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 "move.h"
  20 #include "quota.h"
  21 #include "recovery.h"
  22 #include "replicas.h"
  23 #include "super-io.h"
  24
  25 #include <linux/sort.h>
  26 #include <linux/stat.h>
  27
  28 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  29
  30 /* for -o reconstruct_alloc: */
  31 static void drop_alloc_keys(struct journal_keys *keys)
  32 {
  33         size_t src, dst;
  34
  35         for (src = 0, dst = 0; src < keys->nr; src++)
  36                 if (keys->d[src].btree_id != BTREE_ID_alloc)
  37                         keys->d[dst++] = keys->d[src];
  38
  39         keys->nr = dst;
  40 }
  41
  42 /* iterate over keys read from the journal: */
  43
  44 static int __journal_key_cmp(enum btree_id      l_btree_id,
  45                              unsigned           l_level,
  46                              struct bpos        l_pos,
  47                              struct journal_key *r)
  48 {
  49         return (cmp_int(l_btree_id,     r->btree_id) ?:
  50                 cmp_int(l_level,        r->level) ?:
  51                 bpos_cmp(l_pos, r->k->k.p));
  52 }
  53
  54 static int journal_key_cmp(struct journal_key *l, struct journal_key *r)
  55 {
  56         return (cmp_int(l->btree_id,    r->btree_id) ?:
  57                 cmp_int(l->level,       r->level) ?:
  58                 bpos_cmp(l->k->k.p,     r->k->k.p));
  59 }
  60
  61 static size_t journal_key_search(struct journal_keys *journal_keys,
  62                                  enum btree_id id, unsigned level,
  63                                  struct bpos pos)
  64 {
  65         size_t l = 0, r = journal_keys->nr, m;
  66
  67         while (l < r) {
  68                 m = l + ((r - l) >> 1);
  69                 if (__journal_key_cmp(id, level, pos, &journal_keys->d[m]) > 0)
  70                         l = m + 1;
  71                 else
  72                         r = m;
  73         }
  74
  75         BUG_ON(l < journal_keys->nr &&
  76                __journal_key_cmp(id, level, pos, &journal_keys->d[l]) > 0);
  77
  78         BUG_ON(l &&
  79                __journal_key_cmp(id, level, pos, &journal_keys->d[l - 1]) <= 0);
  80
  81         return l;
  82 }
  83
  84 static void journal_iter_fix(struct bch_fs *c, struct journal_iter *iter, unsigned idx)
  85 {
  86         struct bkey_i *n = iter->keys->d[idx].k;
  87         struct btree_and_journal_iter *biter =
  88                 container_of(iter, struct btree_and_journal_iter, journal);
  89
  90         if (iter->idx > idx ||
  91             (iter->idx == idx &&
  92              biter->last &&
  93              bpos_cmp(n->k.p, biter->unpacked.p) <= 0))
  94                 iter->idx++;
  95 }
  96
  97 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
  98                             unsigned level, struct bkey_i *k)
  99 {
 100         struct journal_key n = {
 101                 .btree_id       = id,
 102                 .level          = level,
 103                 .k              = k,
 104                 .allocated      = true
 105         };
 106         struct journal_keys *keys = &c->journal_keys;
 107         struct journal_iter *iter;
 108         unsigned idx = journal_key_search(keys, id, level, k->k.p);
 109
 110         if (idx < keys->nr &&
 111             journal_key_cmp(&n, &keys->d[idx]) == 0) {
 112                 if (keys->d[idx].allocated)
 113                         kfree(keys->d[idx].k);
 114                 keys->d[idx] = n;
 115                 return 0;
 116         }
 117
 118         if (keys->nr == keys->size) {
 119                 struct journal_keys new_keys = {
 120                         .nr                     = keys->nr,
 121                         .size                   = keys->size * 2,
 122                         .journal_seq_base       = keys->journal_seq_base,
 123                 };
 124
 125                 new_keys.d = kvmalloc(sizeof(new_keys.d[0]) * new_keys.size, GFP_KERNEL);
 126                 if (!new_keys.d) {
 127                         bch_err(c, "%s: error allocating new key array (size %zu)",
 128                                 __func__, new_keys.size);
 129                         return -ENOMEM;
 130                 }
 131
 132                 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
 133                 kvfree(keys->d);
 134                 *keys = new_keys;
 135         }
 136
 137         array_insert_item(keys->d, keys->nr, idx, n);
 138
 139         list_for_each_entry(iter, &c->journal_iters, list)
 140                 journal_iter_fix(c, iter, idx);
 141
 142         return 0;
 143 }
 144
 145 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
 146                             unsigned level, struct bpos pos)
 147 {
 148         struct bkey_i *whiteout =
 149                 kmalloc(sizeof(struct bkey), GFP_KERNEL);
 150         int ret;
 151
 152         if (!whiteout) {
 153                 bch_err(c, "%s: error allocating new key", __func__);
 154                 return -ENOMEM;
 155         }
 156
 157         bkey_init(&whiteout->k);
 158         whiteout->k.p = pos;
 159
 160         ret = bch2_journal_key_insert(c, id, level, whiteout);
 161         if (ret)
 162                 kfree(whiteout);
 163         return ret;
 164 }
 165
 166 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
 167 {
 168         struct journal_key *k = iter->idx - iter->keys->nr
 169                 ? iter->keys->d + iter->idx : NULL;
 170
 171         if (k &&
 172             k->btree_id == iter->btree_id &&
 173             k->level    == iter->level)
 174                 return k->k;
 175
 176         iter->idx = iter->keys->nr;
 177         return NULL;
 178 }
 179
 180 static void bch2_journal_iter_advance(struct journal_iter *iter)
 181 {
 182         if (iter->idx < iter->keys->nr)
 183                 iter->idx++;
 184 }
 185
 186 static void bch2_journal_iter_exit(struct journal_iter *iter)
 187 {
 188         list_del(&iter->list);
 189 }
 190
 191 static void bch2_journal_iter_init(struct bch_fs *c,
 192                                    struct journal_iter *iter,
 193                                    enum btree_id id, unsigned level,
 194                                    struct bpos pos)
 195 {
 196         iter->btree_id  = id;
 197         iter->level     = level;
 198         iter->keys      = &c->journal_keys;
 199         iter->idx       = journal_key_search(&c->journal_keys, id, level, pos);
 200         list_add(&iter->list, &c->journal_iters);
 201 }
 202
 203 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 204 {
 205         return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 206                                                 iter->b, &iter->unpacked);
 207 }
 208
 209 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 210 {
 211         bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 212 }
 213
 214 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 215 {
 216         switch (iter->last) {
 217         case none:
 218                 break;
 219         case btree:
 220                 bch2_journal_iter_advance_btree(iter);
 221                 break;
 222         case journal:
 223                 bch2_journal_iter_advance(&iter->journal);
 224                 break;
 225         }
 226
 227         iter->last = none;
 228 }
 229
 230 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 231 {
 232         struct bkey_s_c ret;
 233
 234         while (1) {
 235                 struct bkey_s_c btree_k         =
 236                         bch2_journal_iter_peek_btree(iter);
 237                 struct bkey_s_c journal_k       =
 238                         bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
 239
 240                 if (btree_k.k && journal_k.k) {
 241                         int cmp = bpos_cmp(btree_k.k->p, journal_k.k->p);
 242
 243                         if (!cmp)
 244                                 bch2_journal_iter_advance_btree(iter);
 245
 246                         iter->last = cmp < 0 ? btree : journal;
 247                 } else if (btree_k.k) {
 248                         iter->last = btree;
 249                 } else if (journal_k.k) {
 250                         iter->last = journal;
 251                 } else {
 252                         iter->last = none;
 253                         return bkey_s_c_null;
 254                 }
 255
 256                 ret = iter->last == journal ? journal_k : btree_k;
 257
 258                 if (iter->b &&
 259                     bpos_cmp(ret.k->p, iter->b->data->max_key) > 0) {
 260                         iter->journal.idx = iter->journal.keys->nr;
 261                         iter->last = none;
 262                         return bkey_s_c_null;
 263                 }
 264
 265                 if (!bkey_deleted(ret.k))
 266                         break;
 267
 268                 bch2_btree_and_journal_iter_advance(iter);
 269         }
 270
 271         return ret;
 272 }
 273
 274 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 275 {
 276         bch2_btree_and_journal_iter_advance(iter);
 277
 278         return bch2_btree_and_journal_iter_peek(iter);
 279 }
 280
 281 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
 282 {
 283         bch2_journal_iter_exit(&iter->journal);
 284 }
 285
 286 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 287                                                 struct bch_fs *c,
 288                                                 struct btree *b)
 289 {
 290         memset(iter, 0, sizeof(*iter));
 291
 292         iter->b = b;
 293         bch2_btree_node_iter_init_from_start(&iter->node_iter, iter->b);
 294         bch2_journal_iter_init(c, &iter->journal,
 295                                b->c.btree_id, b->c.level, b->data->min_key);
 296 }
 297
 298 /* Walk btree, overlaying keys from the journal: */
 299
 300 static void btree_and_journal_iter_prefetch(struct bch_fs *c, struct btree *b,
 301                                            struct btree_and_journal_iter iter)
 302 {
 303         unsigned i = 0, nr = b->c.level > 1 ? 2 : 16;
 304         struct bkey_s_c k;
 305         struct bkey_buf tmp;
 306
 307         BUG_ON(!b->c.level);
 308
 309         bch2_bkey_buf_init(&tmp);
 310
 311         while (i < nr &&
 312                (k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 313                 bch2_bkey_buf_reassemble(&tmp, c, k);
 314
 315                 bch2_btree_node_prefetch(c, NULL, tmp.k,
 316                                         b->c.btree_id, b->c.level - 1);
 317
 318                 bch2_btree_and_journal_iter_advance(&iter);
 319                 i++;
 320         }
 321
 322         bch2_bkey_buf_exit(&tmp, c);
 323 }
 324
 325 static int bch2_btree_and_journal_walk_recurse(struct bch_fs *c, struct btree *b,
 326                                 enum btree_id btree_id,
 327                                 btree_walk_key_fn key_fn)
 328 {
 329         struct btree_and_journal_iter iter;
 330         struct bkey_s_c k;
 331         struct bkey_buf tmp;
 332         struct btree *child;
 333         int ret = 0;
 334
 335         bch2_bkey_buf_init(&tmp);
 336         bch2_btree_and_journal_iter_init_node_iter(&iter, c, b);
 337
 338         while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 339                 if (b->c.level) {
 340                         bch2_bkey_buf_reassemble(&tmp, c, k);
 341
 342                         child = bch2_btree_node_get_noiter(c, tmp.k,
 343                                                 b->c.btree_id, b->c.level - 1,
 344                                                 false);
 345
 346                         ret = PTR_ERR_OR_ZERO(child);
 347                         if (ret)
 348                                 break;
 349
 350                         btree_and_journal_iter_prefetch(c, b, iter);
 351
 352                         ret = bch2_btree_and_journal_walk_recurse(c, child,
 353                                         btree_id, key_fn);
 354                         six_unlock_read(&child->c.lock);
 355                 } else {
 356                         ret = key_fn(c, k);
 357                 }
 358
 359                 if (ret)
 360                         break;
 361
 362                 bch2_btree_and_journal_iter_advance(&iter);
 363         }
 364
 365         bch2_btree_and_journal_iter_exit(&iter);
 366         bch2_bkey_buf_exit(&tmp, c);
 367         return ret;
 368 }
 369
 370 int bch2_btree_and_journal_walk(struct bch_fs *c, enum btree_id btree_id,
 371                                 btree_walk_key_fn key_fn)
 372 {
 373         struct btree *b = c->btree_roots[btree_id].b;
 374         int ret = 0;
 375
 376         if (btree_node_fake(b))
 377                 return 0;
 378
 379         six_lock_read(&b->c.lock, NULL, NULL);
 380         ret = bch2_btree_and_journal_walk_recurse(c, b, btree_id, key_fn);
 381         six_unlock_read(&b->c.lock);
 382
 383         return ret;
 384 }
 385
 386 /* sort and dedup all keys in the journal: */
 387
 388 void bch2_journal_entries_free(struct list_head *list)
 389 {
 390
 391         while (!list_empty(list)) {
 392                 struct journal_replay *i =
 393                         list_first_entry(list, struct journal_replay, list);
 394                 list_del(&i->list);
 395                 kvpfree(i, offsetof(struct journal_replay, j) +
 396                         vstruct_bytes(&i->j));
 397         }
 398 }
 399
 400 /*
 401  * When keys compare equal, oldest compares first:
 402  */
 403 static int journal_sort_key_cmp(const void *_l, const void *_r)
 404 {
 405         const struct journal_key *l = _l;
 406         const struct journal_key *r = _r;
 407
 408         return  cmp_int(l->btree_id,    r->btree_id) ?:
 409                 cmp_int(l->level,       r->level) ?:
 410                 bpos_cmp(l->k->k.p, r->k->k.p) ?:
 411                 cmp_int(l->journal_seq, r->journal_seq) ?:
 412                 cmp_int(l->journal_offset, r->journal_offset);
 413 }
 414
 415 void bch2_journal_keys_free(struct journal_keys *keys)
 416 {
 417         struct journal_key *i;
 418
 419         for (i = keys->d; i < keys->d + keys->nr; i++)
 420                 if (i->allocated)
 421                         kfree(i->k);
 422
 423         kvfree(keys->d);
 424         keys->d = NULL;
 425         keys->nr = 0;
 426 }
 427
 428 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 429 {
 430         struct journal_replay *i;
 431         struct jset_entry *entry;
 432         struct bkey_i *k, *_n;
 433         struct journal_keys keys = { NULL };
 434         struct journal_key *src, *dst;
 435         size_t nr_keys = 0;
 436
 437         if (list_empty(journal_entries))
 438                 return keys;
 439
 440         list_for_each_entry(i, journal_entries, list) {
 441                 if (i->ignore)
 442                         continue;
 443
 444                 if (!keys.journal_seq_base)
 445                         keys.journal_seq_base = le64_to_cpu(i->j.seq);
 446
 447                 for_each_jset_key(k, _n, entry, &i->j)
 448                         nr_keys++;
 449         }
 450
 451         keys.size = roundup_pow_of_two(nr_keys);
 452
 453         keys.d = kvmalloc(sizeof(keys.d[0]) * keys.size, GFP_KERNEL);
 454         if (!keys.d)
 455                 goto err;
 456
 457         list_for_each_entry(i, journal_entries, list) {
 458                 if (i->ignore)
 459                         continue;
 460
 461                 BUG_ON(le64_to_cpu(i->j.seq) - keys.journal_seq_base > U32_MAX);
 462
 463                 for_each_jset_key(k, _n, entry, &i->j)
 464                         keys.d[keys.nr++] = (struct journal_key) {
 465                                 .btree_id       = entry->btree_id,
 466                                 .level          = entry->level,
 467                                 .k              = k,
 468                                 .journal_seq    = le64_to_cpu(i->j.seq) -
 469                                         keys.journal_seq_base,
 470                                 .journal_offset = k->_data - i->j._data,
 471                         };
 472         }
 473
 474         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 475
 476         src = dst = keys.d;
 477         while (src < keys.d + keys.nr) {
 478                 while (src + 1 < keys.d + keys.nr &&
 479                        src[0].btree_id  == src[1].btree_id &&
 480                        src[0].level     == src[1].level &&
 481                        !bpos_cmp(src[0].k->k.p, src[1].k->k.p))
 482                         src++;
 483
 484                 *dst++ = *src++;
 485         }
 486
 487         keys.nr = dst - keys.d;
 488 err:
 489         return keys;
 490 }
 491
 492 /* journal replay: */
 493
 494 static void replay_now_at(struct journal *j, u64 seq)
 495 {
 496         BUG_ON(seq < j->replay_journal_seq);
 497         BUG_ON(seq > j->replay_journal_seq_end);
 498
 499         while (j->replay_journal_seq < seq)
 500                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 501 }
 502
 503 static int __bch2_journal_replay_key(struct btree_trans *trans,
 504                                      enum btree_id id, unsigned level,
 505                                      struct bkey_i *k)
 506 {
 507         struct btree_iter *iter;
 508         int ret;
 509
 510         iter = bch2_trans_get_node_iter(trans, id, k->k.p,
 511                                         BTREE_MAX_DEPTH, level,
 512                                         BTREE_ITER_INTENT|
 513                                         BTREE_ITER_NOT_EXTENTS);
 514         ret   = bch2_btree_iter_traverse(iter) ?:
 515                 bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 516         bch2_trans_iter_put(trans, iter);
 517         return ret;
 518 }
 519
 520 static int bch2_journal_replay_key(struct bch_fs *c, struct journal_key *k)
 521 {
 522         unsigned commit_flags = BTREE_INSERT_NOFAIL|
 523                 BTREE_INSERT_LAZY_RW;
 524
 525         if (!k->allocated)
 526                 commit_flags |= BTREE_INSERT_JOURNAL_REPLAY;
 527
 528         return bch2_trans_do(c, NULL, NULL, commit_flags,
 529                              __bch2_journal_replay_key(&trans, k->btree_id, k->level, k->k));
 530 }
 531
 532 static int __bch2_alloc_replay_key(struct btree_trans *trans, struct bkey_i *k)
 533 {
 534         struct btree_iter *iter;
 535         int ret;
 536
 537         iter = bch2_trans_get_iter(trans, BTREE_ID_alloc, k->k.p,
 538                                    BTREE_ITER_CACHED|
 539                                    BTREE_ITER_CACHED_NOFILL|
 540                                    BTREE_ITER_INTENT);
 541         ret   = bch2_btree_iter_traverse(iter) ?:
 542                 bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 543         bch2_trans_iter_put(trans, iter);
 544         return ret;
 545 }
 546
 547 static int bch2_alloc_replay_key(struct bch_fs *c, struct bkey_i *k)
 548 {
 549         return bch2_trans_do(c, NULL, NULL,
 550                              BTREE_INSERT_NOFAIL|
 551                              BTREE_INSERT_USE_RESERVE|
 552                              BTREE_INSERT_LAZY_RW|
 553                              BTREE_INSERT_JOURNAL_REPLAY,
 554                         __bch2_alloc_replay_key(&trans, k));
 555 }
 556
 557 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 558 {
 559         const struct journal_key *l = _l;
 560         const struct journal_key *r = _r;
 561
 562         return  cmp_int(r->level,       l->level) ?:
 563                 cmp_int(l->journal_seq, r->journal_seq) ?:
 564                 cmp_int(l->btree_id,    r->btree_id) ?:
 565                 bpos_cmp(l->k->k.p,     r->k->k.p);
 566 }
 567
 568 static int bch2_journal_replay(struct bch_fs *c,
 569                                struct journal_keys keys)
 570 {
 571         struct journal *j = &c->journal;
 572         struct journal_key *i;
 573         u64 seq;
 574         int ret;
 575
 576         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
 577
 578         if (keys.nr)
 579                 replay_now_at(j, keys.journal_seq_base);
 580
 581         seq = j->replay_journal_seq;
 582
 583         /*
 584          * First replay updates to the alloc btree - these will only update the
 585          * btree key cache:
 586          */
 587         for_each_journal_key(keys, i) {
 588                 cond_resched();
 589
 590                 if (!i->level && i->btree_id == BTREE_ID_alloc) {
 591                         j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
 592                         ret = bch2_alloc_replay_key(c, i->k);
 593                         if (ret)
 594                                 goto err;
 595                 }
 596         }
 597
 598         /*
 599          * Next replay updates to interior btree nodes:
 600          */
 601         for_each_journal_key(keys, i) {
 602                 cond_resched();
 603
 604                 if (i->level) {
 605                         j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
 606                         ret = bch2_journal_replay_key(c, i);
 607                         if (ret)
 608                                 goto err;
 609                 }
 610         }
 611
 612         /*
 613          * Now that the btree is in a consistent state, we can start journal
 614          * reclaim (which will be flushing entries from the btree key cache back
 615          * to the btree:
 616          */
 617         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
 618         set_bit(JOURNAL_RECLAIM_STARTED, &j->flags);
 619         journal_reclaim_kick(j);
 620
 621         j->replay_journal_seq = seq;
 622
 623         /*
 624          * Now replay leaf node updates:
 625          */
 626         for_each_journal_key(keys, i) {
 627                 cond_resched();
 628
 629                 if (i->level || i->btree_id == BTREE_ID_alloc)
 630                         continue;
 631
 632                 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 633
 634                 ret = bch2_journal_replay_key(c, i);
 635                 if (ret)
 636                         goto err;
 637         }
 638
 639         replay_now_at(j, j->replay_journal_seq_end);
 640         j->replay_journal_seq = 0;
 641
 642         bch2_journal_set_replay_done(j);
 643         bch2_journal_flush_all_pins(j);
 644         return bch2_journal_error(j);
 645 err:
 646         bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
 647                 ret, bch2_btree_ids[i->btree_id], i->level);
 648         return ret;
 649 }
 650
 651 /* journal replay early: */
 652
 653 static int journal_replay_entry_early(struct bch_fs *c,
 654                                       struct jset_entry *entry)
 655 {
 656         int ret = 0;
 657
 658         switch (entry->type) {
 659         case BCH_JSET_ENTRY_btree_root: {
 660                 struct btree_root *r;
 661
 662                 if (entry->btree_id >= BTREE_ID_NR) {
 663                         bch_err(c, "filesystem has unknown btree type %u",
 664                                 entry->btree_id);
 665                         return -EINVAL;
 666                 }
 667
 668                 r = &c->btree_roots[entry->btree_id];
 669
 670                 if (entry->u64s) {
 671                         r->level = entry->level;
 672                         bkey_copy(&r->key, &entry->start[0]);
 673                         r->error = 0;
 674                 } else {
 675                         r->error = -EIO;
 676                 }
 677                 r->alive = true;
 678                 break;
 679         }
 680         case BCH_JSET_ENTRY_usage: {
 681                 struct jset_entry_usage *u =
 682                         container_of(entry, struct jset_entry_usage, entry);
 683
 684                 switch (entry->btree_id) {
 685                 case FS_USAGE_RESERVED:
 686                         if (entry->level < BCH_REPLICAS_MAX)
 687                                 c->usage_base->persistent_reserved[entry->level] =
 688                                         le64_to_cpu(u->v);
 689                         break;
 690                 case FS_USAGE_INODES:
 691                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 692                         break;
 693                 case FS_USAGE_KEY_VERSION:
 694                         atomic64_set(&c->key_version,
 695                                      le64_to_cpu(u->v));
 696                         break;
 697                 }
 698
 699                 break;
 700         }
 701         case BCH_JSET_ENTRY_data_usage: {
 702                 struct jset_entry_data_usage *u =
 703                         container_of(entry, struct jset_entry_data_usage, entry);
 704
 705                 ret = bch2_replicas_set_usage(c, &u->r,
 706                                               le64_to_cpu(u->v));
 707                 break;
 708         }
 709         case BCH_JSET_ENTRY_dev_usage: {
 710                 struct jset_entry_dev_usage *u =
 711                         container_of(entry, struct jset_entry_dev_usage, entry);
 712                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 713                 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
 714                 unsigned nr_types = (bytes - sizeof(struct jset_entry_dev_usage)) /
 715                         sizeof(struct jset_entry_dev_usage_type);
 716                 unsigned i;
 717
 718                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 719                 ca->usage_base->buckets_unavailable     = le64_to_cpu(u->buckets_unavailable);
 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                                 struct list_head *journal)
 761 {
 762         struct journal_replay *i;
 763         struct jset_entry *entry;
 764         int ret;
 765
 766         if (clean) {
 767                 for (entry = clean->start;
 768                      entry != vstruct_end(&clean->field);
 769                      entry = vstruct_next(entry)) {
 770                         ret = journal_replay_entry_early(c, entry);
 771                         if (ret)
 772                                 return ret;
 773                 }
 774         } else {
 775                 list_for_each_entry(i, journal, list) {
 776                         if (i->ignore)
 777                                 continue;
 778
 779                         vstruct_for_each(&i->j, entry) {
 780                                 ret = journal_replay_entry_early(c, entry);
 781                                 if (ret)
 782                                         return ret;
 783                         }
 784                 }
 785         }
 786
 787         bch2_fs_usage_initialize(c);
 788
 789         return 0;
 790 }
 791
 792 /* sb clean section: */
 793
 794 static struct bkey_i *btree_root_find(struct bch_fs *c,
 795                                       struct bch_sb_field_clean *clean,
 796                                       struct jset *j,
 797                                       enum btree_id id, unsigned *level)
 798 {
 799         struct bkey_i *k;
 800         struct jset_entry *entry, *start, *end;
 801
 802         if (clean) {
 803                 start = clean->start;
 804                 end = vstruct_end(&clean->field);
 805         } else {
 806                 start = j->start;
 807                 end = vstruct_last(j);
 808         }
 809
 810         for (entry = start; entry < end; entry = vstruct_next(entry))
 811                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 812                     entry->btree_id == id)
 813                         goto found;
 814
 815         return NULL;
 816 found:
 817         if (!entry->u64s)
 818                 return ERR_PTR(-EINVAL);
 819
 820         k = entry->start;
 821         *level = entry->level;
 822         return k;
 823 }
 824
 825 static int verify_superblock_clean(struct bch_fs *c,
 826                                    struct bch_sb_field_clean **cleanp,
 827                                    struct jset *j)
 828 {
 829         unsigned i;
 830         struct bch_sb_field_clean *clean = *cleanp;
 831         int ret = 0;
 832
 833         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 834                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 835                         le64_to_cpu(clean->journal_seq),
 836                         le64_to_cpu(j->seq))) {
 837                 kfree(clean);
 838                 *cleanp = NULL;
 839                 return 0;
 840         }
 841
 842         for (i = 0; i < BTREE_ID_NR; i++) {
 843                 char buf1[200], buf2[200];
 844                 struct bkey_i *k1, *k2;
 845                 unsigned l1 = 0, l2 = 0;
 846
 847                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 848                 k2 = btree_root_find(c, NULL, j, i, &l2);
 849
 850                 if (!k1 && !k2)
 851                         continue;
 852
 853                 mustfix_fsck_err_on(!k1 || !k2 ||
 854                                     IS_ERR(k1) ||
 855                                     IS_ERR(k2) ||
 856                                     k1->k.u64s != k2->k.u64s ||
 857                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 858                                     l1 != l2, c,
 859                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 860                         "sb:      l=%u %s\n"
 861                         "journal: l=%u %s\n", i,
 862                         l1, (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(k1)), buf1),
 863                         l2, (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(k2)), buf2));
 864         }
 865 fsck_err:
 866         return ret;
 867 }
 868
 869 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 870 {
 871         struct bch_sb_field_clean *clean, *sb_clean;
 872         int ret;
 873
 874         mutex_lock(&c->sb_lock);
 875         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 876
 877         if (fsck_err_on(!sb_clean, c,
 878                         "superblock marked clean but clean section not present")) {
 879                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 880                 c->sb.clean = false;
 881                 mutex_unlock(&c->sb_lock);
 882                 return NULL;
 883         }
 884
 885         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 886                         GFP_KERNEL);
 887         if (!clean) {
 888                 mutex_unlock(&c->sb_lock);
 889                 return ERR_PTR(-ENOMEM);
 890         }
 891
 892         ret = bch2_sb_clean_validate(c, clean, READ);
 893         if (ret) {
 894                 mutex_unlock(&c->sb_lock);
 895                 return ERR_PTR(ret);
 896         }
 897
 898         mutex_unlock(&c->sb_lock);
 899
 900         return clean;
 901 fsck_err:
 902         mutex_unlock(&c->sb_lock);
 903         return ERR_PTR(ret);
 904 }
 905
 906 static int read_btree_roots(struct bch_fs *c)
 907 {
 908         unsigned i;
 909         int ret = 0;
 910
 911         for (i = 0; i < BTREE_ID_NR; i++) {
 912                 struct btree_root *r = &c->btree_roots[i];
 913
 914                 if (!r->alive)
 915                         continue;
 916
 917                 if (i == BTREE_ID_alloc &&
 918                     c->opts.reconstruct_alloc) {
 919                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 920                         continue;
 921                 }
 922
 923                 if (r->error) {
 924                         __fsck_err(c, i == BTREE_ID_alloc
 925                                    ? FSCK_CAN_IGNORE : 0,
 926                                    "invalid btree root %s",
 927                                    bch2_btree_ids[i]);
 928                         if (i == BTREE_ID_alloc)
 929                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 930                 }
 931
 932                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 933                 if (ret) {
 934                         __fsck_err(c, i == BTREE_ID_alloc
 935                                    ? FSCK_CAN_IGNORE : 0,
 936                                    "error reading btree root %s",
 937                                    bch2_btree_ids[i]);
 938                         if (i == BTREE_ID_alloc)
 939                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 940                 }
 941         }
 942
 943         for (i = 0; i < BTREE_ID_NR; i++)
 944                 if (!c->btree_roots[i].b)
 945                         bch2_btree_root_alloc(c, i);
 946 fsck_err:
 947         return ret;
 948 }
 949
 950 int bch2_fs_recovery(struct bch_fs *c)
 951 {
 952         const char *err = "cannot allocate memory";
 953         struct bch_sb_field_clean *clean = NULL;
 954         struct jset *last_journal_entry = NULL;
 955         u64 blacklist_seq, journal_seq;
 956         bool write_sb = false;
 957         int ret = 0;
 958
 959         if (c->sb.clean)
 960                 clean = read_superblock_clean(c);
 961         ret = PTR_ERR_OR_ZERO(clean);
 962         if (ret)
 963                 goto err;
 964
 965         if (c->sb.clean)
 966                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 967                          le64_to_cpu(clean->journal_seq));
 968
 969         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
 970                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
 971                 ret = -EINVAL;
 972                 goto err;
 973         }
 974
 975         if (!c->sb.clean &&
 976             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
 977                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
 978                 ret = -EINVAL;
 979                 goto err;
 980         }
 981
 982         if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
 983                 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
 984                 ret = -EINVAL;
 985                 goto err;
 986
 987         }
 988
 989         if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
 990                 bch_info(c, "alloc_v2 feature bit not set, fsck required");
 991                 c->opts.fsck = true;
 992                 c->opts.fix_errors = FSCK_OPT_YES;
 993         }
 994
 995         if (!c->replicas.entries ||
 996             c->opts.rebuild_replicas) {
 997                 bch_info(c, "building replicas info");
 998                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 999         }
1000
1001         if (c->sb.version < bcachefs_metadata_version_inode_backpointers) {
1002                 bch_info(c, "version prior to inode backpointers, upgrade and fsck required");
1003                 c->opts.version_upgrade = true;
1004                 c->opts.fsck            = true;
1005                 c->opts.fix_errors      = FSCK_OPT_YES;
1006         }
1007
1008         ret = bch2_blacklist_table_initialize(c);
1009         if (ret) {
1010                 bch_err(c, "error initializing blacklist table");
1011                 goto err;
1012         }
1013
1014         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1015                 struct journal_replay *i;
1016
1017                 ret = bch2_journal_read(c, &c->journal_entries,
1018                                         &blacklist_seq, &journal_seq);
1019                 if (ret)
1020                         goto err;
1021
1022                 list_for_each_entry_reverse(i, &c->journal_entries, list)
1023                         if (!i->ignore) {
1024                                 last_journal_entry = &i->j;
1025                                 break;
1026                         }
1027
1028                 if (mustfix_fsck_err_on(c->sb.clean &&
1029                                         last_journal_entry &&
1030                                         !journal_entry_empty(last_journal_entry), c,
1031                                 "filesystem marked clean but journal not empty")) {
1032                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1033                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1034                         c->sb.clean = false;
1035                 }
1036
1037                 if (!last_journal_entry) {
1038                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
1039                         goto use_clean;
1040                 }
1041
1042                 c->journal_keys = journal_keys_sort(&c->journal_entries);
1043                 if (!c->journal_keys.d) {
1044                         ret = -ENOMEM;
1045                         goto err;
1046                 }
1047
1048                 if (c->sb.clean && last_journal_entry) {
1049                         ret = verify_superblock_clean(c, &clean,
1050                                                       last_journal_entry);
1051                         if (ret)
1052                                 goto err;
1053                 }
1054         } else {
1055 use_clean:
1056                 if (!clean) {
1057                         bch_err(c, "no superblock clean section found");
1058                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1059                         goto err;
1060
1061                 }
1062                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1063         }
1064
1065         if (c->opts.reconstruct_alloc) {
1066                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1067                 drop_alloc_keys(&c->journal_keys);
1068         }
1069
1070         ret = journal_replay_early(c, clean, &c->journal_entries);
1071         if (ret)
1072                 goto err;
1073
1074         /*
1075          * After an unclean shutdown, skip then next few journal sequence
1076          * numbers as they may have been referenced by btree writes that
1077          * happened before their corresponding journal writes - those btree
1078          * writes need to be ignored, by skipping and blacklisting the next few
1079          * journal sequence numbers:
1080          */
1081         if (!c->sb.clean)
1082                 journal_seq += 8;
1083
1084         if (blacklist_seq != journal_seq) {
1085                 ret = bch2_journal_seq_blacklist_add(c,
1086                                         blacklist_seq, journal_seq);
1087                 if (ret) {
1088                         bch_err(c, "error creating new journal seq blacklist entry");
1089                         goto err;
1090                 }
1091         }
1092
1093         ret = bch2_fs_journal_start(&c->journal, journal_seq,
1094                                     &c->journal_entries);
1095         if (ret)
1096                 goto err;
1097
1098         ret = read_btree_roots(c);
1099         if (ret)
1100                 goto err;
1101
1102         bch_verbose(c, "starting alloc read");
1103         err = "error reading allocation information";
1104         ret = bch2_alloc_read(c);
1105         if (ret)
1106                 goto err;
1107         bch_verbose(c, "alloc read done");
1108
1109         bch_verbose(c, "starting stripes_read");
1110         err = "error reading stripes";
1111         ret = bch2_stripes_read(c);
1112         if (ret)
1113                 goto err;
1114         bch_verbose(c, "stripes_read done");
1115
1116         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1117
1118         if (c->opts.fsck ||
1119             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)) ||
1120             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_metadata)) ||
1121             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1122                 bool metadata_only = c->opts.norecovery;
1123
1124                 bch_info(c, "starting mark and sweep");
1125                 err = "error in mark and sweep";
1126                 ret = bch2_gc(c, true, metadata_only);
1127                 if (ret)
1128                         goto err;
1129                 bch_verbose(c, "mark and sweep done");
1130         }
1131
1132         bch2_stripes_heap_start(c);
1133
1134         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1135         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1136
1137         /*
1138          * Skip past versions that might have possibly been used (as nonces),
1139          * but hadn't had their pointers written:
1140          */
1141         if (c->sb.encryption_type && !c->sb.clean)
1142                 atomic64_add(1 << 16, &c->key_version);
1143
1144         if (c->opts.norecovery)
1145                 goto out;
1146
1147         bch_verbose(c, "starting journal replay");
1148         err = "journal replay failed";
1149         ret = bch2_journal_replay(c, c->journal_keys);
1150         if (ret)
1151                 goto err;
1152         bch_verbose(c, "journal replay done");
1153
1154         if (test_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags) &&
1155             !c->opts.nochanges) {
1156                 /*
1157                  * note that even when filesystem was clean there might be work
1158                  * to do here, if we ran gc (because of fsck) which recalculated
1159                  * oldest_gen:
1160                  */
1161                 bch_verbose(c, "writing allocation info");
1162                 err = "error writing out alloc info";
1163                 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW) ?:
1164                         bch2_alloc_write(c, BTREE_INSERT_LAZY_RW);
1165                 if (ret) {
1166                         bch_err(c, "error writing alloc info");
1167                         goto err;
1168                 }
1169                 bch_verbose(c, "alloc write done");
1170         }
1171
1172         if (c->opts.fsck) {
1173                 bch_info(c, "starting fsck");
1174                 err = "error in fsck";
1175                 ret = bch2_fsck_full(c);
1176                 if (ret)
1177                         goto err;
1178                 bch_verbose(c, "fsck done");
1179         } else if (!c->sb.clean) {
1180                 bch_verbose(c, "checking for deleted inodes");
1181                 err = "error in recovery";
1182                 ret = bch2_fsck_walk_inodes_only(c);
1183                 if (ret)
1184                         goto err;
1185                 bch_verbose(c, "check inodes done");
1186         }
1187
1188         if (enabled_qtypes(c)) {
1189                 bch_verbose(c, "reading quotas");
1190                 ret = bch2_fs_quota_read(c);
1191                 if (ret)
1192                         goto err;
1193                 bch_verbose(c, "quotas done");
1194         }
1195
1196         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1197             !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1198                 struct bch_move_stats stats = { 0 };
1199
1200                 bch_info(c, "scanning for old btree nodes");
1201                 ret = bch2_fs_read_write(c);
1202                 if (ret)
1203                         goto err;
1204
1205                 ret = bch2_scan_old_btree_nodes(c, &stats);
1206                 if (ret)
1207                         goto err;
1208                 bch_info(c, "scanning for old btree nodes done");
1209         }
1210
1211         mutex_lock(&c->sb_lock);
1212         if (c->opts.version_upgrade) {
1213                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1214                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1215                 write_sb = true;
1216         }
1217
1218         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1219                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1220                 write_sb = true;
1221         }
1222
1223         if (c->opts.fsck &&
1224             !test_bit(BCH_FS_ERROR, &c->flags) &&
1225             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1226                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1227                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1228                 write_sb = true;
1229         }
1230
1231         if (write_sb)
1232                 bch2_write_super(c);
1233         mutex_unlock(&c->sb_lock);
1234
1235         if (c->journal_seq_blacklist_table &&
1236             c->journal_seq_blacklist_table->nr > 128)
1237                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1238
1239         ret = 0;
1240 out:
1241         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1242         bch2_flush_fsck_errs(c);
1243
1244         if (!c->opts.keep_journal) {
1245                 bch2_journal_keys_free(&c->journal_keys);
1246                 bch2_journal_entries_free(&c->journal_entries);
1247         }
1248         kfree(clean);
1249         if (ret)
1250                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1251         else
1252                 bch_verbose(c, "ret %i", ret);
1253         return ret;
1254 err:
1255 fsck_err:
1256         bch2_fs_emergency_read_only(c);
1257         goto out;
1258 }
1259
1260 int bch2_fs_initialize(struct bch_fs *c)
1261 {
1262         struct bch_inode_unpacked root_inode, lostfound_inode;
1263         struct bkey_inode_buf packed_inode;
1264         struct qstr lostfound = QSTR("lost+found");
1265         const char *err = "cannot allocate memory";
1266         struct bch_dev *ca;
1267         LIST_HEAD(journal);
1268         unsigned i;
1269         int ret;
1270
1271         bch_notice(c, "initializing new filesystem");
1272
1273         mutex_lock(&c->sb_lock);
1274         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1275         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1276
1277         if (c->opts.version_upgrade) {
1278                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1279                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1280                 bch2_write_super(c);
1281         }
1282
1283         for_each_online_member(ca, c, i)
1284                 bch2_mark_dev_superblock(c, ca, 0);
1285         mutex_unlock(&c->sb_lock);
1286
1287         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1288         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1289
1290         for (i = 0; i < BTREE_ID_NR; i++)
1291                 bch2_btree_root_alloc(c, i);
1292
1293         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
1294         set_bit(JOURNAL_RECLAIM_STARTED, &c->journal.flags);
1295
1296         err = "unable to allocate journal buckets";
1297         for_each_online_member(ca, c, i) {
1298                 ret = bch2_dev_journal_alloc(ca);
1299                 if (ret) {
1300                         percpu_ref_put(&ca->io_ref);
1301                         goto err;
1302                 }
1303         }
1304
1305         /*
1306          * journal_res_get() will crash if called before this has
1307          * set up the journal.pin FIFO and journal.cur pointer:
1308          */
1309         bch2_fs_journal_start(&c->journal, 1, &journal);
1310         bch2_journal_set_replay_done(&c->journal);
1311
1312         err = "error going read-write";
1313         ret = bch2_fs_read_write_early(c);
1314         if (ret)
1315                 goto err;
1316
1317         /*
1318          * Write out the superblock and journal buckets, now that we can do
1319          * btree updates
1320          */
1321         err = "error marking superblock and journal";
1322         for_each_member_device(ca, c, i) {
1323                 ret = bch2_trans_mark_dev_sb(c, ca);
1324                 if (ret) {
1325                         percpu_ref_put(&ca->ref);
1326                         goto err;
1327                 }
1328         }
1329
1330         bch2_inode_init(c, &root_inode, 0, 0,
1331                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1332         root_inode.bi_inum = BCACHEFS_ROOT_INO;
1333         bch2_inode_pack(c, &packed_inode, &root_inode);
1334         packed_inode.inode.k.p.snapshot = U32_MAX;
1335
1336         err = "error creating root directory";
1337         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1338                                 &packed_inode.inode.k_i,
1339                                 NULL, NULL, 0);
1340         if (ret)
1341                 goto err;
1342
1343         bch2_inode_init_early(c, &lostfound_inode);
1344
1345         err = "error creating lost+found";
1346         ret = bch2_trans_do(c, NULL, NULL, 0,
1347                 bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
1348                                   &root_inode, &lostfound_inode,
1349                                   &lostfound,
1350                                   0, 0, S_IFDIR|0700, 0,
1351                                   NULL, NULL));
1352         if (ret) {
1353                 bch_err(c, "error creating lost+found");
1354                 goto err;
1355         }
1356
1357         if (enabled_qtypes(c)) {
1358                 ret = bch2_fs_quota_read(c);
1359                 if (ret)
1360                         goto err;
1361         }
1362
1363         err = "error writing first journal entry";
1364         ret = bch2_journal_meta(&c->journal);
1365         if (ret)
1366                 goto err;
1367
1368         mutex_lock(&c->sb_lock);
1369         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1370         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1371
1372         bch2_write_super(c);
1373         mutex_unlock(&c->sb_lock);
1374
1375         return 0;
1376 err:
1377         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1378         return ret;
1379 }