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 - better locality of btree access -  but some might fail if
 165          * that would cause a journal 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         /*
 215          * We need to put our btree_trans before calling flush_all_pins(), since
 216          * that will use a btree_trans internally
 217          */
 218         bch2_trans_put(trans);
 219         trans = NULL;
 220
 221         replay_now_at(j, j->replay_journal_seq_end);
 222         j->replay_journal_seq = 0;
 223
 224         bch2_journal_set_replay_done(j);
 225         bch2_journal_flush_all_pins(j);
 226         ret = bch2_journal_error(j);
 227
 228         if (keys->nr && !ret)
 229                 bch2_journal_log_msg(c, "journal replay finished");
 230 err:
 231         if (trans)
 232                 bch2_trans_put(trans);
 233         darray_exit(&keys_sorted);
 234         bch_err_fn(c, ret);
 235         return ret;
 236 }
 237
 238 /* journal replay early: */
 239
 240 static int journal_replay_entry_early(struct bch_fs *c,
 241                                       struct jset_entry *entry)
 242 {
 243         int ret = 0;
 244
 245         switch (entry->type) {
 246         case BCH_JSET_ENTRY_btree_root: {
 247                 struct btree_root *r;
 248
 249                 while (entry->btree_id >= c->btree_roots_extra.nr + BTREE_ID_NR) {
 250                         ret = darray_push(&c->btree_roots_extra, (struct btree_root) { NULL });
 251                         if (ret)
 252                                 return ret;
 253                 }
 254
 255                 r = bch2_btree_id_root(c, entry->btree_id);
 256
 257                 if (entry->u64s) {
 258                         r->level = entry->level;
 259                         bkey_copy(&r->key, (struct bkey_i *) entry->start);
 260                         r->error = 0;
 261                 } else {
 262                         r->error = -EIO;
 263                 }
 264                 r->alive = true;
 265                 break;
 266         }
 267         case BCH_JSET_ENTRY_usage: {
 268                 struct jset_entry_usage *u =
 269                         container_of(entry, struct jset_entry_usage, entry);
 270
 271                 switch (entry->btree_id) {
 272                 case BCH_FS_USAGE_reserved:
 273                         if (entry->level < BCH_REPLICAS_MAX)
 274                                 c->usage_base->persistent_reserved[entry->level] =
 275                                         le64_to_cpu(u->v);
 276                         break;
 277                 case BCH_FS_USAGE_inodes:
 278                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 279                         break;
 280                 case BCH_FS_USAGE_key_version:
 281                         atomic64_set(&c->key_version,
 282                                      le64_to_cpu(u->v));
 283                         break;
 284                 }
 285
 286                 break;
 287         }
 288         case BCH_JSET_ENTRY_data_usage: {
 289                 struct jset_entry_data_usage *u =
 290                         container_of(entry, struct jset_entry_data_usage, entry);
 291
 292                 ret = bch2_replicas_set_usage(c, &u->r,
 293                                               le64_to_cpu(u->v));
 294                 break;
 295         }
 296         case BCH_JSET_ENTRY_dev_usage: {
 297                 struct jset_entry_dev_usage *u =
 298                         container_of(entry, struct jset_entry_dev_usage, entry);
 299                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 300                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 301
 302                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 303
 304                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 305                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 306                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 307                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 308                 }
 309
 310                 break;
 311         }
 312         case BCH_JSET_ENTRY_blacklist: {
 313                 struct jset_entry_blacklist *bl_entry =
 314                         container_of(entry, struct jset_entry_blacklist, entry);
 315
 316                 ret = bch2_journal_seq_blacklist_add(c,
 317                                 le64_to_cpu(bl_entry->seq),
 318                                 le64_to_cpu(bl_entry->seq) + 1);
 319                 break;
 320         }
 321         case BCH_JSET_ENTRY_blacklist_v2: {
 322                 struct jset_entry_blacklist_v2 *bl_entry =
 323                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 324
 325                 ret = bch2_journal_seq_blacklist_add(c,
 326                                 le64_to_cpu(bl_entry->start),
 327                                 le64_to_cpu(bl_entry->end) + 1);
 328                 break;
 329         }
 330         case BCH_JSET_ENTRY_clock: {
 331                 struct jset_entry_clock *clock =
 332                         container_of(entry, struct jset_entry_clock, entry);
 333
 334                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 335         }
 336         }
 337
 338         return ret;
 339 }
 340
 341 static int journal_replay_early(struct bch_fs *c,
 342                                 struct bch_sb_field_clean *clean)
 343 {
 344         struct jset_entry *entry;
 345         int ret;
 346
 347         if (clean) {
 348                 for (entry = clean->start;
 349                      entry != vstruct_end(&clean->field);
 350                      entry = vstruct_next(entry)) {
 351                         ret = journal_replay_entry_early(c, entry);
 352                         if (ret)
 353                                 return ret;
 354                 }
 355         } else {
 356                 struct genradix_iter iter;
 357                 struct journal_replay *i, **_i;
 358
 359                 genradix_for_each(&c->journal_entries, iter, _i) {
 360                         i = *_i;
 361
 362                         if (!i || i->ignore)
 363                                 continue;
 364
 365                         vstruct_for_each(&i->j, entry) {
 366                                 ret = journal_replay_entry_early(c, entry);
 367                                 if (ret)
 368                                         return ret;
 369                         }
 370                 }
 371         }
 372
 373         bch2_fs_usage_initialize(c);
 374
 375         return 0;
 376 }
 377
 378 /* sb clean section: */
 379
 380 static int read_btree_roots(struct bch_fs *c)
 381 {
 382         unsigned i;
 383         int ret = 0;
 384
 385         for (i = 0; i < btree_id_nr_alive(c); i++) {
 386                 struct btree_root *r = bch2_btree_id_root(c, i);
 387
 388                 if (!r->alive)
 389                         continue;
 390
 391                 if (btree_id_is_alloc(i) &&
 392                     c->opts.reconstruct_alloc) {
 393                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 394                         continue;
 395                 }
 396
 397                 if (r->error) {
 398                         __fsck_err(c,
 399                                    btree_id_is_alloc(i)
 400                                    ? FSCK_CAN_IGNORE : 0,
 401                                    btree_root_bkey_invalid,
 402                                    "invalid btree root %s",
 403                                    bch2_btree_id_str(i));
 404                         if (i == BTREE_ID_alloc)
 405                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 406                 }
 407
 408                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 409                 if (ret) {
 410                         fsck_err(c,
 411                                  btree_root_read_error,
 412                                  "error reading btree root %s",
 413                                  bch2_btree_id_str(i));
 414                         if (btree_id_is_alloc(i))
 415                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 416                         ret = 0;
 417                 }
 418         }
 419
 420         for (i = 0; i < BTREE_ID_NR; i++) {
 421                 struct btree_root *r = bch2_btree_id_root(c, i);
 422
 423                 if (!r->b) {
 424                         r->alive = false;
 425                         r->level = 0;
 426                         bch2_btree_root_alloc(c, i);
 427                 }
 428         }
 429 fsck_err:
 430         return ret;
 431 }
 432
 433 static int bch2_initialize_subvolumes(struct bch_fs *c)
 434 {
 435         struct bkey_i_snapshot_tree     root_tree;
 436         struct bkey_i_snapshot          root_snapshot;
 437         struct bkey_i_subvolume         root_volume;
 438         int ret;
 439
 440         bkey_snapshot_tree_init(&root_tree.k_i);
 441         root_tree.k.p.offset            = 1;
 442         root_tree.v.master_subvol       = cpu_to_le32(1);
 443         root_tree.v.root_snapshot       = cpu_to_le32(U32_MAX);
 444
 445         bkey_snapshot_init(&root_snapshot.k_i);
 446         root_snapshot.k.p.offset = U32_MAX;
 447         root_snapshot.v.flags   = 0;
 448         root_snapshot.v.parent  = 0;
 449         root_snapshot.v.subvol  = cpu_to_le32(BCACHEFS_ROOT_SUBVOL);
 450         root_snapshot.v.tree    = cpu_to_le32(1);
 451         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
 452
 453         bkey_subvolume_init(&root_volume.k_i);
 454         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
 455         root_volume.v.flags     = 0;
 456         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
 457         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
 458
 459         ret =   bch2_btree_insert(c, BTREE_ID_snapshot_trees,   &root_tree.k_i, NULL, 0) ?:
 460                 bch2_btree_insert(c, BTREE_ID_snapshots,        &root_snapshot.k_i, NULL, 0) ?:
 461                 bch2_btree_insert(c, BTREE_ID_subvolumes,       &root_volume.k_i, NULL, 0);
 462         if (ret)
 463                 bch_err_fn(c, ret);
 464         return ret;
 465 }
 466
 467 static int __bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
 468 {
 469         struct btree_iter iter;
 470         struct bkey_s_c k;
 471         struct bch_inode_unpacked inode;
 472         int ret;
 473
 474         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
 475                                SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
 476         ret = bkey_err(k);
 477         if (ret)
 478                 return ret;
 479
 480         if (!bkey_is_inode(k.k)) {
 481                 bch_err(trans->c, "root inode not found");
 482                 ret = -BCH_ERR_ENOENT_inode;
 483                 goto err;
 484         }
 485
 486         ret = bch2_inode_unpack(k, &inode);
 487         BUG_ON(ret);
 488
 489         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
 490
 491         ret = bch2_inode_write(trans, &iter, &inode);
 492 err:
 493         bch2_trans_iter_exit(trans, &iter);
 494         return ret;
 495 }
 496
 497 /* set bi_subvol on root inode */
 498 noinline_for_stack
 499 static int bch2_fs_upgrade_for_subvolumes(struct bch_fs *c)
 500 {
 501         int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_lazy_rw,
 502                                 __bch2_fs_upgrade_for_subvolumes(trans));
 503         if (ret)
 504                 bch_err_fn(c, ret);
 505         return ret;
 506 }
 507
 508 const char * const bch2_recovery_passes[] = {
 509 #define x(_fn, _when)   #_fn,
 510         BCH_RECOVERY_PASSES()
 511 #undef x
 512         NULL
 513 };
 514
 515 static int bch2_check_allocations(struct bch_fs *c)
 516 {
 517         return bch2_gc(c, true, c->opts.norecovery);
 518 }
 519
 520 static int bch2_set_may_go_rw(struct bch_fs *c)
 521 {
 522         struct journal_keys *keys = &c->journal_keys;
 523
 524         /*
 525          * After we go RW, the journal keys buffer can't be modified (except for
 526          * setting journal_key->overwritten: it will be accessed by multiple
 527          * threads
 528          */
 529         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 530         keys->gap = keys->nr;
 531
 532         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
 533         if (keys->nr)
 534                 return bch2_fs_read_write_early(c);
 535         return 0;
 536 }
 537
 538 struct recovery_pass_fn {
 539         int             (*fn)(struct bch_fs *);
 540         unsigned        when;
 541 };
 542
 543 static struct recovery_pass_fn recovery_pass_fns[] = {
 544 #define x(_fn, _when)   { .fn = bch2_##_fn, .when = _when },
 545         BCH_RECOVERY_PASSES()
 546 #undef x
 547 };
 548
 549 static void check_version_upgrade(struct bch_fs *c)
 550 {
 551         unsigned latest_compatible = bch2_latest_compatible_version(c->sb.version);
 552         unsigned latest_version = bcachefs_metadata_version_current;
 553         unsigned old_version = c->sb.version_upgrade_complete ?: c->sb.version;
 554         unsigned new_version = 0;
 555         u64 recovery_passes;
 556
 557         if (old_version < bcachefs_metadata_required_upgrade_below) {
 558                 if (c->opts.version_upgrade == BCH_VERSION_UPGRADE_incompatible ||
 559                     latest_compatible < bcachefs_metadata_required_upgrade_below)
 560                         new_version = latest_version;
 561                 else
 562                         new_version = latest_compatible;
 563         } else {
 564                 switch (c->opts.version_upgrade) {
 565                 case BCH_VERSION_UPGRADE_compatible:
 566                         new_version = latest_compatible;
 567                         break;
 568                 case BCH_VERSION_UPGRADE_incompatible:
 569                         new_version = latest_version;
 570                         break;
 571                 case BCH_VERSION_UPGRADE_none:
 572                         new_version = old_version;
 573                         break;
 574                 }
 575         }
 576
 577         if (new_version > old_version) {
 578                 struct printbuf buf = PRINTBUF;
 579
 580                 if (old_version < bcachefs_metadata_required_upgrade_below)
 581                         prt_str(&buf, "Version upgrade required:\n");
 582
 583                 if (old_version != c->sb.version) {
 584                         prt_str(&buf, "Version upgrade from ");
 585                         bch2_version_to_text(&buf, c->sb.version_upgrade_complete);
 586                         prt_str(&buf, " to ");
 587                         bch2_version_to_text(&buf, c->sb.version);
 588                         prt_str(&buf, " incomplete\n");
 589                 }
 590
 591                 prt_printf(&buf, "Doing %s version upgrade from ",
 592                            BCH_VERSION_MAJOR(old_version) != BCH_VERSION_MAJOR(new_version)
 593                            ? "incompatible" : "compatible");
 594                 bch2_version_to_text(&buf, old_version);
 595                 prt_str(&buf, " to ");
 596                 bch2_version_to_text(&buf, new_version);
 597                 prt_newline(&buf);
 598
 599                 recovery_passes = bch2_upgrade_recovery_passes(c, old_version, new_version);
 600                 if (recovery_passes) {
 601                         if ((recovery_passes & RECOVERY_PASS_ALL_FSCK) == RECOVERY_PASS_ALL_FSCK)
 602                                 prt_str(&buf, "fsck required");
 603                         else {
 604                                 prt_str(&buf, "running recovery passes: ");
 605                                 prt_bitflags(&buf, bch2_recovery_passes, recovery_passes);
 606                         }
 607
 608                         c->recovery_passes_explicit |= recovery_passes;
 609                         c->opts.fix_errors = FSCK_FIX_yes;
 610                 }
 611
 612                 bch_info(c, "%s", buf.buf);
 613
 614                 mutex_lock(&c->sb_lock);
 615                 bch2_sb_upgrade(c, new_version);
 616                 mutex_unlock(&c->sb_lock);
 617
 618                 printbuf_exit(&buf);
 619         }
 620 }
 621
 622 u64 bch2_fsck_recovery_passes(void)
 623 {
 624         u64 ret = 0;
 625
 626         for (unsigned i = 0; i < ARRAY_SIZE(recovery_pass_fns); i++)
 627                 if (recovery_pass_fns[i].when & PASS_FSCK)
 628                         ret |= BIT_ULL(i);
 629         return ret;
 630 }
 631
 632 static bool should_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
 633 {
 634         struct recovery_pass_fn *p = recovery_pass_fns + c->curr_recovery_pass;
 635
 636         if (c->opts.norecovery && pass > BCH_RECOVERY_PASS_snapshots_read)
 637                 return false;
 638         if (c->recovery_passes_explicit & BIT_ULL(pass))
 639                 return true;
 640         if ((p->when & PASS_FSCK) && c->opts.fsck)
 641                 return true;
 642         if ((p->when & PASS_UNCLEAN) && !c->sb.clean)
 643                 return true;
 644         if (p->when & PASS_ALWAYS)
 645                 return true;
 646         return false;
 647 }
 648
 649 static int bch2_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
 650 {
 651         int ret;
 652
 653         c->curr_recovery_pass = pass;
 654
 655         if (should_run_recovery_pass(c, pass)) {
 656                 struct recovery_pass_fn *p = recovery_pass_fns + pass;
 657
 658                 if (!(p->when & PASS_SILENT))
 659                         printk(KERN_INFO bch2_log_msg(c, "%s..."),
 660                                bch2_recovery_passes[pass]);
 661                 ret = p->fn(c);
 662                 if (ret)
 663                         return ret;
 664                 if (!(p->when & PASS_SILENT))
 665                         printk(KERN_CONT " done\n");
 666
 667                 c->recovery_passes_complete |= BIT_ULL(pass);
 668         }
 669
 670         return 0;
 671 }
 672
 673 static int bch2_run_recovery_passes(struct bch_fs *c)
 674 {
 675         int ret = 0;
 676
 677         while (c->curr_recovery_pass < ARRAY_SIZE(recovery_pass_fns)) {
 678                 ret = bch2_run_recovery_pass(c, c->curr_recovery_pass);
 679                 if (bch2_err_matches(ret, BCH_ERR_restart_recovery))
 680                         continue;
 681                 if (ret)
 682                         break;
 683                 c->curr_recovery_pass++;
 684         }
 685
 686         return ret;
 687 }
 688
 689 int bch2_fs_recovery(struct bch_fs *c)
 690 {
 691         struct bch_sb_field_clean *clean = NULL;
 692         struct jset *last_journal_entry = NULL;
 693         u64 last_seq = 0, blacklist_seq, journal_seq;
 694         bool write_sb = false;
 695         int ret = 0;
 696
 697         if (c->sb.clean) {
 698                 clean = bch2_read_superblock_clean(c);
 699                 ret = PTR_ERR_OR_ZERO(clean);
 700                 if (ret)
 701                         goto err;
 702
 703                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 704                          le64_to_cpu(clean->journal_seq));
 705         } else {
 706                 bch_info(c, "recovering from unclean shutdown");
 707         }
 708
 709         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
 710                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
 711                 ret = -EINVAL;
 712                 goto err;
 713         }
 714
 715         if (!c->sb.clean &&
 716             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
 717                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
 718                 ret = -EINVAL;
 719                 goto err;
 720         }
 721
 722         if (c->opts.fsck || !(c->opts.nochanges && c->opts.norecovery))
 723                 check_version_upgrade(c);
 724
 725         if (c->opts.fsck && c->opts.norecovery) {
 726                 bch_err(c, "cannot select both norecovery and fsck");
 727                 ret = -EINVAL;
 728                 goto err;
 729         }
 730
 731         ret = bch2_blacklist_table_initialize(c);
 732         if (ret) {
 733                 bch_err(c, "error initializing blacklist table");
 734                 goto err;
 735         }
 736
 737         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
 738                 struct genradix_iter iter;
 739                 struct journal_replay **i;
 740
 741                 bch_verbose(c, "starting journal read");
 742                 ret = bch2_journal_read(c, &last_seq, &blacklist_seq, &journal_seq);
 743                 if (ret)
 744                         goto err;
 745
 746                 /*
 747                  * note: cmd_list_journal needs the blacklist table fully up to date so
 748                  * it can asterisk ignored journal entries:
 749                  */
 750                 if (c->opts.read_journal_only)
 751                         goto out;
 752
 753                 genradix_for_each_reverse(&c->journal_entries, iter, i)
 754                         if (*i && !(*i)->ignore) {
 755                                 last_journal_entry = &(*i)->j;
 756                                 break;
 757                         }
 758
 759                 if (mustfix_fsck_err_on(c->sb.clean &&
 760                                         last_journal_entry &&
 761                                         !journal_entry_empty(last_journal_entry), c,
 762                                 clean_but_journal_not_empty,
 763                                 "filesystem marked clean but journal not empty")) {
 764                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 765                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 766                         c->sb.clean = false;
 767                 }
 768
 769                 if (!last_journal_entry) {
 770                         fsck_err_on(!c->sb.clean, c,
 771                                     dirty_but_no_journal_entries,
 772                                     "no journal entries found");
 773                         if (clean)
 774                                 goto use_clean;
 775
 776                         genradix_for_each_reverse(&c->journal_entries, iter, i)
 777                                 if (*i) {
 778                                         last_journal_entry = &(*i)->j;
 779                                         (*i)->ignore = false;
 780                                         /*
 781                                          * This was probably a NO_FLUSH entry,
 782                                          * so last_seq was garbage - but we know
 783                                          * we're only using a single journal
 784                                          * entry, set it here:
 785                                          */
 786                                         (*i)->j.last_seq = (*i)->j.seq;
 787                                         break;
 788                                 }
 789                 }
 790
 791                 ret = bch2_journal_keys_sort(c);
 792                 if (ret)
 793                         goto err;
 794
 795                 if (c->sb.clean && last_journal_entry) {
 796                         ret = bch2_verify_superblock_clean(c, &clean,
 797                                                       last_journal_entry);
 798                         if (ret)
 799                                 goto err;
 800                 }
 801         } else {
 802 use_clean:
 803                 if (!clean) {
 804                         bch_err(c, "no superblock clean section found");
 805                         ret = -BCH_ERR_fsck_repair_impossible;
 806                         goto err;
 807
 808                 }
 809                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
 810         }
 811
 812         c->journal_replay_seq_start     = last_seq;
 813         c->journal_replay_seq_end       = blacklist_seq - 1;
 814
 815         if (c->opts.reconstruct_alloc) {
 816                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 817                 drop_alloc_keys(&c->journal_keys);
 818         }
 819
 820         zero_out_btree_mem_ptr(&c->journal_keys);
 821
 822         ret = journal_replay_early(c, clean);
 823         if (ret)
 824                 goto err;
 825
 826         /*
 827          * After an unclean shutdown, skip then next few journal sequence
 828          * numbers as they may have been referenced by btree writes that
 829          * happened before their corresponding journal writes - those btree
 830          * writes need to be ignored, by skipping and blacklisting the next few
 831          * journal sequence numbers:
 832          */
 833         if (!c->sb.clean)
 834                 journal_seq += 8;
 835
 836         if (blacklist_seq != journal_seq) {
 837                 ret =   bch2_journal_log_msg(c, "blacklisting entries %llu-%llu",
 838                                              blacklist_seq, journal_seq) ?:
 839                         bch2_journal_seq_blacklist_add(c,
 840                                         blacklist_seq, journal_seq);
 841                 if (ret) {
 842                         bch_err(c, "error creating new journal seq blacklist entry");
 843                         goto err;
 844                 }
 845         }
 846
 847         ret =   bch2_journal_log_msg(c, "starting journal at entry %llu, replaying %llu-%llu",
 848                                      journal_seq, last_seq, blacklist_seq - 1) ?:
 849                 bch2_fs_journal_start(&c->journal, journal_seq);
 850         if (ret)
 851                 goto err;
 852
 853         if (c->opts.reconstruct_alloc)
 854                 bch2_journal_log_msg(c, "dropping alloc info");
 855
 856         /*
 857          * Skip past versions that might have possibly been used (as nonces),
 858          * but hadn't had their pointers written:
 859          */
 860         if (c->sb.encryption_type && !c->sb.clean)
 861                 atomic64_add(1 << 16, &c->key_version);
 862
 863         ret = read_btree_roots(c);
 864         if (ret)
 865                 goto err;
 866
 867         if (c->opts.fsck &&
 868             (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) ||
 869              BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb)))
 870                 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_topology);
 871
 872         ret = bch2_run_recovery_passes(c);
 873         if (ret)
 874                 goto err;
 875
 876         /* If we fixed errors, verify that fs is actually clean now: */
 877         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) &&
 878             test_bit(BCH_FS_ERRORS_FIXED, &c->flags) &&
 879             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags) &&
 880             !test_bit(BCH_FS_ERROR, &c->flags)) {
 881                 bch2_flush_fsck_errs(c);
 882
 883                 bch_info(c, "Fixed errors, running fsck a second time to verify fs is clean");
 884                 clear_bit(BCH_FS_ERRORS_FIXED, &c->flags);
 885
 886                 c->curr_recovery_pass = BCH_RECOVERY_PASS_check_alloc_info;
 887
 888                 ret = bch2_run_recovery_passes(c);
 889                 if (ret)
 890                         goto err;
 891
 892                 if (test_bit(BCH_FS_ERRORS_FIXED, &c->flags) ||
 893                     test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
 894                         bch_err(c, "Second fsck run was not clean");
 895                         set_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags);
 896                 }
 897
 898                 set_bit(BCH_FS_ERRORS_FIXED, &c->flags);
 899         }
 900
 901         if (enabled_qtypes(c)) {
 902                 bch_verbose(c, "reading quotas");
 903                 ret = bch2_fs_quota_read(c);
 904                 if (ret)
 905                         goto err;
 906                 bch_verbose(c, "quotas done");
 907         }
 908
 909         mutex_lock(&c->sb_lock);
 910         if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) != c->sb.version) {
 911                 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, c->sb.version);
 912                 write_sb = true;
 913         }
 914
 915         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
 916                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
 917                 write_sb = true;
 918         }
 919
 920         if (c->opts.fsck &&
 921             !test_bit(BCH_FS_ERROR, &c->flags) &&
 922             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
 923                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
 924                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
 925                 write_sb = true;
 926         }
 927
 928         if (write_sb)
 929                 bch2_write_super(c);
 930         mutex_unlock(&c->sb_lock);
 931
 932         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
 933             c->sb.version_min < bcachefs_metadata_version_btree_ptr_sectors_written) {
 934                 struct bch_move_stats stats;
 935
 936                 bch2_move_stats_init(&stats, "recovery");
 937
 938                 bch_info(c, "scanning for old btree nodes");
 939                 ret =   bch2_fs_read_write(c) ?:
 940                         bch2_scan_old_btree_nodes(c, &stats);
 941                 if (ret)
 942                         goto err;
 943                 bch_info(c, "scanning for old btree nodes done");
 944         }
 945
 946         if (c->journal_seq_blacklist_table &&
 947             c->journal_seq_blacklist_table->nr > 128)
 948                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
 949
 950         ret = 0;
 951 out:
 952         set_bit(BCH_FS_FSCK_DONE, &c->flags);
 953         bch2_flush_fsck_errs(c);
 954
 955         if (!c->opts.keep_journal &&
 956             test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)) {
 957                 bch2_journal_keys_free(&c->journal_keys);
 958                 bch2_journal_entries_free(c);
 959         }
 960         kfree(clean);
 961
 962         if (!ret && test_bit(BCH_FS_NEED_DELETE_DEAD_SNAPSHOTS, &c->flags)) {
 963                 bch2_fs_read_write_early(c);
 964                 bch2_delete_dead_snapshots_async(c);
 965         }
 966
 967         if (ret)
 968                 bch_err_fn(c, ret);
 969         return ret;
 970 err:
 971 fsck_err:
 972         bch2_fs_emergency_read_only(c);
 973         goto out;
 974 }
 975
 976 int bch2_fs_initialize(struct bch_fs *c)
 977 {
 978         struct bch_inode_unpacked root_inode, lostfound_inode;
 979         struct bkey_inode_buf packed_inode;
 980         struct qstr lostfound = QSTR("lost+found");
 981         struct bch_dev *ca;
 982         unsigned i;
 983         int ret;
 984
 985         bch_notice(c, "initializing new filesystem");
 986
 987         mutex_lock(&c->sb_lock);
 988         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
 989         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
 990
 991         bch2_sb_maybe_downgrade(c);
 992
 993         if (c->opts.version_upgrade != BCH_VERSION_UPGRADE_none) {
 994                 bch2_sb_upgrade(c, bcachefs_metadata_version_current);
 995                 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
 996                 bch2_write_super(c);
 997         }
 998         mutex_unlock(&c->sb_lock);
 999
1000         c->curr_recovery_pass = ARRAY_SIZE(recovery_pass_fns);
1001         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1002         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1003
1004         for (i = 0; i < BTREE_ID_NR; i++)
1005                 bch2_btree_root_alloc(c, i);
1006
1007         for_each_member_device(ca, c, i)
1008                 bch2_dev_usage_init(ca);
1009
1010         ret = bch2_fs_journal_alloc(c);
1011         if (ret)
1012                 goto err;
1013
1014         /*
1015          * journal_res_get() will crash if called before this has
1016          * set up the journal.pin FIFO and journal.cur pointer:
1017          */
1018         bch2_fs_journal_start(&c->journal, 1);
1019         bch2_journal_set_replay_done(&c->journal);
1020
1021         ret = bch2_fs_read_write_early(c);
1022         if (ret)
1023                 goto err;
1024
1025         /*
1026          * Write out the superblock and journal buckets, now that we can do
1027          * btree updates
1028          */
1029         bch_verbose(c, "marking superblocks");
1030         ret = bch2_trans_mark_dev_sbs(c);
1031         bch_err_msg(c, ret, "marking superblocks");
1032         if (ret)
1033                 goto err;
1034
1035         for_each_online_member(ca, c, i)
1036                 ca->new_fs_bucket_idx = 0;
1037
1038         ret = bch2_fs_freespace_init(c);
1039         if (ret)
1040                 goto err;
1041
1042         ret = bch2_initialize_subvolumes(c);
1043         if (ret)
1044                 goto err;
1045
1046         bch_verbose(c, "reading snapshots table");
1047         ret = bch2_snapshots_read(c);
1048         if (ret)
1049                 goto err;
1050         bch_verbose(c, "reading snapshots done");
1051
1052         bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL);
1053         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1054         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1055         bch2_inode_pack(&packed_inode, &root_inode);
1056         packed_inode.inode.k.p.snapshot = U32_MAX;
1057
1058         ret = bch2_btree_insert(c, BTREE_ID_inodes, &packed_inode.inode.k_i, NULL, 0);
1059         if (ret) {
1060                 bch_err_msg(c, ret, "creating root directory");
1061                 goto err;
1062         }
1063
1064         bch2_inode_init_early(c, &lostfound_inode);
1065
1066         ret = bch2_trans_do(c, NULL, NULL, 0,
1067                 bch2_create_trans(trans,
1068                                   BCACHEFS_ROOT_SUBVOL_INUM,
1069                                   &root_inode, &lostfound_inode,
1070                                   &lostfound,
1071                                   0, 0, S_IFDIR|0700, 0,
1072                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1073         if (ret) {
1074                 bch_err_msg(c, ret, "creating lost+found");
1075                 goto err;
1076         }
1077
1078         if (enabled_qtypes(c)) {
1079                 ret = bch2_fs_quota_read(c);
1080                 if (ret)
1081                         goto err;
1082         }
1083
1084         ret = bch2_journal_flush(&c->journal);
1085         if (ret) {
1086                 bch_err_msg(c, ret, "writing first journal entry");
1087                 goto err;
1088         }
1089
1090         mutex_lock(&c->sb_lock);
1091         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1092         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1093
1094         bch2_write_super(c);
1095         mutex_unlock(&c->sb_lock);
1096
1097         return 0;
1098 err:
1099         bch_err_fn(ca, ret);
1100         return ret;
1101 }