git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/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 "move.h"
  24 #include "quota.h"
  25 #include "recovery.h"
  26 #include "replicas.h"
  27 #include "sb-clean.h"
  28 #include "subvolume.h"
  29 #include "super-io.h"
  30
  31 #include <linux/sort.h>
  32 #include <linux/stat.h>
  33
  34 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  35
  36 /* for -o reconstruct_alloc: */
  37 static void drop_alloc_keys(struct journal_keys *keys)
  38 {
  39         size_t src, dst;
  40
  41         for (src = 0, dst = 0; src < keys->nr; src++)
  42                 if (keys->d[src].btree_id != BTREE_ID_alloc)
  43                         keys->d[dst++] = keys->d[src];
  44
  45         keys->nr = dst;
  46 }
  47
  48 /*
  49  * Btree node pointers have a field to stack a pointer to the in memory btree
  50  * node; we need to zero out this field when reading in btree nodes, or when
  51  * reading in keys from the journal:
  52  */
  53 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
  54 {
  55         struct journal_key *i;
  56
  57         for (i = keys->d; i < keys->d + keys->nr; i++)
  58                 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
  59                         bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
  60 }
  61
  62 /* journal replay: */
  63
  64 static void replay_now_at(struct journal *j, u64 seq)
  65 {
  66         BUG_ON(seq < j->replay_journal_seq);
  67
  68         seq = min(seq, j->replay_journal_seq_end);
  69
  70         while (j->replay_journal_seq < seq)
  71                 bch2_journal_pin_put(j, j->replay_journal_seq++);
  72 }
  73
  74 static int bch2_journal_replay_key(struct btree_trans *trans,
  75                                    struct journal_key *k)
  76 {
  77         struct btree_iter iter;
  78         unsigned iter_flags =
  79                 BTREE_ITER_INTENT|
  80                 BTREE_ITER_NOT_EXTENTS;
  81         unsigned update_flags = BTREE_TRIGGER_NORUN;
  82         int ret;
  83
  84         /*
  85          * BTREE_UPDATE_KEY_CACHE_RECLAIM disables key cache lookup/update to
  86          * keep the key cache coherent with the underlying btree. Nothing
  87          * besides the allocator is doing updates yet so we don't need key cache
  88          * coherency for non-alloc btrees, and key cache fills for snapshots
  89          * btrees use BTREE_ITER_FILTER_SNAPSHOTS, which isn't available until
  90          * the snapshots recovery pass runs.
  91          */
  92         if (!k->level && k->btree_id == BTREE_ID_alloc)
  93                 iter_flags |= BTREE_ITER_CACHED;
  94         else
  95                 update_flags |= BTREE_UPDATE_KEY_CACHE_RECLAIM;
  96
  97         bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
  98                                   BTREE_MAX_DEPTH, k->level,
  99                                   iter_flags);
 100         ret = bch2_btree_iter_traverse(&iter);
 101         if (ret)
 102                 goto out;
 103
 104         /* Must be checked with btree locked: */
 105         if (k->overwritten)
 106                 goto out;
 107
 108         ret = bch2_trans_update(trans, &iter, k->k, update_flags);
 109 out:
 110         bch2_trans_iter_exit(trans, &iter);
 111         return ret;
 112 }
 113
 114 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 115 {
 116         const struct journal_key *l = *((const struct journal_key **)_l);
 117         const struct journal_key *r = *((const struct journal_key **)_r);
 118
 119         return cmp_int(l->journal_seq, r->journal_seq);
 120 }
 121
 122 static int bch2_journal_replay(struct bch_fs *c)
 123 {
 124         struct journal_keys *keys = &c->journal_keys;
 125         struct journal_key **keys_sorted, *k;
 126         struct journal *j = &c->journal;
 127         u64 start_seq   = c->journal_replay_seq_start;
 128         u64 end_seq     = c->journal_replay_seq_start;
 129         size_t i;
 130         int ret;
 131
 132         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 133         keys->gap = keys->nr;
 134
 135         keys_sorted = kvmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL);
 136         if (!keys_sorted)
 137                 return -BCH_ERR_ENOMEM_journal_replay;
 138
 139         for (i = 0; i < keys->nr; i++)
 140                 keys_sorted[i] = &keys->d[i];
 141
 142         sort(keys_sorted, keys->nr,
 143              sizeof(keys_sorted[0]),
 144              journal_sort_seq_cmp, NULL);
 145
 146         if (keys->nr) {
 147                 ret = bch2_journal_log_msg(c, "Starting journal replay (%zu keys in entries %llu-%llu)",
 148                                            keys->nr, start_seq, end_seq);
 149                 if (ret)
 150                         goto err;
 151         }
 152
 153         for (i = 0; i < keys->nr; i++) {
 154                 k = keys_sorted[i];
 155
 156                 cond_resched();
 157
 158                 replay_now_at(j, k->journal_seq);
 159
 160                 ret = bch2_trans_do(c, NULL, NULL,
 161                                     BTREE_INSERT_LAZY_RW|
 162                                     BTREE_INSERT_NOFAIL|
 163                                     (!k->allocated
 164                                      ? BTREE_INSERT_JOURNAL_REPLAY|BCH_WATERMARK_reclaim
 165                                      : 0),
 166                              bch2_journal_replay_key(&trans, k));
 167                 if (ret) {
 168                         bch_err(c, "journal replay: error while replaying key at btree %s level %u: %s",
 169                                 bch2_btree_ids[k->btree_id], k->level, bch2_err_str(ret));
 170                         goto err;
 171                 }
 172         }
 173
 174         replay_now_at(j, j->replay_journal_seq_end);
 175         j->replay_journal_seq = 0;
 176
 177         bch2_journal_set_replay_done(j);
 178         bch2_journal_flush_all_pins(j);
 179         ret = bch2_journal_error(j);
 180
 181         if (keys->nr && !ret)
 182                 bch2_journal_log_msg(c, "journal replay finished");
 183 err:
 184         kvfree(keys_sorted);
 185
 186         if (ret)
 187                 bch_err_fn(c, ret);
 188         return ret;
 189 }
 190
 191 /* journal replay early: */
 192
 193 static int journal_replay_entry_early(struct bch_fs *c,
 194                                       struct jset_entry *entry)
 195 {
 196         int ret = 0;
 197
 198         switch (entry->type) {
 199         case BCH_JSET_ENTRY_btree_root: {
 200                 struct btree_root *r;
 201
 202                 while (entry->btree_id >= c->btree_roots_extra.nr + BTREE_ID_NR) {
 203                         ret = darray_push(&c->btree_roots_extra, (struct btree_root) { NULL });
 204                         if (ret)
 205                                 return ret;
 206                 }
 207
 208                 r = bch2_btree_id_root(c, entry->btree_id);
 209
 210                 if (entry->u64s) {
 211                         r->level = entry->level;
 212                         bkey_copy(&r->key, &entry->start[0]);
 213                         r->error = 0;
 214                 } else {
 215                         r->error = -EIO;
 216                 }
 217                 r->alive = true;
 218                 break;
 219         }
 220         case BCH_JSET_ENTRY_usage: {
 221                 struct jset_entry_usage *u =
 222                         container_of(entry, struct jset_entry_usage, entry);
 223
 224                 switch (entry->btree_id) {
 225                 case BCH_FS_USAGE_reserved:
 226                         if (entry->level < BCH_REPLICAS_MAX)
 227                                 c->usage_base->persistent_reserved[entry->level] =
 228                                         le64_to_cpu(u->v);
 229                         break;
 230                 case BCH_FS_USAGE_inodes:
 231                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 232                         break;
 233                 case BCH_FS_USAGE_key_version:
 234                         atomic64_set(&c->key_version,
 235                                      le64_to_cpu(u->v));
 236                         break;
 237                 }
 238
 239                 break;
 240         }
 241         case BCH_JSET_ENTRY_data_usage: {
 242                 struct jset_entry_data_usage *u =
 243                         container_of(entry, struct jset_entry_data_usage, entry);
 244
 245                 ret = bch2_replicas_set_usage(c, &u->r,
 246                                               le64_to_cpu(u->v));
 247                 break;
 248         }
 249         case BCH_JSET_ENTRY_dev_usage: {
 250                 struct jset_entry_dev_usage *u =
 251                         container_of(entry, struct jset_entry_dev_usage, entry);
 252                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 253                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 254
 255                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 256
 257                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 258                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 259                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 260                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 261                 }
 262
 263                 break;
 264         }
 265         case BCH_JSET_ENTRY_blacklist: {
 266                 struct jset_entry_blacklist *bl_entry =
 267                         container_of(entry, struct jset_entry_blacklist, entry);
 268
 269                 ret = bch2_journal_seq_blacklist_add(c,
 270                                 le64_to_cpu(bl_entry->seq),
 271                                 le64_to_cpu(bl_entry->seq) + 1);
 272                 break;
 273         }
 274         case BCH_JSET_ENTRY_blacklist_v2: {
 275                 struct jset_entry_blacklist_v2 *bl_entry =
 276                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 277
 278                 ret = bch2_journal_seq_blacklist_add(c,
 279                                 le64_to_cpu(bl_entry->start),
 280                                 le64_to_cpu(bl_entry->end) + 1);
 281                 break;
 282         }
 283         case BCH_JSET_ENTRY_clock: {
 284                 struct jset_entry_clock *clock =
 285                         container_of(entry, struct jset_entry_clock, entry);
 286
 287                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 288         }
 289         }
 290
 291         return ret;
 292 }
 293
 294 static int journal_replay_early(struct bch_fs *c,
 295                                 struct bch_sb_field_clean *clean)
 296 {
 297         struct jset_entry *entry;
 298         int ret;
 299
 300         if (clean) {
 301                 for (entry = clean->start;
 302                      entry != vstruct_end(&clean->field);
 303                      entry = vstruct_next(entry)) {
 304                         ret = journal_replay_entry_early(c, entry);
 305                         if (ret)
 306                                 return ret;
 307                 }
 308         } else {
 309                 struct genradix_iter iter;
 310                 struct journal_replay *i, **_i;
 311
 312                 genradix_for_each(&c->journal_entries, iter, _i) {
 313                         i = *_i;
 314
 315                         if (!i || i->ignore)
 316                                 continue;
 317
 318                         vstruct_for_each(&i->j, entry) {
 319                                 ret = journal_replay_entry_early(c, entry);
 320                                 if (ret)
 321                                         return ret;
 322                         }
 323                 }
 324         }
 325
 326         bch2_fs_usage_initialize(c);
 327
 328         return 0;
 329 }
 330
 331 /* sb clean section: */
 332
 333 static bool btree_id_is_alloc(enum btree_id id)
 334 {
 335         switch (id) {
 336         case BTREE_ID_alloc:
 337         case BTREE_ID_backpointers:
 338         case BTREE_ID_need_discard:
 339         case BTREE_ID_freespace:
 340         case BTREE_ID_bucket_gens:
 341                 return true;
 342         default:
 343                 return false;
 344         }
 345 }
 346
 347 static int read_btree_roots(struct bch_fs *c)
 348 {
 349         unsigned i;
 350         int ret = 0;
 351
 352         for (i = 0; i < btree_id_nr_alive(c); i++) {
 353                 struct btree_root *r = bch2_btree_id_root(c, i);
 354
 355                 if (!r->alive)
 356                         continue;
 357
 358                 if (btree_id_is_alloc(i) &&
 359                     c->opts.reconstruct_alloc) {
 360                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 361                         continue;
 362                 }
 363
 364                 if (r->error) {
 365                         __fsck_err(c, btree_id_is_alloc(i)
 366                                    ? FSCK_CAN_IGNORE : 0,
 367                                    "invalid btree root %s",
 368                                    bch2_btree_ids[i]);
 369                         if (i == BTREE_ID_alloc)
 370                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 371                 }
 372
 373                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 374                 if (ret) {
 375                         __fsck_err(c,
 376                                    btree_id_is_alloc(i)
 377                                    ? FSCK_CAN_IGNORE : 0,
 378                                    "error reading btree root %s",
 379                                    bch2_btree_ids[i]);
 380                         if (btree_id_is_alloc(i))
 381                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 382                 }
 383         }
 384
 385         for (i = 0; i < BTREE_ID_NR; i++) {
 386                 struct btree_root *r = bch2_btree_id_root(c, i);
 387
 388                 if (!r->b) {
 389                         r->alive = false;
 390                         r->level = 0;
 391                         bch2_btree_root_alloc(c, i);
 392                 }
 393         }
 394 fsck_err:
 395         return ret;
 396 }
 397
 398 static int bch2_initialize_subvolumes(struct bch_fs *c)
 399 {
 400         struct bkey_i_snapshot_tree     root_tree;
 401         struct bkey_i_snapshot          root_snapshot;
 402         struct bkey_i_subvolume         root_volume;
 403         int ret;
 404
 405         bkey_snapshot_tree_init(&root_tree.k_i);
 406         root_tree.k.p.offset            = 1;
 407         root_tree.v.master_subvol       = cpu_to_le32(1);
 408         root_tree.v.root_snapshot       = cpu_to_le32(U32_MAX);
 409
 410         bkey_snapshot_init(&root_snapshot.k_i);
 411         root_snapshot.k.p.offset = U32_MAX;
 412         root_snapshot.v.flags   = 0;
 413         root_snapshot.v.parent  = 0;
 414         root_snapshot.v.subvol  = cpu_to_le32(BCACHEFS_ROOT_SUBVOL);
 415         root_snapshot.v.tree    = cpu_to_le32(1);
 416         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
 417
 418         bkey_subvolume_init(&root_volume.k_i);
 419         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
 420         root_volume.v.flags     = 0;
 421         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
 422         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
 423
 424         ret =   bch2_btree_insert(c, BTREE_ID_snapshot_trees,
 425                                   &root_tree.k_i,
 426                                   NULL, NULL, 0) ?:
 427                 bch2_btree_insert(c, BTREE_ID_snapshots,
 428                                   &root_snapshot.k_i,
 429                                   NULL, NULL, 0) ?:
 430                 bch2_btree_insert(c, BTREE_ID_subvolumes,
 431                                   &root_volume.k_i,
 432                                   NULL, NULL, 0);
 433         if (ret)
 434                 bch_err_fn(c, ret);
 435         return ret;
 436 }
 437
 438 static int __bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
 439 {
 440         struct btree_iter iter;
 441         struct bkey_s_c k;
 442         struct bch_inode_unpacked inode;
 443         int ret;
 444
 445         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
 446                                SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
 447         ret = bkey_err(k);
 448         if (ret)
 449                 return ret;
 450
 451         if (!bkey_is_inode(k.k)) {
 452                 bch_err(trans->c, "root inode not found");
 453                 ret = -BCH_ERR_ENOENT_inode;
 454                 goto err;
 455         }
 456
 457         ret = bch2_inode_unpack(k, &inode);
 458         BUG_ON(ret);
 459
 460         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
 461
 462         ret = bch2_inode_write(trans, &iter, &inode);
 463 err:
 464         bch2_trans_iter_exit(trans, &iter);
 465         return ret;
 466 }
 467
 468 /* set bi_subvol on root inode */
 469 noinline_for_stack
 470 static int bch2_fs_upgrade_for_subvolumes(struct bch_fs *c)
 471 {
 472         int ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
 473                                 __bch2_fs_upgrade_for_subvolumes(&trans));
 474         if (ret)
 475                 bch_err_fn(c, ret);
 476         return ret;
 477 }
 478
 479 const char * const bch2_recovery_passes[] = {
 480 #define x(_fn, _when)   #_fn,
 481         BCH_RECOVERY_PASSES()
 482 #undef x
 483         NULL
 484 };
 485
 486 static int bch2_check_allocations(struct bch_fs *c)
 487 {
 488         return bch2_gc(c, true, c->opts.norecovery);
 489 }
 490
 491 static int bch2_set_may_go_rw(struct bch_fs *c)
 492 {
 493         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
 494         return 0;
 495 }
 496
 497 struct recovery_pass_fn {
 498         int             (*fn)(struct bch_fs *);
 499         unsigned        when;
 500 };
 501
 502 static struct recovery_pass_fn recovery_pass_fns[] = {
 503 #define x(_fn, _when)   { .fn = bch2_##_fn, .when = _when },
 504         BCH_RECOVERY_PASSES()
 505 #undef x
 506 };
 507
 508 static void check_version_upgrade(struct bch_fs *c)
 509 {
 510         unsigned latest_compatible = bch2_version_compatible(c->sb.version);
 511         unsigned latest_version = bcachefs_metadata_version_current;
 512         unsigned old_version = c->sb.version_upgrade_complete ?: c->sb.version;
 513         unsigned new_version = 0;
 514         u64 recovery_passes;
 515
 516         if (old_version < bcachefs_metadata_required_upgrade_below) {
 517                 if (c->opts.version_upgrade == BCH_VERSION_UPGRADE_incompatible ||
 518                     latest_compatible < bcachefs_metadata_required_upgrade_below)
 519                         new_version = latest_version;
 520                 else
 521                         new_version = latest_compatible;
 522         } else {
 523                 switch (c->opts.version_upgrade) {
 524                 case BCH_VERSION_UPGRADE_compatible:
 525                         new_version = latest_compatible;
 526                         break;
 527                 case BCH_VERSION_UPGRADE_incompatible:
 528                         new_version = latest_version;
 529                         break;
 530                 case BCH_VERSION_UPGRADE_none:
 531                         new_version = old_version;
 532                         break;
 533                 }
 534         }
 535
 536         if (new_version > old_version) {
 537                 struct printbuf buf = PRINTBUF;
 538
 539                 if (old_version < bcachefs_metadata_required_upgrade_below)
 540                         prt_str(&buf, "Version upgrade required:\n");
 541
 542                 if (old_version != c->sb.version) {
 543                         prt_str(&buf, "Version upgrade from ");
 544                         bch2_version_to_text(&buf, c->sb.version_upgrade_complete);
 545                         prt_str(&buf, " to ");
 546                         bch2_version_to_text(&buf, c->sb.version);
 547                         prt_str(&buf, " incomplete\n");
 548                 }
 549
 550                 prt_printf(&buf, "Doing %s version upgrade from ",
 551                            BCH_VERSION_MAJOR(old_version) != BCH_VERSION_MAJOR(new_version)
 552                            ? "incompatible" : "compatible");
 553                 bch2_version_to_text(&buf, old_version);
 554                 prt_str(&buf, " to ");
 555                 bch2_version_to_text(&buf, new_version);
 556                 prt_newline(&buf);
 557
 558                 recovery_passes = bch2_upgrade_recovery_passes(c, old_version, new_version);
 559                 if (recovery_passes) {
 560                         if ((recovery_passes & RECOVERY_PASS_ALL_FSCK) == RECOVERY_PASS_ALL_FSCK)
 561                                 prt_str(&buf, "fsck required");
 562                         else {
 563                                 prt_str(&buf, "running recovery passses: ");
 564                                 prt_bitflags(&buf, bch2_recovery_passes, recovery_passes);
 565                         }
 566
 567                         c->recovery_passes_explicit |= recovery_passes;
 568                         c->opts.fix_errors = FSCK_FIX_yes;
 569                 }
 570
 571                 bch_info(c, "%s", buf.buf);
 572
 573                 mutex_lock(&c->sb_lock);
 574                 bch2_sb_upgrade(c, new_version);
 575                 mutex_unlock(&c->sb_lock);
 576
 577                 printbuf_exit(&buf);
 578         }
 579 }
 580
 581 u64 bch2_fsck_recovery_passes(void)
 582 {
 583         u64 ret = 0;
 584
 585         for (unsigned i = 0; i < ARRAY_SIZE(recovery_pass_fns); i++)
 586                 if (recovery_pass_fns[i].when & PASS_FSCK)
 587                         ret |= BIT_ULL(i);
 588         return ret;
 589 }
 590
 591 static bool should_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
 592 {
 593         struct recovery_pass_fn *p = recovery_pass_fns + c->curr_recovery_pass;
 594
 595         if (c->opts.norecovery && pass > BCH_RECOVERY_PASS_snapshots_read)
 596                 return false;
 597         if (c->recovery_passes_explicit & BIT_ULL(pass))
 598                 return true;
 599         if ((p->when & PASS_FSCK) && c->opts.fsck)
 600                 return true;
 601         if ((p->when & PASS_UNCLEAN) && !c->sb.clean)
 602                 return true;
 603         if (p->when & PASS_ALWAYS)
 604                 return true;
 605         return false;
 606 }
 607
 608 static int bch2_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
 609 {
 610         int ret;
 611
 612         c->curr_recovery_pass = pass;
 613
 614         if (should_run_recovery_pass(c, pass)) {
 615                 struct recovery_pass_fn *p = recovery_pass_fns + pass;
 616
 617                 if (!(p->when & PASS_SILENT))
 618                         printk(KERN_INFO bch2_log_msg(c, "%s..."),
 619                                bch2_recovery_passes[pass]);
 620                 ret = p->fn(c);
 621                 if (ret)
 622                         return ret;
 623                 if (!(p->when & PASS_SILENT))
 624                         printk(KERN_CONT " done\n");
 625
 626                 c->recovery_passes_complete |= BIT_ULL(pass);
 627         }
 628
 629         return 0;
 630 }
 631
 632 static int bch2_run_recovery_passes(struct bch_fs *c)
 633 {
 634         int ret = 0;
 635
 636         while (c->curr_recovery_pass < ARRAY_SIZE(recovery_pass_fns)) {
 637                 ret = bch2_run_recovery_pass(c, c->curr_recovery_pass);
 638                 if (bch2_err_matches(ret, BCH_ERR_restart_recovery))
 639                         continue;
 640                 if (ret)
 641                         break;
 642                 c->curr_recovery_pass++;
 643         }
 644
 645         return ret;
 646 }
 647
 648 int bch2_fs_recovery(struct bch_fs *c)
 649 {
 650         struct bch_sb_field_clean *clean = NULL;
 651         struct jset *last_journal_entry = NULL;
 652         u64 last_seq, blacklist_seq, journal_seq;
 653         bool write_sb = false;
 654         int ret = 0;
 655
 656         if (c->sb.clean) {
 657                 clean = bch2_read_superblock_clean(c);
 658                 ret = PTR_ERR_OR_ZERO(clean);
 659                 if (ret)
 660                         goto err;
 661
 662                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 663                          le64_to_cpu(clean->journal_seq));
 664         } else {
 665                 bch_info(c, "recovering from unclean shutdown");
 666         }
 667
 668         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
 669                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
 670                 ret = -EINVAL;
 671                 goto err;
 672         }
 673
 674         if (!c->sb.clean &&
 675             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
 676                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
 677                 ret = -EINVAL;
 678                 goto err;
 679         }
 680
 681         if (c->opts.fsck || !(c->opts.nochanges && c->opts.norecovery))
 682                 check_version_upgrade(c);
 683
 684         if (c->opts.fsck && c->opts.norecovery) {
 685                 bch_err(c, "cannot select both norecovery and fsck");
 686                 ret = -EINVAL;
 687                 goto err;
 688         }
 689
 690         ret = bch2_blacklist_table_initialize(c);
 691         if (ret) {
 692                 bch_err(c, "error initializing blacklist table");
 693                 goto err;
 694         }
 695
 696         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
 697                 struct genradix_iter iter;
 698                 struct journal_replay **i;
 699
 700                 bch_verbose(c, "starting journal read");
 701                 ret = bch2_journal_read(c, &last_seq, &blacklist_seq, &journal_seq);
 702                 if (ret)
 703                         goto err;
 704
 705                 /*
 706                  * note: cmd_list_journal needs the blacklist table fully up to date so
 707                  * it can asterisk ignored journal entries:
 708                  */
 709                 if (c->opts.read_journal_only)
 710                         goto out;
 711
 712                 genradix_for_each_reverse(&c->journal_entries, iter, i)
 713                         if (*i && !(*i)->ignore) {
 714                                 last_journal_entry = &(*i)->j;
 715                                 break;
 716                         }
 717
 718                 if (mustfix_fsck_err_on(c->sb.clean &&
 719                                         last_journal_entry &&
 720                                         !journal_entry_empty(last_journal_entry), c,
 721                                 "filesystem marked clean but journal not empty")) {
 722                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 723                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 724                         c->sb.clean = false;
 725                 }
 726
 727                 if (!last_journal_entry) {
 728                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
 729                         if (clean)
 730                                 goto use_clean;
 731
 732                         genradix_for_each_reverse(&c->journal_entries, iter, i)
 733                                 if (*i) {
 734                                         last_journal_entry = &(*i)->j;
 735                                         (*i)->ignore = false;
 736                                         break;
 737                                 }
 738                 }
 739
 740                 ret = bch2_journal_keys_sort(c);
 741                 if (ret)
 742                         goto err;
 743
 744                 if (c->sb.clean && last_journal_entry) {
 745                         ret = bch2_verify_superblock_clean(c, &clean,
 746                                                       last_journal_entry);
 747                         if (ret)
 748                                 goto err;
 749                 }
 750         } else {
 751 use_clean:
 752                 if (!clean) {
 753                         bch_err(c, "no superblock clean section found");
 754                         ret = -BCH_ERR_fsck_repair_impossible;
 755                         goto err;
 756
 757                 }
 758                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
 759         }
 760
 761         c->journal_replay_seq_start     = last_seq;
 762         c->journal_replay_seq_end       = blacklist_seq - 1;;
 763
 764         if (c->opts.reconstruct_alloc) {
 765                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 766                 drop_alloc_keys(&c->journal_keys);
 767         }
 768
 769         zero_out_btree_mem_ptr(&c->journal_keys);
 770
 771         ret = journal_replay_early(c, clean);
 772         if (ret)
 773                 goto err;
 774
 775         /*
 776          * After an unclean shutdown, skip then next few journal sequence
 777          * numbers as they may have been referenced by btree writes that
 778          * happened before their corresponding journal writes - those btree
 779          * writes need to be ignored, by skipping and blacklisting the next few
 780          * journal sequence numbers:
 781          */
 782         if (!c->sb.clean)
 783                 journal_seq += 8;
 784
 785         if (blacklist_seq != journal_seq) {
 786                 ret =   bch2_journal_log_msg(c, "blacklisting entries %llu-%llu",
 787                                              blacklist_seq, journal_seq) ?:
 788                         bch2_journal_seq_blacklist_add(c,
 789                                         blacklist_seq, journal_seq);
 790                 if (ret) {
 791                         bch_err(c, "error creating new journal seq blacklist entry");
 792                         goto err;
 793                 }
 794         }
 795
 796         ret =   bch2_journal_log_msg(c, "starting journal at entry %llu, replaying %llu-%llu",
 797                                      journal_seq, last_seq, blacklist_seq - 1) ?:
 798                 bch2_fs_journal_start(&c->journal, journal_seq);
 799         if (ret)
 800                 goto err;
 801
 802         if (c->opts.reconstruct_alloc)
 803                 bch2_journal_log_msg(c, "dropping alloc info");
 804
 805         /*
 806          * Skip past versions that might have possibly been used (as nonces),
 807          * but hadn't had their pointers written:
 808          */
 809         if (c->sb.encryption_type && !c->sb.clean)
 810                 atomic64_add(1 << 16, &c->key_version);
 811
 812         ret = read_btree_roots(c);
 813         if (ret)
 814                 goto err;
 815
 816         if (c->opts.fsck &&
 817             (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) ||
 818              BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb)))
 819                 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_topology);
 820
 821         ret = bch2_run_recovery_passes(c);
 822         if (ret)
 823                 goto err;
 824
 825         /* If we fixed errors, verify that fs is actually clean now: */
 826         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) &&
 827             test_bit(BCH_FS_ERRORS_FIXED, &c->flags) &&
 828             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags) &&
 829             !test_bit(BCH_FS_ERROR, &c->flags)) {
 830                 bch_info(c, "Fixed errors, running fsck a second time to verify fs is clean");
 831                 clear_bit(BCH_FS_ERRORS_FIXED, &c->flags);
 832
 833                 c->curr_recovery_pass = BCH_RECOVERY_PASS_check_alloc_info;
 834
 835                 ret = bch2_run_recovery_passes(c);
 836                 if (ret)
 837                         goto err;
 838
 839                 if (test_bit(BCH_FS_ERRORS_FIXED, &c->flags) ||
 840                     test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
 841                         bch_err(c, "Second fsck run was not clean");
 842                         set_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags);
 843                 }
 844
 845                 set_bit(BCH_FS_ERRORS_FIXED, &c->flags);
 846         }
 847
 848         if (enabled_qtypes(c)) {
 849                 bch_verbose(c, "reading quotas");
 850                 ret = bch2_fs_quota_read(c);
 851                 if (ret)
 852                         goto err;
 853                 bch_verbose(c, "quotas done");
 854         }
 855
 856         mutex_lock(&c->sb_lock);
 857         if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) != c->sb.version) {
 858                 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, c->sb.version);
 859                 write_sb = true;
 860         }
 861
 862         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
 863                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
 864                 write_sb = true;
 865         }
 866
 867         if (c->opts.fsck &&
 868             !test_bit(BCH_FS_ERROR, &c->flags) &&
 869             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
 870                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
 871                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
 872                 write_sb = true;
 873         }
 874
 875         if (write_sb)
 876                 bch2_write_super(c);
 877         mutex_unlock(&c->sb_lock);
 878
 879         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
 880             c->sb.version_min < bcachefs_metadata_version_btree_ptr_sectors_written) {
 881                 struct bch_move_stats stats;
 882
 883                 bch2_move_stats_init(&stats, "recovery");
 884
 885                 bch_info(c, "scanning for old btree nodes");
 886                 ret =   bch2_fs_read_write(c) ?:
 887                         bch2_scan_old_btree_nodes(c, &stats);
 888                 if (ret)
 889                         goto err;
 890                 bch_info(c, "scanning for old btree nodes done");
 891         }
 892
 893         if (c->journal_seq_blacklist_table &&
 894             c->journal_seq_blacklist_table->nr > 128)
 895                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
 896
 897         ret = 0;
 898 out:
 899         set_bit(BCH_FS_FSCK_DONE, &c->flags);
 900         bch2_flush_fsck_errs(c);
 901
 902         if (!c->opts.keep_journal &&
 903             test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)) {
 904                 bch2_journal_keys_free(&c->journal_keys);
 905                 bch2_journal_entries_free(c);
 906         }
 907         kfree(clean);
 908
 909         if (!ret && test_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags)) {
 910                 bch2_fs_read_write_early(c);
 911                 bch2_delete_dead_snapshots_async(c);
 912         }
 913
 914         if (ret)
 915                 bch_err_fn(c, ret);
 916         return ret;
 917 err:
 918 fsck_err:
 919         bch2_fs_emergency_read_only(c);
 920         goto out;
 921 }
 922
 923 int bch2_fs_initialize(struct bch_fs *c)
 924 {
 925         struct bch_inode_unpacked root_inode, lostfound_inode;
 926         struct bkey_inode_buf packed_inode;
 927         struct qstr lostfound = QSTR("lost+found");
 928         struct bch_dev *ca;
 929         unsigned i;
 930         int ret;
 931
 932         bch_notice(c, "initializing new filesystem");
 933
 934         mutex_lock(&c->sb_lock);
 935         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
 936         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
 937
 938         bch2_sb_maybe_downgrade(c);
 939
 940         if (c->opts.version_upgrade != BCH_VERSION_UPGRADE_none) {
 941                 bch2_sb_upgrade(c, bcachefs_metadata_version_current);
 942                 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
 943                 bch2_write_super(c);
 944         }
 945         mutex_unlock(&c->sb_lock);
 946
 947         c->curr_recovery_pass = ARRAY_SIZE(recovery_pass_fns);
 948         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
 949         set_bit(BCH_FS_FSCK_DONE, &c->flags);
 950
 951         for (i = 0; i < BTREE_ID_NR; i++)
 952                 bch2_btree_root_alloc(c, i);
 953
 954         for_each_online_member(ca, c, i)
 955                 bch2_dev_usage_init(ca);
 956
 957         for_each_online_member(ca, c, i) {
 958                 ret = bch2_dev_journal_alloc(ca);
 959                 if (ret) {
 960                         percpu_ref_put(&ca->io_ref);
 961                         goto err;
 962                 }
 963         }
 964
 965         /*
 966          * journal_res_get() will crash if called before this has
 967          * set up the journal.pin FIFO and journal.cur pointer:
 968          */
 969         bch2_fs_journal_start(&c->journal, 1);
 970         bch2_journal_set_replay_done(&c->journal);
 971
 972         ret = bch2_fs_read_write_early(c);
 973         if (ret)
 974                 goto err;
 975
 976         /*
 977          * Write out the superblock and journal buckets, now that we can do
 978          * btree updates
 979          */
 980         bch_verbose(c, "marking superblocks");
 981         for_each_member_device(ca, c, i) {
 982                 ret = bch2_trans_mark_dev_sb(c, ca);
 983                 if (ret) {
 984                         percpu_ref_put(&ca->ref);
 985                         goto err;
 986                 }
 987
 988                 ca->new_fs_bucket_idx = 0;
 989         }
 990
 991         ret = bch2_fs_freespace_init(c);
 992         if (ret)
 993                 goto err;
 994
 995         ret = bch2_initialize_subvolumes(c);
 996         if (ret)
 997                 goto err;
 998
 999         bch_verbose(c, "reading snapshots table");
1000         ret = bch2_snapshots_read(c);
1001         if (ret)
1002                 goto err;
1003         bch_verbose(c, "reading snapshots done");
1004
1005         bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL);
1006         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1007         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1008         bch2_inode_pack(&packed_inode, &root_inode);
1009         packed_inode.inode.k.p.snapshot = U32_MAX;
1010
1011         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1012                                 &packed_inode.inode.k_i,
1013                                 NULL, NULL, 0);
1014         if (ret) {
1015                 bch_err_msg(c, ret, "creating root directory");
1016                 goto err;
1017         }
1018
1019         bch2_inode_init_early(c, &lostfound_inode);
1020
1021         ret = bch2_trans_do(c, NULL, NULL, 0,
1022                 bch2_create_trans(&trans,
1023                                   BCACHEFS_ROOT_SUBVOL_INUM,
1024                                   &root_inode, &lostfound_inode,
1025                                   &lostfound,
1026                                   0, 0, S_IFDIR|0700, 0,
1027                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1028         if (ret) {
1029                 bch_err_msg(c, ret, "creating lost+found");
1030                 goto err;
1031         }
1032
1033         if (enabled_qtypes(c)) {
1034                 ret = bch2_fs_quota_read(c);
1035                 if (ret)
1036                         goto err;
1037         }
1038
1039         ret = bch2_journal_flush(&c->journal);
1040         if (ret) {
1041                 bch_err_msg(c, ret, "writing first journal entry");
1042                 goto err;
1043         }
1044
1045         mutex_lock(&c->sb_lock);
1046         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1047         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1048
1049         bch2_write_super(c);
1050         mutex_unlock(&c->sb_lock);
1051
1052         return 0;
1053 err:
1054         bch_err_fn(ca, ret);
1055         return ret;
1056 }