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 "logged_ops.h"
  24 #include "move.h"
  25 #include "quota.h"
  26 #include "rebalance.h"
  27 #include "recovery.h"
  28 #include "replicas.h"
  29 #include "sb-clean.h"
  30 #include "snapshot.h"
  31 #include "subvolume.h"
  32 #include "super-io.h"
  33
  34 #include <linux/sort.h>
  35 #include <linux/stat.h>
  36
  37 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  38
  39 static bool btree_id_is_alloc(enum btree_id id)
  40 {
  41         switch (id) {
  42         case BTREE_ID_alloc:
  43         case BTREE_ID_backpointers:
  44         case BTREE_ID_need_discard:
  45         case BTREE_ID_freespace:
  46         case BTREE_ID_bucket_gens:
  47                 return true;
  48         default:
  49                 return false;
  50         }
  51 }
  52
  53 /* for -o reconstruct_alloc: */
  54 static void drop_alloc_keys(struct journal_keys *keys)
  55 {
  56         size_t src, dst;
  57
  58         for (src = 0, dst = 0; src < keys->nr; src++)
  59                 if (!btree_id_is_alloc(keys->d[src].btree_id))
  60                         keys->d[dst++] = keys->d[src];
  61
  62         keys->nr = dst;
  63 }
  64
  65 /*
  66  * Btree node pointers have a field to stack a pointer to the in memory btree
  67  * node; we need to zero out this field when reading in btree nodes, or when
  68  * reading in keys from the journal:
  69  */
  70 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
  71 {
  72         struct journal_key *i;
  73
  74         for (i = keys->d; i < keys->d + keys->nr; i++)
  75                 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
  76                         bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
  77 }
  78
  79 /* journal replay: */
  80
  81 static void replay_now_at(struct journal *j, u64 seq)
  82 {
  83         BUG_ON(seq < j->replay_journal_seq);
  84
  85         seq = min(seq, j->replay_journal_seq_end);
  86
  87         while (j->replay_journal_seq < seq)
  88                 bch2_journal_pin_put(j, j->replay_journal_seq++);
  89 }
  90
  91 static int bch2_journal_replay_key(struct btree_trans *trans,
  92                                    struct journal_key *k)
  93 {
  94         struct btree_iter iter;
  95         unsigned iter_flags =
  96                 BTREE_ITER_INTENT|
  97                 BTREE_ITER_NOT_EXTENTS;
  98         unsigned update_flags = BTREE_TRIGGER_NORUN;
  99         int ret;
 100
 101         if (k->overwritten)
 102                 return 0;
 103
 104         trans->journal_res.seq = k->journal_seq;
 105
 106         /*
 107          * BTREE_UPDATE_KEY_CACHE_RECLAIM disables key cache lookup/update to
 108          * keep the key cache coherent with the underlying btree. Nothing
 109          * besides the allocator is doing updates yet so we don't need key cache
 110          * coherency for non-alloc btrees, and key cache fills for snapshots
 111          * btrees use BTREE_ITER_FILTER_SNAPSHOTS, which isn't available until
 112          * the snapshots recovery pass runs.
 113          */
 114         if (!k->level && k->btree_id == BTREE_ID_alloc)
 115                 iter_flags |= BTREE_ITER_CACHED;
 116         else
 117                 update_flags |= BTREE_UPDATE_KEY_CACHE_RECLAIM;
 118
 119         bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 120                                   BTREE_MAX_DEPTH, k->level,
 121                                   iter_flags);
 122         ret = bch2_btree_iter_traverse(&iter);
 123         if (ret)
 124                 goto out;
 125
 126         /* Must be checked with btree locked: */
 127         if (k->overwritten)
 128                 goto out;
 129
 130         ret = bch2_trans_update(trans, &iter, k->k, update_flags);
 131 out:
 132         bch2_trans_iter_exit(trans, &iter);
 133         return ret;
 134 }
 135
 136 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 137 {
 138         const struct journal_key *l = *((const struct journal_key **)_l);
 139         const struct journal_key *r = *((const struct journal_key **)_r);
 140
 141         return cmp_int(l->journal_seq, r->journal_seq);
 142 }
 143
 144 static int bch2_journal_replay(struct bch_fs *c)
 145 {
 146         struct journal_keys *keys = &c->journal_keys;
 147         DARRAY(struct journal_key *) keys_sorted = { 0 };
 148         struct journal_key **kp;
 149         struct journal *j = &c->journal;
 150         u64 start_seq   = c->journal_replay_seq_start;
 151         u64 end_seq     = c->journal_replay_seq_start;
 152         struct btree_trans *trans = bch2_trans_get(c);
 153         int ret;
 154
 155         if (keys->nr) {
 156                 ret = bch2_journal_log_msg(c, "Starting journal replay (%zu keys in entries %llu-%llu)",
 157                                            keys->nr, start_seq, end_seq);
 158                 if (ret)
 159                         goto err;
 160         }
 161
 162         BUG_ON(!atomic_read(&keys->ref));
 163
 164         /*
 165          * First, attempt to replay keys in sorted order. This is more
 166          * efficient - better locality of btree access -  but some might fail if
 167          * that would cause a journal deadlock.
 168          */
 169         for (size_t i = 0; i < keys->nr; i++) {
 170                 cond_resched();
 171
 172                 struct journal_key *k = keys->d + i;
 173
 174                 /* Skip fastpath if we're low on space in the journal */
 175                 ret = c->journal.watermark ? -1 :
 176                         commit_do(trans, NULL, NULL,
 177                                   BCH_TRANS_COMMIT_no_enospc|
 178                                   BCH_TRANS_COMMIT_journal_reclaim|
 179                                   (!k->allocated ? BCH_TRANS_COMMIT_no_journal_res : 0),
 180                              bch2_journal_replay_key(trans, k));
 181                 BUG_ON(!ret && !k->overwritten);
 182                 if (ret) {
 183                         ret = darray_push(&keys_sorted, k);
 184                         if (ret)
 185                                 goto err;
 186                 }
 187         }
 188
 189         /*
 190          * Now, replay any remaining keys in the order in which they appear in
 191          * the journal, unpinning those journal entries as we go:
 192          */
 193         sort(keys_sorted.data, keys_sorted.nr,
 194              sizeof(keys_sorted.data[0]),
 195              journal_sort_seq_cmp, NULL);
 196
 197         darray_for_each(keys_sorted, kp) {
 198                 cond_resched();
 199
 200                 struct journal_key *k = *kp;
 201
 202                 replay_now_at(j, k->journal_seq);
 203
 204                 ret = commit_do(trans, NULL, NULL,
 205                                 BCH_TRANS_COMMIT_no_enospc|
 206                                 (!k->allocated
 207                                  ? BCH_TRANS_COMMIT_no_journal_res|BCH_WATERMARK_reclaim
 208                                  : 0),
 209                              bch2_journal_replay_key(trans, k));
 210                 bch_err_msg(c, ret, "while replaying key at btree %s level %u:",
 211                             bch2_btree_id_str(k->btree_id), k->level);
 212                 if (ret)
 213                         goto err;
 214
 215                 BUG_ON(!k->overwritten);
 216         }
 217
 218         /*
 219          * We need to put our btree_trans before calling flush_all_pins(), since
 220          * that will use a btree_trans internally
 221          */
 222         bch2_trans_put(trans);
 223         trans = NULL;
 224
 225         if (!c->opts.keep_journal)
 226                 bch2_journal_keys_put_initial(c);
 227
 228         replay_now_at(j, j->replay_journal_seq_end);
 229         j->replay_journal_seq = 0;
 230
 231         bch2_journal_set_replay_done(j);
 232
 233         if (keys->nr)
 234                 bch2_journal_log_msg(c, "journal replay finished");
 235 err:
 236         if (trans)
 237                 bch2_trans_put(trans);
 238         darray_exit(&keys_sorted);
 239         bch_err_fn(c, ret);
 240         return ret;
 241 }
 242
 243 /* journal replay early: */
 244
 245 static int journal_replay_entry_early(struct bch_fs *c,
 246                                       struct jset_entry *entry)
 247 {
 248         int ret = 0;
 249
 250         switch (entry->type) {
 251         case BCH_JSET_ENTRY_btree_root: {
 252                 struct btree_root *r;
 253
 254                 while (entry->btree_id >= c->btree_roots_extra.nr + BTREE_ID_NR) {
 255                         ret = darray_push(&c->btree_roots_extra, (struct btree_root) { NULL });
 256                         if (ret)
 257                                 return ret;
 258                 }
 259
 260                 r = bch2_btree_id_root(c, entry->btree_id);
 261
 262                 if (entry->u64s) {
 263                         r->level = entry->level;
 264                         bkey_copy(&r->key, (struct bkey_i *) entry->start);
 265                         r->error = 0;
 266                 } else {
 267                         r->error = -EIO;
 268                 }
 269                 r->alive = true;
 270                 break;
 271         }
 272         case BCH_JSET_ENTRY_usage: {
 273                 struct jset_entry_usage *u =
 274                         container_of(entry, struct jset_entry_usage, entry);
 275
 276                 switch (entry->btree_id) {
 277                 case BCH_FS_USAGE_reserved:
 278                         if (entry->level < BCH_REPLICAS_MAX)
 279                                 c->usage_base->persistent_reserved[entry->level] =
 280                                         le64_to_cpu(u->v);
 281                         break;
 282                 case BCH_FS_USAGE_inodes:
 283                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 284                         break;
 285                 case BCH_FS_USAGE_key_version:
 286                         atomic64_set(&c->key_version,
 287                                      le64_to_cpu(u->v));
 288                         break;
 289                 }
 290
 291                 break;
 292         }
 293         case BCH_JSET_ENTRY_data_usage: {
 294                 struct jset_entry_data_usage *u =
 295                         container_of(entry, struct jset_entry_data_usage, entry);
 296
 297                 ret = bch2_replicas_set_usage(c, &u->r,
 298                                               le64_to_cpu(u->v));
 299                 break;
 300         }
 301         case BCH_JSET_ENTRY_dev_usage: {
 302                 struct jset_entry_dev_usage *u =
 303                         container_of(entry, struct jset_entry_dev_usage, entry);
 304                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 305                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 306
 307                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 308                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 309                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 310                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 311                 }
 312
 313                 break;
 314         }
 315         case BCH_JSET_ENTRY_blacklist: {
 316                 struct jset_entry_blacklist *bl_entry =
 317                         container_of(entry, struct jset_entry_blacklist, entry);
 318
 319                 ret = bch2_journal_seq_blacklist_add(c,
 320                                 le64_to_cpu(bl_entry->seq),
 321                                 le64_to_cpu(bl_entry->seq) + 1);
 322                 break;
 323         }
 324         case BCH_JSET_ENTRY_blacklist_v2: {
 325                 struct jset_entry_blacklist_v2 *bl_entry =
 326                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 327
 328                 ret = bch2_journal_seq_blacklist_add(c,
 329                                 le64_to_cpu(bl_entry->start),
 330                                 le64_to_cpu(bl_entry->end) + 1);
 331                 break;
 332         }
 333         case BCH_JSET_ENTRY_clock: {
 334                 struct jset_entry_clock *clock =
 335                         container_of(entry, struct jset_entry_clock, entry);
 336
 337                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 338         }
 339         }
 340
 341         return ret;
 342 }
 343
 344 static int journal_replay_early(struct bch_fs *c,
 345                                 struct bch_sb_field_clean *clean)
 346 {
 347         struct jset_entry *entry;
 348         int ret;
 349
 350         if (clean) {
 351                 for (entry = clean->start;
 352                      entry != vstruct_end(&clean->field);
 353                      entry = vstruct_next(entry)) {
 354                         ret = journal_replay_entry_early(c, entry);
 355                         if (ret)
 356                                 return ret;
 357                 }
 358         } else {
 359                 struct genradix_iter iter;
 360                 struct journal_replay *i, **_i;
 361
 362                 genradix_for_each(&c->journal_entries, iter, _i) {
 363                         i = *_i;
 364
 365                         if (!i || i->ignore)
 366                                 continue;
 367
 368                         vstruct_for_each(&i->j, entry) {
 369                                 ret = journal_replay_entry_early(c, entry);
 370                                 if (ret)
 371                                         return ret;
 372                         }
 373                 }
 374         }
 375
 376         bch2_fs_usage_initialize(c);
 377
 378         return 0;
 379 }
 380
 381 /* sb clean section: */
 382
 383 static int read_btree_roots(struct bch_fs *c)
 384 {
 385         unsigned i;
 386         int ret = 0;
 387
 388         for (i = 0; i < btree_id_nr_alive(c); i++) {
 389                 struct btree_root *r = bch2_btree_id_root(c, i);
 390
 391                 if (!r->alive)
 392                         continue;
 393
 394                 if (btree_id_is_alloc(i) &&
 395                     c->opts.reconstruct_alloc) {
 396                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 397                         continue;
 398                 }
 399
 400                 if (r->error) {
 401                         __fsck_err(c,
 402                                    btree_id_is_alloc(i)
 403                                    ? FSCK_CAN_IGNORE : 0,
 404                                    btree_root_bkey_invalid,
 405                                    "invalid btree root %s",
 406                                    bch2_btree_id_str(i));
 407                         if (i == BTREE_ID_alloc)
 408                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 409                 }
 410
 411                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 412                 if (ret) {
 413                         fsck_err(c,
 414                                  btree_root_read_error,
 415                                  "error reading btree root %s",
 416                                  bch2_btree_id_str(i));
 417                         if (btree_id_is_alloc(i))
 418                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 419                         ret = 0;
 420                 }
 421         }
 422
 423         for (i = 0; i < BTREE_ID_NR; i++) {
 424                 struct btree_root *r = bch2_btree_id_root(c, i);
 425
 426                 if (!r->b) {
 427                         r->alive = false;
 428                         r->level = 0;
 429                         bch2_btree_root_alloc(c, i);
 430                 }
 431         }
 432 fsck_err:
 433         return ret;
 434 }
 435
 436 static int bch2_initialize_subvolumes(struct bch_fs *c)
 437 {
 438         struct bkey_i_snapshot_tree     root_tree;
 439         struct bkey_i_snapshot          root_snapshot;
 440         struct bkey_i_subvolume         root_volume;
 441         int ret;
 442
 443         bkey_snapshot_tree_init(&root_tree.k_i);
 444         root_tree.k.p.offset            = 1;
 445         root_tree.v.master_subvol       = cpu_to_le32(1);
 446         root_tree.v.root_snapshot       = cpu_to_le32(U32_MAX);
 447
 448         bkey_snapshot_init(&root_snapshot.k_i);
 449         root_snapshot.k.p.offset = U32_MAX;
 450         root_snapshot.v.flags   = 0;
 451         root_snapshot.v.parent  = 0;
 452         root_snapshot.v.subvol  = cpu_to_le32(BCACHEFS_ROOT_SUBVOL);
 453         root_snapshot.v.tree    = cpu_to_le32(1);
 454         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
 455
 456         bkey_subvolume_init(&root_volume.k_i);
 457         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
 458         root_volume.v.flags     = 0;
 459         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
 460         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
 461
 462         ret =   bch2_btree_insert(c, BTREE_ID_snapshot_trees,   &root_tree.k_i, NULL, 0) ?:
 463                 bch2_btree_insert(c, BTREE_ID_snapshots,        &root_snapshot.k_i, NULL, 0) ?:
 464                 bch2_btree_insert(c, BTREE_ID_subvolumes,       &root_volume.k_i, NULL, 0);
 465         if (ret)
 466                 bch_err_fn(c, ret);
 467         return ret;
 468 }
 469
 470 static int __bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
 471 {
 472         struct btree_iter iter;
 473         struct bkey_s_c k;
 474         struct bch_inode_unpacked inode;
 475         int ret;
 476
 477         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
 478                                SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
 479         ret = bkey_err(k);
 480         if (ret)
 481                 return ret;
 482
 483         if (!bkey_is_inode(k.k)) {
 484                 bch_err(trans->c, "root inode not found");
 485                 ret = -BCH_ERR_ENOENT_inode;
 486                 goto err;
 487         }
 488
 489         ret = bch2_inode_unpack(k, &inode);
 490         BUG_ON(ret);
 491
 492         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
 493
 494         ret = bch2_inode_write(trans, &iter, &inode);
 495 err:
 496         bch2_trans_iter_exit(trans, &iter);
 497         return ret;
 498 }
 499
 500 /* set bi_subvol on root inode */
 501 noinline_for_stack
 502 static int bch2_fs_upgrade_for_subvolumes(struct bch_fs *c)
 503 {
 504         int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_lazy_rw,
 505                                 __bch2_fs_upgrade_for_subvolumes(trans));
 506         if (ret)
 507                 bch_err_fn(c, ret);
 508         return ret;
 509 }
 510
 511 const char * const bch2_recovery_passes[] = {
 512 #define x(_fn, _when)   #_fn,
 513         BCH_RECOVERY_PASSES()
 514 #undef x
 515         NULL
 516 };
 517
 518 static int bch2_check_allocations(struct bch_fs *c)
 519 {
 520         return bch2_gc(c, true, c->opts.norecovery);
 521 }
 522
 523 static int bch2_set_may_go_rw(struct bch_fs *c)
 524 {
 525         struct journal_keys *keys = &c->journal_keys;
 526
 527         /*
 528          * After we go RW, the journal keys buffer can't be modified (except for
 529          * setting journal_key->overwritten: it will be accessed by multiple
 530          * threads
 531          */
 532         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 533         keys->gap = keys->nr;
 534
 535         set_bit(BCH_FS_may_go_rw, &c->flags);
 536         if (keys->nr || c->opts.fsck)
 537                 return bch2_fs_read_write_early(c);
 538         return 0;
 539 }
 540
 541 struct recovery_pass_fn {
 542         int             (*fn)(struct bch_fs *);
 543         unsigned        when;
 544 };
 545
 546 static struct recovery_pass_fn recovery_pass_fns[] = {
 547 #define x(_fn, _when)   { .fn = bch2_##_fn, .when = _when },
 548         BCH_RECOVERY_PASSES()
 549 #undef x
 550 };
 551
 552 static void check_version_upgrade(struct bch_fs *c)
 553 {
 554         unsigned latest_compatible = bch2_latest_compatible_version(c->sb.version);
 555         unsigned latest_version = bcachefs_metadata_version_current;
 556         unsigned old_version = c->sb.version_upgrade_complete ?: c->sb.version;
 557         unsigned new_version = 0;
 558         u64 recovery_passes;
 559
 560         if (old_version < bcachefs_metadata_required_upgrade_below) {
 561                 if (c->opts.version_upgrade == BCH_VERSION_UPGRADE_incompatible ||
 562                     latest_compatible < bcachefs_metadata_required_upgrade_below)
 563                         new_version = latest_version;
 564                 else
 565                         new_version = latest_compatible;
 566         } else {
 567                 switch (c->opts.version_upgrade) {
 568                 case BCH_VERSION_UPGRADE_compatible:
 569                         new_version = latest_compatible;
 570                         break;
 571                 case BCH_VERSION_UPGRADE_incompatible:
 572                         new_version = latest_version;
 573                         break;
 574                 case BCH_VERSION_UPGRADE_none:
 575                         new_version = old_version;
 576                         break;
 577                 }
 578         }
 579
 580         if (new_version > old_version) {
 581                 struct printbuf buf = PRINTBUF;
 582
 583                 if (old_version < bcachefs_metadata_required_upgrade_below)
 584                         prt_str(&buf, "Version upgrade required:\n");
 585
 586                 if (old_version != c->sb.version) {
 587                         prt_str(&buf, "Version upgrade from ");
 588                         bch2_version_to_text(&buf, c->sb.version_upgrade_complete);
 589                         prt_str(&buf, " to ");
 590                         bch2_version_to_text(&buf, c->sb.version);
 591                         prt_str(&buf, " incomplete\n");
 592                 }
 593
 594                 prt_printf(&buf, "Doing %s version upgrade from ",
 595                            BCH_VERSION_MAJOR(old_version) != BCH_VERSION_MAJOR(new_version)
 596                            ? "incompatible" : "compatible");
 597                 bch2_version_to_text(&buf, old_version);
 598                 prt_str(&buf, " to ");
 599                 bch2_version_to_text(&buf, new_version);
 600                 prt_newline(&buf);
 601
 602                 recovery_passes = bch2_upgrade_recovery_passes(c, old_version, new_version);
 603                 if (recovery_passes) {
 604                         if ((recovery_passes & RECOVERY_PASS_ALL_FSCK) == RECOVERY_PASS_ALL_FSCK)
 605                                 prt_str(&buf, "fsck required");
 606                         else {
 607                                 prt_str(&buf, "running recovery passes: ");
 608                                 prt_bitflags(&buf, bch2_recovery_passes, recovery_passes);
 609                         }
 610
 611                         c->recovery_passes_explicit |= recovery_passes;
 612                         c->opts.fix_errors = FSCK_FIX_yes;
 613                 }
 614
 615                 bch_info(c, "%s", buf.buf);
 616
 617                 mutex_lock(&c->sb_lock);
 618                 bch2_sb_upgrade(c, new_version);
 619                 mutex_unlock(&c->sb_lock);
 620
 621                 printbuf_exit(&buf);
 622         }
 623 }
 624
 625 u64 bch2_fsck_recovery_passes(void)
 626 {
 627         u64 ret = 0;
 628
 629         for (unsigned i = 0; i < ARRAY_SIZE(recovery_pass_fns); i++)
 630                 if (recovery_pass_fns[i].when & PASS_FSCK)
 631                         ret |= BIT_ULL(i);
 632         return ret;
 633 }
 634
 635 static bool should_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
 636 {
 637         struct recovery_pass_fn *p = recovery_pass_fns + c->curr_recovery_pass;
 638
 639         if (c->opts.norecovery && pass > BCH_RECOVERY_PASS_snapshots_read)
 640                 return false;
 641         if (c->recovery_passes_explicit & BIT_ULL(pass))
 642                 return true;
 643         if ((p->when & PASS_FSCK) && c->opts.fsck)
 644                 return true;
 645         if ((p->when & PASS_UNCLEAN) && !c->sb.clean)
 646                 return true;
 647         if (p->when & PASS_ALWAYS)
 648                 return true;
 649         return false;
 650 }
 651
 652 static int bch2_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
 653 {
 654         int ret;
 655
 656         c->curr_recovery_pass = pass;
 657
 658         if (should_run_recovery_pass(c, pass)) {
 659                 struct recovery_pass_fn *p = recovery_pass_fns + pass;
 660
 661                 if (!(p->when & PASS_SILENT))
 662                         bch2_print(c, KERN_INFO bch2_log_msg(c, "%s..."),
 663                                    bch2_recovery_passes[pass]);
 664                 ret = p->fn(c);
 665                 if (ret)
 666                         return ret;
 667                 if (!(p->when & PASS_SILENT))
 668                         bch2_print(c, KERN_CONT " done\n");
 669
 670                 c->recovery_passes_complete |= BIT_ULL(pass);
 671         }
 672
 673         return 0;
 674 }
 675
 676 static int bch2_run_recovery_passes(struct bch_fs *c)
 677 {
 678         int ret = 0;
 679
 680         while (c->curr_recovery_pass < ARRAY_SIZE(recovery_pass_fns)) {
 681                 ret = bch2_run_recovery_pass(c, c->curr_recovery_pass);
 682                 if (bch2_err_matches(ret, BCH_ERR_restart_recovery))
 683                         continue;
 684                 if (ret)
 685                         break;
 686                 c->curr_recovery_pass++;
 687         }
 688
 689         return ret;
 690 }
 691
 692 int bch2_fs_recovery(struct bch_fs *c)
 693 {
 694         struct bch_sb_field_clean *clean = NULL;
 695         struct jset *last_journal_entry = NULL;
 696         u64 last_seq = 0, blacklist_seq, journal_seq;
 697         bool write_sb = false;
 698         int ret = 0;
 699
 700         if (c->sb.clean) {
 701                 clean = bch2_read_superblock_clean(c);
 702                 ret = PTR_ERR_OR_ZERO(clean);
 703                 if (ret)
 704                         goto err;
 705
 706                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 707                          le64_to_cpu(clean->journal_seq));
 708         } else {
 709                 bch_info(c, "recovering from unclean shutdown");
 710         }
 711
 712         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
 713                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
 714                 ret = -EINVAL;
 715                 goto err;
 716         }
 717
 718         if (!c->sb.clean &&
 719             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
 720                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
 721                 ret = -EINVAL;
 722                 goto err;
 723         }
 724
 725         if (c->opts.fsck || !(c->opts.nochanges && c->opts.norecovery))
 726                 check_version_upgrade(c);
 727
 728         if (c->opts.fsck && c->opts.norecovery) {
 729                 bch_err(c, "cannot select both norecovery and fsck");
 730                 ret = -EINVAL;
 731                 goto err;
 732         }
 733
 734         ret = bch2_blacklist_table_initialize(c);
 735         if (ret) {
 736                 bch_err(c, "error initializing blacklist table");
 737                 goto err;
 738         }
 739
 740         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
 741                 struct genradix_iter iter;
 742                 struct journal_replay **i;
 743
 744                 bch_verbose(c, "starting journal read");
 745                 ret = bch2_journal_read(c, &last_seq, &blacklist_seq, &journal_seq);
 746                 if (ret)
 747                         goto err;
 748
 749                 /*
 750                  * note: cmd_list_journal needs the blacklist table fully up to date so
 751                  * it can asterisk ignored journal entries:
 752                  */
 753                 if (c->opts.read_journal_only)
 754                         goto out;
 755
 756                 genradix_for_each_reverse(&c->journal_entries, iter, i)
 757                         if (*i && !(*i)->ignore) {
 758                                 last_journal_entry = &(*i)->j;
 759                                 break;
 760                         }
 761
 762                 if (mustfix_fsck_err_on(c->sb.clean &&
 763                                         last_journal_entry &&
 764                                         !journal_entry_empty(last_journal_entry), c,
 765                                 clean_but_journal_not_empty,
 766                                 "filesystem marked clean but journal not empty")) {
 767                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 768                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 769                         c->sb.clean = false;
 770                 }
 771
 772                 if (!last_journal_entry) {
 773                         fsck_err_on(!c->sb.clean, c,
 774                                     dirty_but_no_journal_entries,
 775                                     "no journal entries found");
 776                         if (clean)
 777                                 goto use_clean;
 778
 779                         genradix_for_each_reverse(&c->journal_entries, iter, i)
 780                                 if (*i) {
 781                                         last_journal_entry = &(*i)->j;
 782                                         (*i)->ignore = false;
 783                                         /*
 784                                          * This was probably a NO_FLUSH entry,
 785                                          * so last_seq was garbage - but we know
 786                                          * we're only using a single journal
 787                                          * entry, set it here:
 788                                          */
 789                                         (*i)->j.last_seq = (*i)->j.seq;
 790                                         break;
 791                                 }
 792                 }
 793
 794                 ret = bch2_journal_keys_sort(c);
 795                 if (ret)
 796                         goto err;
 797
 798                 if (c->sb.clean && last_journal_entry) {
 799                         ret = bch2_verify_superblock_clean(c, &clean,
 800                                                       last_journal_entry);
 801                         if (ret)
 802                                 goto err;
 803                 }
 804         } else {
 805 use_clean:
 806                 if (!clean) {
 807                         bch_err(c, "no superblock clean section found");
 808                         ret = -BCH_ERR_fsck_repair_impossible;
 809                         goto err;
 810
 811                 }
 812                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
 813         }
 814
 815         c->journal_replay_seq_start     = last_seq;
 816         c->journal_replay_seq_end       = blacklist_seq - 1;
 817
 818         if (c->opts.reconstruct_alloc) {
 819                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 820                 drop_alloc_keys(&c->journal_keys);
 821         }
 822
 823         zero_out_btree_mem_ptr(&c->journal_keys);
 824
 825         ret = journal_replay_early(c, clean);
 826         if (ret)
 827                 goto err;
 828
 829         /*
 830          * After an unclean shutdown, skip then next few journal sequence
 831          * numbers as they may have been referenced by btree writes that
 832          * happened before their corresponding journal writes - those btree
 833          * writes need to be ignored, by skipping and blacklisting the next few
 834          * journal sequence numbers:
 835          */
 836         if (!c->sb.clean)
 837                 journal_seq += 8;
 838
 839         if (blacklist_seq != journal_seq) {
 840                 ret =   bch2_journal_log_msg(c, "blacklisting entries %llu-%llu",
 841                                              blacklist_seq, journal_seq) ?:
 842                         bch2_journal_seq_blacklist_add(c,
 843                                         blacklist_seq, journal_seq);
 844                 if (ret) {
 845                         bch_err(c, "error creating new journal seq blacklist entry");
 846                         goto err;
 847                 }
 848         }
 849
 850         ret =   bch2_journal_log_msg(c, "starting journal at entry %llu, replaying %llu-%llu",
 851                                      journal_seq, last_seq, blacklist_seq - 1) ?:
 852                 bch2_fs_journal_start(&c->journal, journal_seq);
 853         if (ret)
 854                 goto err;
 855
 856         if (c->opts.reconstruct_alloc)
 857                 bch2_journal_log_msg(c, "dropping alloc info");
 858
 859         /*
 860          * Skip past versions that might have possibly been used (as nonces),
 861          * but hadn't had their pointers written:
 862          */
 863         if (c->sb.encryption_type && !c->sb.clean)
 864                 atomic64_add(1 << 16, &c->key_version);
 865
 866         ret = read_btree_roots(c);
 867         if (ret)
 868                 goto err;
 869
 870         if (c->opts.fsck &&
 871             (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) ||
 872              BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb)))
 873                 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_topology);
 874
 875         ret = bch2_run_recovery_passes(c);
 876         if (ret)
 877                 goto err;
 878
 879         /* If we fixed errors, verify that fs is actually clean now: */
 880         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) &&
 881             test_bit(BCH_FS_errors_fixed, &c->flags) &&
 882             !test_bit(BCH_FS_errors_not_fixed, &c->flags) &&
 883             !test_bit(BCH_FS_error, &c->flags)) {
 884                 bch2_flush_fsck_errs(c);
 885
 886                 bch_info(c, "Fixed errors, running fsck a second time to verify fs is clean");
 887                 clear_bit(BCH_FS_errors_fixed, &c->flags);
 888
 889                 c->curr_recovery_pass = BCH_RECOVERY_PASS_check_alloc_info;
 890
 891                 ret = bch2_run_recovery_passes(c);
 892                 if (ret)
 893                         goto err;
 894
 895                 if (test_bit(BCH_FS_errors_fixed, &c->flags) ||
 896                     test_bit(BCH_FS_errors_not_fixed, &c->flags)) {
 897                         bch_err(c, "Second fsck run was not clean");
 898                         set_bit(BCH_FS_errors_not_fixed, &c->flags);
 899                 }
 900
 901                 set_bit(BCH_FS_errors_fixed, &c->flags);
 902         }
 903
 904         if (enabled_qtypes(c)) {
 905                 bch_verbose(c, "reading quotas");
 906                 ret = bch2_fs_quota_read(c);
 907                 if (ret)
 908                         goto err;
 909                 bch_verbose(c, "quotas done");
 910         }
 911
 912         mutex_lock(&c->sb_lock);
 913         if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) != c->sb.version) {
 914                 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, c->sb.version);
 915                 write_sb = true;
 916         }
 917
 918         if (!test_bit(BCH_FS_error, &c->flags)) {
 919                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
 920                 write_sb = true;
 921         }
 922
 923         if (c->opts.fsck &&
 924             !test_bit(BCH_FS_error, &c->flags) &&
 925             !test_bit(BCH_FS_errors_not_fixed, &c->flags)) {
 926                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
 927                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
 928                 write_sb = true;
 929         }
 930
 931         if (write_sb)
 932                 bch2_write_super(c);
 933         mutex_unlock(&c->sb_lock);
 934
 935         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
 936             c->sb.version_min < bcachefs_metadata_version_btree_ptr_sectors_written) {
 937                 struct bch_move_stats stats;
 938
 939                 bch2_move_stats_init(&stats, "recovery");
 940
 941                 struct printbuf buf = PRINTBUF;
 942                 bch2_version_to_text(&buf, c->sb.version_min);
 943                 bch_info(c, "scanning for old btree nodes: min_version %s", buf.buf);
 944                 printbuf_exit(&buf);
 945
 946                 ret =   bch2_fs_read_write_early(c) ?:
 947                         bch2_scan_old_btree_nodes(c, &stats);
 948                 if (ret)
 949                         goto err;
 950                 bch_info(c, "scanning for old btree nodes done");
 951         }
 952
 953         if (c->journal_seq_blacklist_table &&
 954             c->journal_seq_blacklist_table->nr > 128)
 955                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
 956
 957         ret = 0;
 958 out:
 959         set_bit(BCH_FS_fsck_done, &c->flags);
 960         bch2_flush_fsck_errs(c);
 961
 962         if (!c->opts.keep_journal &&
 963             test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags))
 964                 bch2_journal_keys_put_initial(c);
 965         kfree(clean);
 966
 967         if (!ret && test_bit(BCH_FS_need_delete_dead_snapshots, &c->flags)) {
 968                 bch2_fs_read_write_early(c);
 969                 bch2_delete_dead_snapshots_async(c);
 970         }
 971
 972         if (ret)
 973                 bch_err_fn(c, ret);
 974         return ret;
 975 err:
 976 fsck_err:
 977         bch2_fs_emergency_read_only(c);
 978         goto out;
 979 }
 980
 981 int bch2_fs_initialize(struct bch_fs *c)
 982 {
 983         struct bch_inode_unpacked root_inode, lostfound_inode;
 984         struct bkey_inode_buf packed_inode;
 985         struct qstr lostfound = QSTR("lost+found");
 986         struct bch_dev *ca;
 987         unsigned i;
 988         int ret;
 989
 990         bch_notice(c, "initializing new filesystem");
 991
 992         mutex_lock(&c->sb_lock);
 993         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
 994         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
 995
 996         bch2_sb_maybe_downgrade(c);
 997
 998         if (c->opts.version_upgrade != BCH_VERSION_UPGRADE_none) {
 999                 bch2_sb_upgrade(c, bcachefs_metadata_version_current);
1000                 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
1001                 bch2_write_super(c);
1002         }
1003         mutex_unlock(&c->sb_lock);
1004
1005         c->curr_recovery_pass = ARRAY_SIZE(recovery_pass_fns);
1006         set_bit(BCH_FS_may_go_rw, &c->flags);
1007         set_bit(BCH_FS_fsck_done, &c->flags);
1008
1009         for (i = 0; i < BTREE_ID_NR; i++)
1010                 bch2_btree_root_alloc(c, i);
1011
1012         for_each_member_device(ca, c, i)
1013                 bch2_dev_usage_init(ca);
1014
1015         ret = bch2_fs_journal_alloc(c);
1016         if (ret)
1017                 goto err;
1018
1019         /*
1020          * journal_res_get() will crash if called before this has
1021          * set up the journal.pin FIFO and journal.cur pointer:
1022          */
1023         bch2_fs_journal_start(&c->journal, 1);
1024         bch2_journal_set_replay_done(&c->journal);
1025
1026         ret = bch2_fs_read_write_early(c);
1027         if (ret)
1028                 goto err;
1029
1030         /*
1031          * Write out the superblock and journal buckets, now that we can do
1032          * btree updates
1033          */
1034         bch_verbose(c, "marking superblocks");
1035         ret = bch2_trans_mark_dev_sbs(c);
1036         bch_err_msg(c, ret, "marking superblocks");
1037         if (ret)
1038                 goto err;
1039
1040         for_each_online_member(ca, c, i)
1041                 ca->new_fs_bucket_idx = 0;
1042
1043         ret = bch2_fs_freespace_init(c);
1044         if (ret)
1045                 goto err;
1046
1047         ret = bch2_initialize_subvolumes(c);
1048         if (ret)
1049                 goto err;
1050
1051         bch_verbose(c, "reading snapshots table");
1052         ret = bch2_snapshots_read(c);
1053         if (ret)
1054                 goto err;
1055         bch_verbose(c, "reading snapshots done");
1056
1057         bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL);
1058         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1059         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1060         bch2_inode_pack(&packed_inode, &root_inode);
1061         packed_inode.inode.k.p.snapshot = U32_MAX;
1062
1063         ret = bch2_btree_insert(c, BTREE_ID_inodes, &packed_inode.inode.k_i, NULL, 0);
1064         if (ret) {
1065                 bch_err_msg(c, ret, "creating root directory");
1066                 goto err;
1067         }
1068
1069         bch2_inode_init_early(c, &lostfound_inode);
1070
1071         ret = bch2_trans_do(c, NULL, NULL, 0,
1072                 bch2_create_trans(trans,
1073                                   BCACHEFS_ROOT_SUBVOL_INUM,
1074                                   &root_inode, &lostfound_inode,
1075                                   &lostfound,
1076                                   0, 0, S_IFDIR|0700, 0,
1077                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1078         if (ret) {
1079                 bch_err_msg(c, ret, "creating lost+found");
1080                 goto err;
1081         }
1082
1083         if (enabled_qtypes(c)) {
1084                 ret = bch2_fs_quota_read(c);
1085                 if (ret)
1086                         goto err;
1087         }
1088
1089         ret = bch2_journal_flush(&c->journal);
1090         if (ret) {
1091                 bch_err_msg(c, ret, "writing first journal entry");
1092                 goto err;
1093         }
1094
1095         mutex_lock(&c->sb_lock);
1096         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1097         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1098
1099         bch2_write_super(c);
1100         mutex_unlock(&c->sb_lock);
1101
1102         return 0;
1103 err:
1104         bch_err_fn(ca, ret);
1105         return ret;
1106 }