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 /* for -o reconstruct_alloc: */
  29 static void drop_alloc_keys(struct journal_keys *keys)
  30 {
  31         size_t src, dst;
  32
  33         for (src = 0, dst = 0; src < keys->nr; src++)
  34                 if (keys->d[src].btree_id != BTREE_ID_ALLOC)
  35                         keys->d[dst++] = keys->d[src];
  36
  37         keys->nr = dst;
  38 }
  39
  40 /* iterate over keys read from the journal: */
  41
  42 static struct journal_key *journal_key_search(struct journal_keys *journal_keys,
  43                                               enum btree_id id, unsigned level,
  44                                               struct bpos pos)
  45 {
  46         size_t l = 0, r = journal_keys->nr, m;
  47
  48         while (l < r) {
  49                 m = l + ((r - l) >> 1);
  50                 if ((cmp_int(id,        journal_keys->d[m].btree_id) ?:
  51                      cmp_int(level,     journal_keys->d[m].level) ?:
  52                      bkey_cmp(pos,      journal_keys->d[m].k->k.p)) > 0)
  53                         l = m + 1;
  54                 else
  55                         r = m;
  56         }
  57
  58         BUG_ON(l < journal_keys->nr &&
  59                (cmp_int(id,     journal_keys->d[l].btree_id) ?:
  60                 cmp_int(level,  journal_keys->d[l].level) ?:
  61                 bkey_cmp(pos,   journal_keys->d[l].k->k.p)) > 0);
  62
  63         BUG_ON(l &&
  64                (cmp_int(id,     journal_keys->d[l - 1].btree_id) ?:
  65                 cmp_int(level,  journal_keys->d[l - 1].level) ?:
  66                 bkey_cmp(pos,   journal_keys->d[l - 1].k->k.p)) <= 0);
  67
  68         return l < journal_keys->nr ? journal_keys->d + l : NULL;
  69 }
  70
  71 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
  72 {
  73         if (iter->k &&
  74             iter->k < iter->keys->d + iter->keys->nr &&
  75             iter->k->btree_id   == iter->btree_id &&
  76             iter->k->level      == iter->level)
  77                 return iter->k->k;
  78
  79         iter->k = NULL;
  80         return NULL;
  81 }
  82
  83 static void bch2_journal_iter_advance(struct journal_iter *iter)
  84 {
  85         if (iter->k)
  86                 iter->k++;
  87 }
  88
  89 static void bch2_journal_iter_init(struct journal_iter *iter,
  90                                    struct journal_keys *journal_keys,
  91                                    enum btree_id id, unsigned level,
  92                                    struct bpos pos)
  93 {
  94         iter->btree_id  = id;
  95         iter->level     = level;
  96         iter->keys      = journal_keys;
  97         iter->k         = journal_key_search(journal_keys, id, level, pos);
  98 }
  99
 100 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 101 {
 102         return iter->btree
 103                 ? bch2_btree_iter_peek(iter->btree)
 104                 : bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 105                                                    iter->b, &iter->unpacked);
 106 }
 107
 108 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 109 {
 110         if (iter->btree)
 111                 bch2_btree_iter_next(iter->btree);
 112         else
 113                 bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 114 }
 115
 116 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 117 {
 118         switch (iter->last) {
 119         case none:
 120                 break;
 121         case btree:
 122                 bch2_journal_iter_advance_btree(iter);
 123                 break;
 124         case journal:
 125                 bch2_journal_iter_advance(&iter->journal);
 126                 break;
 127         }
 128
 129         iter->last = none;
 130 }
 131
 132 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 133 {
 134         struct bkey_s_c ret;
 135
 136         while (1) {
 137                 struct bkey_s_c btree_k         =
 138                         bch2_journal_iter_peek_btree(iter);
 139                 struct bkey_s_c journal_k       =
 140                         bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
 141
 142                 if (btree_k.k && journal_k.k) {
 143                         int cmp = bkey_cmp(btree_k.k->p, journal_k.k->p);
 144
 145                         if (!cmp)
 146                                 bch2_journal_iter_advance_btree(iter);
 147
 148                         iter->last = cmp < 0 ? btree : journal;
 149                 } else if (btree_k.k) {
 150                         iter->last = btree;
 151                 } else if (journal_k.k) {
 152                         iter->last = journal;
 153                 } else {
 154                         iter->last = none;
 155                         return bkey_s_c_null;
 156                 }
 157
 158                 ret = iter->last == journal ? journal_k : btree_k;
 159
 160                 if (iter->b &&
 161                     bkey_cmp(ret.k->p, iter->b->data->max_key) > 0) {
 162                         iter->journal.k = NULL;
 163                         iter->last = none;
 164                         return bkey_s_c_null;
 165                 }
 166
 167                 if (!bkey_deleted(ret.k))
 168                         break;
 169
 170                 bch2_btree_and_journal_iter_advance(iter);
 171         }
 172
 173         return ret;
 174 }
 175
 176 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 177 {
 178         bch2_btree_and_journal_iter_advance(iter);
 179
 180         return bch2_btree_and_journal_iter_peek(iter);
 181 }
 182
 183 void bch2_btree_and_journal_iter_init(struct btree_and_journal_iter *iter,
 184                                       struct btree_trans *trans,
 185                                       struct journal_keys *journal_keys,
 186                                       enum btree_id id, struct bpos pos)
 187 {
 188         memset(iter, 0, sizeof(*iter));
 189
 190         iter->btree = bch2_trans_get_iter(trans, id, pos, 0);
 191         bch2_journal_iter_init(&iter->journal, journal_keys, id, 0, pos);
 192 }
 193
 194 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 195                                                 struct journal_keys *journal_keys,
 196                                                 struct btree *b)
 197 {
 198         memset(iter, 0, sizeof(*iter));
 199
 200         iter->b = b;
 201         bch2_btree_node_iter_init_from_start(&iter->node_iter, iter->b);
 202         bch2_journal_iter_init(&iter->journal, journal_keys,
 203                                b->c.btree_id, b->c.level, b->data->min_key);
 204 }
 205
 206 /* Walk btree, overlaying keys from the journal: */
 207
 208 static int bch2_btree_and_journal_walk_recurse(struct bch_fs *c, struct btree *b,
 209                                 struct journal_keys *journal_keys,
 210                                 enum btree_id btree_id,
 211                                 btree_walk_node_fn node_fn,
 212                                 btree_walk_key_fn key_fn)
 213 {
 214         struct btree_and_journal_iter iter;
 215         struct bkey_s_c k;
 216         int ret = 0;
 217
 218         bch2_btree_and_journal_iter_init_node_iter(&iter, journal_keys, b);
 219
 220         while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 221                 ret = key_fn(c, btree_id, b->c.level, k);
 222                 if (ret)
 223                         break;
 224
 225                 if (b->c.level) {
 226                         struct btree *child;
 227                         BKEY_PADDED(k) tmp;
 228
 229                         bkey_reassemble(&tmp.k, k);
 230                         k = bkey_i_to_s_c(&tmp.k);
 231
 232                         bch2_btree_and_journal_iter_advance(&iter);
 233
 234                         if (b->c.level > 0) {
 235                                 child = bch2_btree_node_get_noiter(c, &tmp.k,
 236                                                         b->c.btree_id, b->c.level - 1);
 237                                 ret = PTR_ERR_OR_ZERO(child);
 238                                 if (ret)
 239                                         break;
 240
 241                                 ret   = (node_fn ? node_fn(c, b) : 0) ?:
 242                                         bch2_btree_and_journal_walk_recurse(c, child,
 243                                                 journal_keys, btree_id, node_fn, key_fn);
 244                                 six_unlock_read(&child->c.lock);
 245
 246                                 if (ret)
 247                                         break;
 248                         }
 249                 } else {
 250                         bch2_btree_and_journal_iter_advance(&iter);
 251                 }
 252         }
 253
 254         return ret;
 255 }
 256
 257 int bch2_btree_and_journal_walk(struct bch_fs *c, struct journal_keys *journal_keys,
 258                                 enum btree_id btree_id,
 259                                 btree_walk_node_fn node_fn,
 260                                 btree_walk_key_fn key_fn)
 261 {
 262         struct btree *b = c->btree_roots[btree_id].b;
 263         int ret = 0;
 264
 265         if (btree_node_fake(b))
 266                 return 0;
 267
 268         six_lock_read(&b->c.lock, NULL, NULL);
 269         ret   = (node_fn ? node_fn(c, b) : 0) ?:
 270                 bch2_btree_and_journal_walk_recurse(c, b, journal_keys, btree_id,
 271                                                     node_fn, key_fn) ?:
 272                 key_fn(c, btree_id, b->c.level + 1, bkey_i_to_s_c(&b->key));
 273         six_unlock_read(&b->c.lock);
 274
 275         return ret;
 276 }
 277
 278 /* sort and dedup all keys in the journal: */
 279
 280 void bch2_journal_entries_free(struct list_head *list)
 281 {
 282
 283         while (!list_empty(list)) {
 284                 struct journal_replay *i =
 285                         list_first_entry(list, struct journal_replay, list);
 286                 list_del(&i->list);
 287                 kvpfree(i, offsetof(struct journal_replay, j) +
 288                         vstruct_bytes(&i->j));
 289         }
 290 }
 291
 292 /*
 293  * When keys compare equal, oldest compares first:
 294  */
 295 static int journal_sort_key_cmp(const void *_l, const void *_r)
 296 {
 297         const struct journal_key *l = _l;
 298         const struct journal_key *r = _r;
 299
 300         return  cmp_int(l->btree_id,    r->btree_id) ?:
 301                 cmp_int(l->level,       r->level) ?:
 302                 bkey_cmp(l->k->k.p, r->k->k.p) ?:
 303                 cmp_int(l->journal_seq, r->journal_seq) ?:
 304                 cmp_int(l->journal_offset, r->journal_offset);
 305 }
 306
 307 void bch2_journal_keys_free(struct journal_keys *keys)
 308 {
 309         kvfree(keys->d);
 310         keys->d = NULL;
 311         keys->nr = 0;
 312 }
 313
 314 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 315 {
 316         struct journal_replay *p;
 317         struct jset_entry *entry;
 318         struct bkey_i *k, *_n;
 319         struct journal_keys keys = { NULL };
 320         struct journal_key *src, *dst;
 321         size_t nr_keys = 0;
 322
 323         if (list_empty(journal_entries))
 324                 return keys;
 325
 326         keys.journal_seq_base =
 327                 le64_to_cpu(list_last_entry(journal_entries,
 328                                 struct journal_replay, list)->j.last_seq);
 329
 330         list_for_each_entry(p, journal_entries, list) {
 331                 if (le64_to_cpu(p->j.seq) < keys.journal_seq_base)
 332                         continue;
 333
 334                 for_each_jset_key(k, _n, entry, &p->j)
 335                         nr_keys++;
 336         }
 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(p, journal_entries, list) {
 344                 if (le64_to_cpu(p->j.seq) < keys.journal_seq_base)
 345                         continue;
 346
 347                 for_each_jset_key(k, _n, entry, &p->j)
 348                         keys.d[keys.nr++] = (struct journal_key) {
 349                                 .btree_id       = entry->btree_id,
 350                                 .level          = entry->level,
 351                                 .k              = k,
 352                                 .journal_seq    = le64_to_cpu(p->j.seq) -
 353                                         keys.journal_seq_base,
 354                                 .journal_offset = k->_data - p->j._data,
 355                         };
 356         }
 357
 358         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 359
 360         src = dst = keys.d;
 361         while (src < keys.d + keys.nr) {
 362                 while (src + 1 < keys.d + keys.nr &&
 363                        src[0].btree_id  == src[1].btree_id &&
 364                        src[0].level     == src[1].level &&
 365                        !bkey_cmp(src[0].k->k.p, src[1].k->k.p))
 366                         src++;
 367
 368                 *dst++ = *src++;
 369         }
 370
 371         keys.nr = dst - keys.d;
 372 err:
 373         return keys;
 374 }
 375
 376 /* journal replay: */
 377
 378 static void replay_now_at(struct journal *j, u64 seq)
 379 {
 380         BUG_ON(seq < j->replay_journal_seq);
 381         BUG_ON(seq > j->replay_journal_seq_end);
 382
 383         while (j->replay_journal_seq < seq)
 384                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 385 }
 386
 387 static int bch2_extent_replay_key(struct bch_fs *c, enum btree_id btree_id,
 388                                   struct bkey_i *k)
 389 {
 390         struct btree_trans trans;
 391         struct btree_iter *iter, *split_iter;
 392         /*
 393          * We might cause compressed extents to be split, so we need to pass in
 394          * a disk_reservation:
 395          */
 396         struct disk_reservation disk_res =
 397                 bch2_disk_reservation_init(c, 0);
 398         struct bkey_i *split;
 399         struct bpos atomic_end;
 400         /*
 401          * Some extents aren't equivalent - w.r.t. what the triggers do
 402          * - if they're split:
 403          */
 404         bool remark_if_split = bch2_bkey_sectors_compressed(bkey_i_to_s_c(k)) ||
 405                 k->k.type == KEY_TYPE_reflink_p;
 406         bool remark = false;
 407         int ret;
 408
 409         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 410 retry:
 411         bch2_trans_begin(&trans);
 412
 413         iter = bch2_trans_get_iter(&trans, btree_id,
 414                                    bkey_start_pos(&k->k),
 415                                    BTREE_ITER_INTENT);
 416
 417         do {
 418                 ret = bch2_btree_iter_traverse(iter);
 419                 if (ret)
 420                         goto err;
 421
 422                 atomic_end = bpos_min(k->k.p, iter->l[0].b->key.k.p);
 423
 424                 split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
 425                 ret = PTR_ERR_OR_ZERO(split);
 426                 if (ret)
 427                         goto err;
 428
 429                 if (!remark &&
 430                     remark_if_split &&
 431                     bkey_cmp(atomic_end, k->k.p) < 0) {
 432                         ret = bch2_disk_reservation_add(c, &disk_res,
 433                                         k->k.size *
 434                                         bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(k)),
 435                                         BCH_DISK_RESERVATION_NOFAIL);
 436                         BUG_ON(ret);
 437
 438                         remark = true;
 439                 }
 440
 441                 bkey_copy(split, k);
 442                 bch2_cut_front(iter->pos, split);
 443                 bch2_cut_back(atomic_end, split);
 444
 445                 split_iter = bch2_trans_copy_iter(&trans, iter);
 446                 ret = PTR_ERR_OR_ZERO(split_iter);
 447                 if (ret)
 448                         goto err;
 449
 450                 /*
 451                  * It's important that we don't go through the
 452                  * extent_handle_overwrites() and extent_update_to_keys() path
 453                  * here: journal replay is supposed to treat extents like
 454                  * regular keys
 455                  */
 456                 __bch2_btree_iter_set_pos(split_iter, split->k.p, false);
 457                 bch2_trans_update(&trans, split_iter, split,
 458                                   BTREE_TRIGGER_NORUN);
 459                 bch2_trans_iter_put(&trans, split_iter);
 460
 461                 bch2_btree_iter_set_pos(iter, split->k.p);
 462
 463                 if (remark) {
 464                         ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(split),
 465                                                   0, split->k.size,
 466                                                   BTREE_TRIGGER_INSERT);
 467                         if (ret)
 468                                 goto err;
 469                 }
 470         } while (bkey_cmp(iter->pos, k->k.p) < 0);
 471
 472         if (remark) {
 473                 ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k),
 474                                           0, -((s64) k->k.size),
 475                                           BTREE_TRIGGER_OVERWRITE);
 476                 if (ret)
 477                         goto err;
 478         }
 479
 480         ret = bch2_trans_commit(&trans, &disk_res, NULL,
 481                                 BTREE_INSERT_NOFAIL|
 482                                 BTREE_INSERT_LAZY_RW|
 483                                 BTREE_INSERT_JOURNAL_REPLAY);
 484 err:
 485         bch2_trans_iter_put(&trans, iter);
 486
 487         if (ret == -EINTR)
 488                 goto retry;
 489
 490         bch2_disk_reservation_put(c, &disk_res);
 491
 492         return bch2_trans_exit(&trans) ?: ret;
 493 }
 494
 495 static int __bch2_journal_replay_key(struct btree_trans *trans,
 496                                      enum btree_id id, unsigned level,
 497                                      struct bkey_i *k)
 498 {
 499         struct btree_iter *iter;
 500         int ret;
 501
 502         iter = bch2_trans_get_node_iter(trans, id, k->k.p,
 503                                         BTREE_MAX_DEPTH, level,
 504                                         BTREE_ITER_INTENT);
 505         if (IS_ERR(iter))
 506                 return PTR_ERR(iter);
 507
 508         /*
 509          * iter->flags & BTREE_ITER_IS_EXTENTS triggers the update path to run
 510          * extent_handle_overwrites() and extent_update_to_keys() - but we don't
 511          * want that here, journal replay is supposed to treat extents like
 512          * regular keys:
 513          */
 514         __bch2_btree_iter_set_pos(iter, k->k.p, false);
 515
 516         ret   = bch2_btree_iter_traverse(iter) ?:
 517                 bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 518         bch2_trans_iter_put(trans, iter);
 519         return ret;
 520 }
 521
 522 static int bch2_journal_replay_key(struct bch_fs *c, enum btree_id id,
 523                                    unsigned level, struct bkey_i *k)
 524 {
 525         return bch2_trans_do(c, NULL, NULL,
 526                              BTREE_INSERT_NOFAIL|
 527                              BTREE_INSERT_LAZY_RW|
 528                              BTREE_INSERT_JOURNAL_REPLAY,
 529                              __bch2_journal_replay_key(&trans, id, level, k));
 530 }
 531
 532 static int __bch2_alloc_replay_key(struct btree_trans *trans, struct bkey_i *k)
 533 {
 534         struct btree_iter *iter;
 535         int ret;
 536
 537         iter = bch2_trans_get_iter(trans, BTREE_ID_ALLOC, k->k.p,
 538                                    BTREE_ITER_CACHED|
 539                                    BTREE_ITER_CACHED_NOFILL|
 540                                    BTREE_ITER_INTENT);
 541         ret =   PTR_ERR_OR_ZERO(iter) ?:
 542                 bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 543         bch2_trans_iter_put(trans, iter);
 544         return ret;
 545 }
 546
 547 static int bch2_alloc_replay_key(struct bch_fs *c, struct bkey_i *k)
 548 {
 549         return bch2_trans_do(c, NULL, NULL,
 550                              BTREE_INSERT_NOFAIL|
 551                              BTREE_INSERT_USE_RESERVE|
 552                              BTREE_INSERT_LAZY_RW|
 553                              BTREE_INSERT_JOURNAL_REPLAY,
 554                         __bch2_alloc_replay_key(&trans, k));
 555 }
 556
 557 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 558 {
 559         const struct journal_key *l = _l;
 560         const struct journal_key *r = _r;
 561
 562         return  cmp_int(r->level,       l->level) ?:
 563                 cmp_int(l->journal_seq, r->journal_seq) ?:
 564                 cmp_int(l->btree_id,    r->btree_id) ?:
 565                 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
 620         j->replay_journal_seq = seq;
 621
 622         /*
 623          * Now replay leaf node updates:
 624          */
 625         for_each_journal_key(keys, i) {
 626                 cond_resched();
 627
 628                 if (i->level || i->btree_id == BTREE_ID_ALLOC)
 629                         continue;
 630
 631                 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 632
 633                 ret = i->k->k.size
 634                         ? bch2_extent_replay_key(c, i->btree_id, i->k)
 635                         : bch2_journal_replay_key(c, i->btree_id, i->level, i->k);
 636                 if (ret)
 637                         goto err;
 638         }
 639
 640         replay_now_at(j, j->replay_journal_seq_end);
 641         j->replay_journal_seq = 0;
 642
 643         bch2_journal_set_replay_done(j);
 644         bch2_journal_flush_all_pins(j);
 645         return bch2_journal_error(j);
 646 err:
 647         bch_err(c, "journal replay: error %d while replaying key", ret);
 648         return ret;
 649 }
 650
 651 static bool journal_empty(struct list_head *journal)
 652 {
 653         return list_empty(journal) ||
 654                 journal_entry_empty(&list_last_entry(journal,
 655                                         struct journal_replay, list)->j);
 656 }
 657
 658 static int
 659 verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c,
 660                                                   struct list_head *journal)
 661 {
 662         struct journal_replay *i =
 663                 list_last_entry(journal, struct journal_replay, list);
 664         u64 start_seq   = le64_to_cpu(i->j.last_seq);
 665         u64 end_seq     = le64_to_cpu(i->j.seq);
 666         u64 seq         = start_seq;
 667         int ret = 0;
 668
 669         list_for_each_entry(i, journal, list) {
 670                 if (le64_to_cpu(i->j.seq) < start_seq)
 671                         continue;
 672
 673                 fsck_err_on(seq != le64_to_cpu(i->j.seq), c,
 674                         "journal entries %llu-%llu missing! (replaying %llu-%llu)",
 675                         seq, le64_to_cpu(i->j.seq) - 1,
 676                         start_seq, end_seq);
 677
 678                 seq = le64_to_cpu(i->j.seq);
 679
 680                 fsck_err_on(bch2_journal_seq_is_blacklisted(c, seq, false), c,
 681                             "found blacklisted journal entry %llu", seq);
 682
 683                 do {
 684                         seq++;
 685                 } while (bch2_journal_seq_is_blacklisted(c, seq, false));
 686         }
 687 fsck_err:
 688         return ret;
 689 }
 690
 691 /* journal replay early: */
 692
 693 static int journal_replay_entry_early(struct bch_fs *c,
 694                                       struct jset_entry *entry)
 695 {
 696         int ret = 0;
 697
 698         switch (entry->type) {
 699         case BCH_JSET_ENTRY_btree_root: {
 700                 struct btree_root *r;
 701
 702                 if (entry->btree_id >= BTREE_ID_NR) {
 703                         bch_err(c, "filesystem has unknown btree type %u",
 704                                 entry->btree_id);
 705                         return -EINVAL;
 706                 }
 707
 708                 r = &c->btree_roots[entry->btree_id];
 709
 710                 if (entry->u64s) {
 711                         r->level = entry->level;
 712                         bkey_copy(&r->key, &entry->start[0]);
 713                         r->error = 0;
 714                 } else {
 715                         r->error = -EIO;
 716                 }
 717                 r->alive = true;
 718                 break;
 719         }
 720         case BCH_JSET_ENTRY_usage: {
 721                 struct jset_entry_usage *u =
 722                         container_of(entry, struct jset_entry_usage, entry);
 723
 724                 switch (entry->btree_id) {
 725                 case FS_USAGE_RESERVED:
 726                         if (entry->level < BCH_REPLICAS_MAX)
 727                                 c->usage_base->persistent_reserved[entry->level] =
 728                                         le64_to_cpu(u->v);
 729                         break;
 730                 case FS_USAGE_INODES:
 731                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 732                         break;
 733                 case FS_USAGE_KEY_VERSION:
 734                         atomic64_set(&c->key_version,
 735                                      le64_to_cpu(u->v));
 736                         break;
 737                 }
 738
 739                 break;
 740         }
 741         case BCH_JSET_ENTRY_data_usage: {
 742                 struct jset_entry_data_usage *u =
 743                         container_of(entry, struct jset_entry_data_usage, entry);
 744                 ret = bch2_replicas_set_usage(c, &u->r,
 745                                               le64_to_cpu(u->v));
 746                 break;
 747         }
 748         case BCH_JSET_ENTRY_blacklist: {
 749                 struct jset_entry_blacklist *bl_entry =
 750                         container_of(entry, struct jset_entry_blacklist, entry);
 751
 752                 ret = bch2_journal_seq_blacklist_add(c,
 753                                 le64_to_cpu(bl_entry->seq),
 754                                 le64_to_cpu(bl_entry->seq) + 1);
 755                 break;
 756         }
 757         case BCH_JSET_ENTRY_blacklist_v2: {
 758                 struct jset_entry_blacklist_v2 *bl_entry =
 759                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 760
 761                 ret = bch2_journal_seq_blacklist_add(c,
 762                                 le64_to_cpu(bl_entry->start),
 763                                 le64_to_cpu(bl_entry->end) + 1);
 764                 break;
 765         }
 766         }
 767
 768         return ret;
 769 }
 770
 771 static int journal_replay_early(struct bch_fs *c,
 772                                 struct bch_sb_field_clean *clean,
 773                                 struct list_head *journal)
 774 {
 775         struct jset_entry *entry;
 776         int ret;
 777
 778         if (clean) {
 779                 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
 780                 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
 781
 782                 for (entry = clean->start;
 783                      entry != vstruct_end(&clean->field);
 784                      entry = vstruct_next(entry)) {
 785                         ret = journal_replay_entry_early(c, entry);
 786                         if (ret)
 787                                 return ret;
 788                 }
 789         } else {
 790                 struct journal_replay *i =
 791                         list_last_entry(journal, struct journal_replay, list);
 792
 793                 c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
 794                 c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
 795
 796                 list_for_each_entry(i, journal, list)
 797                         vstruct_for_each(&i->j, entry) {
 798                                 ret = journal_replay_entry_early(c, entry);
 799                                 if (ret)
 800                                         return ret;
 801                         }
 802         }
 803
 804         bch2_fs_usage_initialize(c);
 805
 806         return 0;
 807 }
 808
 809 /* sb clean section: */
 810
 811 static struct bkey_i *btree_root_find(struct bch_fs *c,
 812                                       struct bch_sb_field_clean *clean,
 813                                       struct jset *j,
 814                                       enum btree_id id, unsigned *level)
 815 {
 816         struct bkey_i *k;
 817         struct jset_entry *entry, *start, *end;
 818
 819         if (clean) {
 820                 start = clean->start;
 821                 end = vstruct_end(&clean->field);
 822         } else {
 823                 start = j->start;
 824                 end = vstruct_last(j);
 825         }
 826
 827         for (entry = start; entry < end; entry = vstruct_next(entry))
 828                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 829                     entry->btree_id == id)
 830                         goto found;
 831
 832         return NULL;
 833 found:
 834         if (!entry->u64s)
 835                 return ERR_PTR(-EINVAL);
 836
 837         k = entry->start;
 838         *level = entry->level;
 839         return k;
 840 }
 841
 842 static int verify_superblock_clean(struct bch_fs *c,
 843                                    struct bch_sb_field_clean **cleanp,
 844                                    struct jset *j)
 845 {
 846         unsigned i;
 847         struct bch_sb_field_clean *clean = *cleanp;
 848         int ret = 0;
 849
 850         if (!c->sb.clean || !j)
 851                 return 0;
 852
 853         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 854                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 855                         le64_to_cpu(clean->journal_seq),
 856                         le64_to_cpu(j->seq))) {
 857                 kfree(clean);
 858                 *cleanp = NULL;
 859                 return 0;
 860         }
 861
 862         mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
 863                         "superblock read clock %u doesn't match journal %u after clean shutdown",
 864                         clean->read_clock, j->read_clock);
 865         mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
 866                         "superblock write clock %u doesn't match journal %u after clean shutdown",
 867                         clean->write_clock, j->write_clock);
 868
 869         for (i = 0; i < BTREE_ID_NR; i++) {
 870                 char buf1[200], buf2[200];
 871                 struct bkey_i *k1, *k2;
 872                 unsigned l1 = 0, l2 = 0;
 873
 874                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 875                 k2 = btree_root_find(c, NULL, j, i, &l2);
 876
 877                 if (!k1 && !k2)
 878                         continue;
 879
 880                 mustfix_fsck_err_on(!k1 || !k2 ||
 881                                     IS_ERR(k1) ||
 882                                     IS_ERR(k2) ||
 883                                     k1->k.u64s != k2->k.u64s ||
 884                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 885                                     l1 != l2, c,
 886                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 887                         "sb:      l=%u %s\n"
 888                         "journal: l=%u %s\n", i,
 889                         l1, (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(k1)), buf1),
 890                         l2, (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(k2)), buf2));
 891         }
 892 fsck_err:
 893         return ret;
 894 }
 895
 896 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 897 {
 898         struct bch_sb_field_clean *clean, *sb_clean;
 899         int ret;
 900
 901         mutex_lock(&c->sb_lock);
 902         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 903
 904         if (fsck_err_on(!sb_clean, c,
 905                         "superblock marked clean but clean section not present")) {
 906                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 907                 c->sb.clean = false;
 908                 mutex_unlock(&c->sb_lock);
 909                 return NULL;
 910         }
 911
 912         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 913                         GFP_KERNEL);
 914         if (!clean) {
 915                 mutex_unlock(&c->sb_lock);
 916                 return ERR_PTR(-ENOMEM);
 917         }
 918
 919         if (le16_to_cpu(c->disk_sb.sb->version) <
 920             bcachefs_metadata_version_bkey_renumber)
 921                 bch2_sb_clean_renumber(clean, READ);
 922
 923         mutex_unlock(&c->sb_lock);
 924
 925         return clean;
 926 fsck_err:
 927         mutex_unlock(&c->sb_lock);
 928         return ERR_PTR(ret);
 929 }
 930
 931 static int read_btree_roots(struct bch_fs *c)
 932 {
 933         unsigned i;
 934         int ret = 0;
 935
 936         for (i = 0; i < BTREE_ID_NR; i++) {
 937                 struct btree_root *r = &c->btree_roots[i];
 938
 939                 if (!r->alive)
 940                         continue;
 941
 942                 if (i == BTREE_ID_ALLOC &&
 943                     c->opts.reconstruct_alloc) {
 944                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 945                         continue;
 946                 }
 947
 948                 if (r->error) {
 949                         __fsck_err(c, i == BTREE_ID_ALLOC
 950                                    ? FSCK_CAN_IGNORE : 0,
 951                                    "invalid btree root %s",
 952                                    bch2_btree_ids[i]);
 953                         if (i == BTREE_ID_ALLOC)
 954                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 955                 }
 956
 957                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 958                 if (ret) {
 959                         __fsck_err(c, i == BTREE_ID_ALLOC
 960                                    ? FSCK_CAN_IGNORE : 0,
 961                                    "error reading btree root %s",
 962                                    bch2_btree_ids[i]);
 963                         if (i == BTREE_ID_ALLOC)
 964                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 965                 }
 966         }
 967
 968         for (i = 0; i < BTREE_ID_NR; i++)
 969                 if (!c->btree_roots[i].b)
 970                         bch2_btree_root_alloc(c, i);
 971 fsck_err:
 972         return ret;
 973 }
 974
 975 int bch2_fs_recovery(struct bch_fs *c)
 976 {
 977         const char *err = "cannot allocate memory";
 978         struct bch_sb_field_clean *clean = NULL;
 979         u64 journal_seq;
 980         bool write_sb = false, need_write_alloc = false;
 981         int ret;
 982
 983         if (c->sb.clean)
 984                 clean = read_superblock_clean(c);
 985         ret = PTR_ERR_OR_ZERO(clean);
 986         if (ret)
 987                 goto err;
 988
 989         if (c->sb.clean)
 990                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 991                          le64_to_cpu(clean->journal_seq));
 992
 993         if (!c->replicas.entries ||
 994             c->opts.rebuild_replicas) {
 995                 bch_info(c, "building replicas info");
 996                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 997         }
 998
 999         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1000                 struct jset *j;
1001
1002                 ret = bch2_journal_read(c, &c->journal_entries);
1003                 if (ret)
1004                         goto err;
1005
1006                 if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&c->journal_entries), c,
1007                                 "filesystem marked clean but journal not empty")) {
1008                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
1009                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1010                         c->sb.clean = false;
1011                 }
1012
1013                 if (!c->sb.clean && list_empty(&c->journal_entries)) {
1014                         bch_err(c, "no journal entries found");
1015                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1016                         goto err;
1017                 }
1018
1019                 c->journal_keys = journal_keys_sort(&c->journal_entries);
1020                 if (!c->journal_keys.d) {
1021                         ret = -ENOMEM;
1022                         goto err;
1023                 }
1024
1025                 j = &list_last_entry(&c->journal_entries,
1026                                      struct journal_replay, list)->j;
1027
1028                 ret = verify_superblock_clean(c, &clean, j);
1029                 if (ret)
1030                         goto err;
1031
1032                 journal_seq = le64_to_cpu(j->seq) + 1;
1033         } else {
1034                 journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1035         }
1036
1037         if (!c->sb.clean &&
1038             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1039                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1040                 ret = -EINVAL;
1041                 goto err;
1042         }
1043
1044         if (c->opts.reconstruct_alloc) {
1045                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
1046                 drop_alloc_keys(&c->journal_keys);
1047         }
1048
1049         ret = journal_replay_early(c, clean, &c->journal_entries);
1050         if (ret)
1051                 goto err;
1052
1053         if (!c->sb.clean) {
1054                 ret = bch2_journal_seq_blacklist_add(c,
1055                                                      journal_seq,
1056                                                      journal_seq + 4);
1057                 if (ret) {
1058                         bch_err(c, "error creating new journal seq blacklist entry");
1059                         goto err;
1060                 }
1061
1062                 journal_seq += 4;
1063
1064                 /*
1065                  * The superblock needs to be written before we do any btree
1066                  * node writes: it will be in the read_write() path
1067                  */
1068         }
1069
1070         ret = bch2_blacklist_table_initialize(c);
1071
1072         if (!list_empty(&c->journal_entries)) {
1073                 ret = verify_journal_entries_not_blacklisted_or_missing(c,
1074                                                         &c->journal_entries);
1075                 if (ret)
1076                         goto err;
1077         }
1078
1079         ret = bch2_fs_journal_start(&c->journal, journal_seq,
1080                                     &c->journal_entries);
1081         if (ret)
1082                 goto err;
1083
1084         ret = read_btree_roots(c);
1085         if (ret)
1086                 goto err;
1087
1088         bch_verbose(c, "starting alloc read");
1089         err = "error reading allocation information";
1090         ret = bch2_alloc_read(c, &c->journal_keys);
1091         if (ret)
1092                 goto err;
1093         bch_verbose(c, "alloc read done");
1094
1095         bch_verbose(c, "starting stripes_read");
1096         err = "error reading stripes";
1097         ret = bch2_stripes_read(c, &c->journal_keys);
1098         if (ret)
1099                 goto err;
1100         bch_verbose(c, "stripes_read done");
1101
1102         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1103
1104         if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
1105             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
1106                 /*
1107                  * interior btree node updates aren't consistent with the
1108                  * journal; after an unclean shutdown we have to walk all
1109                  * pointers to metadata:
1110                  */
1111                 bch_info(c, "starting metadata mark and sweep");
1112                 err = "error in mark and sweep";
1113                 ret = bch2_gc(c, &c->journal_keys, true, true);
1114                 if (ret < 0)
1115                         goto err;
1116                 if (ret)
1117                         need_write_alloc = true;
1118                 bch_verbose(c, "mark and sweep done");
1119         }
1120
1121         if (c->opts.fsck ||
1122             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
1123             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1124                 bch_info(c, "starting mark and sweep");
1125                 err = "error in mark and sweep";
1126                 ret = bch2_gc(c, &c->journal_keys, true, false);
1127                 if (ret < 0)
1128                         goto err;
1129                 if (ret)
1130                         need_write_alloc = true;
1131                 bch_verbose(c, "mark and sweep done");
1132         }
1133
1134         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1135         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1136
1137         /*
1138          * Skip past versions that might have possibly been used (as nonces),
1139          * but hadn't had their pointers written:
1140          */
1141         if (c->sb.encryption_type && !c->sb.clean)
1142                 atomic64_add(1 << 16, &c->key_version);
1143
1144         if (c->opts.norecovery)
1145                 goto out;
1146
1147         bch_verbose(c, "starting journal replay");
1148         err = "journal replay failed";
1149         ret = bch2_journal_replay(c, c->journal_keys);
1150         if (ret)
1151                 goto err;
1152         bch_verbose(c, "journal replay done");
1153
1154         if (need_write_alloc && !c->opts.nochanges) {
1155                 /*
1156                  * note that even when filesystem was clean there might be work
1157                  * to do here, if we ran gc (because of fsck) which recalculated
1158                  * oldest_gen:
1159                  */
1160                 bch_verbose(c, "writing allocation info");
1161                 err = "error writing out alloc info";
1162                 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW) ?:
1163                         bch2_alloc_write(c, BTREE_INSERT_LAZY_RW);
1164                 if (ret) {
1165                         bch_err(c, "error writing alloc info");
1166                         goto err;
1167                 }
1168                 bch_verbose(c, "alloc write done");
1169
1170                 set_bit(BCH_FS_ALLOC_WRITTEN, &c->flags);
1171         }
1172
1173         if (!c->sb.clean) {
1174                 if (!(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
1175                         bch_info(c, "checking inode link counts");
1176                         err = "error in recovery";
1177                         ret = bch2_fsck_inode_nlink(c);
1178                         if (ret)
1179                                 goto err;
1180                         bch_verbose(c, "check inodes done");
1181
1182                 } else {
1183                         bch_verbose(c, "checking for deleted inodes");
1184                         err = "error in recovery";
1185                         ret = bch2_fsck_walk_inodes_only(c);
1186                         if (ret)
1187                                 goto err;
1188                         bch_verbose(c, "check inodes done");
1189                 }
1190         }
1191
1192         if (c->opts.fsck) {
1193                 bch_info(c, "starting fsck");
1194                 err = "error in fsck";
1195                 ret = bch2_fsck_full(c);
1196                 if (ret)
1197                         goto err;
1198                 bch_verbose(c, "fsck done");
1199         }
1200
1201         if (enabled_qtypes(c)) {
1202                 bch_verbose(c, "reading quotas");
1203                 ret = bch2_fs_quota_read(c);
1204                 if (ret)
1205                         goto err;
1206                 bch_verbose(c, "quotas done");
1207         }
1208
1209         mutex_lock(&c->sb_lock);
1210         if (c->opts.version_upgrade) {
1211                 if (c->sb.version < bcachefs_metadata_version_new_versioning)
1212                         c->disk_sb.sb->version_min =
1213                                 le16_to_cpu(bcachefs_metadata_version_min);
1214                 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
1215                 c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1216                 write_sb = true;
1217         }
1218
1219         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1220                 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
1221                 write_sb = true;
1222         }
1223
1224         if (c->opts.fsck &&
1225             !test_bit(BCH_FS_ERROR, &c->flags)) {
1226                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1227                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1228                 write_sb = true;
1229         }
1230
1231         if (write_sb)
1232                 bch2_write_super(c);
1233         mutex_unlock(&c->sb_lock);
1234
1235         if (c->journal_seq_blacklist_table &&
1236             c->journal_seq_blacklist_table->nr > 128)
1237                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1238 out:
1239         ret = 0;
1240 err:
1241 fsck_err:
1242         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1243         bch2_flush_fsck_errs(c);
1244
1245         if (!c->opts.keep_journal) {
1246                 bch2_journal_keys_free(&c->journal_keys);
1247                 bch2_journal_entries_free(&c->journal_entries);
1248         }
1249         kfree(clean);
1250         if (ret)
1251                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1252         else
1253                 bch_verbose(c, "ret %i", ret);
1254         return ret;
1255 }
1256
1257 int bch2_fs_initialize(struct bch_fs *c)
1258 {
1259         struct bch_inode_unpacked root_inode, lostfound_inode;
1260         struct bkey_inode_buf packed_inode;
1261         struct qstr lostfound = QSTR("lost+found");
1262         const char *err = "cannot allocate memory";
1263         struct bch_dev *ca;
1264         LIST_HEAD(journal);
1265         unsigned i;
1266         int ret;
1267
1268         bch_notice(c, "initializing new filesystem");
1269
1270         mutex_lock(&c->sb_lock);
1271         for_each_online_member(ca, c, i)
1272                 bch2_mark_dev_superblock(c, ca, 0);
1273         mutex_unlock(&c->sb_lock);
1274
1275         mutex_lock(&c->sb_lock);
1276         c->disk_sb.sb->version = c->disk_sb.sb->version_min =
1277                 le16_to_cpu(bcachefs_metadata_version_current);
1278         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1279         c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1280
1281         bch2_write_super(c);
1282         mutex_unlock(&c->sb_lock);
1283
1284         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1285         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1286
1287         for (i = 0; i < BTREE_ID_NR; i++)
1288                 bch2_btree_root_alloc(c, i);
1289
1290         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
1291         set_bit(JOURNAL_RECLAIM_STARTED, &c->journal.flags);
1292
1293         err = "unable to allocate journal buckets";
1294         for_each_online_member(ca, c, i) {
1295                 ret = bch2_dev_journal_alloc(ca);
1296                 if (ret) {
1297                         percpu_ref_put(&ca->io_ref);
1298                         goto err;
1299                 }
1300         }
1301
1302         /*
1303          * journal_res_get() will crash if called before this has
1304          * set up the journal.pin FIFO and journal.cur pointer:
1305          */
1306         bch2_fs_journal_start(&c->journal, 1, &journal);
1307         bch2_journal_set_replay_done(&c->journal);
1308
1309         err = "error going read-write";
1310         ret = bch2_fs_read_write_early(c);
1311         if (ret)
1312                 goto err;
1313
1314         /*
1315          * Write out the superblock and journal buckets, now that we can do
1316          * btree updates
1317          */
1318         err = "error writing alloc info";
1319         ret = bch2_alloc_write(c, 0);
1320         if (ret)
1321                 goto err;
1322
1323         bch2_inode_init(c, &root_inode, 0, 0,
1324                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1325         root_inode.bi_inum = BCACHEFS_ROOT_INO;
1326         bch2_inode_pack(c, &packed_inode, &root_inode);
1327
1328         err = "error creating root directory";
1329         ret = bch2_btree_insert(c, BTREE_ID_INODES,
1330                                 &packed_inode.inode.k_i,
1331                                 NULL, NULL, 0);
1332         if (ret)
1333                 goto err;
1334
1335         bch2_inode_init_early(c, &lostfound_inode);
1336
1337         err = "error creating lost+found";
1338         ret = bch2_trans_do(c, NULL, NULL, 0,
1339                 bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
1340                                   &root_inode, &lostfound_inode,
1341                                   &lostfound,
1342                                   0, 0, S_IFDIR|0700, 0,
1343                                   NULL, NULL));
1344         if (ret)
1345                 goto err;
1346
1347         if (enabled_qtypes(c)) {
1348                 ret = bch2_fs_quota_read(c);
1349                 if (ret)
1350                         goto err;
1351         }
1352
1353         err = "error writing first journal entry";
1354         ret = bch2_journal_meta(&c->journal);
1355         if (ret)
1356                 goto err;
1357
1358         mutex_lock(&c->sb_lock);
1359         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1360         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1361
1362         bch2_write_super(c);
1363         mutex_unlock(&c->sb_lock);
1364
1365         return 0;
1366 err:
1367         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1368         return ret;
1369 }