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 "bkey_buf.h"
   5 #include "alloc_background.h"
   6 #include "btree_gc.h"
   7 #include "btree_update.h"
   8 #include "btree_update_interior.h"
   9 #include "btree_io.h"
  10 #include "buckets.h"
  11 #include "dirent.h"
  12 #include "ec.h"
  13 #include "error.h"
  14 #include "fs-common.h"
  15 #include "fsck.h"
  16 #include "journal_io.h"
  17 #include "journal_reclaim.h"
  18 #include "journal_seq_blacklist.h"
  19 #include "lru.h"
  20 #include "move.h"
  21 #include "quota.h"
  22 #include "recovery.h"
  23 #include "replicas.h"
  24 #include "subvolume.h"
  25 #include "super-io.h"
  26
  27 #include <linux/sort.h>
  28 #include <linux/stat.h>
  29
  30 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  31
  32 /* for -o reconstruct_alloc: */
  33 static void drop_alloc_keys(struct journal_keys *keys)
  34 {
  35         size_t src, dst;
  36
  37         for (src = 0, dst = 0; src < keys->nr; src++)
  38                 if (keys->d[src].btree_id != BTREE_ID_alloc)
  39                         keys->d[dst++] = keys->d[src];
  40
  41         keys->nr = dst;
  42 }
  43
  44 /*
  45  * Btree node pointers have a field to stack a pointer to the in memory btree
  46  * node; we need to zero out this field when reading in btree nodes, or when
  47  * reading in keys from the journal:
  48  */
  49 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
  50 {
  51         struct journal_key *i;
  52
  53         for (i = keys->d; i < keys->d + keys->nr; i++)
  54                 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
  55                         bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
  56 }
  57
  58 /* iterate over keys read from the journal: */
  59
  60 static int __journal_key_cmp(enum btree_id      l_btree_id,
  61                              unsigned           l_level,
  62                              struct bpos        l_pos,
  63                              const struct journal_key *r)
  64 {
  65         return (cmp_int(l_btree_id,     r->btree_id) ?:
  66                 cmp_int(l_level,        r->level) ?:
  67                 bpos_cmp(l_pos, r->k->k.p));
  68 }
  69
  70 static int journal_key_cmp(const struct journal_key *l, const struct journal_key *r)
  71 {
  72         return __journal_key_cmp(l->btree_id, l->level, l->k->k.p, r);
  73 }
  74
  75 static inline size_t idx_to_pos(struct journal_keys *keys, size_t idx)
  76 {
  77         size_t gap_size = keys->size - keys->nr;
  78
  79         if (idx >= keys->gap)
  80                 idx += gap_size;
  81         return idx;
  82 }
  83
  84 static inline struct journal_key *idx_to_key(struct journal_keys *keys, size_t idx)
  85 {
  86         return keys->d + idx_to_pos(keys, idx);
  87 }
  88
  89 size_t bch2_journal_key_search(struct journal_keys *keys,
  90                                enum btree_id id, unsigned level,
  91                                struct bpos pos)
  92 {
  93         size_t l = 0, r = keys->nr, m;
  94
  95         while (l < r) {
  96                 m = l + ((r - l) >> 1);
  97                 if (__journal_key_cmp(id, level, pos, idx_to_key(keys, m)) > 0)
  98                         l = m + 1;
  99                 else
 100                         r = m;
 101         }
 102
 103         BUG_ON(l < keys->nr &&
 104                __journal_key_cmp(id, level, pos, idx_to_key(keys, l)) > 0);
 105
 106         BUG_ON(l &&
 107                __journal_key_cmp(id, level, pos, idx_to_key(keys, l - 1)) <= 0);
 108
 109         return idx_to_pos(keys, l);
 110 }
 111
 112 struct bkey_i *bch2_journal_keys_peek(struct bch_fs *c, enum btree_id btree_id,
 113                                       unsigned level, struct bpos pos)
 114 {
 115         struct journal_keys *keys = &c->journal_keys;
 116         size_t idx = bch2_journal_key_search(keys, btree_id, level, pos);
 117
 118         while (idx < keys->size &&
 119                keys->d[idx].overwritten) {
 120                 idx++;
 121                 if (idx == keys->gap)
 122                         idx += keys->size - keys->nr;
 123         }
 124
 125         if (idx < keys->size &&
 126             keys->d[idx].btree_id == btree_id &&
 127             keys->d[idx].level == level)
 128                 return keys->d[idx].k;
 129         return NULL;
 130 }
 131
 132 static void journal_iters_fix(struct bch_fs *c)
 133 {
 134         struct journal_keys *keys = &c->journal_keys;
 135         /* The key we just inserted is immediately before the gap: */
 136         struct journal_key *n = &keys->d[keys->gap - 1];
 137         size_t gap_end = keys->gap + (keys->size - keys->nr);
 138         struct btree_and_journal_iter *iter;
 139
 140         /*
 141          * If an iterator points one after the key we just inserted,
 142          * and the key we just inserted compares >= the iterator's position,
 143          * decrement the iterator so it points at the key we just inserted:
 144          */
 145         list_for_each_entry(iter, &c->journal_iters, journal.list)
 146                 if (iter->journal.idx == gap_end &&
 147                     iter->last &&
 148                     iter->b->c.btree_id == n->btree_id &&
 149                     iter->b->c.level    == n->level &&
 150                     bpos_cmp(n->k->k.p, iter->unpacked.p) >= 0)
 151                         iter->journal.idx = keys->gap - 1;
 152 }
 153
 154 static void journal_iters_move_gap(struct bch_fs *c, size_t old_gap, size_t new_gap)
 155 {
 156         struct journal_keys *keys = &c->journal_keys;
 157         struct journal_iter *iter;
 158         size_t gap_size = keys->size - keys->nr;
 159
 160         list_for_each_entry(iter, &c->journal_iters, list) {
 161                 if (iter->idx > old_gap)
 162                         iter->idx -= gap_size;
 163                 if (iter->idx >= new_gap)
 164                         iter->idx += gap_size;
 165         }
 166 }
 167
 168 int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
 169                                  unsigned level, struct bkey_i *k)
 170 {
 171         struct journal_key n = {
 172                 .btree_id       = id,
 173                 .level          = level,
 174                 .k              = k,
 175                 .allocated      = true,
 176                 /*
 177                  * Ensure these keys are done last by journal replay, to unblock
 178                  * journal reclaim:
 179                  */
 180                 .journal_seq    = U32_MAX,
 181         };
 182         struct journal_keys *keys = &c->journal_keys;
 183         size_t idx = bch2_journal_key_search(keys, id, level, k->k.p);
 184
 185         BUG_ON(test_bit(BCH_FS_RW, &c->flags));
 186
 187         if (idx < keys->size &&
 188             journal_key_cmp(&n, &keys->d[idx]) == 0) {
 189                 if (keys->d[idx].allocated)
 190                         kfree(keys->d[idx].k);
 191                 keys->d[idx] = n;
 192                 return 0;
 193         }
 194
 195         if (idx > keys->gap)
 196                 idx -= keys->size - keys->nr;
 197
 198         if (keys->nr == keys->size) {
 199                 struct journal_keys new_keys = {
 200                         .nr                     = keys->nr,
 201                         .size                   = max(keys->size, 8UL) * 2,
 202                         .journal_seq_base       = keys->journal_seq_base,
 203                 };
 204
 205                 new_keys.d = kvmalloc(sizeof(new_keys.d[0]) * new_keys.size, GFP_KERNEL);
 206                 if (!new_keys.d) {
 207                         bch_err(c, "%s: error allocating new key array (size %zu)",
 208                                 __func__, new_keys.size);
 209                         return -ENOMEM;
 210                 }
 211
 212                 /* Since @keys was full, there was no gap: */
 213                 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
 214                 kvfree(keys->d);
 215                 *keys = new_keys;
 216
 217                 /* And now the gap is at the end: */
 218                 keys->gap = keys->nr;
 219         }
 220
 221         journal_iters_move_gap(c, keys->gap, idx);
 222
 223         move_gap(keys->d, keys->nr, keys->size, keys->gap, idx);
 224         keys->gap = idx;
 225
 226         keys->nr++;
 227         keys->d[keys->gap++] = n;
 228
 229         journal_iters_fix(c);
 230
 231         return 0;
 232 }
 233
 234 /*
 235  * Can only be used from the recovery thread while we're still RO - can't be
 236  * used once we've got RW, as journal_keys is at that point used by multiple
 237  * threads:
 238  */
 239 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
 240                             unsigned level, struct bkey_i *k)
 241 {
 242         struct bkey_i *n;
 243         int ret;
 244
 245         n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
 246         if (!n)
 247                 return -ENOMEM;
 248
 249         bkey_copy(n, k);
 250         ret = bch2_journal_key_insert_take(c, id, level, n);
 251         if (ret)
 252                 kfree(n);
 253         return ret;
 254 }
 255
 256 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
 257                             unsigned level, struct bpos pos)
 258 {
 259         struct bkey_i whiteout;
 260
 261         bkey_init(&whiteout.k);
 262         whiteout.k.p = pos;
 263
 264         return bch2_journal_key_insert(c, id, level, &whiteout);
 265 }
 266
 267 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
 268                                   unsigned level, struct bpos pos)
 269 {
 270         struct journal_keys *keys = &c->journal_keys;
 271         size_t idx = bch2_journal_key_search(keys, btree, level, pos);
 272
 273         if (idx < keys->size &&
 274             keys->d[idx].btree_id       == btree &&
 275             keys->d[idx].level          == level &&
 276             !bpos_cmp(keys->d[idx].k->k.p, pos))
 277                 keys->d[idx].overwritten = true;
 278 }
 279
 280 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
 281 {
 282         struct journal_key *k = iter->keys->d + iter->idx;
 283
 284         while (k < iter->keys->d + iter->keys->nr &&
 285                k->btree_id      == iter->btree_id &&
 286                k->level         == iter->level) {
 287                 if (!k->overwritten)
 288                         return k->k;
 289
 290                 iter->idx++;
 291                 k = iter->keys->d + iter->idx;
 292         }
 293
 294         return NULL;
 295 }
 296
 297 static void bch2_journal_iter_advance(struct journal_iter *iter)
 298 {
 299         if (iter->idx < iter->keys->size) {
 300                 iter->idx++;
 301                 if (iter->idx == iter->keys->gap)
 302                         iter->idx += iter->keys->size - iter->keys->nr;
 303         }
 304 }
 305
 306 static void bch2_journal_iter_exit(struct journal_iter *iter)
 307 {
 308         list_del(&iter->list);
 309 }
 310
 311 static void bch2_journal_iter_init(struct bch_fs *c,
 312                                    struct journal_iter *iter,
 313                                    enum btree_id id, unsigned level,
 314                                    struct bpos pos)
 315 {
 316         iter->btree_id  = id;
 317         iter->level     = level;
 318         iter->keys      = &c->journal_keys;
 319         iter->idx       = bch2_journal_key_search(&c->journal_keys, id, level, pos);
 320 }
 321
 322 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 323 {
 324         return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 325                                                 iter->b, &iter->unpacked);
 326 }
 327
 328 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 329 {
 330         bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 331 }
 332
 333 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 334 {
 335         switch (iter->last) {
 336         case none:
 337                 break;
 338         case btree:
 339                 bch2_journal_iter_advance_btree(iter);
 340                 break;
 341         case journal:
 342                 bch2_journal_iter_advance(&iter->journal);
 343                 break;
 344         }
 345
 346         iter->last = none;
 347 }
 348
 349 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 350 {
 351         struct bkey_s_c ret;
 352
 353         while (1) {
 354                 struct bkey_s_c btree_k         =
 355                         bch2_journal_iter_peek_btree(iter);
 356                 struct bkey_s_c journal_k       =
 357                         bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
 358
 359                 if (btree_k.k && journal_k.k) {
 360                         int cmp = bpos_cmp(btree_k.k->p, journal_k.k->p);
 361
 362                         if (!cmp)
 363                                 bch2_journal_iter_advance_btree(iter);
 364
 365                         iter->last = cmp < 0 ? btree : journal;
 366                 } else if (btree_k.k) {
 367                         iter->last = btree;
 368                 } else if (journal_k.k) {
 369                         iter->last = journal;
 370                 } else {
 371                         iter->last = none;
 372                         return bkey_s_c_null;
 373                 }
 374
 375                 ret = iter->last == journal ? journal_k : btree_k;
 376
 377                 if (iter->b &&
 378                     bpos_cmp(ret.k->p, iter->b->data->max_key) > 0) {
 379                         iter->journal.idx = iter->journal.keys->nr;
 380                         iter->last = none;
 381                         return bkey_s_c_null;
 382                 }
 383
 384                 if (!bkey_deleted(ret.k))
 385                         break;
 386
 387                 bch2_btree_and_journal_iter_advance(iter);
 388         }
 389
 390         return ret;
 391 }
 392
 393 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 394 {
 395         bch2_btree_and_journal_iter_advance(iter);
 396
 397         return bch2_btree_and_journal_iter_peek(iter);
 398 }
 399
 400 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
 401 {
 402         bch2_journal_iter_exit(&iter->journal);
 403 }
 404
 405 void __bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 406                                                   struct bch_fs *c,
 407                                                   struct btree *b,
 408                                                   struct btree_node_iter node_iter,
 409                                                   struct bpos pos)
 410 {
 411         memset(iter, 0, sizeof(*iter));
 412
 413         iter->b = b;
 414         iter->node_iter = node_iter;
 415         bch2_journal_iter_init(c, &iter->journal, b->c.btree_id, b->c.level, pos);
 416         INIT_LIST_HEAD(&iter->journal.list);
 417 }
 418
 419 /*
 420  * this version is used by btree_gc before filesystem has gone RW and
 421  * multithreaded, so uses the journal_iters list:
 422  */
 423 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 424                                                 struct bch_fs *c,
 425                                                 struct btree *b)
 426 {
 427         struct btree_node_iter node_iter;
 428
 429         bch2_btree_node_iter_init_from_start(&node_iter, b);
 430         __bch2_btree_and_journal_iter_init_node_iter(iter, c, b, node_iter, b->data->min_key);
 431         list_add(&iter->journal.list, &c->journal_iters);
 432 }
 433
 434 /* sort and dedup all keys in the journal: */
 435
 436 void bch2_journal_entries_free(struct bch_fs *c)
 437 {
 438         struct journal_replay **i;
 439         struct genradix_iter iter;
 440
 441         genradix_for_each(&c->journal_entries, iter, i)
 442                 if (*i)
 443                         kvpfree(*i, offsetof(struct journal_replay, j) +
 444                                 vstruct_bytes(&(*i)->j));
 445         genradix_free(&c->journal_entries);
 446 }
 447
 448 /*
 449  * When keys compare equal, oldest compares first:
 450  */
 451 static int journal_sort_key_cmp(const void *_l, const void *_r)
 452 {
 453         const struct journal_key *l = _l;
 454         const struct journal_key *r = _r;
 455
 456         return  journal_key_cmp(l, r) ?:
 457                 cmp_int(l->journal_seq, r->journal_seq) ?:
 458                 cmp_int(l->journal_offset, r->journal_offset);
 459 }
 460
 461 void bch2_journal_keys_free(struct journal_keys *keys)
 462 {
 463         struct journal_key *i;
 464
 465         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 466         keys->gap = keys->nr;
 467
 468         for (i = keys->d; i < keys->d + keys->nr; i++)
 469                 if (i->allocated)
 470                         kfree(i->k);
 471
 472         kvfree(keys->d);
 473         keys->d = NULL;
 474         keys->nr = keys->gap = keys->size = 0;
 475 }
 476
 477 static int journal_keys_sort(struct bch_fs *c)
 478 {
 479         struct genradix_iter iter;
 480         struct journal_replay *i, **_i;
 481         struct jset_entry *entry;
 482         struct bkey_i *k, *_n;
 483         struct journal_keys *keys = &c->journal_keys;
 484         struct journal_key *src, *dst;
 485         size_t nr_keys = 0;
 486
 487         genradix_for_each(&c->journal_entries, iter, _i) {
 488                 i = *_i;
 489
 490                 if (!i || i->ignore)
 491                         continue;
 492
 493                 if (!keys->journal_seq_base)
 494                         keys->journal_seq_base = le64_to_cpu(i->j.seq);
 495
 496                 for_each_jset_key(k, _n, entry, &i->j)
 497                         nr_keys++;
 498         }
 499
 500         if (!nr_keys)
 501                 return 0;
 502
 503         keys->size = roundup_pow_of_two(nr_keys);
 504
 505         keys->d = kvmalloc(sizeof(keys->d[0]) * keys->size, GFP_KERNEL);
 506         if (!keys->d)
 507                 return -ENOMEM;
 508
 509         genradix_for_each(&c->journal_entries, iter, _i) {
 510                 i = *_i;
 511
 512                 if (!i || i->ignore)
 513                         continue;
 514
 515                 BUG_ON(le64_to_cpu(i->j.seq) - keys->journal_seq_base > U32_MAX);
 516
 517                 for_each_jset_key(k, _n, entry, &i->j)
 518                         keys->d[keys->nr++] = (struct journal_key) {
 519                                 .btree_id       = entry->btree_id,
 520                                 .level          = entry->level,
 521                                 .k              = k,
 522                                 .journal_seq    = le64_to_cpu(i->j.seq) -
 523                                         keys->journal_seq_base,
 524                                 .journal_offset = k->_data - i->j._data,
 525                         };
 526         }
 527
 528         sort(keys->d, keys->nr, sizeof(keys->d[0]), journal_sort_key_cmp, NULL);
 529
 530         src = dst = keys->d;
 531         while (src < keys->d + keys->nr) {
 532                 while (src + 1 < keys->d + keys->nr &&
 533                        src[0].btree_id  == src[1].btree_id &&
 534                        src[0].level     == src[1].level &&
 535                        !bpos_cmp(src[0].k->k.p, src[1].k->k.p))
 536                         src++;
 537
 538                 *dst++ = *src++;
 539         }
 540
 541         keys->nr = dst - keys->d;
 542         keys->gap = keys->nr;
 543         return 0;
 544 }
 545
 546 /* journal replay: */
 547
 548 static void replay_now_at(struct journal *j, u64 seq)
 549 {
 550         BUG_ON(seq < j->replay_journal_seq);
 551         BUG_ON(seq > j->replay_journal_seq_end);
 552
 553         while (j->replay_journal_seq < seq)
 554                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 555 }
 556
 557 static int bch2_journal_replay_key(struct btree_trans *trans,
 558                                    struct journal_key *k)
 559 {
 560         struct btree_iter iter;
 561         unsigned iter_flags =
 562                 BTREE_ITER_INTENT|
 563                 BTREE_ITER_NOT_EXTENTS;
 564         int ret;
 565
 566         if (!k->level && k->btree_id == BTREE_ID_alloc)
 567                 iter_flags |= BTREE_ITER_CACHED;
 568
 569         bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 570                                   BTREE_MAX_DEPTH, k->level,
 571                                   iter_flags);
 572         ret = bch2_btree_iter_traverse(&iter);
 573         if (ret)
 574                 goto out;
 575
 576         /* Must be checked with btree locked: */
 577         if (k->overwritten)
 578                 goto out;
 579
 580         ret = bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
 581 out:
 582         bch2_trans_iter_exit(trans, &iter);
 583         return ret;
 584 }
 585
 586 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 587 {
 588         const struct journal_key *l = *((const struct journal_key **)_l);
 589         const struct journal_key *r = *((const struct journal_key **)_r);
 590
 591         return cmp_int(l->journal_seq, r->journal_seq);
 592 }
 593
 594 static int bch2_journal_replay(struct bch_fs *c)
 595 {
 596         struct journal_keys *keys = &c->journal_keys;
 597         struct journal_key **keys_sorted, *k;
 598         struct journal *j = &c->journal;
 599         size_t i;
 600         int ret;
 601
 602         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 603         keys->gap = keys->nr;
 604
 605         keys_sorted = kvmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL);
 606         if (!keys_sorted)
 607                 return -ENOMEM;
 608
 609         for (i = 0; i < keys->nr; i++)
 610                 keys_sorted[i] = &keys->d[i];
 611
 612         sort(keys_sorted, keys->nr,
 613              sizeof(keys_sorted[0]),
 614              journal_sort_seq_cmp, NULL);
 615
 616         if (keys->nr)
 617                 replay_now_at(j, keys->journal_seq_base);
 618
 619         for (i = 0; i < keys->nr; i++) {
 620                 k = keys_sorted[i];
 621
 622                 cond_resched();
 623
 624                 if (!k->allocated)
 625                         replay_now_at(j, keys->journal_seq_base + k->journal_seq);
 626
 627                 ret = bch2_trans_do(c, NULL, NULL,
 628                                     BTREE_INSERT_LAZY_RW|
 629                                     BTREE_INSERT_NOFAIL|
 630                                     (!k->allocated
 631                                      ? BTREE_INSERT_JOURNAL_REPLAY|JOURNAL_WATERMARK_reserved
 632                                      : 0),
 633                              bch2_journal_replay_key(&trans, k));
 634                 if (ret) {
 635                         bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
 636                                 ret, bch2_btree_ids[k->btree_id], k->level);
 637                         goto err;
 638                 }
 639         }
 640
 641         replay_now_at(j, j->replay_journal_seq_end);
 642         j->replay_journal_seq = 0;
 643
 644         bch2_journal_set_replay_done(j);
 645         bch2_journal_flush_all_pins(j);
 646         ret = bch2_journal_error(j);
 647
 648         if (keys->nr && !ret)
 649                 bch2_journal_log_msg(&c->journal, "journal replay finished");
 650 err:
 651         kvfree(keys_sorted);
 652         return ret;
 653 }
 654
 655 /* journal replay early: */
 656
 657 static int journal_replay_entry_early(struct bch_fs *c,
 658                                       struct jset_entry *entry)
 659 {
 660         int ret = 0;
 661
 662         switch (entry->type) {
 663         case BCH_JSET_ENTRY_btree_root: {
 664                 struct btree_root *r;
 665
 666                 if (entry->btree_id >= BTREE_ID_NR) {
 667                         bch_err(c, "filesystem has unknown btree type %u",
 668                                 entry->btree_id);
 669                         return -EINVAL;
 670                 }
 671
 672                 r = &c->btree_roots[entry->btree_id];
 673
 674                 if (entry->u64s) {
 675                         r->level = entry->level;
 676                         bkey_copy(&r->key, &entry->start[0]);
 677                         r->error = 0;
 678                 } else {
 679                         r->error = -EIO;
 680                 }
 681                 r->alive = true;
 682                 break;
 683         }
 684         case BCH_JSET_ENTRY_usage: {
 685                 struct jset_entry_usage *u =
 686                         container_of(entry, struct jset_entry_usage, entry);
 687
 688                 switch (entry->btree_id) {
 689                 case BCH_FS_USAGE_reserved:
 690                         if (entry->level < BCH_REPLICAS_MAX)
 691                                 c->usage_base->persistent_reserved[entry->level] =
 692                                         le64_to_cpu(u->v);
 693                         break;
 694                 case BCH_FS_USAGE_inodes:
 695                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 696                         break;
 697                 case BCH_FS_USAGE_key_version:
 698                         atomic64_set(&c->key_version,
 699                                      le64_to_cpu(u->v));
 700                         break;
 701                 }
 702
 703                 break;
 704         }
 705         case BCH_JSET_ENTRY_data_usage: {
 706                 struct jset_entry_data_usage *u =
 707                         container_of(entry, struct jset_entry_data_usage, entry);
 708
 709                 ret = bch2_replicas_set_usage(c, &u->r,
 710                                               le64_to_cpu(u->v));
 711                 break;
 712         }
 713         case BCH_JSET_ENTRY_dev_usage: {
 714                 struct jset_entry_dev_usage *u =
 715                         container_of(entry, struct jset_entry_dev_usage, entry);
 716                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 717                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 718
 719                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 720
 721                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 722                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 723                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 724                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 725                 }
 726
 727                 break;
 728         }
 729         case BCH_JSET_ENTRY_blacklist: {
 730                 struct jset_entry_blacklist *bl_entry =
 731                         container_of(entry, struct jset_entry_blacklist, entry);
 732
 733                 ret = bch2_journal_seq_blacklist_add(c,
 734                                 le64_to_cpu(bl_entry->seq),
 735                                 le64_to_cpu(bl_entry->seq) + 1);
 736                 break;
 737         }
 738         case BCH_JSET_ENTRY_blacklist_v2: {
 739                 struct jset_entry_blacklist_v2 *bl_entry =
 740                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 741
 742                 ret = bch2_journal_seq_blacklist_add(c,
 743                                 le64_to_cpu(bl_entry->start),
 744                                 le64_to_cpu(bl_entry->end) + 1);
 745                 break;
 746         }
 747         case BCH_JSET_ENTRY_clock: {
 748                 struct jset_entry_clock *clock =
 749                         container_of(entry, struct jset_entry_clock, entry);
 750
 751                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 752         }
 753         }
 754
 755         return ret;
 756 }
 757
 758 static int journal_replay_early(struct bch_fs *c,
 759                                 struct bch_sb_field_clean *clean)
 760 {
 761         struct jset_entry *entry;
 762         int ret;
 763
 764         if (clean) {
 765                 for (entry = clean->start;
 766                      entry != vstruct_end(&clean->field);
 767                      entry = vstruct_next(entry)) {
 768                         ret = journal_replay_entry_early(c, entry);
 769                         if (ret)
 770                                 return ret;
 771                 }
 772         } else {
 773                 struct genradix_iter iter;
 774                 struct journal_replay *i, **_i;
 775
 776                 genradix_for_each(&c->journal_entries, iter, _i) {
 777                         i = *_i;
 778
 779                         if (!i || i->ignore)
 780                                 continue;
 781
 782                         vstruct_for_each(&i->j, entry) {
 783                                 ret = journal_replay_entry_early(c, entry);
 784                                 if (ret)
 785                                         return ret;
 786                         }
 787                 }
 788         }
 789
 790         bch2_fs_usage_initialize(c);
 791
 792         return 0;
 793 }
 794
 795 /* sb clean section: */
 796
 797 static struct bkey_i *btree_root_find(struct bch_fs *c,
 798                                       struct bch_sb_field_clean *clean,
 799                                       struct jset *j,
 800                                       enum btree_id id, unsigned *level)
 801 {
 802         struct bkey_i *k;
 803         struct jset_entry *entry, *start, *end;
 804
 805         if (clean) {
 806                 start = clean->start;
 807                 end = vstruct_end(&clean->field);
 808         } else {
 809                 start = j->start;
 810                 end = vstruct_last(j);
 811         }
 812
 813         for (entry = start; entry < end; entry = vstruct_next(entry))
 814                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 815                     entry->btree_id == id)
 816                         goto found;
 817
 818         return NULL;
 819 found:
 820         if (!entry->u64s)
 821                 return ERR_PTR(-EINVAL);
 822
 823         k = entry->start;
 824         *level = entry->level;
 825         return k;
 826 }
 827
 828 static int verify_superblock_clean(struct bch_fs *c,
 829                                    struct bch_sb_field_clean **cleanp,
 830                                    struct jset *j)
 831 {
 832         unsigned i;
 833         struct bch_sb_field_clean *clean = *cleanp;
 834         struct printbuf buf1 = PRINTBUF;
 835         struct printbuf buf2 = PRINTBUF;
 836         int ret = 0;
 837
 838         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 839                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 840                         le64_to_cpu(clean->journal_seq),
 841                         le64_to_cpu(j->seq))) {
 842                 kfree(clean);
 843                 *cleanp = NULL;
 844                 return 0;
 845         }
 846
 847         for (i = 0; i < BTREE_ID_NR; i++) {
 848                 struct bkey_i *k1, *k2;
 849                 unsigned l1 = 0, l2 = 0;
 850
 851                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 852                 k2 = btree_root_find(c, NULL, j, i, &l2);
 853
 854                 if (!k1 && !k2)
 855                         continue;
 856
 857                 printbuf_reset(&buf1);
 858                 printbuf_reset(&buf2);
 859
 860                 if (k1)
 861                         bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(k1));
 862                 else
 863                         pr_buf(&buf1, "(none)");
 864
 865                 if (k2)
 866                         bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(k2));
 867                 else
 868                         pr_buf(&buf2, "(none)");
 869
 870                 mustfix_fsck_err_on(!k1 || !k2 ||
 871                                     IS_ERR(k1) ||
 872                                     IS_ERR(k2) ||
 873                                     k1->k.u64s != k2->k.u64s ||
 874                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 875                                     l1 != l2, c,
 876                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 877                         "sb:      l=%u %s\n"
 878                         "journal: l=%u %s\n", i,
 879                         l1, buf1.buf,
 880                         l2, buf2.buf);
 881         }
 882 fsck_err:
 883         printbuf_exit(&buf2);
 884         printbuf_exit(&buf1);
 885         return ret;
 886 }
 887
 888 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 889 {
 890         struct bch_sb_field_clean *clean, *sb_clean;
 891         int ret;
 892
 893         mutex_lock(&c->sb_lock);
 894         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 895
 896         if (fsck_err_on(!sb_clean, c,
 897                         "superblock marked clean but clean section not present")) {
 898                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 899                 c->sb.clean = false;
 900                 mutex_unlock(&c->sb_lock);
 901                 return NULL;
 902         }
 903
 904         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 905                         GFP_KERNEL);
 906         if (!clean) {
 907                 mutex_unlock(&c->sb_lock);
 908                 return ERR_PTR(-ENOMEM);
 909         }
 910
 911         ret = bch2_sb_clean_validate_late(c, clean, READ);
 912         if (ret) {
 913                 mutex_unlock(&c->sb_lock);
 914                 return ERR_PTR(ret);
 915         }
 916
 917         mutex_unlock(&c->sb_lock);
 918
 919         return clean;
 920 fsck_err:
 921         mutex_unlock(&c->sb_lock);
 922         return ERR_PTR(ret);
 923 }
 924
 925 static int read_btree_roots(struct bch_fs *c)
 926 {
 927         unsigned i;
 928         int ret = 0;
 929
 930         for (i = 0; i < BTREE_ID_NR; i++) {
 931                 struct btree_root *r = &c->btree_roots[i];
 932
 933                 if (!r->alive)
 934                         continue;
 935
 936                 if (i == BTREE_ID_alloc &&
 937                     c->opts.reconstruct_alloc) {
 938                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 939                         continue;
 940                 }
 941
 942                 if (r->error) {
 943                         __fsck_err(c, i == BTREE_ID_alloc
 944                                    ? FSCK_CAN_IGNORE : 0,
 945                                    "invalid btree root %s",
 946                                    bch2_btree_ids[i]);
 947                         if (i == BTREE_ID_alloc)
 948                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 949                 }
 950
 951                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 952                 if (ret) {
 953                         __fsck_err(c, i == BTREE_ID_alloc
 954                                    ? FSCK_CAN_IGNORE : 0,
 955                                    "error reading btree root %s",
 956                                    bch2_btree_ids[i]);
 957                         if (i == BTREE_ID_alloc)
 958                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 959                 }
 960         }
 961
 962         for (i = 0; i < BTREE_ID_NR; i++)
 963                 if (!c->btree_roots[i].b)
 964                         bch2_btree_root_alloc(c, i);
 965 fsck_err:
 966         return ret;
 967 }
 968
 969 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
 970 {
 971         struct bkey_i_snapshot  root_snapshot;
 972         struct bkey_i_subvolume root_volume;
 973         int ret;
 974
 975         bkey_snapshot_init(&root_snapshot.k_i);
 976         root_snapshot.k.p.offset = U32_MAX;
 977         root_snapshot.v.flags   = 0;
 978         root_snapshot.v.parent  = 0;
 979         root_snapshot.v.subvol  = BCACHEFS_ROOT_SUBVOL;
 980         root_snapshot.v.pad     = 0;
 981         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
 982
 983         ret = bch2_btree_insert(c, BTREE_ID_snapshots,
 984                                 &root_snapshot.k_i,
 985                                 NULL, NULL, 0);
 986         if (ret)
 987                 return ret;
 988
 989
 990         bkey_subvolume_init(&root_volume.k_i);
 991         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
 992         root_volume.v.flags     = 0;
 993         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
 994         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
 995
 996         ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
 997                                 &root_volume.k_i,
 998                                 NULL, NULL, 0);
 999         if (ret)
1000                 return ret;
1001
1002         return 0;
1003 }
1004
1005 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
1006 {
1007         struct btree_iter iter;
1008         struct bkey_s_c k;
1009         struct bch_inode_unpacked inode;
1010         int ret;
1011
1012         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
1013                              SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
1014         k = bch2_btree_iter_peek_slot(&iter);
1015         ret = bkey_err(k);
1016         if (ret)
1017                 goto err;
1018
1019         if (!bkey_is_inode(k.k)) {
1020                 bch_err(trans->c, "root inode not found");
1021                 ret = -ENOENT;
1022                 goto err;
1023         }
1024
1025         ret = bch2_inode_unpack(k, &inode);
1026         BUG_ON(ret);
1027
1028         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1029
1030         ret = bch2_inode_write(trans, &iter, &inode);
1031 err:
1032         bch2_trans_iter_exit(trans, &iter);
1033         return ret;
1034 }
1035
1036 int bch2_fs_recovery(struct bch_fs *c)
1037 {
1038         const char *err = "cannot allocate memory";
1039         struct bch_sb_field_clean *clean = NULL;
1040         struct jset *last_journal_entry = NULL;
1041         u64 blacklist_seq, journal_seq;
1042         bool write_sb = false;
1043         int ret = 0;
1044
1045         if (c->sb.clean)
1046                 clean = read_superblock_clean(c);
1047         ret = PTR_ERR_OR_ZERO(clean);
1048         if (ret)
1049                 goto err;
1050
1051         if (c->sb.clean)
1052                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
1053                          le64_to_cpu(clean->journal_seq));
1054         else
1055                 bch_info(c, "recovering from unclean shutdown");
1056
1057         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
1058                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
1059                 ret = -EINVAL;
1060                 goto err;
1061         }
1062
1063         if (!c->sb.clean &&
1064             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1065                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1066                 ret = -EINVAL;
1067                 goto err;
1068         }
1069
1070         if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1071                 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
1072                 ret = -EINVAL;
1073                 goto err;
1074         }
1075
1076         if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
1077                 bch_info(c, "alloc_v2 feature bit not set, fsck required");
1078                 c->opts.fsck = true;
1079                 c->opts.fix_errors = FSCK_OPT_YES;
1080         }
1081
1082         if (!c->replicas.entries ||
1083             c->opts.rebuild_replicas) {
1084                 bch_info(c, "building replicas info");
1085                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1086         }
1087
1088         if (!c->opts.nochanges) {
1089                 if (c->sb.version < bcachefs_metadata_version_new_data_types) {
1090                         bch_info(c, "version prior to new_data_types, upgrade and fsck required");
1091                         c->opts.version_upgrade = true;
1092                         c->opts.fsck            = true;
1093                         c->opts.fix_errors      = FSCK_OPT_YES;
1094                 }
1095         }
1096
1097         ret = bch2_blacklist_table_initialize(c);
1098         if (ret) {
1099                 bch_err(c, "error initializing blacklist table");
1100                 goto err;
1101         }
1102
1103         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1104                 struct genradix_iter iter;
1105                 struct journal_replay **i;
1106
1107                 bch_verbose(c, "starting journal read");
1108                 ret = bch2_journal_read(c, &blacklist_seq, &journal_seq);
1109                 if (ret)
1110                         goto err;
1111
1112                 genradix_for_each_reverse(&c->journal_entries, iter, i)
1113                         if (*i && !(*i)->ignore) {
1114                                 last_journal_entry = &(*i)->j;
1115                                 break;
1116                         }
1117
1118                 if (mustfix_fsck_err_on(c->sb.clean &&
1119                                         last_journal_entry &&
1120                                         !journal_entry_empty(last_journal_entry), c,
1121                                 "filesystem marked clean but journal not empty")) {
1122                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1123                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1124                         c->sb.clean = false;
1125                 }
1126
1127                 if (!last_journal_entry) {
1128                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
1129                         goto use_clean;
1130                 }
1131
1132                 ret = journal_keys_sort(c);
1133                 if (ret)
1134                         goto err;
1135
1136                 if (c->sb.clean && last_journal_entry) {
1137                         ret = verify_superblock_clean(c, &clean,
1138                                                       last_journal_entry);
1139                         if (ret)
1140                                 goto err;
1141                 }
1142         } else {
1143 use_clean:
1144                 if (!clean) {
1145                         bch_err(c, "no superblock clean section found");
1146                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1147                         goto err;
1148
1149                 }
1150                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1151         }
1152
1153         if (c->opts.read_journal_only)
1154                 goto out;
1155
1156         if (c->opts.reconstruct_alloc) {
1157                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1158                 drop_alloc_keys(&c->journal_keys);
1159         }
1160
1161         zero_out_btree_mem_ptr(&c->journal_keys);
1162
1163         ret = journal_replay_early(c, clean);
1164         if (ret)
1165                 goto err;
1166
1167         /*
1168          * After an unclean shutdown, skip then next few journal sequence
1169          * numbers as they may have been referenced by btree writes that
1170          * happened before their corresponding journal writes - those btree
1171          * writes need to be ignored, by skipping and blacklisting the next few
1172          * journal sequence numbers:
1173          */
1174         if (!c->sb.clean)
1175                 journal_seq += 8;
1176
1177         if (blacklist_seq != journal_seq) {
1178                 ret = bch2_journal_seq_blacklist_add(c,
1179                                         blacklist_seq, journal_seq);
1180                 if (ret) {
1181                         bch_err(c, "error creating new journal seq blacklist entry");
1182                         goto err;
1183                 }
1184         }
1185
1186         ret = bch2_fs_journal_start(&c->journal, journal_seq);
1187         if (ret)
1188                 goto err;
1189
1190         ret = read_btree_roots(c);
1191         if (ret)
1192                 goto err;
1193
1194         bch_verbose(c, "starting alloc read");
1195         err = "error reading allocation information";
1196
1197         down_read(&c->gc_lock);
1198         ret = bch2_alloc_read(c);
1199         up_read(&c->gc_lock);
1200
1201         if (ret)
1202                 goto err;
1203         bch_verbose(c, "alloc read done");
1204
1205         bch_verbose(c, "starting stripes_read");
1206         err = "error reading stripes";
1207         ret = bch2_stripes_read(c);
1208         if (ret)
1209                 goto err;
1210         bch_verbose(c, "stripes_read done");
1211
1212         /*
1213          * If we're not running fsck, this ensures bch2_fsck_err() calls are
1214          * instead interpreted as bch2_inconsistent_err() calls:
1215          */
1216         if (!c->opts.fsck)
1217                 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1218
1219         if (c->opts.fsck ||
1220             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)) ||
1221             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_metadata)) ||
1222             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1223                 bool metadata_only = c->opts.norecovery;
1224
1225                 bch_info(c, "checking allocations");
1226                 err = "error checking allocations";
1227                 ret = bch2_gc(c, true, metadata_only);
1228                 if (ret)
1229                         goto err;
1230                 bch_verbose(c, "done checking allocations");
1231         }
1232
1233         if (c->opts.fsck) {
1234                 bch_info(c, "checking need_discard and freespace btrees");
1235                 err = "error checking need_discard and freespace btrees";
1236                 ret = bch2_check_alloc_info(c);
1237                 if (ret)
1238                         goto err;
1239
1240                 ret = bch2_check_lrus(c, true);
1241                 if (ret)
1242                         goto err;
1243                 bch_verbose(c, "done checking need_discard and freespace btrees");
1244         }
1245
1246         bch2_stripes_heap_start(c);
1247
1248         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1249         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1250         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1251
1252         /*
1253          * Skip past versions that might have possibly been used (as nonces),
1254          * but hadn't had their pointers written:
1255          */
1256         if (c->sb.encryption_type && !c->sb.clean)
1257                 atomic64_add(1 << 16, &c->key_version);
1258
1259         if (c->opts.norecovery)
1260                 goto out;
1261
1262         bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1263         err = "journal replay failed";
1264         ret = bch2_journal_replay(c);
1265         if (ret)
1266                 goto err;
1267         if (c->opts.verbose || !c->sb.clean)
1268                 bch_info(c, "journal replay done");
1269
1270         err = "error initializing freespace";
1271         ret = bch2_fs_freespace_init(c);
1272         if (ret)
1273                 goto err;
1274
1275         if (c->opts.fsck) {
1276                 bch_info(c, "checking alloc to lru refs");
1277                 err = "error checking alloc to lru refs";
1278                 ret = bch2_check_alloc_to_lru_refs(c);
1279                 if (ret)
1280                         goto err;
1281
1282                 ret = bch2_check_lrus(c, true);
1283                 if (ret)
1284                         goto err;
1285                 bch_verbose(c, "done checking alloc to lru refs");
1286         }
1287
1288         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1289                 bch2_fs_lazy_rw(c);
1290
1291                 err = "error creating root snapshot node";
1292                 ret = bch2_fs_initialize_subvolumes(c);
1293                 if (ret)
1294                         goto err;
1295         }
1296
1297         bch_verbose(c, "reading snapshots table");
1298         err = "error reading snapshots table";
1299         ret = bch2_fs_snapshots_start(c);
1300         if (ret)
1301                 goto err;
1302         bch_verbose(c, "reading snapshots done");
1303
1304         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1305                 /* set bi_subvol on root inode */
1306                 err = "error upgrade root inode for subvolumes";
1307                 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1308                                     bch2_fs_upgrade_for_subvolumes(&trans));
1309                 if (ret)
1310                         goto err;
1311         }
1312
1313         if (c->opts.fsck) {
1314                 bch_info(c, "starting fsck");
1315                 err = "error in fsck";
1316                 ret = bch2_fsck_full(c);
1317                 if (ret)
1318                         goto err;
1319                 bch_verbose(c, "fsck done");
1320         } else if (!c->sb.clean) {
1321                 bch_verbose(c, "checking for deleted inodes");
1322                 err = "error in recovery";
1323                 ret = bch2_fsck_walk_inodes_only(c);
1324                 if (ret)
1325                         goto err;
1326                 bch_verbose(c, "check inodes done");
1327         }
1328
1329         if (enabled_qtypes(c)) {
1330                 bch_verbose(c, "reading quotas");
1331                 ret = bch2_fs_quota_read(c);
1332                 if (ret)
1333                         goto err;
1334                 bch_verbose(c, "quotas done");
1335         }
1336
1337         mutex_lock(&c->sb_lock);
1338         if (c->opts.version_upgrade) {
1339                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1340                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1341                 write_sb = true;
1342         }
1343
1344         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1345                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1346                 write_sb = true;
1347         }
1348
1349         if (c->opts.fsck &&
1350             !test_bit(BCH_FS_ERROR, &c->flags) &&
1351             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1352                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1353                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1354                 write_sb = true;
1355         }
1356
1357         if (write_sb)
1358                 bch2_write_super(c);
1359         mutex_unlock(&c->sb_lock);
1360
1361         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1362             !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1363             le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) {
1364                 struct bch_move_stats stats;
1365
1366                 bch_move_stats_init(&stats, "recovery");
1367
1368                 bch_info(c, "scanning for old btree nodes");
1369                 ret = bch2_fs_read_write(c);
1370                 if (ret)
1371                         goto err;
1372
1373                 ret = bch2_scan_old_btree_nodes(c, &stats);
1374                 if (ret)
1375                         goto err;
1376                 bch_info(c, "scanning for old btree nodes done");
1377         }
1378
1379         if (c->journal_seq_blacklist_table &&
1380             c->journal_seq_blacklist_table->nr > 128)
1381                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1382
1383         ret = 0;
1384 out:
1385         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1386         bch2_flush_fsck_errs(c);
1387
1388         if (!c->opts.keep_journal) {
1389                 bch2_journal_keys_free(&c->journal_keys);
1390                 bch2_journal_entries_free(c);
1391         }
1392         kfree(clean);
1393         if (ret)
1394                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1395         else
1396                 bch_verbose(c, "ret %i", ret);
1397         return ret;
1398 err:
1399 fsck_err:
1400         bch2_fs_emergency_read_only(c);
1401         goto out;
1402 }
1403
1404 int bch2_fs_initialize(struct bch_fs *c)
1405 {
1406         struct bch_inode_unpacked root_inode, lostfound_inode;
1407         struct bkey_inode_buf packed_inode;
1408         struct qstr lostfound = QSTR("lost+found");
1409         const char *err = "cannot allocate memory";
1410         struct bch_dev *ca;
1411         unsigned i;
1412         int ret;
1413
1414         bch_notice(c, "initializing new filesystem");
1415
1416         mutex_lock(&c->sb_lock);
1417         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1418         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1419
1420         if (c->opts.version_upgrade) {
1421                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1422                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1423                 bch2_write_super(c);
1424         }
1425         mutex_unlock(&c->sb_lock);
1426
1427         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1428         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1429         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1430
1431         for (i = 0; i < BTREE_ID_NR; i++)
1432                 bch2_btree_root_alloc(c, i);
1433
1434         for_each_online_member(ca, c, i)
1435                 bch2_dev_usage_init(ca);
1436
1437         err = "unable to allocate journal buckets";
1438         for_each_online_member(ca, c, i) {
1439                 ret = bch2_dev_journal_alloc(ca);
1440                 if (ret) {
1441                         percpu_ref_put(&ca->io_ref);
1442                         goto err;
1443                 }
1444         }
1445
1446         /*
1447          * journal_res_get() will crash if called before this has
1448          * set up the journal.pin FIFO and journal.cur pointer:
1449          */
1450         bch2_fs_journal_start(&c->journal, 1);
1451         bch2_journal_set_replay_done(&c->journal);
1452
1453         err = "error going read-write";
1454         ret = bch2_fs_read_write_early(c);
1455         if (ret)
1456                 goto err;
1457
1458         /*
1459          * Write out the superblock and journal buckets, now that we can do
1460          * btree updates
1461          */
1462         bch_verbose(c, "marking superblocks");
1463         err = "error marking superblock and journal";
1464         for_each_member_device(ca, c, i) {
1465                 ret = bch2_trans_mark_dev_sb(c, ca);
1466                 if (ret) {
1467                         percpu_ref_put(&ca->ref);
1468                         goto err;
1469                 }
1470
1471                 ca->new_fs_bucket_idx = 0;
1472         }
1473
1474         bch_verbose(c, "initializing freespace");
1475         err = "error initializing freespace";
1476         ret = bch2_fs_freespace_init(c);
1477         if (ret)
1478                 goto err;
1479
1480         err = "error creating root snapshot node";
1481         ret = bch2_fs_initialize_subvolumes(c);
1482         if (ret)
1483                 goto err;
1484
1485         bch_verbose(c, "reading snapshots table");
1486         err = "error reading snapshots table";
1487         ret = bch2_fs_snapshots_start(c);
1488         if (ret)
1489                 goto err;
1490         bch_verbose(c, "reading snapshots done");
1491
1492         bch2_inode_init(c, &root_inode, 0, 0,
1493                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1494         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1495         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1496         bch2_inode_pack(c, &packed_inode, &root_inode);
1497         packed_inode.inode.k.p.snapshot = U32_MAX;
1498
1499         err = "error creating root directory";
1500         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1501                                 &packed_inode.inode.k_i,
1502                                 NULL, NULL, 0);
1503         if (ret)
1504                 goto err;
1505
1506         bch2_inode_init_early(c, &lostfound_inode);
1507
1508         err = "error creating lost+found";
1509         ret = bch2_trans_do(c, NULL, NULL, 0,
1510                 bch2_create_trans(&trans,
1511                                   BCACHEFS_ROOT_SUBVOL_INUM,
1512                                   &root_inode, &lostfound_inode,
1513                                   &lostfound,
1514                                   0, 0, S_IFDIR|0700, 0,
1515                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1516         if (ret) {
1517                 bch_err(c, "error creating lost+found");
1518                 goto err;
1519         }
1520
1521         if (enabled_qtypes(c)) {
1522                 ret = bch2_fs_quota_read(c);
1523                 if (ret)
1524                         goto err;
1525         }
1526
1527         err = "error writing first journal entry";
1528         ret = bch2_journal_flush(&c->journal);
1529         if (ret)
1530                 goto err;
1531
1532         mutex_lock(&c->sb_lock);
1533         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1534         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1535
1536         bch2_write_super(c);
1537         mutex_unlock(&c->sb_lock);
1538
1539         return 0;
1540 err:
1541         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1542         return ret;
1543 }