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         /*
 163          * First, attempt to replay keys in sorted order. This is more
 164          * efficient, but some might fail if that would cause a journal
 165          * deadlock.
 166          */
 167         for (size_t i = 0; i < keys->nr; i++) {
 168                 cond_resched();
 169
 170                 struct journal_key *k = keys->d + i;
 171
 172                 ret = commit_do(trans, NULL, NULL,
 173                                 BCH_TRANS_COMMIT_no_enospc|
 174                                 BCH_TRANS_COMMIT_journal_reclaim|
 175                                 (!k->allocated ? BCH_TRANS_COMMIT_no_journal_res : 0),
 176                              bch2_journal_replay_key(trans, k));
 177                 BUG_ON(!ret && !k->overwritten);
 178                 if (ret) {
 179                         ret = darray_push(&keys_sorted, k);
 180                         if (ret)
 181                                 goto err;
 182                 }
 183         }
 184
 185         /*
 186          * Now, replay any remaining keys in the order in which they appear in
 187          * the journal, unpinning those journal entries as we go:
 188          */
 189         sort(keys_sorted.data, keys_sorted.nr,
 190              sizeof(keys_sorted.data[0]),
 191              journal_sort_seq_cmp, NULL);
 192
 193         darray_for_each(keys_sorted, kp) {
 194                 cond_resched();
 195
 196                 struct journal_key *k = *kp;
 197
 198                 replay_now_at(j, k->journal_seq);
 199
 200                 ret = commit_do(trans, NULL, NULL,
 201                                 BCH_TRANS_COMMIT_no_enospc|
 202                                 (!k->allocated
 203                                  ? BCH_TRANS_COMMIT_no_journal_res|BCH_WATERMARK_reclaim
 204                                  : 0),
 205                              bch2_journal_replay_key(trans, k));
 206                 bch_err_msg(c, ret, "while replaying key at btree %s level %u:",
 207                             bch2_btree_id_str(k->btree_id), k->level);
 208                 if (ret)
 209                         goto err;
 210
 211                 BUG_ON(!k->overwritten);
 212         }
 213
 214         bch2_trans_put(trans);
 215         trans = NULL;
 216
 217         replay_now_at(j, j->replay_journal_seq_end);
 218         j->replay_journal_seq = 0;
 219
 220         bch2_journal_set_replay_done(j);
 221         bch2_journal_flush_all_pins(j);
 222         ret = bch2_journal_error(j);
 223
 224         if (keys->nr && !ret)
 225                 bch2_journal_log_msg(c, "journal replay finished");
 226 err:
 227         if (trans)
 228                 bch2_trans_put(trans);
 229         darray_exit(&keys_sorted);
 230         bch_err_fn(c, ret);
 231         return ret;
 232 }
 233
 234 /* journal replay early: */
 235
 236 static int journal_replay_entry_early(struct bch_fs *c,
 237                                       struct jset_entry *entry)
 238 {
 239         int ret = 0;
 240
 241         switch (entry->type) {
 242         case BCH_JSET_ENTRY_btree_root: {
 243                 struct btree_root *r;
 244
 245                 while (entry->btree_id >= c->btree_roots_extra.nr + BTREE_ID_NR) {
 246                         ret = darray_push(&c->btree_roots_extra, (struct btree_root) { NULL });
 247                         if (ret)
 248                                 return ret;
 249                 }
 250
 251                 r = bch2_btree_id_root(c, entry->btree_id);
 252
 253                 if (entry->u64s) {
 254                         r->level = entry->level;
 255                         bkey_copy(&r->key, (struct bkey_i *) entry->start);
 256                         r->error = 0;
 257                 } else {
 258                         r->error = -EIO;
 259                 }
 260                 r->alive = true;
 261                 break;
 262         }
 263         case BCH_JSET_ENTRY_usage: {
 264                 struct jset_entry_usage *u =
 265                         container_of(entry, struct jset_entry_usage, entry);
 266
 267                 switch (entry->btree_id) {
 268                 case BCH_FS_USAGE_reserved:
 269                         if (entry->level < BCH_REPLICAS_MAX)
 270                                 c->usage_base->persistent_reserved[entry->level] =
 271                                         le64_to_cpu(u->v);
 272                         break;
 273                 case BCH_FS_USAGE_inodes:
 274                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 275                         break;
 276                 case BCH_FS_USAGE_key_version:
 277                         atomic64_set(&c->key_version,
 278                                      le64_to_cpu(u->v));
 279                         break;
 280                 }
 281
 282                 break;
 283         }
 284         case BCH_JSET_ENTRY_data_usage: {
 285                 struct jset_entry_data_usage *u =
 286                         container_of(entry, struct jset_entry_data_usage, entry);
 287
 288                 ret = bch2_replicas_set_usage(c, &u->r,
 289                                               le64_to_cpu(u->v));
 290                 break;
 291         }
 292         case BCH_JSET_ENTRY_dev_usage: {
 293                 struct jset_entry_dev_usage *u =
 294                         container_of(entry, struct jset_entry_dev_usage, entry);
 295                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 296                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 297
 298                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 299
 300                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 301                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 302                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 303                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 304                 }
 305
 306                 break;
 307         }
 308         case BCH_JSET_ENTRY_blacklist: {
 309                 struct jset_entry_blacklist *bl_entry =
 310                         container_of(entry, struct jset_entry_blacklist, entry);
 311
 312                 ret = bch2_journal_seq_blacklist_add(c,
 313                                 le64_to_cpu(bl_entry->seq),
 314                                 le64_to_cpu(bl_entry->seq) + 1);
 315                 break;
 316         }
 317         case BCH_JSET_ENTRY_blacklist_v2: {
 318                 struct jset_entry_blacklist_v2 *bl_entry =
 319                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 320
 321                 ret = bch2_journal_seq_blacklist_add(c,
 322                                 le64_to_cpu(bl_entry->start),
 323                                 le64_to_cpu(bl_entry->end) + 1);
 324                 break;
 325         }
 326         case BCH_JSET_ENTRY_clock: {
 327                 struct jset_entry_clock *clock =
 328                         container_of(entry, struct jset_entry_clock, entry);
 329
 330                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 331         }
 332         }
 333
 334         return ret;
 335 }
 336
 337 static int journal_replay_early(struct bch_fs *c,
 338                                 struct bch_sb_field_clean *clean)
 339 {
 340         struct jset_entry *entry;
 341         int ret;
 342
 343         if (clean) {
 344                 for (entry = clean->start;
 345                      entry != vstruct_end(&clean->field);
 346                      entry = vstruct_next(entry)) {
 347                         ret = journal_replay_entry_early(c, entry);
 348                         if (ret)
 349                                 return ret;
 350                 }
 351         } else {
 352                 struct genradix_iter iter;
 353                 struct journal_replay *i, **_i;
 354
 355                 genradix_for_each(&c->journal_entries, iter, _i) {
 356                         i = *_i;
 357
 358                         if (!i || i->ignore)
 359                                 continue;
 360
 361                         vstruct_for_each(&i->j, entry) {
 362                                 ret = journal_replay_entry_early(c, entry);
 363                                 if (ret)
 364                                         return ret;
 365                         }
 366                 }
 367         }
 368
 369         bch2_fs_usage_initialize(c);
 370
 371         return 0;
 372 }
 373
 374 /* sb clean section: */
 375
 376 static int read_btree_roots(struct bch_fs *c)
 377 {
 378         unsigned i;
 379         int ret = 0;
 380
 381         for (i = 0; i < btree_id_nr_alive(c); i++) {
 382                 struct btree_root *r = bch2_btree_id_root(c, i);
 383
 384                 if (!r->alive)
 385                         continue;
 386
 387                 if (btree_id_is_alloc(i) &&
 388                     c->opts.reconstruct_alloc) {
 389                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 390                         continue;
 391                 }
 392
 393                 if (r->error) {
 394                         __fsck_err(c,
 395                                    btree_id_is_alloc(i)
 396                                    ? FSCK_CAN_IGNORE : 0,
 397                                    btree_root_bkey_invalid,
 398                                    "invalid btree root %s",
 399                                    bch2_btree_id_str(i));
 400                         if (i == BTREE_ID_alloc)
 401                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 402                 }
 403
 404                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 405                 if (ret) {
 406                         fsck_err(c,
 407                                  btree_root_read_error,
 408                                  "error reading btree root %s",
 409                                  bch2_btree_id_str(i));
 410                         if (btree_id_is_alloc(i))
 411                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 412                         ret = 0;
 413                 }
 414         }
 415
 416         for (i = 0; i < BTREE_ID_NR; i++) {
 417                 struct btree_root *r = bch2_btree_id_root(c, i);
 418
 419                 if (!r->b) {
 420                         r->alive = false;
 421                         r->level = 0;
 422                         bch2_btree_root_alloc(c, i);
 423                 }
 424         }
 425 fsck_err:
 426         return ret;
 427 }
 428
 429 static int bch2_initialize_subvolumes(struct bch_fs *c)
 430 {
 431         struct bkey_i_snapshot_tree     root_tree;
 432         struct bkey_i_snapshot          root_snapshot;
 433         struct bkey_i_subvolume         root_volume;
 434         int ret;
 435
 436         bkey_snapshot_tree_init(&root_tree.k_i);
 437         root_tree.k.p.offset            = 1;
 438         root_tree.v.master_subvol       = cpu_to_le32(1);
 439         root_tree.v.root_snapshot       = cpu_to_le32(U32_MAX);
 440
 441         bkey_snapshot_init(&root_snapshot.k_i);
 442         root_snapshot.k.p.offset = U32_MAX;
 443         root_snapshot.v.flags   = 0;
 444         root_snapshot.v.parent  = 0;
 445         root_snapshot.v.subvol  = cpu_to_le32(BCACHEFS_ROOT_SUBVOL);
 446         root_snapshot.v.tree    = cpu_to_le32(1);
 447         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
 448
 449         bkey_subvolume_init(&root_volume.k_i);
 450         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
 451         root_volume.v.flags     = 0;
 452         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
 453         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
 454
 455         ret =   bch2_btree_insert(c, BTREE_ID_snapshot_trees,   &root_tree.k_i, NULL, 0) ?:
 456                 bch2_btree_insert(c, BTREE_ID_snapshots,        &root_snapshot.k_i, NULL, 0) ?:
 457                 bch2_btree_insert(c, BTREE_ID_subvolumes,       &root_volume.k_i, NULL, 0);
 458         if (ret)
 459                 bch_err_fn(c, ret);
 460         return ret;
 461 }
 462
 463 static int __bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
 464 {
 465         struct btree_iter iter;
 466         struct bkey_s_c k;
 467         struct bch_inode_unpacked inode;
 468         int ret;
 469
 470         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
 471                                SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
 472         ret = bkey_err(k);
 473         if (ret)
 474                 return ret;
 475
 476         if (!bkey_is_inode(k.k)) {
 477                 bch_err(trans->c, "root inode not found");
 478                 ret = -BCH_ERR_ENOENT_inode;
 479                 goto err;
 480         }
 481
 482         ret = bch2_inode_unpack(k, &inode);
 483         BUG_ON(ret);
 484
 485         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
 486
 487         ret = bch2_inode_write(trans, &iter, &inode);
 488 err:
 489         bch2_trans_iter_exit(trans, &iter);
 490         return ret;
 491 }
 492
 493 /* set bi_subvol on root inode */
 494 noinline_for_stack
 495 static int bch2_fs_upgrade_for_subvolumes(struct bch_fs *c)
 496 {
 497         int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_lazy_rw,
 498                                 __bch2_fs_upgrade_for_subvolumes(trans));
 499         if (ret)
 500                 bch_err_fn(c, ret);
 501         return ret;
 502 }
 503
 504 const char * const bch2_recovery_passes[] = {
 505 #define x(_fn, _when)   #_fn,
 506         BCH_RECOVERY_PASSES()
 507 #undef x
 508         NULL
 509 };
 510
 511 static int bch2_check_allocations(struct bch_fs *c)
 512 {
 513         return bch2_gc(c, true, c->opts.norecovery);
 514 }
 515
 516 static int bch2_set_may_go_rw(struct bch_fs *c)
 517 {
 518         struct journal_keys *keys = &c->journal_keys;
 519
 520         /*
 521          * After we go RW, the journal keys buffer can't be modified (except for
 522          * setting journal_key->overwritten: it will be accessed by multiple
 523          * threads
 524          */
 525         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 526         keys->gap = keys->nr;
 527
 528         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
 529         if (keys->nr)
 530                 return bch2_fs_read_write_early(c);
 531         return 0;
 532 }
 533
 534 struct recovery_pass_fn {
 535         int             (*fn)(struct bch_fs *);
 536         unsigned        when;
 537 };
 538
 539 static struct recovery_pass_fn recovery_pass_fns[] = {
 540 #define x(_fn, _when)   { .fn = bch2_##_fn, .when = _when },
 541         BCH_RECOVERY_PASSES()
 542 #undef x
 543 };
 544
 545 static void check_version_upgrade(struct bch_fs *c)
 546 {
 547         unsigned latest_compatible = bch2_latest_compatible_version(c->sb.version);
 548         unsigned latest_version = bcachefs_metadata_version_current;
 549         unsigned old_version = c->sb.version_upgrade_complete ?: c->sb.version;
 550         unsigned new_version = 0;
 551         u64 recovery_passes;
 552
 553         if (old_version < bcachefs_metadata_required_upgrade_below) {
 554                 if (c->opts.version_upgrade == BCH_VERSION_UPGRADE_incompatible ||
 555                     latest_compatible < bcachefs_metadata_required_upgrade_below)
 556                         new_version = latest_version;
 557                 else
 558                         new_version = latest_compatible;
 559         } else {
 560                 switch (c->opts.version_upgrade) {
 561                 case BCH_VERSION_UPGRADE_compatible:
 562                         new_version = latest_compatible;
 563                         break;
 564                 case BCH_VERSION_UPGRADE_incompatible:
 565                         new_version = latest_version;
 566                         break;
 567                 case BCH_VERSION_UPGRADE_none:
 568                         new_version = old_version;
 569                         break;
 570                 }
 571         }
 572
 573         if (new_version > old_version) {
 574                 struct printbuf buf = PRINTBUF;
 575
 576                 if (old_version < bcachefs_metadata_required_upgrade_below)
 577                         prt_str(&buf, "Version upgrade required:\n");
 578
 579                 if (old_version != c->sb.version) {
 580                         prt_str(&buf, "Version upgrade from ");
 581                         bch2_version_to_text(&buf, c->sb.version_upgrade_complete);
 582                         prt_str(&buf, " to ");
 583                         bch2_version_to_text(&buf, c->sb.version);
 584                         prt_str(&buf, " incomplete\n");
 585                 }
 586
 587                 prt_printf(&buf, "Doing %s version upgrade from ",
 588                            BCH_VERSION_MAJOR(old_version) != BCH_VERSION_MAJOR(new_version)
 589                            ? "incompatible" : "compatible");
 590                 bch2_version_to_text(&buf, old_version);
 591                 prt_str(&buf, " to ");
 592                 bch2_version_to_text(&buf, new_version);
 593                 prt_newline(&buf);
 594
 595                 recovery_passes = bch2_upgrade_recovery_passes(c, old_version, new_version);
 596                 if (recovery_passes) {
 597                         if ((recovery_passes & RECOVERY_PASS_ALL_FSCK) == RECOVERY_PASS_ALL_FSCK)
 598                                 prt_str(&buf, "fsck required");
 599                         else {
 600                                 prt_str(&buf, "running recovery passes: ");
 601                                 prt_bitflags(&buf, bch2_recovery_passes, recovery_passes);
 602                         }
 603
 604                         c->recovery_passes_explicit |= recovery_passes;
 605                         c->opts.fix_errors = FSCK_FIX_yes;
 606                 }
 607
 608                 bch_info(c, "%s", buf.buf);
 609
 610                 mutex_lock(&c->sb_lock);
 611                 bch2_sb_upgrade(c, new_version);
 612                 mutex_unlock(&c->sb_lock);
 613
 614                 printbuf_exit(&buf);
 615         }
 616 }
 617
 618 u64 bch2_fsck_recovery_passes(void)
 619 {
 620         u64 ret = 0;
 621
 622         for (unsigned i = 0; i < ARRAY_SIZE(recovery_pass_fns); i++)
 623                 if (recovery_pass_fns[i].when & PASS_FSCK)
 624                         ret |= BIT_ULL(i);
 625         return ret;
 626 }
 627
 628 static bool should_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
 629 {
 630         struct recovery_pass_fn *p = recovery_pass_fns + c->curr_recovery_pass;
 631
 632         if (c->opts.norecovery && pass > BCH_RECOVERY_PASS_snapshots_read)
 633                 return false;
 634         if (c->recovery_passes_explicit & BIT_ULL(pass))
 635                 return true;
 636         if ((p->when & PASS_FSCK) && c->opts.fsck)
 637                 return true;
 638         if ((p->when & PASS_UNCLEAN) && !c->sb.clean)
 639                 return true;
 640         if (p->when & PASS_ALWAYS)
 641                 return true;
 642         return false;
 643 }
 644
 645 static int bch2_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
 646 {
 647         int ret;
 648
 649         c->curr_recovery_pass = pass;
 650
 651         if (should_run_recovery_pass(c, pass)) {
 652                 struct recovery_pass_fn *p = recovery_pass_fns + pass;
 653
 654                 if (!(p->when & PASS_SILENT))
 655                         printk(KERN_INFO bch2_log_msg(c, "%s..."),
 656                                bch2_recovery_passes[pass]);
 657                 ret = p->fn(c);
 658                 if (ret)
 659                         return ret;
 660                 if (!(p->when & PASS_SILENT))
 661                         printk(KERN_CONT " done\n");
 662
 663                 c->recovery_passes_complete |= BIT_ULL(pass);
 664         }
 665
 666         return 0;
 667 }
 668
 669 static int bch2_run_recovery_passes(struct bch_fs *c)
 670 {
 671         int ret = 0;
 672
 673         while (c->curr_recovery_pass < ARRAY_SIZE(recovery_pass_fns)) {
 674                 ret = bch2_run_recovery_pass(c, c->curr_recovery_pass);
 675                 if (bch2_err_matches(ret, BCH_ERR_restart_recovery))
 676                         continue;
 677                 if (ret)
 678                         break;
 679                 c->curr_recovery_pass++;
 680         }
 681
 682         return ret;
 683 }
 684
 685 int bch2_fs_recovery(struct bch_fs *c)
 686 {
 687         struct bch_sb_field_clean *clean = NULL;
 688         struct jset *last_journal_entry = NULL;
 689         u64 last_seq = 0, blacklist_seq, journal_seq;
 690         bool write_sb = false;
 691         int ret = 0;
 692
 693         if (c->sb.clean) {
 694                 clean = bch2_read_superblock_clean(c);
 695                 ret = PTR_ERR_OR_ZERO(clean);
 696                 if (ret)
 697                         goto err;
 698
 699                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 700                          le64_to_cpu(clean->journal_seq));
 701         } else {
 702                 bch_info(c, "recovering from unclean shutdown");
 703         }
 704
 705         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
 706                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
 707                 ret = -EINVAL;
 708                 goto err;
 709         }
 710
 711         if (!c->sb.clean &&
 712             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
 713                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
 714                 ret = -EINVAL;
 715                 goto err;
 716         }
 717
 718         if (c->opts.fsck || !(c->opts.nochanges && c->opts.norecovery))
 719                 check_version_upgrade(c);
 720
 721         if (c->opts.fsck && c->opts.norecovery) {
 722                 bch_err(c, "cannot select both norecovery and fsck");
 723                 ret = -EINVAL;
 724                 goto err;
 725         }
 726
 727         ret = bch2_blacklist_table_initialize(c);
 728         if (ret) {
 729                 bch_err(c, "error initializing blacklist table");
 730                 goto err;
 731         }
 732
 733         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
 734                 struct genradix_iter iter;
 735                 struct journal_replay **i;
 736
 737                 bch_verbose(c, "starting journal read");
 738                 ret = bch2_journal_read(c, &last_seq, &blacklist_seq, &journal_seq);
 739                 if (ret)
 740                         goto err;
 741
 742                 /*
 743                  * note: cmd_list_journal needs the blacklist table fully up to date so
 744                  * it can asterisk ignored journal entries:
 745                  */
 746                 if (c->opts.read_journal_only)
 747                         goto out;
 748
 749                 genradix_for_each_reverse(&c->journal_entries, iter, i)
 750                         if (*i && !(*i)->ignore) {
 751                                 last_journal_entry = &(*i)->j;
 752                                 break;
 753                         }
 754
 755                 if (mustfix_fsck_err_on(c->sb.clean &&
 756                                         last_journal_entry &&
 757                                         !journal_entry_empty(last_journal_entry), c,
 758                                 clean_but_journal_not_empty,
 759                                 "filesystem marked clean but journal not empty")) {
 760                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 761                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 762                         c->sb.clean = false;
 763                 }
 764
 765                 if (!last_journal_entry) {
 766                         fsck_err_on(!c->sb.clean, c,
 767                                     dirty_but_no_journal_entries,
 768                                     "no journal entries found");
 769                         if (clean)
 770                                 goto use_clean;
 771
 772                         genradix_for_each_reverse(&c->journal_entries, iter, i)
 773                                 if (*i) {
 774                                         last_journal_entry = &(*i)->j;
 775                                         (*i)->ignore = false;
 776                                         /*
 777                                          * This was probably a NO_FLUSH entry,
 778                                          * so last_seq was garbage - but we know
 779                                          * we're only using a single journal
 780                                          * entry, set it here:
 781                                          */
 782                                         (*i)->j.last_seq = (*i)->j.seq;
 783                                         break;
 784                                 }
 785                 }
 786
 787                 ret = bch2_journal_keys_sort(c);
 788                 if (ret)
 789                         goto err;
 790
 791                 if (c->sb.clean && last_journal_entry) {
 792                         ret = bch2_verify_superblock_clean(c, &clean,
 793                                                       last_journal_entry);
 794                         if (ret)
 795                                 goto err;
 796                 }
 797         } else {
 798 use_clean:
 799                 if (!clean) {
 800                         bch_err(c, "no superblock clean section found");
 801                         ret = -BCH_ERR_fsck_repair_impossible;
 802                         goto err;
 803
 804                 }
 805                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
 806         }
 807
 808         c->journal_replay_seq_start     = last_seq;
 809         c->journal_replay_seq_end       = blacklist_seq - 1;
 810
 811         if (c->opts.reconstruct_alloc) {
 812                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 813                 drop_alloc_keys(&c->journal_keys);
 814         }
 815
 816         zero_out_btree_mem_ptr(&c->journal_keys);
 817
 818         ret = journal_replay_early(c, clean);
 819         if (ret)
 820                 goto err;
 821
 822         /*
 823          * After an unclean shutdown, skip then next few journal sequence
 824          * numbers as they may have been referenced by btree writes that
 825          * happened before their corresponding journal writes - those btree
 826          * writes need to be ignored, by skipping and blacklisting the next few
 827          * journal sequence numbers:
 828          */
 829         if (!c->sb.clean)
 830                 journal_seq += 8;
 831
 832         if (blacklist_seq != journal_seq) {
 833                 ret =   bch2_journal_log_msg(c, "blacklisting entries %llu-%llu",
 834                                              blacklist_seq, journal_seq) ?:
 835                         bch2_journal_seq_blacklist_add(c,
 836                                         blacklist_seq, journal_seq);
 837                 if (ret) {
 838                         bch_err(c, "error creating new journal seq blacklist entry");
 839                         goto err;
 840                 }
 841         }
 842
 843         ret =   bch2_journal_log_msg(c, "starting journal at entry %llu, replaying %llu-%llu",
 844                                      journal_seq, last_seq, blacklist_seq - 1) ?:
 845                 bch2_fs_journal_start(&c->journal, journal_seq);
 846         if (ret)
 847                 goto err;
 848
 849         if (c->opts.reconstruct_alloc)
 850                 bch2_journal_log_msg(c, "dropping alloc info");
 851
 852         /*
 853          * Skip past versions that might have possibly been used (as nonces),
 854          * but hadn't had their pointers written:
 855          */
 856         if (c->sb.encryption_type && !c->sb.clean)
 857                 atomic64_add(1 << 16, &c->key_version);
 858
 859         ret = read_btree_roots(c);
 860         if (ret)
 861                 goto err;
 862
 863         if (c->opts.fsck &&
 864             (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) ||
 865              BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb)))
 866                 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_topology);
 867
 868         ret = bch2_run_recovery_passes(c);
 869         if (ret)
 870                 goto err;
 871
 872         /* If we fixed errors, verify that fs is actually clean now: */
 873         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) &&
 874             test_bit(BCH_FS_ERRORS_FIXED, &c->flags) &&
 875             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags) &&
 876             !test_bit(BCH_FS_ERROR, &c->flags)) {
 877                 bch_info(c, "Fixed errors, running fsck a second time to verify fs is clean");
 878                 clear_bit(BCH_FS_ERRORS_FIXED, &c->flags);
 879
 880                 c->curr_recovery_pass = BCH_RECOVERY_PASS_check_alloc_info;
 881
 882                 ret = bch2_run_recovery_passes(c);
 883                 if (ret)
 884                         goto err;
 885
 886                 if (test_bit(BCH_FS_ERRORS_FIXED, &c->flags) ||
 887                     test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
 888                         bch_err(c, "Second fsck run was not clean");
 889                         set_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags);
 890                 }
 891
 892                 set_bit(BCH_FS_ERRORS_FIXED, &c->flags);
 893         }
 894
 895         if (enabled_qtypes(c)) {
 896                 bch_verbose(c, "reading quotas");
 897                 ret = bch2_fs_quota_read(c);
 898                 if (ret)
 899                         goto err;
 900                 bch_verbose(c, "quotas done");
 901         }
 902
 903         mutex_lock(&c->sb_lock);
 904         if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) != c->sb.version) {
 905                 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, c->sb.version);
 906                 write_sb = true;
 907         }
 908
 909         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
 910                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
 911                 write_sb = true;
 912         }
 913
 914         if (c->opts.fsck &&
 915             !test_bit(BCH_FS_ERROR, &c->flags) &&
 916             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
 917                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
 918                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
 919                 write_sb = true;
 920         }
 921
 922         if (write_sb)
 923                 bch2_write_super(c);
 924         mutex_unlock(&c->sb_lock);
 925
 926         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
 927             c->sb.version_min < bcachefs_metadata_version_btree_ptr_sectors_written) {
 928                 struct bch_move_stats stats;
 929
 930                 bch2_move_stats_init(&stats, "recovery");
 931
 932                 bch_info(c, "scanning for old btree nodes");
 933                 ret =   bch2_fs_read_write(c) ?:
 934                         bch2_scan_old_btree_nodes(c, &stats);
 935                 if (ret)
 936                         goto err;
 937                 bch_info(c, "scanning for old btree nodes done");
 938         }
 939
 940         if (c->journal_seq_blacklist_table &&
 941             c->journal_seq_blacklist_table->nr > 128)
 942                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
 943
 944         ret = 0;
 945 out:
 946         set_bit(BCH_FS_FSCK_DONE, &c->flags);
 947         bch2_flush_fsck_errs(c);
 948
 949         if (!c->opts.keep_journal &&
 950             test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)) {
 951                 bch2_journal_keys_free(&c->journal_keys);
 952                 bch2_journal_entries_free(c);
 953         }
 954         kfree(clean);
 955
 956         if (!ret && test_bit(BCH_FS_NEED_DELETE_DEAD_SNAPSHOTS, &c->flags)) {
 957                 bch2_fs_read_write_early(c);
 958                 bch2_delete_dead_snapshots_async(c);
 959         }
 960
 961         if (ret)
 962                 bch_err_fn(c, ret);
 963         return ret;
 964 err:
 965 fsck_err:
 966         bch2_fs_emergency_read_only(c);
 967         goto out;
 968 }
 969
 970 int bch2_fs_initialize(struct bch_fs *c)
 971 {
 972         struct bch_inode_unpacked root_inode, lostfound_inode;
 973         struct bkey_inode_buf packed_inode;
 974         struct qstr lostfound = QSTR("lost+found");
 975         struct bch_dev *ca;
 976         unsigned i;
 977         int ret;
 978
 979         bch_notice(c, "initializing new filesystem");
 980
 981         mutex_lock(&c->sb_lock);
 982         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
 983         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
 984
 985         bch2_sb_maybe_downgrade(c);
 986
 987         if (c->opts.version_upgrade != BCH_VERSION_UPGRADE_none) {
 988                 bch2_sb_upgrade(c, bcachefs_metadata_version_current);
 989                 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
 990                 bch2_write_super(c);
 991         }
 992         mutex_unlock(&c->sb_lock);
 993
 994         c->curr_recovery_pass = ARRAY_SIZE(recovery_pass_fns);
 995         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
 996         set_bit(BCH_FS_FSCK_DONE, &c->flags);
 997
 998         for (i = 0; i < BTREE_ID_NR; i++)
 999                 bch2_btree_root_alloc(c, i);
1000
1001         for_each_member_device(ca, c, i)
1002                 bch2_dev_usage_init(ca);
1003
1004         ret = bch2_fs_journal_alloc(c);
1005         if (ret)
1006                 goto err;
1007
1008         /*
1009          * journal_res_get() will crash if called before this has
1010          * set up the journal.pin FIFO and journal.cur pointer:
1011          */
1012         bch2_fs_journal_start(&c->journal, 1);
1013         bch2_journal_set_replay_done(&c->journal);
1014
1015         ret = bch2_fs_read_write_early(c);
1016         if (ret)
1017                 goto err;
1018
1019         /*
1020          * Write out the superblock and journal buckets, now that we can do
1021          * btree updates
1022          */
1023         bch_verbose(c, "marking superblocks");
1024         ret = bch2_trans_mark_dev_sbs(c);
1025         bch_err_msg(c, ret, "marking superblocks");
1026         if (ret)
1027                 goto err;
1028
1029         for_each_online_member(ca, c, i)
1030                 ca->new_fs_bucket_idx = 0;
1031
1032         ret = bch2_fs_freespace_init(c);
1033         if (ret)
1034                 goto err;
1035
1036         ret = bch2_initialize_subvolumes(c);
1037         if (ret)
1038                 goto err;
1039
1040         bch_verbose(c, "reading snapshots table");
1041         ret = bch2_snapshots_read(c);
1042         if (ret)
1043                 goto err;
1044         bch_verbose(c, "reading snapshots done");
1045
1046         bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL);
1047         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1048         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1049         bch2_inode_pack(&packed_inode, &root_inode);
1050         packed_inode.inode.k.p.snapshot = U32_MAX;
1051
1052         ret = bch2_btree_insert(c, BTREE_ID_inodes, &packed_inode.inode.k_i, NULL, 0);
1053         if (ret) {
1054                 bch_err_msg(c, ret, "creating root directory");
1055                 goto err;
1056         }
1057
1058         bch2_inode_init_early(c, &lostfound_inode);
1059
1060         ret = bch2_trans_do(c, NULL, NULL, 0,
1061                 bch2_create_trans(trans,
1062                                   BCACHEFS_ROOT_SUBVOL_INUM,
1063                                   &root_inode, &lostfound_inode,
1064                                   &lostfound,
1065                                   0, 0, S_IFDIR|0700, 0,
1066                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1067         if (ret) {
1068                 bch_err_msg(c, ret, "creating lost+found");
1069                 goto err;
1070         }
1071
1072         if (enabled_qtypes(c)) {
1073                 ret = bch2_fs_quota_read(c);
1074                 if (ret)
1075                         goto err;
1076         }
1077
1078         ret = bch2_journal_flush(&c->journal);
1079         if (ret) {
1080                 bch_err_msg(c, ret, "writing first journal entry");
1081                 goto err;
1082         }
1083
1084         mutex_lock(&c->sb_lock);
1085         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1086         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1087
1088         bch2_write_super(c);
1089         mutex_unlock(&c->sb_lock);
1090
1091         return 0;
1092 err:
1093         bch_err_fn(ca, ret);
1094         return ret;
1095 }