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