GFP_KERNEL);
 138         if (!keys_sorted)
 139                 return -BCH_ERR_ENOMEM_journal_replay;
 140
 141         for (i = 0; i < keys->nr; i++)
 142                 keys_sorted[i] = &keys->d[i];
 143
 144         sort(keys_sorted, keys->nr,
 145              sizeof(keys_sorted[0]),
 146              journal_sort_seq_cmp, NULL);
 147
 148         if (keys->nr) {
 149                 ret = bch2_journal_log_msg(c, "Starting journal replay (%zu keys in entries %llu-%llu)",
 150                                            keys->nr, start_seq, end_seq);
 151                 if (ret)
 152                         goto err;
 153         }
 154
 155         for (i = 0; i < keys->nr; i++) {
 156                 k = keys_sorted[i];
 157
 158                 cond_resched();
 159
 160                 replay_now_at(j, k->journal_seq);
 161
 162                 ret = bch2_trans_do(c, NULL, NULL,
 163                                     BTREE_INSERT_LAZY_RW|
 164                                     BTREE_INSERT_NOFAIL|
 165                                     (!k->allocated
 166                                      ? BTREE_INSERT_JOURNAL_REPLAY|BCH_WATERMARK_reclaim
 167                                      : 0),
 168                              bch2_journal_replay_key(&trans, k));
 169                 if (ret) {
 170                         bch_err(c, "journal replay: error while replaying key at btree %s level %u: %s",
 171                                 bch2_btree_ids[k->btree_id], k->level, bch2_err_str(ret));
 172                         goto err;
 173                 }
 174         }
 175
 176         replay_now_at(j, j->replay_journal_seq_end);
 177         j->replay_journal_seq = 0;
 178
 179         bch2_journal_set_replay_done(j);
 180         bch2_journal_flush_all_pins(j);
 181         ret = bch2_journal_error(j);
 182
 183         if (keys->nr && !ret)
 184                 bch2_journal_log_msg(c, "journal replay finished");
 185 err:
 186         kvfree(keys_sorted);
 187
 188         if (ret)
 189                 bch_err_fn(c, ret);
 190         return ret;
 191 }
 192
 193 /* journal replay early: */
 194
 195 static int journal_replay_entry_early(struct bch_fs *c,
 196                                       struct jset_entry *entry)
 197 {
 198         int ret = 0;
 199
 200         switch (entry->type) {
 201         case BCH_JSET_ENTRY_btree_root: {
 202                 struct btree_root *r;
 203
 204                 while (entry->btree_id >= c->btree_roots_extra.nr + BTREE_ID_NR) {
 205                         ret = darray_push(&c->btree_roots_extra, (struct btree_root) { NULL });
 206                         if (ret)
 207                                 return ret;
 208                 }
 209
 210                 r = bch2_btree_id_root(c, entry->btree_id);
 211
 212                 if (entry->u64s) {
 213                         r->level = entry->level;
 214                         bkey_copy(&r->key, &entry->start[0]);
 215                         r->error = 0;
 216                 } else {
 217                         r->error = -EIO;
 218                 }
 219                 r->alive = true;
 220                 break;
 221         }
 222         case BCH_JSET_ENTRY_usage: {
 223                 struct jset_entry_usage *u =
 224                         container_of(entry, struct jset_entry_usage, entry);
 225
 226                 switch (entry->btree_id) {
 227                 case BCH_FS_USAGE_reserved:
 228                         if (entry->level < BCH_REPLICAS_MAX)
 229                                 c->usage_base->persistent_reserved[entry->level] =
 230                                         le64_to_cpu(u->v);
 231                         break;
 232                 case BCH_FS_USAGE_inodes:
 233                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 234                         break;
 235                 case BCH_FS_USAGE_key_version:
 236                         atomic64_set(&c->key_version,
 237                                      le64_to_cpu(u->v));
 238                         break;
 239                 }
 240
 241                 break;
 242         }
 243         case BCH_JSET_ENTRY_data_usage: {
 244                 struct jset_entry_data_usage *u =
 245                         container_of(entry, struct jset_entry_data_usage, entry);
 246
 247                 ret = bch2_replicas_set_usage(c, &u->r,
 248                                               le64_to_cpu(u->v));
 249                 break;
 250         }
 251         case BCH_JSET_ENTRY_dev_usage: {
 252                 struct jset_entry_dev_usage *u =
 253                         container_of(entry, struct jset_entry_dev_usage, entry);
 254                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 255                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 256
 257                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 258
 259                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 260                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 261                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 262                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 263                 }
 264
 265                 break;
 266         }
 267         case BCH_JSET_ENTRY_blacklist: {
 268                 struct jset_entry_blacklist *bl_entry =
 269                         container_of(entry, struct jset_entry_blacklist, entry);
 270
 271                 ret = bch2_journal_seq_blacklist_add(c,
 272                                 le64_to_cpu(bl_entry->seq),
 273                                 le64_to_cpu(bl_entry->seq) + 1);
 274                 break;
 275         }
 276         case BCH_JSET_ENTRY_blacklist_v2: {
 277                 struct jset_entry_blacklist_v2 *bl_entry =
 278                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 279
 280                 ret = bch2_journal_seq_blacklist_add(c,
 281                                 le64_to_cpu(bl_entry->start),
 282                                 le64_to_cpu(bl_entry->end) + 1);
 283                 break;
 284         }
 285         case BCH_JSET_ENTRY_clock: {
 286                 struct jset_entry_clock *clock =
 287                         container_of(entry, struct jset_entry_clock, entry);
 288
 289                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 290         }
 291         }
 292
 293         return ret;
 294 }
 295
 296 static int journal_replay_early(struct bch_fs *c,
 297                                 struct bch_sb_field_clean *clean)
 298 {
 299         struct jset_entry *entry;
 300         int ret;
 301
 302         if (clean) {
 303                 for (entry = clean->start;
 304                      entry != vstruct_end(&clean->field);
 305                      entry = vstruct_next(entry)) {
 306                         ret = journal_replay_entry_early(c, entry);
 307                         if (ret)
 308                                 return ret;
 309                 }
 310         } else {
 311                 struct genradix_iter iter;
 312                 struct journal_replay *i, **_i;
 313
 314                 genradix_for_each(&c->journal_entries, iter, _i) {
 315                         i = *_i;
 316
 317                         if (!i || i->ignore)
 318                                 continue;
 319
 320                         vstruct_for_each(&i->j, entry) {
 321                                 ret = journal_replay_entry_early(c, entry);
 322                                 if (ret)
 323                                         return ret;
 324                         }
 325                 }
 326         }
 327
 328         bch2_fs_usage_initialize(c);
 329
 330         return 0;
 331 }
 332
 333 /* sb clean section: */
 334
 335 static bool btree_id_is_alloc(enum btree_id id)
 336 {
 337         switch (id) {
 338         case BTREE_ID_alloc:
 339         case BTREE_ID_backpointers:
 340         case BTREE_ID_need_discard:
 341         case BTREE_ID_freespace:
 342         case BTREE_ID_bucket_gens:
 343                 return true;
 344         default:
 345                 return false;
 346         }
 347 }
 348
 349 static int read_btree_roots(struct bch_fs *c)
 350 {
 351         unsigned i;
 352         int ret = 0;
 353
 354         for (i = 0; i < btree_id_nr_alive(c); i++) {
 355                 struct btree_root *r = bch2_btree_id_root(c, i);
 356
 357                 if (!r->alive)
 358                         continue;
 359
 360                 if (btree_id_is_alloc(i) &&
 361                     c->opts.reconstruct_alloc) {
 362                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 363                         continue;
 364                 }
 365
 366                 if (r->error) {
 367                         __fsck_err(c, btree_id_is_alloc(i)
 368                                    ? FSCK_CAN_IGNORE : 0,
 369                                    "invalid btree root %s",
 370                                    bch2_btree_ids[i]);
 371                         if (i == BTREE_ID_alloc)
 372                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 373                 }
 374
 375                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 376                 if (ret) {
 377                         __fsck_err(c,
 378                                    btree_id_is_alloc(i)
 379                                    ? FSCK_CAN_IGNORE : 0,
 380                                    "error reading btree root %s",
 381                                    bch2_btree_ids[i]);
 382                         if (btree_id_is_alloc(i))
 383                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 384                 }
 385         }
 386
 387         for (i = 0; i < BTREE_ID_NR; i++) {
 388                 struct btree_root *r = bch2_btree_id_root(c, i);
 389
 390                 if (!r->b) {
 391                         r->alive = false;
 392                         r->level = 0;
 393                         bch2_btree_root_alloc(c, i);
 394                 }
 395         }
 396 fsck_err:
 397         return ret;
 398 }
 399
 400 static int bch2_initialize_subvolumes(struct bch_fs *c)
 401 {
 402         struct bkey_i_snapshot_tree     root_tree;
 403         struct bkey_i_snapshot          root_snapshot;
 404         struct bkey_i_subvolume         root_volume;
 405         int ret;
 406
 407         bkey_snapshot_tree_init(&root_tree.k_i);
 408         root_tree.k.p.offset            = 1;
 409         root_tree.v.master_subvol       = cpu_to_le32(1);
 410         root_tree.v.root_snapshot       = cpu_to_le32(U32_MAX);
 411
 412         bkey_snapshot_init(&root_snapshot.k_i);
 413         root_snapshot.k.p.offset = U32_MAX;
 414         root_snapshot.v.flags   = 0;
 415         root_snapshot.v.parent  = 0;
 416         root_snapshot.v.subvol  = cpu_to_le32(BCACHEFS_ROOT_SUBVOL);
 417         root_snapshot.v.tree    = cpu_to_le32(1);
 418         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
 419
 420         bkey_subvolume_init(&root_volume.k_i);
 421         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
 422         root_volume.v.flags     = 0;
 423         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
 424         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
 425
 426         ret =   bch2_btree_insert(c, BTREE_ID_snapshot_trees,
 427                                   &root_tree.k_i,
 428                                   NULL, NULL, 0) ?:
 429                 bch2_btree_insert(c, BTREE_ID_snapshots,
 430                                   &root_snapshot.k_i,
 431                                   NULL, NULL, 0) ?:
 432                 bch2_btree_insert(c, BTREE_ID_subvolumes,
 433                                   &root_volume.k_i,
 434                                   NULL, NULL, 0);
 435         if (ret)
 436                 bch_err_fn(c, ret);
 437         return ret;
 438 }
 439
 440 static int __bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
 441 {
 442         struct btree_iter iter;
 443         struct bkey_s_c k;
 444         struct bch_inode_unpacked inode;
 445         int ret;
 446
 447         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
 448                                SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
 449         ret = bkey_err(k);
 450         if (ret)
 451                 return ret;
 452
 453         if (!bkey_is_inode(k.k)) {
 454                 bch_err(trans->c, "root inode not found");
 455                 ret = -BCH_ERR_ENOENT_inode;
 456                 goto err;
 457         }
 458
 459         ret = bch2_inode_unpack(k, &inode);
 460         BUG_ON(ret);
 461
 462         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
 463
 464         ret = bch2_inode_write(trans, &iter, &inode);
 465 err:
 466         bch2_trans_iter_exit(trans, &iter);
 467         return ret;
 468 }
 469
 470 /* set bi_subvol on root inode */
 471 noinline_for_stack
 472 static int bch2_fs_upgrade_for_subvolumes(struct bch_fs *c)
 473 {
 474         int ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
 475                                 __bch2_fs_upgrade_for_subvolumes(&trans));
 476         if (ret)
 477                 bch_err_fn(c, ret);
 478         return ret;
 479 }
 480
 481 const char * const bch2_recovery_passes[] = {
 482 #define x(_fn, _when)   #_fn,
 483         BCH_RECOVERY_PASSES()
 484 #undef x
 485         NULL
 486 };
 487
 488 static int bch2_check_allocations(struct bch_fs *c)
 489 {
 490         return bch2_gc(c, true, c->opts.norecovery);
 491 }
 492
 493 static int bch2_set_may_go_rw(struct bch_fs *c)
 494 {
 495         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
 496         return 0;
 497 }
 498
 499 struct recovery_pass_fn {
 500         int             (*fn)(struct bch_fs *);
 501         unsigned        when;
 502 };
 503
 504 static struct recovery_pass_fn recovery_pass_fns[] = {
 505 #define x(_fn, _when)   { .fn = bch2_##_fn, .when = _when },
 506         BCH_RECOVERY_PASSES()
 507 #undef x
 508 };
 509
 510 static void check_version_upgrade(struct bch_fs *c)
 511 {
 512         unsigned latest_compatible = bch2_latest_compatible_version(c->sb.version);
 513         unsigned latest_version = bcachefs_metadata_version_current;
 514         unsigned old_version = c->sb.version_upgrade_complete ?: c->sb.version;
 515         unsigned new_version = 0;
 516         u64 recovery_passes;
 517
 518         if (old_version < bcachefs_metadata_required_upgrade_below) {
 519                 if (c->opts.version_upgrade == BCH_VERSION_UPGRADE_incompatible ||
 520                     latest_compatible < bcachefs_metadata_required_upgrade_below)
 521                         new_version = latest_version;
 522                 else
 523                         new_version = latest_compatible;
 524         } else {
 525                 switch (c->opts.version_upgrade) {
 526                 case BCH_VERSION_UPGRADE_compatible:
 527                         new_version = latest_compatible;
 528                         break;
 529                 case BCH_VERSION_UPGRADE_incompatible:
 530                         new_version = latest_version;
 531                         break;
 532                 case BCH_VERSION_UPGRADE_none:
 533                         new_version = old_version;
 534                         break;
 535                 }
 536         }
 537
 538         if (new_version > old_version) {
 539                 struct printbuf buf = PRINTBUF;
 540
 541                 if (old_version < bcachefs_metadata_required_upgrade_below)
 542                         prt_str(&buf, "Version upgrade required:\n");
 543
 544                 if (old_version != c->sb.version) {
 545                         prt_str(&buf, "Version upgrade from ");
 546                         bch2_version_to_text(&buf, c->sb.version_upgrade_complete);
 547                         prt_str(&buf, " to ");
 548                         bch2_version_to_text(&buf, c->sb.version);
 549                         prt_str(&buf, " incomplete\n");
 550                 }
 551
 552                 prt_printf(&buf, "Doing %s version upgrade from ",
 553                            BCH_VERSION_MAJOR(old_version) != BCH_VERSION_MAJOR(new_version)
 554                            ? "incompatible" : "compatible");
 555                 bch2_version_to_text(&buf, old_version);
 556                 prt_str(&buf, " to ");
 557                 bch2_version_to_text(&buf, new_version);
 558                 prt_newline(&buf);
 559
 560                 recovery_passes = bch2_upgrade_recovery_passes(c, old_version, new_version);
 561                 if (recovery_passes) {
 562                         if ((recovery_passes & RECOVERY_PASS_ALL_FSCK) == RECOVERY_PASS_ALL_FSCK)
 563                                 prt_str(&buf, "fsck required");
 564                         else {
 565                                 prt_str(&buf, "running recovery passes: ");
 566                                 prt_bitflags(&buf, bch2_recovery_passes, recovery_passes);
 567                         }
 568
 569                         c->recovery_passes_explicit |= recovery_passes;
 570                         c->opts.fix_errors = FSCK_FIX_yes;
 571                 }
 572
 573                 bch_info(c, "%s", buf.buf);
 574
 575                 mutex_lock(&c->sb_lock);
 576                 bch2_sb_upgrade(c, new_version);
 577                 mutex_unlock(&c->sb_lock);
 578
 579                 printbuf_exit(&buf);
 580         }
 581 }
 582
 583 u64 bch2_fsck_recovery_passes(void)
 584 {
 585         u64 ret = 0;
 586
 587         for (unsigned i = 0; i < ARRAY_SIZE(recovery_pass_fns); i++)
 588                 if (recovery_pass_fns[i].when & PASS_FSCK)
 589                         ret |= BIT_ULL(i);
 590         return ret;
 591 }
 592
 593 static bool should_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
 594 {
 595         struct recovery_pass_fn *p = recovery_pass_fns + c->curr_recovery_pass;
 596
 597         if (c->opts.norecovery && pass > BCH_RECOVERY_PASS_snapshots_read)
 598                 return false;
 599         if (c->recovery_passes_explicit & BIT_ULL(pass))
 600                 return true;
 601         if ((p->when & PASS_FSCK) && c->opts.fsck)
 602                 return true;
 603         if ((p->when & PASS_UNCLEAN) && !c->sb.clean)
 604                 return true;
 605         if (p->when & PASS_ALWAYS)
 606                 return true;
 607         return false;
 608 }
 609
 610 static int bch2_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
 611 {
 612         int ret;
 613
 614         c->curr_recovery_pass = pass;
 615
 616         if (should_run_recovery_pass(c, pass)) {
 617                 struct recovery_pass_fn *p = recovery_pass_fns + pass;
 618
 619                 if (!(p->when & PASS_SILENT))
 620                         printk(KERN_INFO bch2_log_msg(c, "%s..."),
 621                                bch2_recovery_passes[pass]);
 622                 ret = p->fn(c);
 623                 if (ret)
 624                         return ret;
 625                 if (!(p->when & PASS_SILENT))
 626                         printk(KERN_CONT " done\n");
 627
 628                 c->recovery_passes_complete |= BIT_ULL(pass);
 629         }
 630
 631         return 0;
 632 }
 633
 634 static int bch2_run_recovery_passes(struct bch_fs *c)
 635 {
 636         int ret = 0;
 637
 638         while (c->curr_recovery_pass < ARRAY_SIZE(recovery_pass_fns)) {
 639                 ret = bch2_run_recovery_pass(c, c->curr_recovery_pass);
 640                 if (bch2_err_matches(ret, BCH_ERR_restart_recovery))
 641                         continue;
 642                 if (ret)
 643                         break;
 644                 c->curr_recovery_pass++;
 645         }
 646
 647         return ret;
 648 }
 649
 650 int bch2_fs_recovery(struct bch_fs *c)
 651 {
 652         struct bch_sb_field_clean *clean = NULL;
 653         struct jset *last_journal_entry = NULL;
 654         u64 last_seq, blacklist_seq, journal_seq;
 655         bool write_sb = false;
 656         int ret = 0;
 657
 658         if (c->sb.clean) {
 659                 clean = bch2_read_superblock_clean(c);
 660                 ret = PTR_ERR_OR_ZERO(clean);
 661                 if (ret)
 662                         goto err;
 663
 664                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 665                          le64_to_cpu(clean->journal_seq));
 666         } else {
 667                 bch_info(c, "recovering from unclean shutdown");
 668         }
 669
 670         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
 671                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
 672                 ret = -EINVAL;
 673                 goto err;
 674         }
 675
 676         if (!c->sb.clean &&
 677             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
 678                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
 679                 ret = -EINVAL;
 680                 goto err;
 681         }
 682
 683         if (c->opts.fsck || !(c->opts.nochanges && c->opts.norecovery))
 684                 check_version_upgrade(c);
 685
 686         if (c->opts.fsck && c->opts.norecovery) {
 687                 bch_err(c, "cannot select both norecovery and fsck");
 688                 ret = -EINVAL;
 689                 goto err;
 690         }
 691
 692         ret = bch2_blacklist_table_initialize(c);
 693         if (ret) {
 694                 bch_err(c, "error initializing blacklist table");
 695                 goto err;
 696         }
 697
 698         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
 699                 struct genradix_iter iter;
 700                 struct journal_replay **i;
 701
 702                 bch_verbose(c, "starting journal read");
 703                 ret = bch2_journal_read(c, &last_seq, &blacklist_seq, &journal_seq);
 704                 if (ret)
 705                         goto err;
 706
 707                 /*
 708                  * note: cmd_list_journal needs the blacklist table fully up to date so
 709                  * it can asterisk ignored journal entries:
 710                  */
 711                 if (c->opts.read_journal_only)
 712                         goto out;
 713
 714                 genradix_for_each_reverse(&c->journal_entries, iter, i)
 715                         if (*i && !(*i)->ignore) {
 716                                 last_journal_entry = &(*i)->j;
 717                                 break;
 718                         }
 719
 720                 if (mustfix_fsck_err_on(c->sb.clean &&
 721                                         last_journal_entry &&
 722                                         !journal_entry_empty(last_journal_entry), c,
 723                                 "filesystem marked clean but journal not empty")) {
 724                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 725                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 726                         c->sb.clean = false;
 727                 }
 728
 729                 if (!last_journal_entry) {
 730                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
 731                         if (clean)
 732                                 goto use_clean;
 733
 734                         genradix_for_each_reverse(&c->journal_entries, iter, i)
 735                                 if (*i) {
 736                                         last_journal_entry = &(*i)->j;
 737                                         (*i)->ignore = false;
 738                                         break;
 739                                 }
 740                 }
 741
 742                 ret = bch2_journal_keys_sort(c);
 743                 if (ret)
 744                         goto err;
 745
 746                 if (c->sb.clean && last_journal_entry) {
 747                         ret = bch2_verify_superblock_clean(c, &clean,
 748                                                       last_journal_entry);
 749                         if (ret)
 750                                 goto err;
 751                 }
 752         } else {
 753 use_clean:
 754                 if (!clean) {
 755                         bch_err(c, "no superblock clean section found");
 756                         ret = -BCH_ERR_fsck_repair_impossible;
 757                         goto err;
 758
 759                 }
 760                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
 761         }
 762
 763         c->journal_replay_seq_start     = last_seq;
 764         c->journal_replay_seq_end       = blacklist_seq - 1;
 765
 766         if (c->opts.reconstruct_alloc) {
 767                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 768                 drop_alloc_keys(&c->journal_keys);
 769         }
 770
 771         zero_out_btree_mem_ptr(&c->journal_keys);
 772
 773         ret = journal_replay_early(c, clean);
 774         if (ret)
 775                 goto err;
 776
 777         /*
 778          * After an unclean shutdown, skip then next few journal sequence
 779          * numbers as they may have been referenced by btree writes that
 780          * happened before their corresponding journal writes - those btree
 781          * writes need to be ignored, by skipping and blacklisting the next few
 782          * journal sequence numbers:
 783          */
 784         if (!c->sb.clean)
 785                 journal_seq += 8;
 786
 787         if (blacklist_seq != journal_seq) {
 788                 ret =   bch2_journal_log_msg(c, "blacklisting entries %llu-%llu",
 789                                              blacklist_seq, journal_seq) ?:
 790                         bch2_journal_seq_blacklist_add(c,
 791                                         blacklist_seq, journal_seq);
 792                 if (ret) {
 793                         bch_err(c, "error creating new journal seq blacklist entry");
 794                         goto err;
 795                 }
 796         }
 797
 798         ret =   bch2_journal_log_msg(c, "starting journal at entry %llu, replaying %llu-%llu",
 799                                      journal_seq, last_seq, blacklist_seq - 1) ?:
 800                 bch2_fs_journal_start(&c->journal, journal_seq);
 801         if (ret)
 802                 goto err;
 803
 804         if (c->opts.reconstruct_alloc)
 805                 bch2_journal_log_msg(c, "dropping alloc info");
 806
 807         /*
 808          * Skip past versions that might have possibly been used (as nonces),
 809          * but hadn't had their pointers written:
 810          */
 811         if (c->sb.encryption_type && !c->sb.clean)
 812                 atomic64_add(1 << 16, &c->key_version);
 813
 814         ret = read_btree_roots(c);
 815         if (ret)
 816                 goto err;
 817
 818         if (c->opts.fsck &&
 819             (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) ||
 820              BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb)))
 821                 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_topology);
 822
 823         ret = bch2_run_recovery_passes(c);
 824         if (ret)
 825                 goto err;
 826
 827         /* If we fixed errors, verify that fs is actually clean now: */
 828         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) &&
 829             test_bit(BCH_FS_ERRORS_FIXED, &c->flags) &&
 830             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags) &&
 831             !test_bit(BCH_FS_ERROR, &c->flags)) {
 832                 bch_info(c, "Fixed errors, running fsck a second time to verify fs is clean");
 833                 clear_bit(BCH_FS_ERRORS_FIXED, &c->flags);
 834
 835                 c->curr_recovery_pass = BCH_RECOVERY_PASS_check_alloc_info;
 836
 837                 ret = bch2_run_recovery_passes(c);
 838                 if (ret)
 839                         goto err;
 840
 841                 if (test_bit(BCH_FS_ERRORS_FIXED, &c->flags) ||
 842                     test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
 843                         bch_err(c, "Second fsck run was not clean");
 844                         set_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags);
 845                 }
 846
 847                 set_bit(BCH_FS_ERRORS_FIXED, &c->flags);
 848         }
 849
 850         if (enabled_qtypes(c)) {
 851                 bch_verbose(c, "reading quotas");
 852                 ret = bch2_fs_quota_read(c);
 853                 if (ret)
 854                         goto err;
 855                 bch_verbose(c, "quotas done");
 856         }
 857
 858         mutex_lock(&c->sb_lock);
 859         if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) != c->sb.version) {
 860                 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, c->sb.version);
 861                 write_sb = true;
 862         }
 863
 864         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
 865                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
 866                 write_sb = true;
 867         }
 868
 869         if (c->opts.fsck &&
 870             !test_bit(BCH_FS_ERROR, &c->flags) &&
 871             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
 872                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
 873                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
 874                 write_sb = true;
 875         }
 876
 877         if (write_sb)
 878                 bch2_write_super(c);
 879         mutex_unlock(&c->sb_lock);
 880
 881         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
 882             c->sb.version_min < bcachefs_metadata_version_btree_ptr_sectors_written) {
 883                 struct bch_move_stats stats;
 884
 885                 bch2_move_stats_init(&stats, "recovery");
 886
 887                 bch_info(c, "scanning for old btree nodes");
 888                 ret =   bch2_fs_read_write(c) ?:
 889                         bch2_scan_old_btree_nodes(c, &stats);
 890                 if (ret)
 891                         goto err;
 892                 bch_info(c, "scanning for old btree nodes done");
 893         }
 894
 895         if (c->journal_seq_blacklist_table &&
 896             c->journal_seq_blacklist_table->nr > 128)
 897                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
 898
 899         ret = 0;
 900 out:
 901         set_bit(BCH_FS_FSCK_DONE, &c->flags);
 902         bch2_flush_fsck_errs(c);
 903
 904         if (!c->opts.keep_journal &&
 905             test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)) {
 906                 bch2_journal_keys_free(&c->journal_keys);
 907                 bch2_journal_entries_free(c);
 908         }
 909         kfree(clean);
 910
 911         if (!ret && test_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags)) {
 912                 bch2_fs_read_write_early(c);
 913                 bch2_delete_dead_snapshots_async(c);
 914         }
 915
 916         if (ret)
 917                 bch_err_fn(c, ret);
 918         return ret;
 919 err:
 920 fsck_err:
 921         bch2_fs_emergency_read_only(c);
 922         goto out;
 923 }
 924
 925 int bch2_fs_initialize(struct bch_fs *c)
 926 {
 927         struct bch_inode_unpacked root_inode, lostfound_inode;
 928         struct bkey_inode_buf packed_inode;
 929         struct qstr lostfound = QSTR("lost+found");
 930         struct bch_dev *ca;
 931         unsigned i;
 932         int ret;
 933
 934         bch_notice(c, "initializing new filesystem");
 935
 936         mutex_lock(&c->sb_lock);
 937         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
 938         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
 939
 940         bch2_sb_maybe_downgrade(c);
 941
 942         if (c->opts.version_upgrade != BCH_VERSION_UPGRADE_none) {
 943                 bch2_sb_upgrade(c, bcachefs_metadata_version_current);
 944                 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
 945                 bch2_write_super(c);
 946         }
 947         mutex_unlock(&c->sb_lock);
 948
 949         c->curr_recovery_pass = ARRAY_SIZE(recovery_pass_fns);
 950         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
 951         set_bit(BCH_FS_FSCK_DONE, &c->flags);
 952
 953         for (i = 0; i < BTREE_ID_NR; i++)
 954                 bch2_btree_root_alloc(c, i);
 955
 956         for_each_online_member(ca, c, i)
 957                 bch2_dev_usage_init(ca);
 958
 959         for_each_online_member(ca, c, i) {
 960                 ret = bch2_dev_journal_alloc(ca);
 961                 if (ret) {
 962                         percpu_ref_put(&ca->io_ref);
 963                         goto err;
 964                 }
 965         }
 966
 967         /*
 968          * journal_res_get() will crash if called before this has
 969          * set up the journal.pin FIFO and journal.cur pointer:
 970          */
 971         bch2_fs_journal_start(&c->journal, 1);
 972         bch2_journal_set_replay_done(&c->journal);
 973
 974         ret = bch2_fs_read_write_early(c);
 975         if (ret)
 976                 goto err;
 977
 978         /*
 979          * Write out the superblock and journal buckets, now that we can do
 980          * btree updates
 981          */
 982         bch_verbose(c, "marking superblocks");
 983         for_each_member_device(ca, c, i) {
 984                 ret = bch2_trans_mark_dev_sb(c, ca);
 985                 if (ret) {
 986                         percpu_ref_put(&ca->ref);
 987                         goto err;
 988                 }
 989
 990                 ca->new_fs_bucket_idx = 0;
 991         }
 992
 993         ret = bch2_fs_freespace_init(c);
 994         if (ret)
 995                 goto err;
 996
 997         ret = bch2_initialize_subvolumes(c);
 998         if (ret)
 999                 goto err;
1000
1001         bch_verbose(c, "reading snapshots table");
1002         ret = bch2_snapshots_read(c);
1003         if (ret)
1004                 goto err;
1005         bch_verbose(c, "reading snapshots done");
1006
1007         bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL);
1008         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1009         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1010         bch2_inode_pack(&packed_inode, &root_inode);
1011         packed_inode.inode.k.p.snapshot = U32_MAX;
1012
1013         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1014                                 &packed_inode.inode.k_i,
1015                                 NULL, NULL, 0);
1016         if (ret) {
1017                 bch_err_msg(c, ret, "creating root directory");
1018                 goto err;
1019         }
1020
1021         bch2_inode_init_early(c, &lostfound_inode);
1022
1023         ret = bch2_trans_do(c, NULL, NULL, 0,
1024                 bch2_create_trans(&trans,
1025                                   BCACHEFS_ROOT_SUBVOL_INUM,
1026                                   &root_inode, &lostfound_inode,
1027                                   &lostfound,
1028                                   0, 0, S_IFDIR|0700, 0,
1029                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1030         if (ret) {
1031                 bch_err_msg(c, ret, "creating lost+found");
1032                 goto err;
1033         }
1034
1035         if (enabled_qtypes(c)) {
1036                 ret = bch2_fs_quota_read(c);
1037                 if (ret)
1038                         goto err;
1039         }
1040
1041         ret = bch2_journal_flush(&c->journal);
1042         if (ret) {
1043                 bch_err_msg(c, ret, "writing first journal entry");
1044                 goto err;
1045         }
1046
1047         mutex_lock(&c->sb_lock);
1048         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1049         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1050
1051         bch2_write_super(c);
1052         mutex_unlock(&c->sb_lock);
1053
1054         return 0;
1055 err:
1056         bch_err_fn(ca, ret);
1057         return ret;
1058 }