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 "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 /* iterate over keys read from the journal: */
  29
  30 static struct journal_key *journal_key_search(struct journal_keys *journal_keys,
  31                                               enum btree_id id, unsigned level,
  32                                               struct bpos pos)
  33 {
  34         size_t l = 0, r = journal_keys->nr, m;
  35
  36         while (l < r) {
  37                 m = l + ((r - l) >> 1);
  38                 if ((cmp_int(id,        journal_keys->d[m].btree_id) ?:
  39                      cmp_int(level,     journal_keys->d[m].level) ?:
  40                      bkey_cmp(pos,      journal_keys->d[m].k->k.p)) > 0)
  41                         l = m + 1;
  42                 else
  43                         r = m;
  44         }
  45
  46         BUG_ON(l < journal_keys->nr &&
  47                (cmp_int(id,     journal_keys->d[l].btree_id) ?:
  48                 cmp_int(level,  journal_keys->d[l].level) ?:
  49                 bkey_cmp(pos,   journal_keys->d[l].k->k.p)) > 0);
  50
  51         BUG_ON(l &&
  52                (cmp_int(id,     journal_keys->d[l - 1].btree_id) ?:
  53                 cmp_int(level,  journal_keys->d[l - 1].level) ?:
  54                 bkey_cmp(pos,   journal_keys->d[l - 1].k->k.p)) <= 0);
  55
  56         return l < journal_keys->nr ? journal_keys->d + l : NULL;
  57 }
  58
  59 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
  60 {
  61         if (iter->k &&
  62             iter->k < iter->keys->d + iter->keys->nr &&
  63             iter->k->btree_id   == iter->btree_id &&
  64             iter->k->level      == iter->level)
  65                 return iter->k->k;
  66
  67         iter->k = NULL;
  68         return NULL;
  69 }
  70
  71 static void bch2_journal_iter_advance(struct journal_iter *iter)
  72 {
  73         if (iter->k)
  74                 iter->k++;
  75 }
  76
  77 static void bch2_journal_iter_init(struct journal_iter *iter,
  78                                    struct journal_keys *journal_keys,
  79                                    enum btree_id id, unsigned level,
  80                                    struct bpos pos)
  81 {
  82         iter->btree_id  = id;
  83         iter->level     = level;
  84         iter->keys      = journal_keys;
  85         iter->k         = journal_key_search(journal_keys, id, level, pos);
  86 }
  87
  88 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
  89 {
  90         return iter->btree
  91                 ? bch2_btree_iter_peek(iter->btree)
  92                 : bch2_btree_node_iter_peek_unpack(&iter->node_iter,
  93                                                    iter->b, &iter->unpacked);
  94 }
  95
  96 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
  97 {
  98         if (iter->btree)
  99                 bch2_btree_iter_next(iter->btree);
 100         else
 101                 bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 102 }
 103
 104 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 105 {
 106         switch (iter->last) {
 107         case none:
 108                 break;
 109         case btree:
 110                 bch2_journal_iter_advance_btree(iter);
 111                 break;
 112         case journal:
 113                 bch2_journal_iter_advance(&iter->journal);
 114                 break;
 115         }
 116
 117         iter->last = none;
 118 }
 119
 120 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 121 {
 122         struct bkey_s_c ret;
 123
 124         while (1) {
 125                 struct bkey_s_c btree_k         =
 126                         bch2_journal_iter_peek_btree(iter);
 127                 struct bkey_s_c journal_k       =
 128                         bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
 129
 130                 if (btree_k.k && journal_k.k) {
 131                         int cmp = bkey_cmp(btree_k.k->p, journal_k.k->p);
 132
 133                         if (!cmp)
 134                                 bch2_journal_iter_advance_btree(iter);
 135
 136                         iter->last = cmp < 0 ? btree : journal;
 137                 } else if (btree_k.k) {
 138                         iter->last = btree;
 139                 } else if (journal_k.k) {
 140                         iter->last = journal;
 141                 } else {
 142                         iter->last = none;
 143                         return bkey_s_c_null;
 144                 }
 145
 146                 ret = iter->last == journal ? journal_k : btree_k;
 147
 148                 if (iter->b &&
 149                     bkey_cmp(ret.k->p, iter->b->data->max_key) > 0) {
 150                         iter->journal.k = NULL;
 151                         iter->last = none;
 152                         return bkey_s_c_null;
 153                 }
 154
 155                 if (!bkey_deleted(ret.k))
 156                         break;
 157
 158                 bch2_btree_and_journal_iter_advance(iter);
 159         }
 160
 161         return ret;
 162 }
 163
 164 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 165 {
 166         bch2_btree_and_journal_iter_advance(iter);
 167
 168         return bch2_btree_and_journal_iter_peek(iter);
 169 }
 170
 171 void bch2_btree_and_journal_iter_init(struct btree_and_journal_iter *iter,
 172                                       struct btree_trans *trans,
 173                                       struct journal_keys *journal_keys,
 174                                       enum btree_id id, struct bpos pos)
 175 {
 176         memset(iter, 0, sizeof(*iter));
 177
 178         iter->btree = bch2_trans_get_iter(trans, id, pos, 0);
 179         bch2_journal_iter_init(&iter->journal, journal_keys, id, 0, pos);
 180 }
 181
 182 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 183                                                 struct journal_keys *journal_keys,
 184                                                 struct btree *b)
 185 {
 186         memset(iter, 0, sizeof(*iter));
 187
 188         iter->b = b;
 189         bch2_btree_node_iter_init_from_start(&iter->node_iter, iter->b);
 190         bch2_journal_iter_init(&iter->journal, journal_keys,
 191                                b->c.btree_id, b->c.level, b->data->min_key);
 192 }
 193
 194 /* Walk btree, overlaying keys from the journal: */
 195
 196 static int bch2_btree_and_journal_walk_recurse(struct bch_fs *c, struct btree *b,
 197                                 struct journal_keys *journal_keys,
 198                                 enum btree_id btree_id,
 199                                 btree_walk_node_fn node_fn,
 200                                 btree_walk_key_fn key_fn)
 201 {
 202         struct btree_and_journal_iter iter;
 203         struct bkey_s_c k;
 204         int ret = 0;
 205
 206         bch2_btree_and_journal_iter_init_node_iter(&iter, journal_keys, b);
 207
 208         while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 209                 ret = key_fn(c, btree_id, b->c.level, k);
 210                 if (ret)
 211                         break;
 212
 213                 if (b->c.level) {
 214                         struct btree *child;
 215                         BKEY_PADDED(k) tmp;
 216
 217                         bkey_reassemble(&tmp.k, k);
 218                         k = bkey_i_to_s_c(&tmp.k);
 219
 220                         bch2_btree_and_journal_iter_advance(&iter);
 221
 222                         if (b->c.level > 0) {
 223                                 child = bch2_btree_node_get_noiter(c, &tmp.k,
 224                                                         b->c.btree_id, b->c.level - 1);
 225                                 ret = PTR_ERR_OR_ZERO(child);
 226                                 if (ret)
 227                                         break;
 228
 229                                 ret   = (node_fn ? node_fn(c, b) : 0) ?:
 230                                         bch2_btree_and_journal_walk_recurse(c, child,
 231                                                 journal_keys, btree_id, node_fn, key_fn);
 232                                 six_unlock_read(&child->c.lock);
 233
 234                                 if (ret)
 235                                         break;
 236                         }
 237                 } else {
 238                         bch2_btree_and_journal_iter_advance(&iter);
 239                 }
 240         }
 241
 242         return ret;
 243 }
 244
 245 int bch2_btree_and_journal_walk(struct bch_fs *c, struct journal_keys *journal_keys,
 246                                 enum btree_id btree_id,
 247                                 btree_walk_node_fn node_fn,
 248                                 btree_walk_key_fn key_fn)
 249 {
 250         struct btree *b = c->btree_roots[btree_id].b;
 251         int ret = 0;
 252
 253         if (btree_node_fake(b))
 254                 return 0;
 255
 256         six_lock_read(&b->c.lock, NULL, NULL);
 257         ret   = (node_fn ? node_fn(c, b) : 0) ?:
 258                 bch2_btree_and_journal_walk_recurse(c, b, journal_keys, btree_id,
 259                                                     node_fn, key_fn) ?:
 260                 key_fn(c, btree_id, b->c.level + 1, bkey_i_to_s_c(&b->key));
 261         six_unlock_read(&b->c.lock);
 262
 263         return ret;
 264 }
 265
 266 /* sort and dedup all keys in the journal: */
 267
 268 void bch2_journal_entries_free(struct list_head *list)
 269 {
 270
 271         while (!list_empty(list)) {
 272                 struct journal_replay *i =
 273                         list_first_entry(list, struct journal_replay, list);
 274                 list_del(&i->list);
 275                 kvpfree(i, offsetof(struct journal_replay, j) +
 276                         vstruct_bytes(&i->j));
 277         }
 278 }
 279
 280 /*
 281  * When keys compare equal, oldest compares first:
 282  */
 283 static int journal_sort_key_cmp(const void *_l, const void *_r)
 284 {
 285         const struct journal_key *l = _l;
 286         const struct journal_key *r = _r;
 287
 288         return  cmp_int(l->btree_id,    r->btree_id) ?:
 289                 cmp_int(l->level,       r->level) ?:
 290                 bkey_cmp(l->k->k.p, r->k->k.p) ?:
 291                 cmp_int(l->journal_seq, r->journal_seq) ?:
 292                 cmp_int(l->journal_offset, r->journal_offset);
 293 }
 294
 295 void bch2_journal_keys_free(struct journal_keys *keys)
 296 {
 297         kvfree(keys->d);
 298         keys->d = NULL;
 299         keys->nr = 0;
 300 }
 301
 302 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 303 {
 304         struct journal_replay *p;
 305         struct jset_entry *entry;
 306         struct bkey_i *k, *_n;
 307         struct journal_keys keys = { NULL };
 308         struct journal_key *src, *dst;
 309         size_t nr_keys = 0;
 310
 311         if (list_empty(journal_entries))
 312                 return keys;
 313
 314         keys.journal_seq_base =
 315                 le64_to_cpu(list_last_entry(journal_entries,
 316                                 struct journal_replay, list)->j.last_seq);
 317
 318         list_for_each_entry(p, journal_entries, list) {
 319                 if (le64_to_cpu(p->j.seq) < keys.journal_seq_base)
 320                         continue;
 321
 322                 for_each_jset_key(k, _n, entry, &p->j)
 323                         nr_keys++;
 324         }
 325
 326
 327         keys.d = kvmalloc(sizeof(keys.d[0]) * nr_keys, GFP_KERNEL);
 328         if (!keys.d)
 329                 goto err;
 330
 331         list_for_each_entry(p, journal_entries, list) {
 332                 if (le64_to_cpu(p->j.seq) < keys.journal_seq_base)
 333                         continue;
 334
 335                 for_each_jset_key(k, _n, entry, &p->j)
 336                         keys.d[keys.nr++] = (struct journal_key) {
 337                                 .btree_id       = entry->btree_id,
 338                                 .level          = entry->level,
 339                                 .k              = k,
 340                                 .journal_seq    = le64_to_cpu(p->j.seq) -
 341                                         keys.journal_seq_base,
 342                                 .journal_offset = k->_data - p->j._data,
 343                         };
 344         }
 345
 346         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 347
 348         src = dst = keys.d;
 349         while (src < keys.d + keys.nr) {
 350                 while (src + 1 < keys.d + keys.nr &&
 351                        src[0].btree_id  == src[1].btree_id &&
 352                        src[0].level     == src[1].level &&
 353                        !bkey_cmp(src[0].k->k.p, src[1].k->k.p))
 354                         src++;
 355
 356                 *dst++ = *src++;
 357         }
 358
 359         keys.nr = dst - keys.d;
 360 err:
 361         return keys;
 362 }
 363
 364 /* journal replay: */
 365
 366 static void replay_now_at(struct journal *j, u64 seq)
 367 {
 368         BUG_ON(seq < j->replay_journal_seq);
 369         BUG_ON(seq > j->replay_journal_seq_end);
 370
 371         while (j->replay_journal_seq < seq)
 372                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 373 }
 374
 375 static int bch2_extent_replay_key(struct bch_fs *c, enum btree_id btree_id,
 376                                   struct bkey_i *k)
 377 {
 378         struct btree_trans trans;
 379         struct btree_iter *iter, *split_iter;
 380         /*
 381          * We might cause compressed extents to be split, so we need to pass in
 382          * a disk_reservation:
 383          */
 384         struct disk_reservation disk_res =
 385                 bch2_disk_reservation_init(c, 0);
 386         struct bkey_i *split;
 387         struct bpos atomic_end;
 388         /*
 389          * Some extents aren't equivalent - w.r.t. what the triggers do
 390          * - if they're split:
 391          */
 392         bool remark_if_split = bch2_bkey_sectors_compressed(bkey_i_to_s_c(k)) ||
 393                 k->k.type == KEY_TYPE_reflink_p;
 394         bool remark = false;
 395         int ret;
 396
 397         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 398 retry:
 399         bch2_trans_begin(&trans);
 400
 401         iter = bch2_trans_get_iter(&trans, btree_id,
 402                                    bkey_start_pos(&k->k),
 403                                    BTREE_ITER_INTENT);
 404
 405         do {
 406                 ret = bch2_btree_iter_traverse(iter);
 407                 if (ret)
 408                         goto err;
 409
 410                 atomic_end = bpos_min(k->k.p, iter->l[0].b->key.k.p);
 411
 412                 split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
 413                 ret = PTR_ERR_OR_ZERO(split);
 414                 if (ret)
 415                         goto err;
 416
 417                 if (!remark &&
 418                     remark_if_split &&
 419                     bkey_cmp(atomic_end, k->k.p) < 0) {
 420                         ret = bch2_disk_reservation_add(c, &disk_res,
 421                                         k->k.size *
 422                                         bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(k)),
 423                                         BCH_DISK_RESERVATION_NOFAIL);
 424                         BUG_ON(ret);
 425
 426                         remark = true;
 427                 }
 428
 429                 bkey_copy(split, k);
 430                 bch2_cut_front(iter->pos, split);
 431                 bch2_cut_back(atomic_end, split);
 432
 433                 split_iter = bch2_trans_copy_iter(&trans, iter);
 434                 ret = PTR_ERR_OR_ZERO(split_iter);
 435                 if (ret)
 436                         goto err;
 437
 438                 /*
 439                  * It's important that we don't go through the
 440                  * extent_handle_overwrites() and extent_update_to_keys() path
 441                  * here: journal replay is supposed to treat extents like
 442                  * regular keys
 443                  */
 444                 __bch2_btree_iter_set_pos(split_iter, split->k.p, false);
 445                 bch2_trans_update(&trans, split_iter, split, !remark
 446                                   ? BTREE_TRIGGER_NORUN
 447                                   : BTREE_TRIGGER_NOOVERWRITES);
 448
 449                 bch2_btree_iter_set_pos(iter, split->k.p);
 450         } while (bkey_cmp(iter->pos, k->k.p) < 0);
 451
 452         if (remark) {
 453                 ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k),
 454                                           0, -((s64) k->k.size),
 455                                           BTREE_TRIGGER_OVERWRITE);
 456                 if (ret)
 457                         goto err;
 458         }
 459
 460         ret = bch2_trans_commit(&trans, &disk_res, NULL,
 461                                 BTREE_INSERT_NOFAIL|
 462                                 BTREE_INSERT_LAZY_RW|
 463                                 BTREE_INSERT_JOURNAL_REPLAY);
 464 err:
 465         if (ret == -EINTR)
 466                 goto retry;
 467
 468         bch2_disk_reservation_put(c, &disk_res);
 469
 470         return bch2_trans_exit(&trans) ?: ret;
 471 }
 472
 473 static int __bch2_journal_replay_key(struct btree_trans *trans,
 474                                      enum btree_id id, unsigned level,
 475                                      struct bkey_i *k)
 476 {
 477         struct btree_iter *iter;
 478         int ret;
 479
 480         iter = bch2_trans_get_node_iter(trans, id, k->k.p,
 481                                         BTREE_MAX_DEPTH, level,
 482                                         BTREE_ITER_INTENT);
 483         if (IS_ERR(iter))
 484                 return PTR_ERR(iter);
 485
 486         /*
 487          * iter->flags & BTREE_ITER_IS_EXTENTS triggers the update path to run
 488          * extent_handle_overwrites() and extent_update_to_keys() - but we don't
 489          * want that here, journal replay is supposed to treat extents like
 490          * regular keys:
 491          */
 492         __bch2_btree_iter_set_pos(iter, k->k.p, false);
 493
 494         ret   = bch2_btree_iter_traverse(iter) ?:
 495                 bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 496         bch2_trans_iter_put(trans, iter);
 497         return ret;
 498 }
 499
 500 static int bch2_journal_replay_key(struct bch_fs *c, enum btree_id id,
 501                                    unsigned level, struct bkey_i *k)
 502 {
 503         return bch2_trans_do(c, NULL, NULL,
 504                              BTREE_INSERT_NOFAIL|
 505                              BTREE_INSERT_LAZY_RW|
 506                              BTREE_INSERT_JOURNAL_REPLAY,
 507                              __bch2_journal_replay_key(&trans, id, level, k));
 508 }
 509
 510 static int __bch2_alloc_replay_key(struct btree_trans *trans, struct bkey_i *k)
 511 {
 512         struct btree_iter *iter;
 513         int ret;
 514
 515         iter = bch2_trans_get_iter(trans, BTREE_ID_ALLOC, k->k.p,
 516                                    BTREE_ITER_CACHED|
 517                                    BTREE_ITER_CACHED_NOFILL|
 518                                    BTREE_ITER_INTENT);
 519         ret =   PTR_ERR_OR_ZERO(iter) ?:
 520                 bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 521         bch2_trans_iter_put(trans, iter);
 522         return ret;
 523 }
 524
 525 static int bch2_alloc_replay_key(struct bch_fs *c, struct bkey_i *k)
 526 {
 527         return bch2_trans_do(c, NULL, NULL,
 528                              BTREE_INSERT_NOFAIL|
 529                              BTREE_INSERT_USE_RESERVE|
 530                              BTREE_INSERT_LAZY_RW|
 531                              BTREE_INSERT_JOURNAL_REPLAY,
 532                         __bch2_alloc_replay_key(&trans, k));
 533 }
 534
 535 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 536 {
 537         const struct journal_key *l = _l;
 538         const struct journal_key *r = _r;
 539
 540         return  cmp_int(r->level,       l->level) ?:
 541                 cmp_int(l->journal_seq, r->journal_seq) ?:
 542                 cmp_int(l->btree_id,    r->btree_id) ?:
 543                 bkey_cmp(l->k->k.p,     r->k->k.p);
 544 }
 545
 546 static int bch2_journal_replay(struct bch_fs *c,
 547                                struct journal_keys keys)
 548 {
 549         struct journal *j = &c->journal;
 550         struct journal_key *i;
 551         u64 seq;
 552         int ret;
 553
 554         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
 555
 556         if (keys.nr)
 557                 replay_now_at(j, keys.journal_seq_base);
 558
 559         seq = j->replay_journal_seq;
 560
 561         /*
 562          * First replay updates to the alloc btree - these will only update the
 563          * btree key cache:
 564          */
 565         for_each_journal_key(keys, i) {
 566                 cond_resched();
 567
 568                 if (!i->level && i->btree_id == BTREE_ID_ALLOC) {
 569                         j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
 570                         ret = bch2_alloc_replay_key(c, i->k);
 571                         if (ret)
 572                                 goto err;
 573                 }
 574         }
 575
 576         /*
 577          * Next replay updates to interior btree nodes:
 578          */
 579         for_each_journal_key(keys, i) {
 580                 cond_resched();
 581
 582                 if (i->level) {
 583                         j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
 584                         ret = bch2_journal_replay_key(c, i->btree_id, i->level, i->k);
 585                         if (ret)
 586                                 goto err;
 587                 }
 588         }
 589
 590         /*
 591          * Now that the btree is in a consistent state, we can start journal
 592          * reclaim (which will be flushing entries from the btree key cache back
 593          * to the btree:
 594          */
 595         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
 596         set_bit(JOURNAL_RECLAIM_STARTED, &j->flags);
 597
 598         j->replay_journal_seq = seq;
 599
 600         /*
 601          * Now replay leaf node updates:
 602          */
 603         for_each_journal_key(keys, i) {
 604                 cond_resched();
 605
 606                 if (i->level || i->btree_id == BTREE_ID_ALLOC)
 607                         continue;
 608
 609                 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 610
 611                 ret = i->k->k.size
 612                         ? bch2_extent_replay_key(c, i->btree_id, i->k)
 613                         : bch2_journal_replay_key(c, i->btree_id, i->level, i->k);
 614                 if (ret)
 615                         goto err;
 616         }
 617
 618         replay_now_at(j, j->replay_journal_seq_end);
 619         j->replay_journal_seq = 0;
 620
 621         bch2_journal_set_replay_done(j);
 622         bch2_journal_flush_all_pins(j);
 623         return bch2_journal_error(j);
 624 err:
 625         bch_err(c, "journal replay: error %d while replaying key", ret);
 626         return ret;
 627 }
 628
 629 static bool journal_empty(struct list_head *journal)
 630 {
 631         return list_empty(journal) ||
 632                 journal_entry_empty(&list_last_entry(journal,
 633                                         struct journal_replay, list)->j);
 634 }
 635
 636 static int
 637 verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c,
 638                                                   struct list_head *journal)
 639 {
 640         struct journal_replay *i =
 641                 list_last_entry(journal, struct journal_replay, list);
 642         u64 start_seq   = le64_to_cpu(i->j.last_seq);
 643         u64 end_seq     = le64_to_cpu(i->j.seq);
 644         u64 seq         = start_seq;
 645         int ret = 0;
 646
 647         list_for_each_entry(i, journal, list) {
 648                 if (le64_to_cpu(i->j.seq) < start_seq)
 649                         continue;
 650
 651                 fsck_err_on(seq != le64_to_cpu(i->j.seq), c,
 652                         "journal entries %llu-%llu missing! (replaying %llu-%llu)",
 653                         seq, le64_to_cpu(i->j.seq) - 1,
 654                         start_seq, end_seq);
 655
 656                 seq = le64_to_cpu(i->j.seq);
 657
 658                 fsck_err_on(bch2_journal_seq_is_blacklisted(c, seq, false), c,
 659                             "found blacklisted journal entry %llu", seq);
 660
 661                 do {
 662                         seq++;
 663                 } while (bch2_journal_seq_is_blacklisted(c, seq, false));
 664         }
 665 fsck_err:
 666         return ret;
 667 }
 668
 669 /* journal replay early: */
 670
 671 static int journal_replay_entry_early(struct bch_fs *c,
 672                                       struct jset_entry *entry)
 673 {
 674         int ret = 0;
 675
 676         switch (entry->type) {
 677         case BCH_JSET_ENTRY_btree_root: {
 678                 struct btree_root *r;
 679
 680                 if (entry->btree_id >= BTREE_ID_NR) {
 681                         bch_err(c, "filesystem has unknown btree type %u",
 682                                 entry->btree_id);
 683                         return -EINVAL;
 684                 }
 685
 686                 r = &c->btree_roots[entry->btree_id];
 687
 688                 if (entry->u64s) {
 689                         r->level = entry->level;
 690                         bkey_copy(&r->key, &entry->start[0]);
 691                         r->error = 0;
 692                 } else {
 693                         r->error = -EIO;
 694                 }
 695                 r->alive = true;
 696                 break;
 697         }
 698         case BCH_JSET_ENTRY_usage: {
 699                 struct jset_entry_usage *u =
 700                         container_of(entry, struct jset_entry_usage, entry);
 701
 702                 switch (entry->btree_id) {
 703                 case FS_USAGE_RESERVED:
 704                         if (entry->level < BCH_REPLICAS_MAX)
 705                                 c->usage_base->persistent_reserved[entry->level] =
 706                                         le64_to_cpu(u->v);
 707                         break;
 708                 case FS_USAGE_INODES:
 709                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 710                         break;
 711                 case FS_USAGE_KEY_VERSION:
 712                         atomic64_set(&c->key_version,
 713                                      le64_to_cpu(u->v));
 714                         break;
 715                 }
 716
 717                 break;
 718         }
 719         case BCH_JSET_ENTRY_data_usage: {
 720                 struct jset_entry_data_usage *u =
 721                         container_of(entry, struct jset_entry_data_usage, entry);
 722                 ret = bch2_replicas_set_usage(c, &u->r,
 723                                               le64_to_cpu(u->v));
 724                 break;
 725         }
 726         case BCH_JSET_ENTRY_blacklist: {
 727                 struct jset_entry_blacklist *bl_entry =
 728                         container_of(entry, struct jset_entry_blacklist, entry);
 729
 730                 ret = bch2_journal_seq_blacklist_add(c,
 731                                 le64_to_cpu(bl_entry->seq),
 732                                 le64_to_cpu(bl_entry->seq) + 1);
 733                 break;
 734         }
 735         case BCH_JSET_ENTRY_blacklist_v2: {
 736                 struct jset_entry_blacklist_v2 *bl_entry =
 737                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 738
 739                 ret = bch2_journal_seq_blacklist_add(c,
 740                                 le64_to_cpu(bl_entry->start),
 741                                 le64_to_cpu(bl_entry->end) + 1);
 742                 break;
 743         }
 744         }
 745
 746         return ret;
 747 }
 748
 749 static int journal_replay_early(struct bch_fs *c,
 750                                 struct bch_sb_field_clean *clean,
 751                                 struct list_head *journal)
 752 {
 753         struct jset_entry *entry;
 754         int ret;
 755
 756         if (clean) {
 757                 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
 758                 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
 759
 760                 for (entry = clean->start;
 761                      entry != vstruct_end(&clean->field);
 762                      entry = vstruct_next(entry)) {
 763                         ret = journal_replay_entry_early(c, entry);
 764                         if (ret)
 765                                 return ret;
 766                 }
 767         } else {
 768                 struct journal_replay *i =
 769                         list_last_entry(journal, struct journal_replay, list);
 770
 771                 c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
 772                 c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
 773
 774                 list_for_each_entry(i, journal, list)
 775                         vstruct_for_each(&i->j, entry) {
 776                                 ret = journal_replay_entry_early(c, entry);
 777                                 if (ret)
 778                                         return ret;
 779                         }
 780         }
 781
 782         bch2_fs_usage_initialize(c);
 783
 784         return 0;
 785 }
 786
 787 /* sb clean section: */
 788
 789 static struct bkey_i *btree_root_find(struct bch_fs *c,
 790                                       struct bch_sb_field_clean *clean,
 791                                       struct jset *j,
 792                                       enum btree_id id, unsigned *level)
 793 {
 794         struct bkey_i *k;
 795         struct jset_entry *entry, *start, *end;
 796
 797         if (clean) {
 798                 start = clean->start;
 799                 end = vstruct_end(&clean->field);
 800         } else {
 801                 start = j->start;
 802                 end = vstruct_last(j);
 803         }
 804
 805         for (entry = start; entry < end; entry = vstruct_next(entry))
 806                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 807                     entry->btree_id == id)
 808                         goto found;
 809
 810         return NULL;
 811 found:
 812         if (!entry->u64s)
 813                 return ERR_PTR(-EINVAL);
 814
 815         k = entry->start;
 816         *level = entry->level;
 817         return k;
 818 }
 819
 820 static int verify_superblock_clean(struct bch_fs *c,
 821                                    struct bch_sb_field_clean **cleanp,
 822                                    struct jset *j)
 823 {
 824         unsigned i;
 825         struct bch_sb_field_clean *clean = *cleanp;
 826         int ret = 0;
 827
 828         if (!c->sb.clean || !j)
 829                 return 0;
 830
 831         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 832                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 833                         le64_to_cpu(clean->journal_seq),
 834                         le64_to_cpu(j->seq))) {
 835                 kfree(clean);
 836                 *cleanp = NULL;
 837                 return 0;
 838         }
 839
 840         mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
 841                         "superblock read clock doesn't match journal after clean shutdown");
 842         mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
 843                         "superblock read clock doesn't match journal after clean shutdown");
 844
 845         for (i = 0; i < BTREE_ID_NR; i++) {
 846                 char buf1[200], buf2[200];
 847                 struct bkey_i *k1, *k2;
 848                 unsigned l1 = 0, l2 = 0;
 849
 850                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 851                 k2 = btree_root_find(c, NULL, j, i, &l2);
 852
 853                 if (!k1 && !k2)
 854                         continue;
 855
 856                 mustfix_fsck_err_on(!k1 || !k2 ||
 857                                     IS_ERR(k1) ||
 858                                     IS_ERR(k2) ||
 859                                     k1->k.u64s != k2->k.u64s ||
 860                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 861                                     l1 != l2, c,
 862                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 863                         "sb:      l=%u %s\n"
 864                         "journal: l=%u %s\n", i,
 865                         l1, (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(k1)), buf1),
 866                         l2, (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(k2)), buf2));
 867         }
 868 fsck_err:
 869         return ret;
 870 }
 871
 872 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 873 {
 874         struct bch_sb_field_clean *clean, *sb_clean;
 875         int ret;
 876
 877         mutex_lock(&c->sb_lock);
 878         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 879
 880         if (fsck_err_on(!sb_clean, c,
 881                         "superblock marked clean but clean section not present")) {
 882                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 883                 c->sb.clean = false;
 884                 mutex_unlock(&c->sb_lock);
 885                 return NULL;
 886         }
 887
 888         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 889                         GFP_KERNEL);
 890         if (!clean) {
 891                 mutex_unlock(&c->sb_lock);
 892                 return ERR_PTR(-ENOMEM);
 893         }
 894
 895         if (le16_to_cpu(c->disk_sb.sb->version) <
 896             bcachefs_metadata_version_bkey_renumber)
 897                 bch2_sb_clean_renumber(clean, READ);
 898
 899         mutex_unlock(&c->sb_lock);
 900
 901         return clean;
 902 fsck_err:
 903         mutex_unlock(&c->sb_lock);
 904         return ERR_PTR(ret);
 905 }
 906
 907 static int read_btree_roots(struct bch_fs *c)
 908 {
 909         unsigned i;
 910         int ret = 0;
 911
 912         for (i = 0; i < BTREE_ID_NR; i++) {
 913                 struct btree_root *r = &c->btree_roots[i];
 914
 915                 if (!r->alive)
 916                         continue;
 917
 918                 if (i == BTREE_ID_ALLOC &&
 919                     c->opts.reconstruct_alloc) {
 920                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 921                         continue;
 922                 }
 923
 924
 925                 if (r->error) {
 926                         __fsck_err(c, i == BTREE_ID_ALLOC
 927                                    ? FSCK_CAN_IGNORE : 0,
 928                                    "invalid btree root %s",
 929                                    bch2_btree_ids[i]);
 930                         if (i == BTREE_ID_ALLOC)
 931                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 932                 }
 933
 934                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 935                 if (ret) {
 936                         __fsck_err(c, i == BTREE_ID_ALLOC
 937                                    ? FSCK_CAN_IGNORE : 0,
 938                                    "error reading btree root %s",
 939                                    bch2_btree_ids[i]);
 940                         if (i == BTREE_ID_ALLOC)
 941                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 942                 }
 943         }
 944
 945         for (i = 0; i < BTREE_ID_NR; i++)
 946                 if (!c->btree_roots[i].b)
 947                         bch2_btree_root_alloc(c, i);
 948 fsck_err:
 949         return ret;
 950 }
 951
 952 int bch2_fs_recovery(struct bch_fs *c)
 953 {
 954         const char *err = "cannot allocate memory";
 955         struct bch_sb_field_clean *clean = NULL;
 956         u64 journal_seq;
 957         bool wrote = false, write_sb = false;
 958         int ret;
 959
 960         if (c->sb.clean)
 961                 clean = read_superblock_clean(c);
 962         ret = PTR_ERR_OR_ZERO(clean);
 963         if (ret)
 964                 goto err;
 965
 966         if (c->sb.clean)
 967                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 968                          le64_to_cpu(clean->journal_seq));
 969
 970         if (!c->replicas.entries) {
 971                 bch_info(c, "building replicas info");
 972                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 973         }
 974
 975         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
 976                 struct jset *j;
 977
 978                 ret = bch2_journal_read(c, &c->journal_entries);
 979                 if (ret)
 980                         goto err;
 981
 982                 if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&c->journal_entries), c,
 983                                 "filesystem marked clean but journal not empty")) {
 984                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 985                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 986                         c->sb.clean = false;
 987                 }
 988
 989                 if (!c->sb.clean && list_empty(&c->journal_entries)) {
 990                         bch_err(c, "no journal entries found");
 991                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
 992                         goto err;
 993                 }
 994
 995                 c->journal_keys = journal_keys_sort(&c->journal_entries);
 996                 if (!c->journal_keys.d) {
 997                         ret = -ENOMEM;
 998                         goto err;
 999                 }
1000
1001                 j = &list_last_entry(&c->journal_entries,
1002                                      struct journal_replay, list)->j;
1003
1004                 ret = verify_superblock_clean(c, &clean, j);
1005                 if (ret)
1006                         goto err;
1007
1008                 journal_seq = le64_to_cpu(j->seq) + 1;
1009         } else {
1010                 journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1011         }
1012
1013         if (!c->sb.clean &&
1014             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1015                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1016                 ret = -EINVAL;
1017                 goto err;
1018         }
1019
1020         ret = journal_replay_early(c, clean, &c->journal_entries);
1021         if (ret)
1022                 goto err;
1023
1024         if (!c->sb.clean) {
1025                 ret = bch2_journal_seq_blacklist_add(c,
1026                                                      journal_seq,
1027                                                      journal_seq + 4);
1028                 if (ret) {
1029                         bch_err(c, "error creating new journal seq blacklist entry");
1030                         goto err;
1031                 }
1032
1033                 journal_seq += 4;
1034         }
1035
1036         ret = bch2_blacklist_table_initialize(c);
1037
1038         if (!list_empty(&c->journal_entries)) {
1039                 ret = verify_journal_entries_not_blacklisted_or_missing(c,
1040                                                         &c->journal_entries);
1041                 if (ret)
1042                         goto err;
1043         }
1044
1045         ret = bch2_fs_journal_start(&c->journal, journal_seq,
1046                                     &c->journal_entries);
1047         if (ret)
1048                 goto err;
1049
1050         ret = read_btree_roots(c);
1051         if (ret)
1052                 goto err;
1053
1054         bch_verbose(c, "starting alloc read");
1055         err = "error reading allocation information";
1056         ret = bch2_alloc_read(c, &c->journal_keys);
1057         if (ret)
1058                 goto err;
1059         bch_verbose(c, "alloc read done");
1060
1061         bch_verbose(c, "starting stripes_read");
1062         err = "error reading stripes";
1063         ret = bch2_stripes_read(c, &c->journal_keys);
1064         if (ret)
1065                 goto err;
1066         bch_verbose(c, "stripes_read done");
1067
1068         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1069
1070         if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
1071             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
1072                 /*
1073                  * interior btree node updates aren't consistent with the
1074                  * journal; after an unclean shutdown we have to walk all
1075                  * pointers to metadata:
1076                  */
1077                 bch_info(c, "starting metadata mark and sweep");
1078                 err = "error in mark and sweep";
1079                 ret = bch2_gc(c, &c->journal_keys, true, true);
1080                 if (ret)
1081                         goto err;
1082                 bch_verbose(c, "mark and sweep done");
1083         }
1084
1085         if (c->opts.fsck ||
1086             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
1087             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1088                 bch_info(c, "starting mark and sweep");
1089                 err = "error in mark and sweep";
1090                 ret = bch2_gc(c, &c->journal_keys, true, false);
1091                 if (ret)
1092                         goto err;
1093                 bch_verbose(c, "mark and sweep done");
1094         }
1095
1096         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1097         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1098
1099         /*
1100          * Skip past versions that might have possibly been used (as nonces),
1101          * but hadn't had their pointers written:
1102          */
1103         if (c->sb.encryption_type && !c->sb.clean)
1104                 atomic64_add(1 << 16, &c->key_version);
1105
1106         if (c->opts.norecovery)
1107                 goto out;
1108
1109         bch_verbose(c, "starting journal replay");
1110         err = "journal replay failed";
1111         ret = bch2_journal_replay(c, c->journal_keys);
1112         if (ret)
1113                 goto err;
1114         bch_verbose(c, "journal replay done");
1115
1116         if (!c->opts.nochanges) {
1117                 /*
1118                  * note that even when filesystem was clean there might be work
1119                  * to do here, if we ran gc (because of fsck) which recalculated
1120                  * oldest_gen:
1121                  */
1122                 bch_verbose(c, "writing allocation info");
1123                 err = "error writing out alloc info";
1124                 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW, &wrote) ?:
1125                         bch2_alloc_write(c, BTREE_INSERT_LAZY_RW, &wrote);
1126                 if (ret) {
1127                         bch_err(c, "error writing alloc info");
1128                         goto err;
1129                 }
1130                 bch_verbose(c, "alloc write done");
1131
1132                 set_bit(BCH_FS_ALLOC_WRITTEN, &c->flags);
1133         }
1134
1135         if (!c->sb.clean) {
1136                 if (!(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
1137                         bch_info(c, "checking inode link counts");
1138                         err = "error in recovery";
1139                         ret = bch2_fsck_inode_nlink(c);
1140                         if (ret)
1141                                 goto err;
1142                         bch_verbose(c, "check inodes done");
1143
1144                 } else {
1145                         bch_verbose(c, "checking for deleted inodes");
1146                         err = "error in recovery";
1147                         ret = bch2_fsck_walk_inodes_only(c);
1148                         if (ret)
1149                                 goto err;
1150                         bch_verbose(c, "check inodes done");
1151                 }
1152         }
1153
1154         if (c->opts.fsck) {
1155                 bch_info(c, "starting fsck");
1156                 err = "error in fsck";
1157                 ret = bch2_fsck_full(c);
1158                 if (ret)
1159                         goto err;
1160                 bch_verbose(c, "fsck done");
1161         }
1162
1163         if (enabled_qtypes(c)) {
1164                 bch_verbose(c, "reading quotas");
1165                 ret = bch2_fs_quota_read(c);
1166                 if (ret)
1167                         goto err;
1168                 bch_verbose(c, "quotas done");
1169         }
1170
1171         mutex_lock(&c->sb_lock);
1172         if (c->opts.version_upgrade) {
1173                 if (c->sb.version < bcachefs_metadata_version_new_versioning)
1174                         c->disk_sb.sb->version_min =
1175                                 le16_to_cpu(bcachefs_metadata_version_min);
1176                 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
1177                 c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1178                 write_sb = true;
1179         }
1180
1181         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1182                 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
1183                 write_sb = true;
1184         }
1185
1186         if (c->opts.fsck &&
1187             !test_bit(BCH_FS_ERROR, &c->flags)) {
1188                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1189                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1190                 write_sb = true;
1191         }
1192
1193         if (write_sb)
1194                 bch2_write_super(c);
1195         mutex_unlock(&c->sb_lock);
1196
1197         if (c->journal_seq_blacklist_table &&
1198             c->journal_seq_blacklist_table->nr > 128)
1199                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1200 out:
1201         ret = 0;
1202 err:
1203 fsck_err:
1204         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1205         bch2_flush_fsck_errs(c);
1206
1207         if (!c->opts.keep_journal) {
1208                 bch2_journal_keys_free(&c->journal_keys);
1209                 bch2_journal_entries_free(&c->journal_entries);
1210         }
1211         kfree(clean);
1212         if (ret)
1213                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1214         else
1215                 bch_verbose(c, "ret %i", ret);
1216         return ret;
1217 }
1218
1219 int bch2_fs_initialize(struct bch_fs *c)
1220 {
1221         struct bch_inode_unpacked root_inode, lostfound_inode;
1222         struct bkey_inode_buf packed_inode;
1223         struct qstr lostfound = QSTR("lost+found");
1224         const char *err = "cannot allocate memory";
1225         struct bch_dev *ca;
1226         LIST_HEAD(journal);
1227         unsigned i;
1228         int ret;
1229
1230         bch_notice(c, "initializing new filesystem");
1231
1232         mutex_lock(&c->sb_lock);
1233         for_each_online_member(ca, c, i)
1234                 bch2_mark_dev_superblock(c, ca, 0);
1235         mutex_unlock(&c->sb_lock);
1236
1237         mutex_lock(&c->sb_lock);
1238         c->disk_sb.sb->version = c->disk_sb.sb->version_min =
1239                 le16_to_cpu(bcachefs_metadata_version_current);
1240         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1241         c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1242
1243         bch2_write_super(c);
1244         mutex_unlock(&c->sb_lock);
1245
1246         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1247         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1248
1249         for (i = 0; i < BTREE_ID_NR; i++)
1250                 bch2_btree_root_alloc(c, i);
1251
1252         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
1253         set_bit(JOURNAL_RECLAIM_STARTED, &c->journal.flags);
1254
1255         err = "unable to allocate journal buckets";
1256         for_each_online_member(ca, c, i) {
1257                 ret = bch2_dev_journal_alloc(ca);
1258                 if (ret) {
1259                         percpu_ref_put(&ca->io_ref);
1260                         goto err;
1261                 }
1262         }
1263
1264         /*
1265          * journal_res_get() will crash if called before this has
1266          * set up the journal.pin FIFO and journal.cur pointer:
1267          */
1268         bch2_fs_journal_start(&c->journal, 1, &journal);
1269         bch2_journal_set_replay_done(&c->journal);
1270
1271         bch2_inode_init(c, &root_inode, 0, 0,
1272                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1273         root_inode.bi_inum = BCACHEFS_ROOT_INO;
1274         bch2_inode_pack(&packed_inode, &root_inode);
1275
1276         err = "error creating root directory";
1277         ret = bch2_btree_insert(c, BTREE_ID_INODES,
1278                                 &packed_inode.inode.k_i,
1279                                 NULL, NULL, BTREE_INSERT_LAZY_RW);
1280         if (ret)
1281                 goto err;
1282
1283         bch2_inode_init_early(c, &lostfound_inode);
1284
1285         err = "error creating lost+found";
1286         ret = bch2_trans_do(c, NULL, NULL, 0,
1287                 bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
1288                                   &root_inode, &lostfound_inode,
1289                                   &lostfound,
1290                                   0, 0, S_IFDIR|0700, 0,
1291                                   NULL, NULL));
1292         if (ret)
1293                 goto err;
1294
1295         if (enabled_qtypes(c)) {
1296                 ret = bch2_fs_quota_read(c);
1297                 if (ret)
1298                         goto err;
1299         }
1300
1301         err = "error writing first journal entry";
1302         ret = bch2_journal_meta(&c->journal);
1303         if (ret)
1304                 goto err;
1305
1306         mutex_lock(&c->sb_lock);
1307         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1308         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1309
1310         bch2_write_super(c);
1311         mutex_unlock(&c->sb_lock);
1312
1313         return 0;
1314 err:
1315         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1316         return ret;
1317 }