git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/recovery.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "bkey_buf.h"
   5 #include "alloc_background.h"
   6 #include "btree_gc.h"
   7 #include "btree_update.h"
   8 #include "btree_update_interior.h"
   9 #include "btree_io.h"
  10 #include "buckets.h"
  11 #include "dirent.h"
  12 #include "ec.h"
  13 #include "error.h"
  14 #include "fs-common.h"
  15 #include "fsck.h"
  16 #include "journal_io.h"
  17 #include "journal_reclaim.h"
  18 #include "journal_seq_blacklist.h"
  19 #include "quota.h"
  20 #include "recovery.h"
  21 #include "replicas.h"
  22 #include "super-io.h"
  23
  24 #include <linux/sort.h>
  25 #include <linux/stat.h>
  26
  27 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  28
  29 /* for -o reconstruct_alloc: */
  30 static void drop_alloc_keys(struct journal_keys *keys)
  31 {
  32         size_t src, dst;
  33
  34         for (src = 0, dst = 0; src < keys->nr; src++)
  35                 if (keys->d[src].btree_id != BTREE_ID_ALLOC)
  36                         keys->d[dst++] = keys->d[src];
  37
  38         keys->nr = dst;
  39 }
  40
  41 /* iterate over keys read from the journal: */
  42
  43 static struct journal_key *journal_key_search(struct journal_keys *journal_keys,
  44                                               enum btree_id id, unsigned level,
  45                                               struct bpos pos)
  46 {
  47         size_t l = 0, r = journal_keys->nr, m;
  48
  49         while (l < r) {
  50                 m = l + ((r - l) >> 1);
  51                 if ((cmp_int(id,        journal_keys->d[m].btree_id) ?:
  52                      cmp_int(level,     journal_keys->d[m].level) ?:
  53                      bkey_cmp(pos,      journal_keys->d[m].k->k.p)) > 0)
  54                         l = m + 1;
  55                 else
  56                         r = m;
  57         }
  58
  59         BUG_ON(l < journal_keys->nr &&
  60                (cmp_int(id,     journal_keys->d[l].btree_id) ?:
  61                 cmp_int(level,  journal_keys->d[l].level) ?:
  62                 bkey_cmp(pos,   journal_keys->d[l].k->k.p)) > 0);
  63
  64         BUG_ON(l &&
  65                (cmp_int(id,     journal_keys->d[l - 1].btree_id) ?:
  66                 cmp_int(level,  journal_keys->d[l - 1].level) ?:
  67                 bkey_cmp(pos,   journal_keys->d[l - 1].k->k.p)) <= 0);
  68
  69         return l < journal_keys->nr ? journal_keys->d + l : NULL;
  70 }
  71
  72 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
  73 {
  74         if (iter->k &&
  75             iter->k < iter->keys->d + iter->keys->nr &&
  76             iter->k->btree_id   == iter->btree_id &&
  77             iter->k->level      == iter->level)
  78                 return iter->k->k;
  79
  80         iter->k = NULL;
  81         return NULL;
  82 }
  83
  84 static void bch2_journal_iter_advance(struct journal_iter *iter)
  85 {
  86         if (iter->k)
  87                 iter->k++;
  88 }
  89
  90 static void bch2_journal_iter_init(struct journal_iter *iter,
  91                                    struct journal_keys *journal_keys,
  92                                    enum btree_id id, unsigned level,
  93                                    struct bpos pos)
  94 {
  95         iter->btree_id  = id;
  96         iter->level     = level;
  97         iter->keys      = journal_keys;
  98         iter->k         = journal_key_search(journal_keys, id, level, pos);
  99 }
 100
 101 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 102 {
 103         return iter->btree
 104                 ? bch2_btree_iter_peek(iter->btree)
 105                 : bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 106                                                    iter->b, &iter->unpacked);
 107 }
 108
 109 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 110 {
 111         if (iter->btree)
 112                 bch2_btree_iter_next(iter->btree);
 113         else
 114                 bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 115 }
 116
 117 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 118 {
 119         switch (iter->last) {
 120         case none:
 121                 break;
 122         case btree:
 123                 bch2_journal_iter_advance_btree(iter);
 124                 break;
 125         case journal:
 126                 bch2_journal_iter_advance(&iter->journal);
 127                 break;
 128         }
 129
 130         iter->last = none;
 131 }
 132
 133 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 134 {
 135         struct bkey_s_c ret;
 136
 137         while (1) {
 138                 struct bkey_s_c btree_k         =
 139                         bch2_journal_iter_peek_btree(iter);
 140                 struct bkey_s_c journal_k       =
 141                         bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
 142
 143                 if (btree_k.k && journal_k.k) {
 144                         int cmp = bkey_cmp(btree_k.k->p, journal_k.k->p);
 145
 146                         if (!cmp)
 147                                 bch2_journal_iter_advance_btree(iter);
 148
 149                         iter->last = cmp < 0 ? btree : journal;
 150                 } else if (btree_k.k) {
 151                         iter->last = btree;
 152                 } else if (journal_k.k) {
 153                         iter->last = journal;
 154                 } else {
 155                         iter->last = none;
 156                         return bkey_s_c_null;
 157                 }
 158
 159                 ret = iter->last == journal ? journal_k : btree_k;
 160
 161                 if (iter->b &&
 162                     bkey_cmp(ret.k->p, iter->b->data->max_key) > 0) {
 163                         iter->journal.k = NULL;
 164                         iter->last = none;
 165                         return bkey_s_c_null;
 166                 }
 167
 168                 if (!bkey_deleted(ret.k))
 169                         break;
 170
 171                 bch2_btree_and_journal_iter_advance(iter);
 172         }
 173
 174         return ret;
 175 }
 176
 177 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 178 {
 179         bch2_btree_and_journal_iter_advance(iter);
 180
 181         return bch2_btree_and_journal_iter_peek(iter);
 182 }
 183
 184 void bch2_btree_and_journal_iter_init(struct btree_and_journal_iter *iter,
 185                                       struct btree_trans *trans,
 186                                       struct journal_keys *journal_keys,
 187                                       enum btree_id id, struct bpos pos)
 188 {
 189         memset(iter, 0, sizeof(*iter));
 190
 191         iter->btree = bch2_trans_get_iter(trans, id, pos, BTREE_ITER_PREFETCH);
 192         bch2_journal_iter_init(&iter->journal, journal_keys, id, 0, pos);
 193 }
 194
 195 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 196                                                 struct journal_keys *journal_keys,
 197                                                 struct btree *b)
 198 {
 199         memset(iter, 0, sizeof(*iter));
 200
 201         iter->b = b;
 202         bch2_btree_node_iter_init_from_start(&iter->node_iter, iter->b);
 203         bch2_journal_iter_init(&iter->journal, journal_keys,
 204                                b->c.btree_id, b->c.level, b->data->min_key);
 205 }
 206
 207 /* Walk btree, overlaying keys from the journal: */
 208
 209 static int bch2_btree_and_journal_walk_recurse(struct bch_fs *c, struct btree *b,
 210                                 struct journal_keys *journal_keys,
 211                                 enum btree_id btree_id,
 212                                 btree_walk_node_fn node_fn,
 213                                 btree_walk_key_fn key_fn)
 214 {
 215         struct btree_and_journal_iter iter;
 216         struct bkey_s_c k;
 217         int ret = 0;
 218
 219         bch2_btree_and_journal_iter_init_node_iter(&iter, journal_keys, b);
 220
 221         while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 222                 ret = key_fn(c, btree_id, b->c.level, k);
 223                 if (ret)
 224                         break;
 225
 226                 if (b->c.level) {
 227                         struct btree *child;
 228                         struct bkey_buf tmp;
 229
 230                         bch2_bkey_buf_init(&tmp);
 231                         bch2_bkey_buf_reassemble(&tmp, c, k);
 232                         k = bkey_i_to_s_c(tmp.k);
 233
 234                         bch2_btree_and_journal_iter_advance(&iter);
 235
 236                         child = bch2_btree_node_get_noiter(c, tmp.k,
 237                                                 b->c.btree_id, b->c.level - 1);
 238                         bch2_bkey_buf_exit(&tmp, c);
 239
 240                         ret = PTR_ERR_OR_ZERO(child);
 241                         if (ret)
 242                                 break;
 243
 244                         ret   = (node_fn ? node_fn(c, b) : 0) ?:
 245                                 bch2_btree_and_journal_walk_recurse(c, child,
 246                                         journal_keys, btree_id, node_fn, key_fn);
 247                         six_unlock_read(&child->c.lock);
 248
 249                         if (ret)
 250                                 break;
 251                 } else {
 252                         bch2_btree_and_journal_iter_advance(&iter);
 253                 }
 254         }
 255
 256         return ret;
 257 }
 258
 259 int bch2_btree_and_journal_walk(struct bch_fs *c, struct journal_keys *journal_keys,
 260                                 enum btree_id btree_id,
 261                                 btree_walk_node_fn node_fn,
 262                                 btree_walk_key_fn key_fn)
 263 {
 264         struct btree *b = c->btree_roots[btree_id].b;
 265         int ret = 0;
 266
 267         if (btree_node_fake(b))
 268                 return 0;
 269
 270         six_lock_read(&b->c.lock, NULL, NULL);
 271         ret   = (node_fn ? node_fn(c, b) : 0) ?:
 272                 bch2_btree_and_journal_walk_recurse(c, b, journal_keys, btree_id,
 273                                                     node_fn, key_fn) ?:
 274                 key_fn(c, btree_id, b->c.level + 1, bkey_i_to_s_c(&b->key));
 275         six_unlock_read(&b->c.lock);
 276
 277         return ret;
 278 }
 279
 280 /* sort and dedup all keys in the journal: */
 281
 282 void bch2_journal_entries_free(struct list_head *list)
 283 {
 284
 285         while (!list_empty(list)) {
 286                 struct journal_replay *i =
 287                         list_first_entry(list, struct journal_replay, list);
 288                 list_del(&i->list);
 289                 kvpfree(i, offsetof(struct journal_replay, j) +
 290                         vstruct_bytes(&i->j));
 291         }
 292 }
 293
 294 /*
 295  * When keys compare equal, oldest compares first:
 296  */
 297 static int journal_sort_key_cmp(const void *_l, const void *_r)
 298 {
 299         const struct journal_key *l = _l;
 300         const struct journal_key *r = _r;
 301
 302         return  cmp_int(l->btree_id,    r->btree_id) ?:
 303                 cmp_int(l->level,       r->level) ?:
 304                 bkey_cmp(l->k->k.p, r->k->k.p) ?:
 305                 cmp_int(l->journal_seq, r->journal_seq) ?:
 306                 cmp_int(l->journal_offset, r->journal_offset);
 307 }
 308
 309 void bch2_journal_keys_free(struct journal_keys *keys)
 310 {
 311         kvfree(keys->d);
 312         keys->d = NULL;
 313         keys->nr = 0;
 314 }
 315
 316 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 317 {
 318         struct journal_replay *i;
 319         struct jset_entry *entry;
 320         struct bkey_i *k, *_n;
 321         struct journal_keys keys = { NULL };
 322         struct journal_key *src, *dst;
 323         size_t nr_keys = 0;
 324
 325         if (list_empty(journal_entries))
 326                 return keys;
 327
 328         list_for_each_entry(i, journal_entries, list) {
 329                 if (i->ignore)
 330                         continue;
 331
 332                 if (!keys.journal_seq_base)
 333                         keys.journal_seq_base = le64_to_cpu(i->j.seq);
 334
 335                 for_each_jset_key(k, _n, entry, &i->j)
 336                         nr_keys++;
 337         }
 338
 339         keys.d = kvmalloc(sizeof(keys.d[0]) * nr_keys, GFP_KERNEL);
 340         if (!keys.d)
 341                 goto err;
 342
 343         list_for_each_entry(i, journal_entries, list) {
 344                 if (i->ignore)
 345                         continue;
 346
 347                 BUG_ON(le64_to_cpu(i->j.seq) - keys.journal_seq_base > U32_MAX);
 348
 349                 for_each_jset_key(k, _n, entry, &i->j)
 350                         keys.d[keys.nr++] = (struct journal_key) {
 351                                 .btree_id       = entry->btree_id,
 352                                 .level          = entry->level,
 353                                 .k              = k,
 354                                 .journal_seq    = le64_to_cpu(i->j.seq) -
 355                                         keys.journal_seq_base,
 356                                 .journal_offset = k->_data - i->j._data,
 357                         };
 358         }
 359
 360         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 361
 362         src = dst = keys.d;
 363         while (src < keys.d + keys.nr) {
 364                 while (src + 1 < keys.d + keys.nr &&
 365                        src[0].btree_id  == src[1].btree_id &&
 366                        src[0].level     == src[1].level &&
 367                        !bkey_cmp(src[0].k->k.p, src[1].k->k.p))
 368                         src++;
 369
 370                 *dst++ = *src++;
 371         }
 372
 373         keys.nr = dst - keys.d;
 374 err:
 375         return keys;
 376 }
 377
 378 /* journal replay: */
 379
 380 static void replay_now_at(struct journal *j, u64 seq)
 381 {
 382         BUG_ON(seq < j->replay_journal_seq);
 383         BUG_ON(seq > j->replay_journal_seq_end);
 384
 385         while (j->replay_journal_seq < seq)
 386                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 387 }
 388
 389 static int bch2_extent_replay_key(struct bch_fs *c, enum btree_id btree_id,
 390                                   struct bkey_i *k)
 391 {
 392         struct btree_trans trans;
 393         struct btree_iter *iter, *split_iter;
 394         /*
 395          * We might cause compressed extents to be split, so we need to pass in
 396          * a disk_reservation:
 397          */
 398         struct disk_reservation disk_res =
 399                 bch2_disk_reservation_init(c, 0);
 400         struct bkey_i *split;
 401         struct bpos atomic_end;
 402         /*
 403          * Some extents aren't equivalent - w.r.t. what the triggers do
 404          * - if they're split:
 405          */
 406         bool remark_if_split = bch2_bkey_sectors_compressed(bkey_i_to_s_c(k)) ||
 407                 k->k.type == KEY_TYPE_reflink_p;
 408         bool remark = false;
 409         int ret;
 410
 411         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 412 retry:
 413         bch2_trans_begin(&trans);
 414
 415         iter = bch2_trans_get_iter(&trans, btree_id,
 416                                    bkey_start_pos(&k->k),
 417                                    BTREE_ITER_INTENT);
 418
 419         do {
 420                 ret = bch2_btree_iter_traverse(iter);
 421                 if (ret)
 422                         goto err;
 423
 424                 atomic_end = bpos_min(k->k.p, iter->l[0].b->key.k.p);
 425
 426                 split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
 427                 ret = PTR_ERR_OR_ZERO(split);
 428                 if (ret)
 429                         goto err;
 430
 431                 if (!remark &&
 432                     remark_if_split &&
 433                     bkey_cmp(atomic_end, k->k.p) < 0) {
 434                         ret = bch2_disk_reservation_add(c, &disk_res,
 435                                         k->k.size *
 436                                         bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(k)),
 437                                         BCH_DISK_RESERVATION_NOFAIL);
 438                         BUG_ON(ret);
 439
 440                         remark = true;
 441                 }
 442
 443                 bkey_copy(split, k);
 444                 bch2_cut_front(iter->pos, split);
 445                 bch2_cut_back(atomic_end, split);
 446
 447                 split_iter = bch2_trans_copy_iter(&trans, iter);
 448
 449                 /*
 450                  * It's important that we don't go through the
 451                  * extent_handle_overwrites() and extent_update_to_keys() path
 452                  * here: journal replay is supposed to treat extents like
 453                  * regular keys
 454                  */
 455                 __bch2_btree_iter_set_pos(split_iter, split->k.p, false);
 456                 bch2_trans_update(&trans, split_iter, split,
 457                                   BTREE_TRIGGER_NORUN);
 458                 bch2_trans_iter_put(&trans, split_iter);
 459
 460                 bch2_btree_iter_set_pos(iter, split->k.p);
 461
 462                 if (remark) {
 463                         ret = bch2_trans_mark_key(&trans,
 464                                                   bkey_s_c_null,
 465                                                   bkey_i_to_s_c(split),
 466                                                   0, split->k.size,
 467                                                   BTREE_TRIGGER_INSERT);
 468                         if (ret)
 469                                 goto err;
 470                 }
 471         } while (bkey_cmp(iter->pos, k->k.p) < 0);
 472
 473         if (remark) {
 474                 ret = bch2_trans_mark_key(&trans,
 475                                           bkey_i_to_s_c(k),
 476                                           bkey_s_c_null,
 477                                           0, -((s64) k->k.size),
 478                                           BTREE_TRIGGER_OVERWRITE);
 479                 if (ret)
 480                         goto err;
 481         }
 482
 483         ret = bch2_trans_commit(&trans, &disk_res, NULL,
 484                                 BTREE_INSERT_NOFAIL|
 485                                 BTREE_INSERT_LAZY_RW|
 486                                 BTREE_INSERT_JOURNAL_REPLAY);
 487 err:
 488         bch2_trans_iter_put(&trans, iter);
 489
 490         if (ret == -EINTR)
 491                 goto retry;
 492
 493         bch2_disk_reservation_put(c, &disk_res);
 494
 495         return bch2_trans_exit(&trans) ?: ret;
 496 }
 497
 498 static int __bch2_journal_replay_key(struct btree_trans *trans,
 499                                      enum btree_id id, unsigned level,
 500                                      struct bkey_i *k)
 501 {
 502         struct btree_iter *iter;
 503         int ret;
 504
 505         iter = bch2_trans_get_node_iter(trans, id, k->k.p,
 506                                         BTREE_MAX_DEPTH, level,
 507                                         BTREE_ITER_INTENT);
 508
 509         /*
 510          * iter->flags & BTREE_ITER_IS_EXTENTS triggers the update path to run
 511          * extent_handle_overwrites() and extent_update_to_keys() - but we don't
 512          * want that here, journal replay is supposed to treat extents like
 513          * regular keys:
 514          */
 515         __bch2_btree_iter_set_pos(iter, k->k.p, false);
 516
 517         ret   = bch2_btree_iter_traverse(iter) ?:
 518                 bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 519         bch2_trans_iter_put(trans, iter);
 520         return ret;
 521 }
 522
 523 static int bch2_journal_replay_key(struct bch_fs *c, enum btree_id id,
 524                                    unsigned level, struct bkey_i *k)
 525 {
 526         return bch2_trans_do(c, NULL, NULL,
 527                              BTREE_INSERT_NOFAIL|
 528                              BTREE_INSERT_LAZY_RW|
 529                              BTREE_INSERT_JOURNAL_REPLAY,
 530                              __bch2_journal_replay_key(&trans, id, level, k));
 531 }
 532
 533 static int __bch2_alloc_replay_key(struct btree_trans *trans, struct bkey_i *k)
 534 {
 535         struct btree_iter *iter;
 536         int ret;
 537
 538         iter = bch2_trans_get_iter(trans, BTREE_ID_ALLOC, k->k.p,
 539                                    BTREE_ITER_CACHED|
 540                                    BTREE_ITER_CACHED_NOFILL|
 541                                    BTREE_ITER_INTENT);
 542         ret = bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 543         bch2_trans_iter_put(trans, iter);
 544         return ret;
 545 }
 546
 547 static int bch2_alloc_replay_key(struct bch_fs *c, struct bkey_i *k)
 548 {
 549         return bch2_trans_do(c, NULL, NULL,
 550                              BTREE_INSERT_NOFAIL|
 551                              BTREE_INSERT_USE_RESERVE|
 552                              BTREE_INSERT_LAZY_RW|
 553                              BTREE_INSERT_JOURNAL_REPLAY,
 554                         __bch2_alloc_replay_key(&trans, k));
 555 }
 556
 557 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 558 {
 559         const struct journal_key *l = _l;
 560         const struct journal_key *r = _r;
 561
 562         return  cmp_int(r->level,       l->level) ?:
 563                 cmp_int(l->journal_seq, r->journal_seq) ?:
 564                 cmp_int(l->btree_id,    r->btree_id) ?:
 565                 bkey_cmp(l->k->k.p,     r->k->k.p);
 566 }
 567
 568 static int bch2_journal_replay(struct bch_fs *c,
 569                                struct journal_keys keys)
 570 {
 571         struct journal *j = &c->journal;
 572         struct journal_key *i;
 573         u64 seq;
 574         int ret;
 575
 576         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
 577
 578         if (keys.nr)
 579                 replay_now_at(j, keys.journal_seq_base);
 580
 581         seq = j->replay_journal_seq;
 582
 583         /*
 584          * First replay updates to the alloc btree - these will only update the
 585          * btree key cache:
 586          */
 587         for_each_journal_key(keys, i) {
 588                 cond_resched();
 589
 590                 if (!i->level && i->btree_id == BTREE_ID_ALLOC) {
 591                         j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
 592                         ret = bch2_alloc_replay_key(c, i->k);
 593                         if (ret)
 594                                 goto err;
 595                 }
 596         }
 597
 598         /*
 599          * Next replay updates to interior btree nodes:
 600          */
 601         for_each_journal_key(keys, i) {
 602                 cond_resched();
 603
 604                 if (i->level) {
 605                         j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
 606                         ret = bch2_journal_replay_key(c, i->btree_id, i->level, i->k);
 607                         if (ret)
 608                                 goto err;
 609                 }
 610         }
 611
 612         /*
 613          * Now that the btree is in a consistent state, we can start journal
 614          * reclaim (which will be flushing entries from the btree key cache back
 615          * to the btree:
 616          */
 617         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
 618         set_bit(JOURNAL_RECLAIM_STARTED, &j->flags);
 619         journal_reclaim_kick(j);
 620
 621         j->replay_journal_seq = seq;
 622
 623         /*
 624          * Now replay leaf node updates:
 625          */
 626         for_each_journal_key(keys, i) {
 627                 cond_resched();
 628
 629                 if (i->level || i->btree_id == BTREE_ID_ALLOC)
 630                         continue;
 631
 632                 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 633
 634                 ret = i->k->k.size
 635                         ? bch2_extent_replay_key(c, i->btree_id, i->k)
 636                         : bch2_journal_replay_key(c, i->btree_id, i->level, i->k);
 637                 if (ret)
 638                         goto err;
 639         }
 640
 641         replay_now_at(j, j->replay_journal_seq_end);
 642         j->replay_journal_seq = 0;
 643
 644         bch2_journal_set_replay_done(j);
 645         bch2_journal_flush_all_pins(j);
 646         return bch2_journal_error(j);
 647 err:
 648         bch_err(c, "journal replay: error %d while replaying key", ret);
 649         return ret;
 650 }
 651
 652 /* journal replay early: */
 653
 654 static int journal_replay_entry_early(struct bch_fs *c,
 655                                       struct jset_entry *entry)
 656 {
 657         int ret = 0;
 658
 659         switch (entry->type) {
 660         case BCH_JSET_ENTRY_btree_root: {
 661                 struct btree_root *r;
 662
 663                 if (entry->btree_id >= BTREE_ID_NR) {
 664                         bch_err(c, "filesystem has unknown btree type %u",
 665                                 entry->btree_id);
 666                         return -EINVAL;
 667                 }
 668
 669                 r = &c->btree_roots[entry->btree_id];
 670
 671                 if (entry->u64s) {
 672                         r->level = entry->level;
 673                         bkey_copy(&r->key, &entry->start[0]);
 674                         r->error = 0;
 675                 } else {
 676                         r->error = -EIO;
 677                 }
 678                 r->alive = true;
 679                 break;
 680         }
 681         case BCH_JSET_ENTRY_usage: {
 682                 struct jset_entry_usage *u =
 683                         container_of(entry, struct jset_entry_usage, entry);
 684
 685                 switch (entry->btree_id) {
 686                 case FS_USAGE_RESERVED:
 687                         if (entry->level < BCH_REPLICAS_MAX)
 688                                 c->usage_base->persistent_reserved[entry->level] =
 689                                         le64_to_cpu(u->v);
 690                         break;
 691                 case FS_USAGE_INODES:
 692                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 693                         break;
 694                 case FS_USAGE_KEY_VERSION:
 695                         atomic64_set(&c->key_version,
 696                                      le64_to_cpu(u->v));
 697                         break;
 698                 }
 699
 700                 break;
 701         }
 702         case BCH_JSET_ENTRY_data_usage: {
 703                 struct jset_entry_data_usage *u =
 704                         container_of(entry, struct jset_entry_data_usage, entry);
 705                 ret = bch2_replicas_set_usage(c, &u->r,
 706                                               le64_to_cpu(u->v));
 707                 break;
 708         }
 709         case BCH_JSET_ENTRY_blacklist: {
 710                 struct jset_entry_blacklist *bl_entry =
 711                         container_of(entry, struct jset_entry_blacklist, entry);
 712
 713                 ret = bch2_journal_seq_blacklist_add(c,
 714                                 le64_to_cpu(bl_entry->seq),
 715                                 le64_to_cpu(bl_entry->seq) + 1);
 716                 break;
 717         }
 718         case BCH_JSET_ENTRY_blacklist_v2: {
 719                 struct jset_entry_blacklist_v2 *bl_entry =
 720                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 721
 722                 ret = bch2_journal_seq_blacklist_add(c,
 723                                 le64_to_cpu(bl_entry->start),
 724                                 le64_to_cpu(bl_entry->end) + 1);
 725                 break;
 726         }
 727         }
 728
 729         return ret;
 730 }
 731
 732 static int journal_replay_early(struct bch_fs *c,
 733                                 struct bch_sb_field_clean *clean,
 734                                 struct list_head *journal)
 735 {
 736         struct journal_replay *i;
 737         struct jset_entry *entry;
 738         int ret;
 739
 740         if (clean) {
 741                 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
 742                 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
 743
 744                 for (entry = clean->start;
 745                      entry != vstruct_end(&clean->field);
 746                      entry = vstruct_next(entry)) {
 747                         ret = journal_replay_entry_early(c, entry);
 748                         if (ret)
 749                                 return ret;
 750                 }
 751         } else {
 752                 list_for_each_entry(i, journal, list) {
 753                         if (i->ignore)
 754                                 continue;
 755
 756                         c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
 757                         c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
 758
 759                         vstruct_for_each(&i->j, entry) {
 760                                 ret = journal_replay_entry_early(c, entry);
 761                                 if (ret)
 762                                         return ret;
 763                         }
 764                 }
 765         }
 766
 767         bch2_fs_usage_initialize(c);
 768
 769         return 0;
 770 }
 771
 772 /* sb clean section: */
 773
 774 static struct bkey_i *btree_root_find(struct bch_fs *c,
 775                                       struct bch_sb_field_clean *clean,
 776                                       struct jset *j,
 777                                       enum btree_id id, unsigned *level)
 778 {
 779         struct bkey_i *k;
 780         struct jset_entry *entry, *start, *end;
 781
 782         if (clean) {
 783                 start = clean->start;
 784                 end = vstruct_end(&clean->field);
 785         } else {
 786                 start = j->start;
 787                 end = vstruct_last(j);
 788         }
 789
 790         for (entry = start; entry < end; entry = vstruct_next(entry))
 791                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 792                     entry->btree_id == id)
 793                         goto found;
 794
 795         return NULL;
 796 found:
 797         if (!entry->u64s)
 798                 return ERR_PTR(-EINVAL);
 799
 800         k = entry->start;
 801         *level = entry->level;
 802         return k;
 803 }
 804
 805 static int verify_superblock_clean(struct bch_fs *c,
 806                                    struct bch_sb_field_clean **cleanp,
 807                                    struct jset *j)
 808 {
 809         unsigned i;
 810         struct bch_sb_field_clean *clean = *cleanp;
 811         int ret = 0;
 812
 813         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 814                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 815                         le64_to_cpu(clean->journal_seq),
 816                         le64_to_cpu(j->seq))) {
 817                 kfree(clean);
 818                 *cleanp = NULL;
 819                 return 0;
 820         }
 821
 822         mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
 823                         "superblock read clock %u doesn't match journal %u after clean shutdown",
 824                         clean->read_clock, j->read_clock);
 825         mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
 826                         "superblock write clock %u doesn't match journal %u after clean shutdown",
 827                         clean->write_clock, j->write_clock);
 828
 829         for (i = 0; i < BTREE_ID_NR; i++) {
 830                 char buf1[200], buf2[200];
 831                 struct bkey_i *k1, *k2;
 832                 unsigned l1 = 0, l2 = 0;
 833
 834                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 835                 k2 = btree_root_find(c, NULL, j, i, &l2);
 836
 837                 if (!k1 && !k2)
 838                         continue;
 839
 840                 mustfix_fsck_err_on(!k1 || !k2 ||
 841                                     IS_ERR(k1) ||
 842                                     IS_ERR(k2) ||
 843                                     k1->k.u64s != k2->k.u64s ||
 844                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 845                                     l1 != l2, c,
 846                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 847                         "sb:      l=%u %s\n"
 848                         "journal: l=%u %s\n", i,
 849                         l1, (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(k1)), buf1),
 850                         l2, (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(k2)), buf2));
 851         }
 852 fsck_err:
 853         return ret;
 854 }
 855
 856 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 857 {
 858         struct bch_sb_field_clean *clean, *sb_clean;
 859         int ret;
 860
 861         mutex_lock(&c->sb_lock);
 862         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 863
 864         if (fsck_err_on(!sb_clean, c,
 865                         "superblock marked clean but clean section not present")) {
 866                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 867                 c->sb.clean = false;
 868                 mutex_unlock(&c->sb_lock);
 869                 return NULL;
 870         }
 871
 872         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 873                         GFP_KERNEL);
 874         if (!clean) {
 875                 mutex_unlock(&c->sb_lock);
 876                 return ERR_PTR(-ENOMEM);
 877         }
 878
 879         if (le16_to_cpu(c->disk_sb.sb->version) <
 880             bcachefs_metadata_version_bkey_renumber)
 881                 bch2_sb_clean_renumber(clean, READ);
 882
 883         mutex_unlock(&c->sb_lock);
 884
 885         return clean;
 886 fsck_err:
 887         mutex_unlock(&c->sb_lock);
 888         return ERR_PTR(ret);
 889 }
 890
 891 static int read_btree_roots(struct bch_fs *c)
 892 {
 893         unsigned i;
 894         int ret = 0;
 895
 896         for (i = 0; i < BTREE_ID_NR; i++) {
 897                 struct btree_root *r = &c->btree_roots[i];
 898
 899                 if (!r->alive)
 900                         continue;
 901
 902                 if (i == BTREE_ID_ALLOC &&
 903                     c->opts.reconstruct_alloc) {
 904                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 905                         continue;
 906                 }
 907
 908                 if (r->error) {
 909                         __fsck_err(c, i == BTREE_ID_ALLOC
 910                                    ? FSCK_CAN_IGNORE : 0,
 911                                    "invalid btree root %s",
 912                                    bch2_btree_ids[i]);
 913                         if (i == BTREE_ID_ALLOC)
 914                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 915                 }
 916
 917                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 918                 if (ret) {
 919                         __fsck_err(c, i == BTREE_ID_ALLOC
 920                                    ? FSCK_CAN_IGNORE : 0,
 921                                    "error reading btree root %s",
 922                                    bch2_btree_ids[i]);
 923                         if (i == BTREE_ID_ALLOC)
 924                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 925                 }
 926         }
 927
 928         for (i = 0; i < BTREE_ID_NR; i++)
 929                 if (!c->btree_roots[i].b)
 930                         bch2_btree_root_alloc(c, i);
 931 fsck_err:
 932         return ret;
 933 }
 934
 935 int bch2_fs_recovery(struct bch_fs *c)
 936 {
 937         const char *err = "cannot allocate memory";
 938         struct bch_sb_field_clean *clean = NULL;
 939         struct jset *last_journal_entry = NULL;
 940         u64 blacklist_seq, journal_seq;
 941         bool write_sb = false;
 942         int ret;
 943
 944         if (c->sb.clean)
 945                 clean = read_superblock_clean(c);
 946         ret = PTR_ERR_OR_ZERO(clean);
 947         if (ret)
 948                 goto err;
 949
 950         if (c->sb.clean)
 951                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 952                          le64_to_cpu(clean->journal_seq));
 953
 954         if (!c->replicas.entries ||
 955             c->opts.rebuild_replicas) {
 956                 bch_info(c, "building replicas info");
 957                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 958         }
 959
 960         ret = bch2_blacklist_table_initialize(c);
 961         if (ret) {
 962                 bch_err(c, "error initializing blacklist table");
 963                 goto err;
 964         }
 965
 966         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
 967                 struct journal_replay *i;
 968
 969                 ret = bch2_journal_read(c, &c->journal_entries,
 970                                         &blacklist_seq, &journal_seq);
 971                 if (ret)
 972                         goto err;
 973
 974                 list_for_each_entry_reverse(i, &c->journal_entries, list)
 975                         if (!i->ignore) {
 976                                 last_journal_entry = &i->j;
 977                                 break;
 978                         }
 979
 980                 if (mustfix_fsck_err_on(c->sb.clean &&
 981                                         last_journal_entry &&
 982                                         !journal_entry_empty(last_journal_entry), 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 (!last_journal_entry) {
 990                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
 991                         goto use_clean;
 992                 }
 993
 994                 c->journal_keys = journal_keys_sort(&c->journal_entries);
 995                 if (!c->journal_keys.d) {
 996                         ret = -ENOMEM;
 997                         goto err;
 998                 }
 999
1000                 if (c->sb.clean && last_journal_entry) {
1001                         ret = verify_superblock_clean(c, &clean,
1002                                                       last_journal_entry);
1003                         if (ret)
1004                                 goto err;
1005                 }
1006         } else {
1007 use_clean:
1008                 if (!clean) {
1009                         bch_err(c, "no superblock clean section found");
1010                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1011                         goto err;
1012
1013                 }
1014                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1015         }
1016
1017         if (!c->sb.clean &&
1018             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1019                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1020                 ret = -EINVAL;
1021                 goto err;
1022         }
1023
1024         if (c->opts.reconstruct_alloc) {
1025                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
1026                 drop_alloc_keys(&c->journal_keys);
1027         }
1028
1029         ret = journal_replay_early(c, clean, &c->journal_entries);
1030         if (ret)
1031                 goto err;
1032
1033         /*
1034          * After an unclean shutdown, skip then next few journal sequence
1035          * numbers as they may have been referenced by btree writes that
1036          * happened before their corresponding journal writes - those btree
1037          * writes need to be ignored, by skipping and blacklisting the next few
1038          * journal sequence numbers:
1039          */
1040         if (!c->sb.clean)
1041                 journal_seq += 8;
1042
1043         if (blacklist_seq != journal_seq) {
1044                 ret = bch2_journal_seq_blacklist_add(c,
1045                                         blacklist_seq, journal_seq);
1046                 if (ret) {
1047                         bch_err(c, "error creating new journal seq blacklist entry");
1048                         goto err;
1049                 }
1050         }
1051
1052         ret = bch2_fs_journal_start(&c->journal, journal_seq,
1053                                     &c->journal_entries);
1054         if (ret)
1055                 goto err;
1056
1057         ret = read_btree_roots(c);
1058         if (ret)
1059                 goto err;
1060
1061         bch_verbose(c, "starting alloc read");
1062         err = "error reading allocation information";
1063         ret = bch2_alloc_read(c, &c->journal_keys);
1064         if (ret)
1065                 goto err;
1066         bch_verbose(c, "alloc read done");
1067
1068         bch_verbose(c, "starting stripes_read");
1069         err = "error reading stripes";
1070         ret = bch2_stripes_read(c, &c->journal_keys);
1071         if (ret)
1072                 goto err;
1073         bch_verbose(c, "stripes_read done");
1074
1075         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1076
1077         if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
1078             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
1079                 /*
1080                  * interior btree node updates aren't consistent with the
1081                  * journal; after an unclean shutdown we have to walk all
1082                  * pointers to metadata:
1083                  */
1084                 bch_info(c, "starting metadata mark and sweep");
1085                 err = "error in mark and sweep";
1086                 ret = bch2_gc(c, &c->journal_keys, true, true);
1087                 if (ret)
1088                         goto err;
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)
1099                         goto err;
1100                 bch_verbose(c, "mark and sweep done");
1101         }
1102
1103         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1104         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1105
1106         /*
1107          * Skip past versions that might have possibly been used (as nonces),
1108          * but hadn't had their pointers written:
1109          */
1110         if (c->sb.encryption_type && !c->sb.clean)
1111                 atomic64_add(1 << 16, &c->key_version);
1112
1113         if (c->opts.norecovery)
1114                 goto out;
1115
1116         bch_verbose(c, "starting journal replay");
1117         err = "journal replay failed";
1118         ret = bch2_journal_replay(c, c->journal_keys);
1119         if (ret)
1120                 goto err;
1121         bch_verbose(c, "journal replay done");
1122
1123         if (test_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags) &&
1124             !c->opts.nochanges) {
1125                 /*
1126                  * note that even when filesystem was clean there might be work
1127                  * to do here, if we ran gc (because of fsck) which recalculated
1128                  * oldest_gen:
1129                  */
1130                 bch_verbose(c, "writing allocation info");
1131                 err = "error writing out alloc info";
1132                 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW) ?:
1133                         bch2_alloc_write(c, BTREE_INSERT_LAZY_RW);
1134                 if (ret) {
1135                         bch_err(c, "error writing alloc info");
1136                         goto err;
1137                 }
1138                 bch_verbose(c, "alloc write done");
1139         }
1140
1141         if (!c->sb.clean) {
1142                 if (!(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
1143                         bch_info(c, "checking inode link counts");
1144                         err = "error in recovery";
1145                         ret = bch2_fsck_inode_nlink(c);
1146                         if (ret)
1147                                 goto err;
1148                         bch_verbose(c, "check inodes done");
1149
1150                 } else {
1151                         bch_verbose(c, "checking for deleted inodes");
1152                         err = "error in recovery";
1153                         ret = bch2_fsck_walk_inodes_only(c);
1154                         if (ret)
1155                                 goto err;
1156                         bch_verbose(c, "check inodes done");
1157                 }
1158         }
1159
1160         if (c->opts.fsck) {
1161                 bch_info(c, "starting fsck");
1162                 err = "error in fsck";
1163                 ret = bch2_fsck_full(c);
1164                 if (ret)
1165                         goto err;
1166                 bch_verbose(c, "fsck done");
1167         }
1168
1169         if (enabled_qtypes(c)) {
1170                 bch_verbose(c, "reading quotas");
1171                 ret = bch2_fs_quota_read(c);
1172                 if (ret)
1173                         goto err;
1174                 bch_verbose(c, "quotas done");
1175         }
1176
1177         mutex_lock(&c->sb_lock);
1178         if (c->opts.version_upgrade) {
1179                 if (c->sb.version < bcachefs_metadata_version_new_versioning)
1180                         c->disk_sb.sb->version_min =
1181                                 le16_to_cpu(bcachefs_metadata_version_min);
1182                 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
1183                 c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1184                 write_sb = true;
1185         }
1186
1187         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1188                 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
1189                 write_sb = true;
1190         }
1191
1192         if (c->opts.fsck &&
1193             !test_bit(BCH_FS_ERROR, &c->flags)) {
1194                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1195                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1196                 write_sb = true;
1197         }
1198
1199         if (write_sb)
1200                 bch2_write_super(c);
1201         mutex_unlock(&c->sb_lock);
1202
1203         if (c->journal_seq_blacklist_table &&
1204             c->journal_seq_blacklist_table->nr > 128)
1205                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1206 out:
1207         ret = 0;
1208 err:
1209 fsck_err:
1210         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1211         bch2_flush_fsck_errs(c);
1212
1213         if (!c->opts.keep_journal) {
1214                 bch2_journal_keys_free(&c->journal_keys);
1215                 bch2_journal_entries_free(&c->journal_entries);
1216         }
1217         kfree(clean);
1218         if (ret)
1219                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1220         else
1221                 bch_verbose(c, "ret %i", ret);
1222         return ret;
1223 }
1224
1225 int bch2_fs_initialize(struct bch_fs *c)
1226 {
1227         struct bch_inode_unpacked root_inode, lostfound_inode;
1228         struct bkey_inode_buf packed_inode;
1229         struct qstr lostfound = QSTR("lost+found");
1230         const char *err = "cannot allocate memory";
1231         struct bch_dev *ca;
1232         LIST_HEAD(journal);
1233         unsigned i;
1234         int ret;
1235
1236         bch_notice(c, "initializing new filesystem");
1237
1238         mutex_lock(&c->sb_lock);
1239         for_each_online_member(ca, c, i)
1240                 bch2_mark_dev_superblock(c, ca, 0);
1241         mutex_unlock(&c->sb_lock);
1242
1243         mutex_lock(&c->sb_lock);
1244         c->disk_sb.sb->version = c->disk_sb.sb->version_min =
1245                 le16_to_cpu(bcachefs_metadata_version_current);
1246         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1247         c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1248
1249         bch2_write_super(c);
1250         mutex_unlock(&c->sb_lock);
1251
1252         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1253         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1254
1255         for (i = 0; i < BTREE_ID_NR; i++)
1256                 bch2_btree_root_alloc(c, i);
1257
1258         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
1259         set_bit(JOURNAL_RECLAIM_STARTED, &c->journal.flags);
1260
1261         err = "unable to allocate journal buckets";
1262         for_each_online_member(ca, c, i) {
1263                 ret = bch2_dev_journal_alloc(ca);
1264                 if (ret) {
1265                         percpu_ref_put(&ca->io_ref);
1266                         goto err;
1267                 }
1268         }
1269
1270         /*
1271          * journal_res_get() will crash if called before this has
1272          * set up the journal.pin FIFO and journal.cur pointer:
1273          */
1274         bch2_fs_journal_start(&c->journal, 1, &journal);
1275         bch2_journal_set_replay_done(&c->journal);
1276
1277         err = "error going read-write";
1278         ret = bch2_fs_read_write_early(c);
1279         if (ret)
1280                 goto err;
1281
1282         /*
1283          * Write out the superblock and journal buckets, now that we can do
1284          * btree updates
1285          */
1286         err = "error writing alloc info";
1287         ret = bch2_alloc_write(c, 0);
1288         if (ret)
1289                 goto err;
1290
1291         bch2_inode_init(c, &root_inode, 0, 0,
1292                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1293         root_inode.bi_inum = BCACHEFS_ROOT_INO;
1294         bch2_inode_pack(c, &packed_inode, &root_inode);
1295
1296         err = "error creating root directory";
1297         ret = bch2_btree_insert(c, BTREE_ID_INODES,
1298                                 &packed_inode.inode.k_i,
1299                                 NULL, NULL, 0);
1300         if (ret)
1301                 goto err;
1302
1303         bch2_inode_init_early(c, &lostfound_inode);
1304
1305         err = "error creating lost+found";
1306         ret = bch2_trans_do(c, NULL, NULL, 0,
1307                 bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
1308                                   &root_inode, &lostfound_inode,
1309                                   &lostfound,
1310                                   0, 0, S_IFDIR|0700, 0,
1311                                   NULL, NULL));
1312         if (ret)
1313                 goto err;
1314
1315         if (enabled_qtypes(c)) {
1316                 ret = bch2_fs_quota_read(c);
1317                 if (ret)
1318                         goto err;
1319         }
1320
1321         err = "error writing first journal entry";
1322         ret = bch2_journal_meta(&c->journal);
1323         if (ret)
1324                 goto err;
1325
1326         mutex_lock(&c->sb_lock);
1327         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1328         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1329
1330         bch2_write_super(c);
1331         mutex_unlock(&c->sb_lock);
1332
1333         return 0;
1334 err:
1335         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1336         return ret;
1337 }