fs/bcachefs/recovery.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "alloc_background.h"
   5 #include "btree_gc.h"
   6 #include "btree_update.h"
   7 #include "btree_update_interior.h"
   8 #include "btree_io.h"
   9 #include "buckets.h"
  10 #include "dirent.h"
  11 #include "ec.h"
  12 #include "error.h"
  13 #include "fs-common.h"
  14 #include "fsck.h"
  15 #include "journal_io.h"
  16 #include "journal_reclaim.h"
  17 #include "journal_seq_blacklist.h"
  18 #include "quota.h"
  19 #include "recovery.h"
  20 #include "replicas.h"
  21 #include "super-io.h"
  22
  23 #include <linux/sort.h>
  24 #include <linux/stat.h>
  25
  26 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  27
  28 /* for -o reconstruct_alloc: */
  29 static void drop_alloc_keys(struct journal_keys *keys)
  30 {
  31         size_t src, dst;
  32
  33         for (src = 0, dst = 0; src < keys->nr; src++)
  34                 if (keys->d[src].btree_id != BTREE_ID_ALLOC)
  35                         keys->d[dst++] = keys->d[src];
  36
  37         keys->nr = dst;
  38 }
  39
  40 /* iterate over keys read from the journal: */
  41
  42 static struct journal_key *journal_key_search(struct journal_keys *journal_keys,
  43                                               enum btree_id id, unsigned level,
  44                                               struct bpos pos)
  45 {
  46         size_t l = 0, r = journal_keys->nr, m;
  47
  48         while (l < r) {
  49                 m = l + ((r - l) >> 1);
  50                 if ((cmp_int(id,        journal_keys->d[m].btree_id) ?:
  51                      cmp_int(level,     journal_keys->d[m].level) ?:
  52                      bkey_cmp(pos,      journal_keys->d[m].k->k.p)) > 0)
  53                         l = m + 1;
  54                 else
  55                         r = m;
  56         }
  57
  58         BUG_ON(l < journal_keys->nr &&
  59                (cmp_int(id,     journal_keys->d[l].btree_id) ?:
  60                 cmp_int(level,  journal_keys->d[l].level) ?:
  61                 bkey_cmp(pos,   journal_keys->d[l].k->k.p)) > 0);
  62
  63         BUG_ON(l &&
  64                (cmp_int(id,     journal_keys->d[l - 1].btree_id) ?:
  65                 cmp_int(level,  journal_keys->d[l - 1].level) ?:
  66                 bkey_cmp(pos,   journal_keys->d[l - 1].k->k.p)) <= 0);
  67
  68         return l < journal_keys->nr ? journal_keys->d + l : NULL;
  69 }
  70
  71 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
  72 {
  73         if (iter->k &&
  74             iter->k < iter->keys->d + iter->keys->nr &&
  75             iter->k->btree_id   == iter->btree_id &&
  76             iter->k->level      == iter->level)
  77                 return iter->k->k;
  78
  79         iter->k = NULL;
  80         return NULL;
  81 }
  82
  83 static void bch2_journal_iter_advance(struct journal_iter *iter)
  84 {
  85         if (iter->k)
  86                 iter->k++;
  87 }
  88
  89 static void bch2_journal_iter_init(struct journal_iter *iter,
  90                                    struct journal_keys *journal_keys,
  91                                    enum btree_id id, unsigned level,
  92                                    struct bpos pos)
  93 {
  94         iter->btree_id  = id;
  95         iter->level     = level;
  96         iter->keys      = journal_keys;
  97         iter->k         = journal_key_search(journal_keys, id, level, pos);
  98 }
  99
 100 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 101 {
 102         return iter->btree
 103                 ? bch2_btree_iter_peek(iter->btree)
 104                 : bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 105                                                    iter->b, &iter->unpacked);
 106 }
 107
 108 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 109 {
 110         if (iter->btree)
 111                 bch2_btree_iter_next(iter->btree);
 112         else
 113                 bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 114 }
 115
 116 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 117 {
 118         switch (iter->last) {
 119         case none:
 120                 break;
 121         case btree:
 122                 bch2_journal_iter_advance_btree(iter);
 123                 break;
 124         case journal:
 125                 bch2_journal_iter_advance(&iter->journal);
 126                 break;
 127         }
 128
 129         iter->last = none;
 130 }
 131
 132 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 133 {
 134         struct bkey_s_c ret;
 135
 136         while (1) {
 137                 struct bkey_s_c btree_k         =
 138                         bch2_journal_iter_peek_btree(iter);
 139                 struct bkey_s_c journal_k       =
 140                         bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
 141
 142                 if (btree_k.k && journal_k.k) {
 143                         int cmp = bkey_cmp(btree_k.k->p, journal_k.k->p);
 144
 145                         if (!cmp)
 146                                 bch2_journal_iter_advance_btree(iter);
 147
 148                         iter->last = cmp < 0 ? btree : journal;
 149                 } else if (btree_k.k) {
 150                         iter->last = btree;
 151                 } else if (journal_k.k) {
 152                         iter->last = journal;
 153                 } else {
 154                         iter->last = none;
 155                         return bkey_s_c_null;
 156                 }
 157
 158                 ret = iter->last == journal ? journal_k : btree_k;
 159
 160                 if (iter->b &&
 161                     bkey_cmp(ret.k->p, iter->b->data->max_key) > 0) {
 162                         iter->journal.k = NULL;
 163                         iter->last = none;
 164                         return bkey_s_c_null;
 165                 }
 166
 167                 if (!bkey_deleted(ret.k))
 168                         break;
 169
 170                 bch2_btree_and_journal_iter_advance(iter);
 171         }
 172
 173         return ret;
 174 }
 175
 176 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 177 {
 178         bch2_btree_and_journal_iter_advance(iter);
 179
 180         return bch2_btree_and_journal_iter_peek(iter);
 181 }
 182
 183 void bch2_btree_and_journal_iter_init(struct btree_and_journal_iter *iter,
 184                                       struct btree_trans *trans,
 185                                       struct journal_keys *journal_keys,
 186                                       enum btree_id id, struct bpos pos)
 187 {
 188         memset(iter, 0, sizeof(*iter));
 189
 190         iter->btree = bch2_trans_get_iter(trans, id, pos, BTREE_ITER_PREFETCH);
 191         bch2_journal_iter_init(&iter->journal, journal_keys, id, 0, pos);
 192 }
 193
 194 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 195                                                 struct journal_keys *journal_keys,
 196                                                 struct btree *b)
 197 {
 198         memset(iter, 0, sizeof(*iter));
 199
 200         iter->b = b;
 201         bch2_btree_node_iter_init_from_start(&iter->node_iter, iter->b);
 202         bch2_journal_iter_init(&iter->journal, journal_keys,
 203                                b->c.btree_id, b->c.level, b->data->min_key);
 204 }
 205
 206 /* Walk btree, overlaying keys from the journal: */
 207
 208 static int bch2_btree_and_journal_walk_recurse(struct bch_fs *c, struct btree *b,
 209                                 struct journal_keys *journal_keys,
 210                                 enum btree_id btree_id,
 211                                 btree_walk_node_fn node_fn,
 212                                 btree_walk_key_fn key_fn)
 213 {
 214         struct btree_and_journal_iter iter;
 215         struct bkey_s_c k;
 216         int ret = 0;
 217
 218         bch2_btree_and_journal_iter_init_node_iter(&iter, journal_keys, b);
 219
 220         while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 221                 ret = key_fn(c, btree_id, b->c.level, k);
 222                 if (ret)
 223                         break;
 224
 225                 if (b->c.level) {
 226                         struct btree *child;
 227                         BKEY_PADDED(k) tmp;
 228
 229                         bkey_reassemble(&tmp.k, k);
 230                         k = bkey_i_to_s_c(&tmp.k);
 231
 232                         bch2_btree_and_journal_iter_advance(&iter);
 233
 234                         if (b->c.level > 0) {
 235                                 child = bch2_btree_node_get_noiter(c, &tmp.k,
 236                                                         b->c.btree_id, b->c.level - 1);
 237                                 ret = PTR_ERR_OR_ZERO(child);
 238                                 if (ret)
 239                                         break;
 240
 241                                 ret   = (node_fn ? node_fn(c, b) : 0) ?:
 242                                         bch2_btree_and_journal_walk_recurse(c, child,
 243                                                 journal_keys, btree_id, node_fn, key_fn);
 244                                 six_unlock_read(&child->c.lock);
 245
 246                                 if (ret)
 247                                         break;
 248                         }
 249                 } else {
 250                         bch2_btree_and_journal_iter_advance(&iter);
 251                 }
 252         }
 253
 254         return ret;
 255 }
 256
 257 int bch2_btree_and_journal_walk(struct bch_fs *c, struct journal_keys *journal_keys,
 258                                 enum btree_id btree_id,
 259                                 btree_walk_node_fn node_fn,
 260                                 btree_walk_key_fn key_fn)
 261 {
 262         struct btree *b = c->btree_roots[btree_id].b;
 263         int ret = 0;
 264
 265         if (btree_node_fake(b))
 266                 return 0;
 267
 268         six_lock_read(&b->c.lock, NULL, NULL);
 269         ret   = (node_fn ? node_fn(c, b) : 0) ?:
 270                 bch2_btree_and_journal_walk_recurse(c, b, journal_keys, btree_id,
 271                                                     node_fn, key_fn) ?:
 272                 key_fn(c, btree_id, b->c.level + 1, bkey_i_to_s_c(&b->key));
 273         six_unlock_read(&b->c.lock);
 274
 275         return ret;
 276 }
 277
 278 /* sort and dedup all keys in the journal: */
 279
 280 void bch2_journal_entries_free(struct list_head *list)
 281 {
 282
 283         while (!list_empty(list)) {
 284                 struct journal_replay *i =
 285                         list_first_entry(list, struct journal_replay, list);
 286                 list_del(&i->list);
 287                 kvpfree(i, offsetof(struct journal_replay, j) +
 288                         vstruct_bytes(&i->j));
 289         }
 290 }
 291
 292 /*
 293  * When keys compare equal, oldest compares first:
 294  */
 295 static int journal_sort_key_cmp(const void *_l, const void *_r)
 296 {
 297         const struct journal_key *l = _l;
 298         const struct journal_key *r = _r;
 299
 300         return  cmp_int(l->btree_id,    r->btree_id) ?:
 301                 cmp_int(l->level,       r->level) ?:
 302                 bkey_cmp(l->k->k.p, r->k->k.p) ?:
 303                 cmp_int(l->journal_seq, r->journal_seq) ?:
 304                 cmp_int(l->journal_offset, r->journal_offset);
 305 }
 306
 307 void bch2_journal_keys_free(struct journal_keys *keys)
 308 {
 309         kvfree(keys->d);
 310         keys->d = NULL;
 311         keys->nr = 0;
 312 }
 313
 314 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 315 {
 316         struct journal_replay *i;
 317         struct jset_entry *entry;
 318         struct bkey_i *k, *_n;
 319         struct journal_keys keys = { NULL };
 320         struct journal_key *src, *dst;
 321         size_t nr_keys = 0;
 322
 323         if (list_empty(journal_entries))
 324                 return keys;
 325
 326         list_for_each_entry(i, journal_entries, list) {
 327                 if (i->ignore)
 328                         continue;
 329
 330                 if (!keys.journal_seq_base)
 331                         keys.journal_seq_base = le64_to_cpu(i->j.seq);
 332
 333                 for_each_jset_key(k, _n, entry, &i->j)
 334                         nr_keys++;
 335         }
 336
 337         keys.d = kvmalloc(sizeof(keys.d[0]) * nr_keys, GFP_KERNEL);
 338         if (!keys.d)
 339                 goto err;
 340
 341         list_for_each_entry(i, journal_entries, list) {
 342                 if (i->ignore)
 343                         continue;
 344
 345                 BUG_ON(le64_to_cpu(i->j.seq) - keys.journal_seq_base > U32_MAX);
 346
 347                 for_each_jset_key(k, _n, entry, &i->j)
 348                         keys.d[keys.nr++] = (struct journal_key) {
 349                                 .btree_id       = entry->btree_id,
 350                                 .level          = entry->level,
 351                                 .k              = k,
 352                                 .journal_seq    = le64_to_cpu(i->j.seq) -
 353                                         keys.journal_seq_base,
 354                                 .journal_offset = k->_data - i->j._data,
 355                         };
 356         }
 357
 358         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 359
 360         src = dst = keys.d;
 361         while (src < keys.d + keys.nr) {
 362                 while (src + 1 < keys.d + keys.nr &&
 363                        src[0].btree_id  == src[1].btree_id &&
 364                        src[0].level     == src[1].level &&
 365                        !bkey_cmp(src[0].k->k.p, src[1].k->k.p))
 366                         src++;
 367
 368                 *dst++ = *src++;
 369         }
 370
 371         keys.nr = dst - keys.d;
 372 err:
 373         return keys;
 374 }
 375
 376 /* journal replay: */
 377
 378 static void replay_now_at(struct journal *j, u64 seq)
 379 {
 380         BUG_ON(seq < j->replay_journal_seq);
 381         BUG_ON(seq > j->replay_journal_seq_end);
 382
 383         while (j->replay_journal_seq < seq)
 384                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 385 }
 386
 387 static int bch2_extent_replay_key(struct bch_fs *c, enum btree_id btree_id,
 388                                   struct bkey_i *k)
 389 {
 390         struct btree_trans trans;
 391         struct btree_iter *iter, *split_iter;
 392         /*
 393          * We might cause compressed extents to be split, so we need to pass in
 394          * a disk_reservation:
 395          */
 396         struct disk_reservation disk_res =
 397                 bch2_disk_reservation_init(c, 0);
 398         struct bkey_i *split;
 399         struct bpos atomic_end;
 400         /*
 401          * Some extents aren't equivalent - w.r.t. what the triggers do
 402          * - if they're split:
 403          */
 404         bool remark_if_split = bch2_bkey_sectors_compressed(bkey_i_to_s_c(k)) ||
 405                 k->k.type == KEY_TYPE_reflink_p;
 406         bool remark = false;
 407         int ret;
 408
 409         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 410 retry:
 411         bch2_trans_begin(&trans);
 412
 413         iter = bch2_trans_get_iter(&trans, btree_id,
 414                                    bkey_start_pos(&k->k),
 415                                    BTREE_ITER_INTENT);
 416
 417         do {
 418                 ret = bch2_btree_iter_traverse(iter);
 419                 if (ret)
 420                         goto err;
 421
 422                 atomic_end = bpos_min(k->k.p, iter->l[0].b->key.k.p);
 423
 424                 split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
 425                 ret = PTR_ERR_OR_ZERO(split);
 426                 if (ret)
 427                         goto err;
 428
 429                 if (!remark &&
 430                     remark_if_split &&
 431                     bkey_cmp(atomic_end, k->k.p) < 0) {
 432                         ret = bch2_disk_reservation_add(c, &disk_res,
 433                                         k->k.size *
 434                                         bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(k)),
 435                                         BCH_DISK_RESERVATION_NOFAIL);
 436                         BUG_ON(ret);
 437
 438                         remark = true;
 439                 }
 440
 441                 bkey_copy(split, k);
 442                 bch2_cut_front(iter->pos, split);
 443                 bch2_cut_back(atomic_end, split);
 444
 445                 split_iter = bch2_trans_copy_iter(&trans, iter);
 446
 447                 /*
 448                  * It's important that we don't go through the
 449                  * extent_handle_overwrites() and extent_update_to_keys() path
 450                  * here: journal replay is supposed to treat extents like
 451                  * regular keys
 452                  */
 453                 __bch2_btree_iter_set_pos(split_iter, split->k.p, false);
 454                 bch2_trans_update(&trans, split_iter, split,
 455                                   BTREE_TRIGGER_NORUN);
 456                 bch2_trans_iter_put(&trans, split_iter);
 457
 458                 bch2_btree_iter_set_pos(iter, split->k.p);
 459
 460                 if (remark) {
 461                         ret = bch2_trans_mark_key(&trans,
 462                                                   bkey_s_c_null,
 463                                                   bkey_i_to_s_c(split),
 464                                                   0, split->k.size,
 465                                                   BTREE_TRIGGER_INSERT);
 466                         if (ret)
 467                                 goto err;
 468                 }
 469         } while (bkey_cmp(iter->pos, k->k.p) < 0);
 470
 471         if (remark) {
 472                 ret = bch2_trans_mark_key(&trans,
 473                                           bkey_i_to_s_c(k),
 474                                           bkey_s_c_null,
 475                                           0, -((s64) k->k.size),
 476                                           BTREE_TRIGGER_OVERWRITE);
 477                 if (ret)
 478                         goto err;
 479         }
 480
 481         ret = bch2_trans_commit(&trans, &disk_res, NULL,
 482                                 BTREE_INSERT_NOFAIL|
 483                                 BTREE_INSERT_LAZY_RW|
 484                                 BTREE_INSERT_JOURNAL_REPLAY);
 485 err:
 486         bch2_trans_iter_put(&trans, iter);
 487
 488         if (ret == -EINTR)
 489                 goto retry;
 490
 491         bch2_disk_reservation_put(c, &disk_res);
 492
 493         return bch2_trans_exit(&trans) ?: ret;
 494 }
 495
 496 static int __bch2_journal_replay_key(struct btree_trans *trans,
 497                                      enum btree_id id, unsigned level,
 498                                      struct bkey_i *k)
 499 {
 500         struct btree_iter *iter;
 501         int ret;
 502
 503         iter = bch2_trans_get_node_iter(trans, id, k->k.p,
 504                                         BTREE_MAX_DEPTH, level,
 505                                         BTREE_ITER_INTENT);
 506
 507         /*
 508          * iter->flags & BTREE_ITER_IS_EXTENTS triggers the update path to run
 509          * extent_handle_overwrites() and extent_update_to_keys() - but we don't
 510          * want that here, journal replay is supposed to treat extents like
 511          * regular keys:
 512          */
 513         __bch2_btree_iter_set_pos(iter, k->k.p, false);
 514
 515         ret   = bch2_btree_iter_traverse(iter) ?:
 516                 bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 517         bch2_trans_iter_put(trans, iter);
 518         return ret;
 519 }
 520
 521 static int bch2_journal_replay_key(struct bch_fs *c, enum btree_id id,
 522                                    unsigned level, struct bkey_i *k)
 523 {
 524         return bch2_trans_do(c, NULL, NULL,
 525                              BTREE_INSERT_NOFAIL|
 526                              BTREE_INSERT_LAZY_RW|
 527                              BTREE_INSERT_JOURNAL_REPLAY,
 528                              __bch2_journal_replay_key(&trans, id, level, k));
 529 }
 530
 531 static int __bch2_alloc_replay_key(struct btree_trans *trans, struct bkey_i *k)
 532 {
 533         struct btree_iter *iter;
 534         int ret;
 535
 536         iter = bch2_trans_get_iter(trans, BTREE_ID_ALLOC, k->k.p,
 537                                    BTREE_ITER_CACHED|
 538                                    BTREE_ITER_CACHED_NOFILL|
 539                                    BTREE_ITER_INTENT);
 540         ret = bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 541         bch2_trans_iter_put(trans, iter);
 542         return ret;
 543 }
 544
 545 static int bch2_alloc_replay_key(struct bch_fs *c, struct bkey_i *k)
 546 {
 547         return bch2_trans_do(c, NULL, NULL,
 548                              BTREE_INSERT_NOFAIL|
 549                              BTREE_INSERT_USE_RESERVE|
 550                              BTREE_INSERT_LAZY_RW|
 551                              BTREE_INSERT_JOURNAL_REPLAY,
 552                         __bch2_alloc_replay_key(&trans, k));
 553 }
 554
 555 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 556 {
 557         const struct journal_key *l = _l;
 558         const struct journal_key *r = _r;
 559
 560         return  cmp_int(r->level,       l->level) ?:
 561                 cmp_int(l->journal_seq, r->journal_seq) ?:
 562                 cmp_int(l->btree_id,    r->btree_id) ?:
 563                 bkey_cmp(l->k->k.p,     r->k->k.p);
 564 }
 565
 566 static int bch2_journal_replay(struct bch_fs *c,
 567                                struct journal_keys keys)
 568 {
 569         struct journal *j = &c->journal;
 570         struct journal_key *i;
 571         u64 seq;
 572         int ret;
 573
 574         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
 575
 576         if (keys.nr)
 577                 replay_now_at(j, keys.journal_seq_base);
 578
 579         seq = j->replay_journal_seq;
 580
 581         /*
 582          * First replay updates to the alloc btree - these will only update the
 583          * btree key cache:
 584          */
 585         for_each_journal_key(keys, i) {
 586                 cond_resched();
 587
 588                 if (!i->level && i->btree_id == BTREE_ID_ALLOC) {
 589                         j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
 590                         ret = bch2_alloc_replay_key(c, i->k);
 591                         if (ret)
 592                                 goto err;
 593                 }
 594         }
 595
 596         /*
 597          * Next replay updates to interior btree nodes:
 598          */
 599         for_each_journal_key(keys, i) {
 600                 cond_resched();
 601
 602                 if (i->level) {
 603                         j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
 604                         ret = bch2_journal_replay_key(c, i->btree_id, i->level, i->k);
 605                         if (ret)
 606                                 goto err;
 607                 }
 608         }
 609
 610         /*
 611          * Now that the btree is in a consistent state, we can start journal
 612          * reclaim (which will be flushing entries from the btree key cache back
 613          * to the btree:
 614          */
 615         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
 616         set_bit(JOURNAL_RECLAIM_STARTED, &j->flags);
 617         journal_reclaim_kick(j);
 618
 619         j->replay_journal_seq = seq;
 620
 621         /*
 622          * Now replay leaf node updates:
 623          */
 624         for_each_journal_key(keys, i) {
 625                 cond_resched();
 626
 627                 if (i->level || i->btree_id == BTREE_ID_ALLOC)
 628                         continue;
 629
 630                 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 631
 632                 ret = i->k->k.size
 633                         ? bch2_extent_replay_key(c, i->btree_id, i->k)
 634                         : bch2_journal_replay_key(c, i->btree_id, i->level, i->k);
 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", ret);
 647         return ret;
 648 }
 649
 650 /* journal replay early: */
 651
 652 static int journal_replay_entry_early(struct bch_fs *c,
 653                                       struct jset_entry *entry)
 654 {
 655         int ret = 0;
 656
 657         switch (entry->type) {
 658         case BCH_JSET_ENTRY_btree_root: {
 659                 struct btree_root *r;
 660
 661                 if (entry->btree_id >= BTREE_ID_NR) {
 662                         bch_err(c, "filesystem has unknown btree type %u",
 663                                 entry->btree_id);
 664                         return -EINVAL;
 665                 }
 666
 667                 r = &c->btree_roots[entry->btree_id];
 668
 669                 if (entry->u64s) {
 670                         r->level = entry->level;
 671                         bkey_copy(&r->key, &entry->start[0]);
 672                         r->error = 0;
 673                 } else {
 674                         r->error = -EIO;
 675                 }
 676                 r->alive = true;
 677                 break;
 678         }
 679         case BCH_JSET_ENTRY_usage: {
 680                 struct jset_entry_usage *u =
 681                         container_of(entry, struct jset_entry_usage, entry);
 682
 683                 switch (entry->btree_id) {
 684                 case FS_USAGE_RESERVED:
 685                         if (entry->level < BCH_REPLICAS_MAX)
 686                                 c->usage_base->persistent_reserved[entry->level] =
 687                                         le64_to_cpu(u->v);
 688                         break;
 689                 case FS_USAGE_INODES:
 690                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 691                         break;
 692                 case FS_USAGE_KEY_VERSION:
 693                         atomic64_set(&c->key_version,
 694                                      le64_to_cpu(u->v));
 695                         break;
 696                 }
 697
 698                 break;
 699         }
 700         case BCH_JSET_ENTRY_data_usage: {
 701                 struct jset_entry_data_usage *u =
 702                         container_of(entry, struct jset_entry_data_usage, entry);
 703                 ret = bch2_replicas_set_usage(c, &u->r,
 704                                               le64_to_cpu(u->v));
 705                 break;
 706         }
 707         case BCH_JSET_ENTRY_blacklist: {
 708                 struct jset_entry_blacklist *bl_entry =
 709                         container_of(entry, struct jset_entry_blacklist, entry);
 710
 711                 ret = bch2_journal_seq_blacklist_add(c,
 712                                 le64_to_cpu(bl_entry->seq),
 713                                 le64_to_cpu(bl_entry->seq) + 1);
 714                 break;
 715         }
 716         case BCH_JSET_ENTRY_blacklist_v2: {
 717                 struct jset_entry_blacklist_v2 *bl_entry =
 718                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 719
 720                 ret = bch2_journal_seq_blacklist_add(c,
 721                                 le64_to_cpu(bl_entry->start),
 722                                 le64_to_cpu(bl_entry->end) + 1);
 723                 break;
 724         }
 725         }
 726
 727         return ret;
 728 }
 729
 730 static int journal_replay_early(struct bch_fs *c,
 731                                 struct bch_sb_field_clean *clean,
 732                                 struct list_head *journal)
 733 {
 734         struct journal_replay *i;
 735         struct jset_entry *entry;
 736         int ret;
 737
 738         if (clean) {
 739                 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
 740                 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
 741
 742                 for (entry = clean->start;
 743                      entry != vstruct_end(&clean->field);
 744                      entry = vstruct_next(entry)) {
 745                         ret = journal_replay_entry_early(c, entry);
 746                         if (ret)
 747                                 return ret;
 748                 }
 749         } else {
 750                 list_for_each_entry(i, journal, list) {
 751                         if (i->ignore)
 752                                 continue;
 753
 754                         c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
 755                         c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
 756
 757                         vstruct_for_each(&i->j, entry) {
 758                                 ret = journal_replay_entry_early(c, entry);
 759                                 if (ret)
 760                                         return ret;
 761                         }
 762                 }
 763         }
 764
 765         bch2_fs_usage_initialize(c);
 766
 767         return 0;
 768 }
 769
 770 /* sb clean section: */
 771
 772 static struct bkey_i *btree_root_find(struct bch_fs *c,
 773                                       struct bch_sb_field_clean *clean,
 774                                       struct jset *j,
 775                                       enum btree_id id, unsigned *level)
 776 {
 777         struct bkey_i *k;
 778         struct jset_entry *entry, *start, *end;
 779
 780         if (clean) {
 781                 start = clean->start;
 782                 end = vstruct_end(&clean->field);
 783         } else {
 784                 start = j->start;
 785                 end = vstruct_last(j);
 786         }
 787
 788         for (entry = start; entry < end; entry = vstruct_next(entry))
 789                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 790                     entry->btree_id == id)
 791                         goto found;
 792
 793         return NULL;
 794 found:
 795         if (!entry->u64s)
 796                 return ERR_PTR(-EINVAL);
 797
 798         k = entry->start;
 799         *level = entry->level;
 800         return k;
 801 }
 802
 803 static int verify_superblock_clean(struct bch_fs *c,
 804                                    struct bch_sb_field_clean **cleanp,
 805                                    struct jset *j)
 806 {
 807         unsigned i;
 808         struct bch_sb_field_clean *clean = *cleanp;
 809         int ret = 0;
 810
 811         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 812                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 813                         le64_to_cpu(clean->journal_seq),
 814                         le64_to_cpu(j->seq))) {
 815                 kfree(clean);
 816                 *cleanp = NULL;
 817                 return 0;
 818         }
 819
 820         mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
 821                         "superblock read clock %u doesn't match journal %u after clean shutdown",
 822                         clean->read_clock, j->read_clock);
 823         mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
 824                         "superblock write clock %u doesn't match journal %u after clean shutdown",
 825                         clean->write_clock, j->write_clock);
 826
 827         for (i = 0; i < BTREE_ID_NR; i++) {
 828                 char buf1[200], buf2[200];
 829                 struct bkey_i *k1, *k2;
 830                 unsigned l1 = 0, l2 = 0;
 831
 832                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 833                 k2 = btree_root_find(c, NULL, j, i, &l2);
 834
 835                 if (!k1 && !k2)
 836                         continue;
 837
 838                 mustfix_fsck_err_on(!k1 || !k2 ||
 839                                     IS_ERR(k1) ||
 840                                     IS_ERR(k2) ||
 841                                     k1->k.u64s != k2->k.u64s ||
 842                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 843                                     l1 != l2, c,
 844                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 845                         "sb:      l=%u %s\n"
 846                         "journal: l=%u %s\n", i,
 847                         l1, (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(k1)), buf1),
 848                         l2, (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(k2)), buf2));
 849         }
 850 fsck_err:
 851         return ret;
 852 }
 853
 854 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 855 {
 856         struct bch_sb_field_clean *clean, *sb_clean;
 857         int ret;
 858
 859         mutex_lock(&c->sb_lock);
 860         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 861
 862         if (fsck_err_on(!sb_clean, c,
 863                         "superblock marked clean but clean section not present")) {
 864                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 865                 c->sb.clean = false;
 866                 mutex_unlock(&c->sb_lock);
 867                 return NULL;
 868         }
 869
 870         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 871                         GFP_KERNEL);
 872         if (!clean) {
 873                 mutex_unlock(&c->sb_lock);
 874                 return ERR_PTR(-ENOMEM);
 875         }
 876
 877         if (le16_to_cpu(c->disk_sb.sb->version) <
 878             bcachefs_metadata_version_bkey_renumber)
 879                 bch2_sb_clean_renumber(clean, READ);
 880
 881         mutex_unlock(&c->sb_lock);
 882
 883         return clean;
 884 fsck_err:
 885         mutex_unlock(&c->sb_lock);
 886         return ERR_PTR(ret);
 887 }
 888
 889 static int read_btree_roots(struct bch_fs *c)
 890 {
 891         unsigned i;
 892         int ret = 0;
 893
 894         for (i = 0; i < BTREE_ID_NR; i++) {
 895                 struct btree_root *r = &c->btree_roots[i];
 896
 897                 if (!r->alive)
 898                         continue;
 899
 900                 if (i == BTREE_ID_ALLOC &&
 901                     c->opts.reconstruct_alloc) {
 902                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 903                         continue;
 904                 }
 905
 906                 if (r->error) {
 907                         __fsck_err(c, i == BTREE_ID_ALLOC
 908                                    ? FSCK_CAN_IGNORE : 0,
 909                                    "invalid btree root %s",
 910                                    bch2_btree_ids[i]);
 911                         if (i == BTREE_ID_ALLOC)
 912                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 913                 }
 914
 915                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 916                 if (ret) {
 917                         __fsck_err(c, i == BTREE_ID_ALLOC
 918                                    ? FSCK_CAN_IGNORE : 0,
 919                                    "error reading btree root %s",
 920                                    bch2_btree_ids[i]);
 921                         if (i == BTREE_ID_ALLOC)
 922                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 923                 }
 924         }
 925
 926         for (i = 0; i < BTREE_ID_NR; i++)
 927                 if (!c->btree_roots[i].b)
 928                         bch2_btree_root_alloc(c, i);
 929 fsck_err:
 930         return ret;
 931 }
 932
 933 int bch2_fs_recovery(struct bch_fs *c)
 934 {
 935         const char *err = "cannot allocate memory";
 936         struct bch_sb_field_clean *clean = NULL;
 937         struct jset *last_journal_entry = NULL;
 938         u64 blacklist_seq, journal_seq;
 939         bool write_sb = false, need_write_alloc = false;
 940         int ret;
 941
 942         if (c->sb.clean)
 943                 clean = read_superblock_clean(c);
 944         ret = PTR_ERR_OR_ZERO(clean);
 945         if (ret)
 946                 goto err;
 947
 948         if (c->sb.clean)
 949                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 950                          le64_to_cpu(clean->journal_seq));
 951
 952         if (!c->replicas.entries ||
 953             c->opts.rebuild_replicas) {
 954                 bch_info(c, "building replicas info");
 955                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 956         }
 957
 958         ret = bch2_blacklist_table_initialize(c);
 959         if (ret) {
 960                 bch_err(c, "error initializing blacklist table");
 961                 goto err;
 962         }
 963
 964         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
 965                 struct journal_replay *i;
 966
 967                 ret = bch2_journal_read(c, &c->journal_entries,
 968                                         &blacklist_seq, &journal_seq);
 969                 if (ret)
 970                         goto err;
 971
 972                 list_for_each_entry_reverse(i, &c->journal_entries, list)
 973                         if (!i->ignore) {
 974                                 last_journal_entry = &i->j;
 975                                 break;
 976                         }
 977
 978                 if (mustfix_fsck_err_on(c->sb.clean &&
 979                                         last_journal_entry &&
 980                                         !journal_entry_empty(last_journal_entry), c,
 981                                 "filesystem marked clean but journal not empty")) {
 982                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 983                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 984                         c->sb.clean = false;
 985                 }
 986
 987                 if (!last_journal_entry) {
 988                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
 989                         goto use_clean;
 990                 }
 991
 992                 c->journal_keys = journal_keys_sort(&c->journal_entries);
 993                 if (!c->journal_keys.d) {
 994                         ret = -ENOMEM;
 995                         goto err;
 996                 }
 997
 998                 if (c->sb.clean && last_journal_entry) {
 999                         ret = verify_superblock_clean(c, &clean,
1000                                                       last_journal_entry);
1001                         if (ret)
1002                                 goto err;
1003                 }
1004         } else {
1005 use_clean:
1006                 if (!clean) {
1007                         bch_err(c, "no superblock clean section found");
1008                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1009                         goto err;
1010
1011                 }
1012                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1013         }
1014
1015         if (!c->sb.clean &&
1016             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1017                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1018                 ret = -EINVAL;
1019                 goto err;
1020         }
1021
1022         if (c->opts.reconstruct_alloc) {
1023                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
1024                 drop_alloc_keys(&c->journal_keys);
1025         }
1026
1027         ret = journal_replay_early(c, clean, &c->journal_entries);
1028         if (ret)
1029                 goto err;
1030
1031         /*
1032          * After an unclean shutdown, skip then next few journal sequence
1033          * numbers as they may have been referenced by btree writes that
1034          * happened before their corresponding journal writes - those btree
1035          * writes need to be ignored, by skipping and blacklisting the next few
1036          * journal sequence numbers:
1037          */
1038         if (!c->sb.clean)
1039                 journal_seq += 8;
1040
1041         if (blacklist_seq != journal_seq) {
1042                 ret = bch2_journal_seq_blacklist_add(c,
1043                                         blacklist_seq, journal_seq);
1044                 if (ret) {
1045                         bch_err(c, "error creating new journal seq blacklist entry");
1046                         goto err;
1047                 }
1048         }
1049
1050         ret = bch2_fs_journal_start(&c->journal, journal_seq,
1051                                     &c->journal_entries);
1052         if (ret)
1053                 goto err;
1054
1055         ret = read_btree_roots(c);
1056         if (ret)
1057                 goto err;
1058
1059         bch_verbose(c, "starting alloc read");
1060         err = "error reading allocation information";
1061         ret = bch2_alloc_read(c, &c->journal_keys);
1062         if (ret)
1063                 goto err;
1064         bch_verbose(c, "alloc read done");
1065
1066         bch_verbose(c, "starting stripes_read");
1067         err = "error reading stripes";
1068         ret = bch2_stripes_read(c, &c->journal_keys);
1069         if (ret)
1070                 goto err;
1071         bch_verbose(c, "stripes_read done");
1072
1073         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1074
1075         if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
1076             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
1077                 /*
1078                  * interior btree node updates aren't consistent with the
1079                  * journal; after an unclean shutdown we have to walk all
1080                  * pointers to metadata:
1081                  */
1082                 bch_info(c, "starting metadata mark and sweep");
1083                 err = "error in mark and sweep";
1084                 ret = bch2_gc(c, &c->journal_keys, true, true);
1085                 if (ret < 0)
1086                         goto err;
1087                 if (ret)
1088                         need_write_alloc = true;
1089                 bch_verbose(c, "mark and sweep done");
1090         }
1091
1092         if (c->opts.fsck ||
1093             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
1094             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1095                 bch_info(c, "starting mark and sweep");
1096                 err = "error in mark and sweep";
1097                 ret = bch2_gc(c, &c->journal_keys, true, false);
1098                 if (ret < 0)
1099                         goto err;
1100                 if (ret)
1101                         need_write_alloc = true;
1102                 bch_verbose(c, "mark and sweep done");
1103         }
1104
1105         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1106         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1107
1108         /*
1109          * Skip past versions that might have possibly been used (as nonces),
1110          * but hadn't had their pointers written:
1111          */
1112         if (c->sb.encryption_type && !c->sb.clean)
1113                 atomic64_add(1 << 16, &c->key_version);
1114
1115         if (c->opts.norecovery)
1116                 goto out;
1117
1118         bch_verbose(c, "starting journal replay");
1119         err = "journal replay failed";
1120         ret = bch2_journal_replay(c, c->journal_keys);
1121         if (ret)
1122                 goto err;
1123         bch_verbose(c, "journal replay done");
1124
1125         if (need_write_alloc && !c->opts.nochanges) {
1126                 /*
1127                  * note that even when filesystem was clean there might be work
1128                  * to do here, if we ran gc (because of fsck) which recalculated
1129                  * oldest_gen:
1130                  */
1131                 bch_verbose(c, "writing allocation info");
1132                 err = "error writing out alloc info";
1133                 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW) ?:
1134                         bch2_alloc_write(c, BTREE_INSERT_LAZY_RW);
1135                 if (ret) {
1136                         bch_err(c, "error writing alloc info");
1137                         goto err;
1138                 }
1139                 bch_verbose(c, "alloc write done");
1140
1141                 set_bit(BCH_FS_ALLOC_WRITTEN, &c->flags);
1142         }
1143
1144         if (!c->sb.clean) {
1145                 if (!(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
1146                         bch_info(c, "checking inode link counts");
1147                         err = "error in recovery";
1148                         ret = bch2_fsck_inode_nlink(c);
1149                         if (ret)
1150                                 goto err;
1151                         bch_verbose(c, "check inodes done");
1152
1153                 } else {
1154                         bch_verbose(c, "checking for deleted inodes");
1155                         err = "error in recovery";
1156                         ret = bch2_fsck_walk_inodes_only(c);
1157                         if (ret)
1158                                 goto err;
1159                         bch_verbose(c, "check inodes done");
1160                 }
1161         }
1162
1163         if (c->opts.fsck) {
1164                 bch_info(c, "starting fsck");
1165                 err = "error in fsck";
1166                 ret = bch2_fsck_full(c);
1167                 if (ret)
1168                         goto err;
1169                 bch_verbose(c, "fsck done");
1170         }
1171
1172         if (enabled_qtypes(c)) {
1173                 bch_verbose(c, "reading quotas");
1174                 ret = bch2_fs_quota_read(c);
1175                 if (ret)
1176                         goto err;
1177                 bch_verbose(c, "quotas done");
1178         }
1179
1180         mutex_lock(&c->sb_lock);
1181         if (c->opts.version_upgrade) {
1182                 if (c->sb.version < bcachefs_metadata_version_new_versioning)
1183                         c->disk_sb.sb->version_min =
1184                                 le16_to_cpu(bcachefs_metadata_version_min);
1185                 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
1186                 c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1187                 write_sb = true;
1188         }
1189
1190         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1191                 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
1192                 write_sb = true;
1193         }
1194
1195         if (c->opts.fsck &&
1196             !test_bit(BCH_FS_ERROR, &c->flags)) {
1197                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1198                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1199                 write_sb = true;
1200         }
1201
1202         if (write_sb)
1203                 bch2_write_super(c);
1204         mutex_unlock(&c->sb_lock);
1205
1206         if (c->journal_seq_blacklist_table &&
1207             c->journal_seq_blacklist_table->nr > 128)
1208                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1209 out:
1210         ret = 0;
1211 err:
1212 fsck_err:
1213         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1214         bch2_flush_fsck_errs(c);
1215
1216         if (!c->opts.keep_journal) {
1217                 bch2_journal_keys_free(&c->journal_keys);
1218                 bch2_journal_entries_free(&c->journal_entries);
1219         }
1220         kfree(clean);
1221         if (ret)
1222                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1223         else
1224                 bch_verbose(c, "ret %i", ret);
1225         return ret;
1226 }
1227
1228 int bch2_fs_initialize(struct bch_fs *c)
1229 {
1230         struct bch_inode_unpacked root_inode, lostfound_inode;
1231         struct bkey_inode_buf packed_inode;
1232         struct qstr lostfound = QSTR("lost+found");
1233         const char *err = "cannot allocate memory";
1234         struct bch_dev *ca;
1235         LIST_HEAD(journal);
1236         unsigned i;
1237         int ret;
1238
1239         bch_notice(c, "initializing new filesystem");
1240
1241         mutex_lock(&c->sb_lock);
1242         for_each_online_member(ca, c, i)
1243                 bch2_mark_dev_superblock(c, ca, 0);
1244         mutex_unlock(&c->sb_lock);
1245
1246         mutex_lock(&c->sb_lock);
1247         c->disk_sb.sb->version = c->disk_sb.sb->version_min =
1248                 le16_to_cpu(bcachefs_metadata_version_current);
1249         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1250         c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1251
1252         bch2_write_super(c);
1253         mutex_unlock(&c->sb_lock);
1254
1255         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1256         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1257
1258         for (i = 0; i < BTREE_ID_NR; i++)
1259                 bch2_btree_root_alloc(c, i);
1260
1261         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
1262         set_bit(JOURNAL_RECLAIM_STARTED, &c->journal.flags);
1263
1264         err = "unable to allocate journal buckets";
1265         for_each_online_member(ca, c, i) {
1266                 ret = bch2_dev_journal_alloc(ca);
1267                 if (ret) {
1268                         percpu_ref_put(&ca->io_ref);
1269                         goto err;
1270                 }
1271         }
1272
1273         /*
1274          * journal_res_get() will crash if called before this has
1275          * set up the journal.pin FIFO and journal.cur pointer:
1276          */
1277         bch2_fs_journal_start(&c->journal, 1, &journal);
1278         bch2_journal_set_replay_done(&c->journal);
1279
1280         err = "error going read-write";
1281         ret = bch2_fs_read_write_early(c);
1282         if (ret)
1283                 goto err;
1284
1285         /*
1286          * Write out the superblock and journal buckets, now that we can do
1287          * btree updates
1288          */
1289         err = "error writing alloc info";
1290         ret = bch2_alloc_write(c, 0);
1291         if (ret)
1292                 goto err;
1293
1294         bch2_inode_init(c, &root_inode, 0, 0,
1295                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1296         root_inode.bi_inum = BCACHEFS_ROOT_INO;
1297         bch2_inode_pack(c, &packed_inode, &root_inode);
1298
1299         err = "error creating root directory";
1300         ret = bch2_btree_insert(c, BTREE_ID_INODES,
1301                                 &packed_inode.inode.k_i,
1302                                 NULL, NULL, 0);
1303         if (ret)
1304                 goto err;
1305
1306         bch2_inode_init_early(c, &lostfound_inode);
1307
1308         err = "error creating lost+found";
1309         ret = bch2_trans_do(c, NULL, NULL, 0,
1310                 bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
1311                                   &root_inode, &lostfound_inode,
1312                                   &lostfound,
1313                                   0, 0, S_IFDIR|0700, 0,
1314                                   NULL, NULL));
1315         if (ret)
1316                 goto err;
1317
1318         if (enabled_qtypes(c)) {
1319                 ret = bch2_fs_quota_read(c);
1320                 if (ret)
1321                         goto err;
1322         }
1323
1324         err = "error writing first journal entry";
1325         ret = bch2_journal_meta(&c->journal);
1326         if (ret)
1327                 goto err;
1328
1329         mutex_lock(&c->sb_lock);
1330         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1331         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1332
1333         bch2_write_super(c);
1334         mutex_unlock(&c->sb_lock);
1335
1336         return 0;
1337 err:
1338         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1339         return ret;
1340 }