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

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "bkey_buf.h"
   5 #include "alloc_background.h"
   6 #include "btree_gc.h"
   7 #include "btree_update.h"
   8 #include "btree_update_interior.h"
   9 #include "btree_io.h"
  10 #include "buckets.h"
  11 #include "dirent.h"
  12 #include "ec.h"
  13 #include "error.h"
  14 #include "fs-common.h"
  15 #include "fsck.h"
  16 #include "journal_io.h"
  17 #include "journal_reclaim.h"
  18 #include "journal_seq_blacklist.h"
  19 #include "move.h"
  20 #include "quota.h"
  21 #include "recovery.h"
  22 #include "replicas.h"
  23 #include "subvolume.h"
  24 #include "super-io.h"
  25
  26 #include <linux/sort.h>
  27 #include <linux/stat.h>
  28
  29 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  30
  31 /* for -o reconstruct_alloc: */
  32 static void drop_alloc_keys(struct journal_keys *keys)
  33 {
  34         size_t src, dst;
  35
  36         for (src = 0, dst = 0; src < keys->nr; src++)
  37                 if (keys->d[src].btree_id != BTREE_ID_alloc)
  38                         keys->d[dst++] = keys->d[src];
  39
  40         keys->nr = dst;
  41 }
  42
  43 /*
  44  * Btree node pointers have a field to stack a pointer to the in memory btree
  45  * node; we need to zero out this field when reading in btree nodes, or when
  46  * reading in keys from the journal:
  47  */
  48 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
  49 {
  50         struct journal_key *i;
  51
  52         for (i = keys->d; i < keys->d + keys->nr; i++)
  53                 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
  54                         bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
  55 }
  56
  57 /* iterate over keys read from the journal: */
  58
  59 static int __journal_key_cmp(enum btree_id      l_btree_id,
  60                              unsigned           l_level,
  61                              struct bpos        l_pos,
  62                              const struct journal_key *r)
  63 {
  64         return (cmp_int(l_btree_id,     r->btree_id) ?:
  65                 cmp_int(l_level,        r->level) ?:
  66                 bpos_cmp(l_pos, r->k->k.p));
  67 }
  68
  69 static int journal_key_cmp(const struct journal_key *l, const struct journal_key *r)
  70 {
  71         return __journal_key_cmp(l->btree_id, l->level, l->k->k.p, r);
  72 }
  73
  74 size_t bch2_journal_key_search(struct journal_keys *journal_keys,
  75                                enum btree_id id, unsigned level,
  76                                struct bpos pos)
  77 {
  78         size_t l = 0, r = journal_keys->nr, m;
  79
  80         while (l < r) {
  81                 m = l + ((r - l) >> 1);
  82                 if (__journal_key_cmp(id, level, pos, &journal_keys->d[m]) > 0)
  83                         l = m + 1;
  84                 else
  85                         r = m;
  86         }
  87
  88         BUG_ON(l < journal_keys->nr &&
  89                __journal_key_cmp(id, level, pos, &journal_keys->d[l]) > 0);
  90
  91         BUG_ON(l &&
  92                __journal_key_cmp(id, level, pos, &journal_keys->d[l - 1]) <= 0);
  93
  94         return l;
  95 }
  96
  97 static void journal_iter_fix(struct bch_fs *c, struct journal_iter *iter, unsigned idx)
  98 {
  99         struct bkey_i *n = iter->keys->d[idx].k;
 100         struct btree_and_journal_iter *biter =
 101                 container_of(iter, struct btree_and_journal_iter, journal);
 102
 103         if (iter->idx > idx ||
 104             (iter->idx == idx &&
 105              biter->last &&
 106              bpos_cmp(n->k.p, biter->unpacked.p) <= 0))
 107                 iter->idx++;
 108 }
 109
 110 int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
 111                                  unsigned level, struct bkey_i *k)
 112 {
 113         struct journal_key n = {
 114                 .btree_id       = id,
 115                 .level          = level,
 116                 .k              = k,
 117                 .allocated      = true,
 118                 /*
 119                  * Ensure these keys are done last by journal replay, to unblock
 120                  * journal reclaim:
 121                  */
 122                 .journal_seq    = U32_MAX,
 123         };
 124         struct journal_keys *keys = &c->journal_keys;
 125         struct journal_iter *iter;
 126         size_t idx = bch2_journal_key_search(keys, id, level, k->k.p);
 127
 128         BUG_ON(test_bit(BCH_FS_RW, &c->flags));
 129
 130         if (idx < keys->nr &&
 131             journal_key_cmp(&n, &keys->d[idx]) == 0) {
 132                 if (keys->d[idx].allocated)
 133                         kfree(keys->d[idx].k);
 134                 keys->d[idx] = n;
 135                 return 0;
 136         }
 137
 138         if (keys->nr == keys->size) {
 139                 struct journal_keys new_keys = {
 140                         .nr                     = keys->nr,
 141                         .size                   = keys->size * 2,
 142                         .journal_seq_base       = keys->journal_seq_base,
 143                 };
 144
 145                 new_keys.d = kvmalloc(sizeof(new_keys.d[0]) * new_keys.size, GFP_KERNEL);
 146                 if (!new_keys.d) {
 147                         bch_err(c, "%s: error allocating new key array (size %zu)",
 148                                 __func__, new_keys.size);
 149                         return -ENOMEM;
 150                 }
 151
 152                 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
 153                 kvfree(keys->d);
 154                 *keys = new_keys;
 155         }
 156
 157         array_insert_item(keys->d, keys->nr, idx, n);
 158
 159         list_for_each_entry(iter, &c->journal_iters, list)
 160                 journal_iter_fix(c, iter, idx);
 161
 162         return 0;
 163 }
 164
 165 /*
 166  * Can only be used from the recovery thread while we're still RO - can't be
 167  * used once we've got RW, as journal_keys is at that point used by multiple
 168  * threads:
 169  */
 170 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
 171                             unsigned level, struct bkey_i *k)
 172 {
 173         struct bkey_i *n;
 174         int ret;
 175
 176         n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
 177         if (!n)
 178                 return -ENOMEM;
 179
 180         bkey_copy(n, k);
 181         ret = bch2_journal_key_insert_take(c, id, level, n);
 182         if (ret)
 183                 kfree(n);
 184         return ret;
 185 }
 186
 187 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
 188                             unsigned level, struct bpos pos)
 189 {
 190         struct bkey_i whiteout;
 191
 192         bkey_init(&whiteout.k);
 193         whiteout.k.p = pos;
 194
 195         return bch2_journal_key_insert(c, id, level, &whiteout);
 196 }
 197
 198 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
 199                                   unsigned level, struct bpos pos)
 200 {
 201         struct journal_keys *keys = &c->journal_keys;
 202         size_t idx = bch2_journal_key_search(keys, btree, level, pos);
 203
 204         if (idx < keys->nr &&
 205             keys->d[idx].btree_id       == btree &&
 206             keys->d[idx].level          == level &&
 207             !bpos_cmp(keys->d[idx].k->k.p, pos))
 208                 keys->d[idx].overwritten = true;
 209 }
 210
 211 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
 212 {
 213         struct journal_key *k = iter->keys->d + iter->idx;
 214
 215         while (k < iter->keys->d + iter->keys->nr &&
 216                k->btree_id      == iter->btree_id &&
 217                k->level         == iter->level) {
 218                 if (!k->overwritten)
 219                         return k->k;
 220
 221                 iter->idx++;
 222                 k = iter->keys->d + iter->idx;
 223         }
 224
 225         return NULL;
 226 }
 227
 228 static void bch2_journal_iter_advance(struct journal_iter *iter)
 229 {
 230         if (iter->idx < iter->keys->nr)
 231                 iter->idx++;
 232 }
 233
 234 static void bch2_journal_iter_exit(struct journal_iter *iter)
 235 {
 236         list_del(&iter->list);
 237 }
 238
 239 static void bch2_journal_iter_init(struct bch_fs *c,
 240                                    struct journal_iter *iter,
 241                                    enum btree_id id, unsigned level,
 242                                    struct bpos pos)
 243 {
 244         iter->btree_id  = id;
 245         iter->level     = level;
 246         iter->keys      = &c->journal_keys;
 247         iter->idx       = bch2_journal_key_search(&c->journal_keys, id, level, pos);
 248 }
 249
 250 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 251 {
 252         return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 253                                                 iter->b, &iter->unpacked);
 254 }
 255
 256 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 257 {
 258         bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 259 }
 260
 261 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 262 {
 263         switch (iter->last) {
 264         case none:
 265                 break;
 266         case btree:
 267                 bch2_journal_iter_advance_btree(iter);
 268                 break;
 269         case journal:
 270                 bch2_journal_iter_advance(&iter->journal);
 271                 break;
 272         }
 273
 274         iter->last = none;
 275 }
 276
 277 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 278 {
 279         struct bkey_s_c ret;
 280
 281         while (1) {
 282                 struct bkey_s_c btree_k         =
 283                         bch2_journal_iter_peek_btree(iter);
 284                 struct bkey_s_c journal_k       =
 285                         bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
 286
 287                 if (btree_k.k && journal_k.k) {
 288                         int cmp = bpos_cmp(btree_k.k->p, journal_k.k->p);
 289
 290                         if (!cmp)
 291                                 bch2_journal_iter_advance_btree(iter);
 292
 293                         iter->last = cmp < 0 ? btree : journal;
 294                 } else if (btree_k.k) {
 295                         iter->last = btree;
 296                 } else if (journal_k.k) {
 297                         iter->last = journal;
 298                 } else {
 299                         iter->last = none;
 300                         return bkey_s_c_null;
 301                 }
 302
 303                 ret = iter->last == journal ? journal_k : btree_k;
 304
 305                 if (iter->b &&
 306                     bpos_cmp(ret.k->p, iter->b->data->max_key) > 0) {
 307                         iter->journal.idx = iter->journal.keys->nr;
 308                         iter->last = none;
 309                         return bkey_s_c_null;
 310                 }
 311
 312                 if (!bkey_deleted(ret.k))
 313                         break;
 314
 315                 bch2_btree_and_journal_iter_advance(iter);
 316         }
 317
 318         return ret;
 319 }
 320
 321 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 322 {
 323         bch2_btree_and_journal_iter_advance(iter);
 324
 325         return bch2_btree_and_journal_iter_peek(iter);
 326 }
 327
 328 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
 329 {
 330         bch2_journal_iter_exit(&iter->journal);
 331 }
 332
 333 void __bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 334                                                   struct bch_fs *c,
 335                                                   struct btree *b,
 336                                                   struct btree_node_iter node_iter,
 337                                                   struct bpos pos)
 338 {
 339         memset(iter, 0, sizeof(*iter));
 340
 341         iter->b = b;
 342         iter->node_iter = node_iter;
 343         bch2_journal_iter_init(c, &iter->journal, b->c.btree_id, b->c.level, pos);
 344         INIT_LIST_HEAD(&iter->journal.list);
 345 }
 346
 347 /*
 348  * this version is used by btree_gc before filesystem has gone RW and
 349  * multithreaded, so uses the journal_iters list:
 350  */
 351 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 352                                                 struct bch_fs *c,
 353                                                 struct btree *b)
 354 {
 355         struct btree_node_iter node_iter;
 356
 357         bch2_btree_node_iter_init_from_start(&node_iter, b);
 358         __bch2_btree_and_journal_iter_init_node_iter(iter, c, b, node_iter, b->data->min_key);
 359         list_add(&iter->journal.list, &c->journal_iters);
 360 }
 361
 362 /* sort and dedup all keys in the journal: */
 363
 364 void bch2_journal_entries_free(struct list_head *list)
 365 {
 366
 367         while (!list_empty(list)) {
 368                 struct journal_replay *i =
 369                         list_first_entry(list, struct journal_replay, list);
 370                 list_del(&i->list);
 371                 kvpfree(i, offsetof(struct journal_replay, j) +
 372                         vstruct_bytes(&i->j));
 373         }
 374 }
 375
 376 /*
 377  * When keys compare equal, oldest compares first:
 378  */
 379 static int journal_sort_key_cmp(const void *_l, const void *_r)
 380 {
 381         const struct journal_key *l = _l;
 382         const struct journal_key *r = _r;
 383
 384         return  journal_key_cmp(l, r) ?:
 385                 cmp_int(l->journal_seq, r->journal_seq) ?:
 386                 cmp_int(l->journal_offset, r->journal_offset);
 387 }
 388
 389 void bch2_journal_keys_free(struct journal_keys *keys)
 390 {
 391         struct journal_key *i;
 392
 393         for (i = keys->d; i < keys->d + keys->nr; i++)
 394                 if (i->allocated)
 395                         kfree(i->k);
 396
 397         kvfree(keys->d);
 398         keys->d = NULL;
 399         keys->nr = 0;
 400 }
 401
 402 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 403 {
 404         struct journal_replay *i;
 405         struct jset_entry *entry;
 406         struct bkey_i *k, *_n;
 407         struct journal_keys keys = { NULL };
 408         struct journal_key *src, *dst;
 409         size_t nr_keys = 0;
 410
 411         if (list_empty(journal_entries))
 412                 return keys;
 413
 414         list_for_each_entry(i, journal_entries, list) {
 415                 if (i->ignore)
 416                         continue;
 417
 418                 if (!keys.journal_seq_base)
 419                         keys.journal_seq_base = le64_to_cpu(i->j.seq);
 420
 421                 for_each_jset_key(k, _n, entry, &i->j)
 422                         nr_keys++;
 423         }
 424
 425         keys.size = roundup_pow_of_two(nr_keys);
 426
 427         keys.d = kvmalloc(sizeof(keys.d[0]) * keys.size, GFP_KERNEL);
 428         if (!keys.d)
 429                 goto err;
 430
 431         list_for_each_entry(i, journal_entries, list) {
 432                 if (i->ignore)
 433                         continue;
 434
 435                 BUG_ON(le64_to_cpu(i->j.seq) - keys.journal_seq_base > U32_MAX);
 436
 437                 for_each_jset_key(k, _n, entry, &i->j)
 438                         keys.d[keys.nr++] = (struct journal_key) {
 439                                 .btree_id       = entry->btree_id,
 440                                 .level          = entry->level,
 441                                 .k              = k,
 442                                 .journal_seq    = le64_to_cpu(i->j.seq) -
 443                                         keys.journal_seq_base,
 444                                 .journal_offset = k->_data - i->j._data,
 445                         };
 446         }
 447
 448         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 449
 450         src = dst = keys.d;
 451         while (src < keys.d + keys.nr) {
 452                 while (src + 1 < keys.d + keys.nr &&
 453                        src[0].btree_id  == src[1].btree_id &&
 454                        src[0].level     == src[1].level &&
 455                        !bpos_cmp(src[0].k->k.p, src[1].k->k.p))
 456                         src++;
 457
 458                 *dst++ = *src++;
 459         }
 460
 461         keys.nr = dst - keys.d;
 462 err:
 463         return keys;
 464 }
 465
 466 /* journal replay: */
 467
 468 static void replay_now_at(struct journal *j, u64 seq)
 469 {
 470         BUG_ON(seq < j->replay_journal_seq);
 471         BUG_ON(seq > j->replay_journal_seq_end);
 472
 473         while (j->replay_journal_seq < seq)
 474                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 475 }
 476
 477 static int bch2_journal_replay_key(struct btree_trans *trans,
 478                                    struct journal_key *k)
 479 {
 480         struct btree_iter iter;
 481         unsigned iter_flags =
 482                 BTREE_ITER_INTENT|
 483                 BTREE_ITER_NOT_EXTENTS;
 484         int ret;
 485
 486         if (!k->level && k->btree_id == BTREE_ID_alloc)
 487                 iter_flags |= BTREE_ITER_CACHED;
 488
 489         bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 490                                   BTREE_MAX_DEPTH, k->level,
 491                                   iter_flags);
 492         ret = bch2_btree_iter_traverse(&iter);
 493         if (ret)
 494                 goto out;
 495
 496         /* Must be checked with btree locked: */
 497         if (k->overwritten)
 498                 goto out;
 499
 500         ret = bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
 501 out:
 502         bch2_trans_iter_exit(trans, &iter);
 503         return ret;
 504 }
 505
 506 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 507 {
 508         const struct journal_key *l = *((const struct journal_key **)_l);
 509         const struct journal_key *r = *((const struct journal_key **)_r);
 510
 511         return cmp_int(l->journal_seq, r->journal_seq);
 512 }
 513
 514 static int bch2_journal_replay(struct bch_fs *c)
 515 {
 516         struct journal_keys *keys = &c->journal_keys;
 517         struct journal_key **keys_sorted, *k;
 518         struct journal *j = &c->journal;
 519         size_t i;
 520         int ret;
 521
 522         keys_sorted = kmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL);
 523         if (!keys_sorted)
 524                 return -ENOMEM;
 525
 526         for (i = 0; i < keys->nr; i++)
 527                 keys_sorted[i] = &keys->d[i];
 528
 529         sort(keys_sorted, keys->nr,
 530              sizeof(keys_sorted[0]),
 531              journal_sort_seq_cmp, NULL);
 532
 533         if (keys->nr) {
 534                 bch_verbose(c, "starting journal replay, %zu keys", keys->nr);
 535                 replay_now_at(j, keys->journal_seq_base);
 536         }
 537
 538         for (i = 0; i < keys->nr; i++) {
 539                 k = keys_sorted[i];
 540
 541                 cond_resched();
 542
 543                 if (!k->allocated)
 544                         replay_now_at(j, keys->journal_seq_base + k->journal_seq);
 545
 546                 ret = bch2_trans_do(c, NULL, NULL,
 547                                     BTREE_INSERT_LAZY_RW|
 548                                     BTREE_INSERT_NOFAIL|
 549                                     BTREE_INSERT_JOURNAL_RESERVED|
 550                                     (!k->allocated ? BTREE_INSERT_JOURNAL_REPLAY : 0),
 551                              bch2_journal_replay_key(&trans, k));
 552                 if (ret) {
 553                         bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
 554                                 ret, bch2_btree_ids[k->btree_id], k->level);
 555                         goto err;
 556                 }
 557         }
 558
 559         replay_now_at(j, j->replay_journal_seq_end);
 560         j->replay_journal_seq = 0;
 561
 562         bch2_journal_set_replay_done(j);
 563         bch2_journal_flush_all_pins(j);
 564         ret = bch2_journal_error(j);
 565 err:
 566         kfree(keys_sorted);
 567         return ret;
 568 }
 569
 570 /* journal replay early: */
 571
 572 static int journal_replay_entry_early(struct bch_fs *c,
 573                                       struct jset_entry *entry)
 574 {
 575         int ret = 0;
 576
 577         switch (entry->type) {
 578         case BCH_JSET_ENTRY_btree_root: {
 579                 struct btree_root *r;
 580
 581                 if (entry->btree_id >= BTREE_ID_NR) {
 582                         bch_err(c, "filesystem has unknown btree type %u",
 583                                 entry->btree_id);
 584                         return -EINVAL;
 585                 }
 586
 587                 r = &c->btree_roots[entry->btree_id];
 588
 589                 if (entry->u64s) {
 590                         r->level = entry->level;
 591                         bkey_copy(&r->key, &entry->start[0]);
 592                         r->error = 0;
 593                 } else {
 594                         r->error = -EIO;
 595                 }
 596                 r->alive = true;
 597                 break;
 598         }
 599         case BCH_JSET_ENTRY_usage: {
 600                 struct jset_entry_usage *u =
 601                         container_of(entry, struct jset_entry_usage, entry);
 602
 603                 switch (entry->btree_id) {
 604                 case BCH_FS_USAGE_reserved:
 605                         if (entry->level < BCH_REPLICAS_MAX)
 606                                 c->usage_base->persistent_reserved[entry->level] =
 607                                         le64_to_cpu(u->v);
 608                         break;
 609                 case BCH_FS_USAGE_inodes:
 610                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 611                         break;
 612                 case BCH_FS_USAGE_key_version:
 613                         atomic64_set(&c->key_version,
 614                                      le64_to_cpu(u->v));
 615                         break;
 616                 }
 617
 618                 break;
 619         }
 620         case BCH_JSET_ENTRY_data_usage: {
 621                 struct jset_entry_data_usage *u =
 622                         container_of(entry, struct jset_entry_data_usage, entry);
 623
 624                 ret = bch2_replicas_set_usage(c, &u->r,
 625                                               le64_to_cpu(u->v));
 626                 break;
 627         }
 628         case BCH_JSET_ENTRY_dev_usage: {
 629                 struct jset_entry_dev_usage *u =
 630                         container_of(entry, struct jset_entry_dev_usage, entry);
 631                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 632                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 633
 634                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 635                 ca->usage_base->buckets_unavailable     = le64_to_cpu(u->buckets_unavailable);
 636
 637                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 638                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 639                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 640                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 641                 }
 642
 643                 break;
 644         }
 645         case BCH_JSET_ENTRY_blacklist: {
 646                 struct jset_entry_blacklist *bl_entry =
 647                         container_of(entry, struct jset_entry_blacklist, entry);
 648
 649                 ret = bch2_journal_seq_blacklist_add(c,
 650                                 le64_to_cpu(bl_entry->seq),
 651                                 le64_to_cpu(bl_entry->seq) + 1);
 652                 break;
 653         }
 654         case BCH_JSET_ENTRY_blacklist_v2: {
 655                 struct jset_entry_blacklist_v2 *bl_entry =
 656                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 657
 658                 ret = bch2_journal_seq_blacklist_add(c,
 659                                 le64_to_cpu(bl_entry->start),
 660                                 le64_to_cpu(bl_entry->end) + 1);
 661                 break;
 662         }
 663         case BCH_JSET_ENTRY_clock: {
 664                 struct jset_entry_clock *clock =
 665                         container_of(entry, struct jset_entry_clock, entry);
 666
 667                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 668         }
 669         }
 670
 671         return ret;
 672 }
 673
 674 static int journal_replay_early(struct bch_fs *c,
 675                                 struct bch_sb_field_clean *clean,
 676                                 struct list_head *journal)
 677 {
 678         struct journal_replay *i;
 679         struct jset_entry *entry;
 680         int ret;
 681
 682         if (clean) {
 683                 for (entry = clean->start;
 684                      entry != vstruct_end(&clean->field);
 685                      entry = vstruct_next(entry)) {
 686                         ret = journal_replay_entry_early(c, entry);
 687                         if (ret)
 688                                 return ret;
 689                 }
 690         } else {
 691                 list_for_each_entry(i, journal, list) {
 692                         if (i->ignore)
 693                                 continue;
 694
 695                         vstruct_for_each(&i->j, entry) {
 696                                 ret = journal_replay_entry_early(c, entry);
 697                                 if (ret)
 698                                         return ret;
 699                         }
 700                 }
 701         }
 702
 703         bch2_fs_usage_initialize(c);
 704
 705         return 0;
 706 }
 707
 708 /* sb clean section: */
 709
 710 static struct bkey_i *btree_root_find(struct bch_fs *c,
 711                                       struct bch_sb_field_clean *clean,
 712                                       struct jset *j,
 713                                       enum btree_id id, unsigned *level)
 714 {
 715         struct bkey_i *k;
 716         struct jset_entry *entry, *start, *end;
 717
 718         if (clean) {
 719                 start = clean->start;
 720                 end = vstruct_end(&clean->field);
 721         } else {
 722                 start = j->start;
 723                 end = vstruct_last(j);
 724         }
 725
 726         for (entry = start; entry < end; entry = vstruct_next(entry))
 727                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 728                     entry->btree_id == id)
 729                         goto found;
 730
 731         return NULL;
 732 found:
 733         if (!entry->u64s)
 734                 return ERR_PTR(-EINVAL);
 735
 736         k = entry->start;
 737         *level = entry->level;
 738         return k;
 739 }
 740
 741 static int verify_superblock_clean(struct bch_fs *c,
 742                                    struct bch_sb_field_clean **cleanp,
 743                                    struct jset *j)
 744 {
 745         unsigned i;
 746         struct bch_sb_field_clean *clean = *cleanp;
 747         int ret = 0;
 748
 749         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 750                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 751                         le64_to_cpu(clean->journal_seq),
 752                         le64_to_cpu(j->seq))) {
 753                 kfree(clean);
 754                 *cleanp = NULL;
 755                 return 0;
 756         }
 757
 758         for (i = 0; i < BTREE_ID_NR; i++) {
 759                 char buf1[200], buf2[200];
 760                 struct bkey_i *k1, *k2;
 761                 unsigned l1 = 0, l2 = 0;
 762
 763                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 764                 k2 = btree_root_find(c, NULL, j, i, &l2);
 765
 766                 if (!k1 && !k2)
 767                         continue;
 768
 769                 mustfix_fsck_err_on(!k1 || !k2 ||
 770                                     IS_ERR(k1) ||
 771                                     IS_ERR(k2) ||
 772                                     k1->k.u64s != k2->k.u64s ||
 773                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 774                                     l1 != l2, c,
 775                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 776                         "sb:      l=%u %s\n"
 777                         "journal: l=%u %s\n", i,
 778                         l1, (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(k1)), buf1),
 779                         l2, (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(k2)), buf2));
 780         }
 781 fsck_err:
 782         return ret;
 783 }
 784
 785 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 786 {
 787         struct bch_sb_field_clean *clean, *sb_clean;
 788         int ret;
 789
 790         mutex_lock(&c->sb_lock);
 791         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 792
 793         if (fsck_err_on(!sb_clean, c,
 794                         "superblock marked clean but clean section not present")) {
 795                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 796                 c->sb.clean = false;
 797                 mutex_unlock(&c->sb_lock);
 798                 return NULL;
 799         }
 800
 801         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 802                         GFP_KERNEL);
 803         if (!clean) {
 804                 mutex_unlock(&c->sb_lock);
 805                 return ERR_PTR(-ENOMEM);
 806         }
 807
 808         ret = bch2_sb_clean_validate(c, clean, READ);
 809         if (ret) {
 810                 mutex_unlock(&c->sb_lock);
 811                 return ERR_PTR(ret);
 812         }
 813
 814         mutex_unlock(&c->sb_lock);
 815
 816         return clean;
 817 fsck_err:
 818         mutex_unlock(&c->sb_lock);
 819         return ERR_PTR(ret);
 820 }
 821
 822 static int read_btree_roots(struct bch_fs *c)
 823 {
 824         unsigned i;
 825         int ret = 0;
 826
 827         for (i = 0; i < BTREE_ID_NR; i++) {
 828                 struct btree_root *r = &c->btree_roots[i];
 829
 830                 if (!r->alive)
 831                         continue;
 832
 833                 if (i == BTREE_ID_alloc &&
 834                     c->opts.reconstruct_alloc) {
 835                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 836                         continue;
 837                 }
 838
 839                 if (r->error) {
 840                         __fsck_err(c, i == BTREE_ID_alloc
 841                                    ? FSCK_CAN_IGNORE : 0,
 842                                    "invalid btree root %s",
 843                                    bch2_btree_ids[i]);
 844                         if (i == BTREE_ID_alloc)
 845                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 846                 }
 847
 848                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 849                 if (ret) {
 850                         __fsck_err(c, i == BTREE_ID_alloc
 851                                    ? FSCK_CAN_IGNORE : 0,
 852                                    "error reading btree root %s",
 853                                    bch2_btree_ids[i]);
 854                         if (i == BTREE_ID_alloc)
 855                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 856                 }
 857         }
 858
 859         for (i = 0; i < BTREE_ID_NR; i++)
 860                 if (!c->btree_roots[i].b)
 861                         bch2_btree_root_alloc(c, i);
 862 fsck_err:
 863         return ret;
 864 }
 865
 866 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
 867 {
 868         struct bkey_i_snapshot  root_snapshot;
 869         struct bkey_i_subvolume root_volume;
 870         int ret;
 871
 872         bkey_snapshot_init(&root_snapshot.k_i);
 873         root_snapshot.k.p.offset = U32_MAX;
 874         root_snapshot.v.flags   = 0;
 875         root_snapshot.v.parent  = 0;
 876         root_snapshot.v.subvol  = BCACHEFS_ROOT_SUBVOL;
 877         root_snapshot.v.pad     = 0;
 878         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
 879
 880         ret = bch2_btree_insert(c, BTREE_ID_snapshots,
 881                                 &root_snapshot.k_i,
 882                                 NULL, NULL, 0);
 883         if (ret)
 884                 return ret;
 885
 886
 887         bkey_subvolume_init(&root_volume.k_i);
 888         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
 889         root_volume.v.flags     = 0;
 890         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
 891         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
 892
 893         ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
 894                                 &root_volume.k_i,
 895                                 NULL, NULL, 0);
 896         if (ret)
 897                 return ret;
 898
 899         return 0;
 900 }
 901
 902 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
 903 {
 904         struct bch_fs *c = trans->c;
 905         struct btree_iter iter;
 906         struct bkey_s_c k;
 907         struct bch_inode_unpacked inode;
 908         int ret;
 909
 910         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
 911                              SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
 912         k = bch2_btree_iter_peek_slot(&iter);
 913         ret = bkey_err(k);
 914         if (ret)
 915                 goto err;
 916
 917         if (!bkey_is_inode(k.k)) {
 918                 bch_err(c, "root inode not found");
 919                 ret = -ENOENT;
 920                 goto err;
 921         }
 922
 923         ret = bch2_inode_unpack(k, &inode);
 924         BUG_ON(ret);
 925
 926         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
 927
 928         ret = bch2_inode_write(trans, &iter, &inode);
 929 err:
 930         bch2_trans_iter_exit(trans, &iter);
 931         return ret;
 932 }
 933
 934 int bch2_fs_recovery(struct bch_fs *c)
 935 {
 936         const char *err = "cannot allocate memory";
 937         struct bch_sb_field_clean *clean = NULL;
 938         struct jset *last_journal_entry = NULL;
 939         u64 blacklist_seq, journal_seq;
 940         bool write_sb = false;
 941         int ret = 0;
 942
 943         if (c->sb.clean)
 944                 clean = read_superblock_clean(c);
 945         ret = PTR_ERR_OR_ZERO(clean);
 946         if (ret)
 947                 goto err;
 948
 949         if (c->sb.clean)
 950                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 951                          le64_to_cpu(clean->journal_seq));
 952         else
 953                 bch_info(c, "recovering from unclean shutdown");
 954
 955         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
 956                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
 957                 ret = -EINVAL;
 958                 goto err;
 959         }
 960
 961         if (!c->sb.clean &&
 962             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
 963                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
 964                 ret = -EINVAL;
 965                 goto err;
 966         }
 967
 968         if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
 969                 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
 970                 ret = -EINVAL;
 971                 goto err;
 972         }
 973
 974         if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
 975                 bch_info(c, "alloc_v2 feature bit not set, fsck required");
 976                 c->opts.fsck = true;
 977                 c->opts.fix_errors = FSCK_OPT_YES;
 978         }
 979
 980         if (!c->replicas.entries ||
 981             c->opts.rebuild_replicas) {
 982                 bch_info(c, "building replicas info");
 983                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 984         }
 985
 986         if (!c->opts.nochanges) {
 987                 if (c->sb.version < bcachefs_metadata_version_inode_backpointers) {
 988                         bch_info(c, "version prior to inode backpointers, upgrade and fsck required");
 989                         c->opts.version_upgrade = true;
 990                         c->opts.fsck            = true;
 991                         c->opts.fix_errors      = FSCK_OPT_YES;
 992                 } else if (c->sb.version < bcachefs_metadata_version_subvol_dirent) {
 993                         bch_info(c, "filesystem version is prior to subvol_dirent - upgrading");
 994                         c->opts.version_upgrade = true;
 995                         c->opts.fsck            = true;
 996                 } else if (c->sb.version < bcachefs_metadata_version_inode_v2) {
 997                         bch_info(c, "filesystem version is prior to inode_v2 - upgrading");
 998                         c->opts.version_upgrade = true;
 999                 }
1000         }
1001
1002         ret = bch2_blacklist_table_initialize(c);
1003         if (ret) {
1004                 bch_err(c, "error initializing blacklist table");
1005                 goto err;
1006         }
1007
1008         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1009                 struct journal_replay *i;
1010
1011                 ret = bch2_journal_read(c, &c->journal_entries,
1012                                         &blacklist_seq, &journal_seq);
1013                 if (ret)
1014                         goto err;
1015
1016                 list_for_each_entry_reverse(i, &c->journal_entries, list)
1017                         if (!i->ignore) {
1018                                 last_journal_entry = &i->j;
1019                                 break;
1020                         }
1021
1022                 if (mustfix_fsck_err_on(c->sb.clean &&
1023                                         last_journal_entry &&
1024                                         !journal_entry_empty(last_journal_entry), c,
1025                                 "filesystem marked clean but journal not empty")) {
1026                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1027                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1028                         c->sb.clean = false;
1029                 }
1030
1031                 if (!last_journal_entry) {
1032                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
1033                         goto use_clean;
1034                 }
1035
1036                 c->journal_keys = journal_keys_sort(&c->journal_entries);
1037                 if (!c->journal_keys.d) {
1038                         ret = -ENOMEM;
1039                         goto err;
1040                 }
1041
1042                 if (c->sb.clean && last_journal_entry) {
1043                         ret = verify_superblock_clean(c, &clean,
1044                                                       last_journal_entry);
1045                         if (ret)
1046                                 goto err;
1047                 }
1048         } else {
1049 use_clean:
1050                 if (!clean) {
1051                         bch_err(c, "no superblock clean section found");
1052                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1053                         goto err;
1054
1055                 }
1056                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1057         }
1058
1059         if (c->opts.reconstruct_alloc) {
1060                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1061                 drop_alloc_keys(&c->journal_keys);
1062         }
1063
1064         zero_out_btree_mem_ptr(&c->journal_keys);
1065
1066         ret = journal_replay_early(c, clean, &c->journal_entries);
1067         if (ret)
1068                 goto err;
1069
1070         if (blacklist_seq != journal_seq) {
1071                 ret = bch2_journal_seq_blacklist_add(c,
1072                                         blacklist_seq, journal_seq);
1073                 if (ret) {
1074                         bch_err(c, "error creating new journal seq blacklist entry");
1075                         goto err;
1076                 }
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
1091         down_read(&c->gc_lock);
1092         ret = bch2_alloc_read(c, false, false);
1093         up_read(&c->gc_lock);
1094
1095         if (ret)
1096                 goto err;
1097         bch_verbose(c, "alloc read done");
1098
1099         bch_verbose(c, "starting stripes_read");
1100         err = "error reading stripes";
1101         ret = bch2_stripes_read(c);
1102         if (ret)
1103                 goto err;
1104         bch_verbose(c, "stripes_read done");
1105
1106         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1107
1108         /*
1109          * If we're not running fsck, this ensures bch2_fsck_err() calls are
1110          * instead interpreted as bch2_inconsistent_err() calls:
1111          */
1112         if (!c->opts.fsck)
1113                 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1114
1115         if (c->opts.fsck ||
1116             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)) ||
1117             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_metadata)) ||
1118             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1119                 bool metadata_only = c->opts.norecovery;
1120
1121                 bch_info(c, "starting mark and sweep");
1122                 err = "error in mark and sweep";
1123                 ret = bch2_gc(c, true, metadata_only);
1124                 if (ret)
1125                         goto err;
1126                 bch_verbose(c, "mark and sweep done");
1127         }
1128
1129         bch2_stripes_heap_start(c);
1130
1131         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1132         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1133
1134         /*
1135          * Skip past versions that might have possibly been used (as nonces),
1136          * but hadn't had their pointers written:
1137          */
1138         if (c->sb.encryption_type && !c->sb.clean)
1139                 atomic64_add(1 << 16, &c->key_version);
1140
1141         if (c->opts.norecovery)
1142                 goto out;
1143
1144         bch_verbose(c, "starting journal replay");
1145         err = "journal replay failed";
1146         ret = bch2_journal_replay(c);
1147         if (ret)
1148                 goto err;
1149         if (c->opts.verbose || !c->sb.clean)
1150                 bch_info(c, "journal replay done");
1151
1152         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1153                 bch2_fs_lazy_rw(c);
1154
1155                 err = "error creating root snapshot node";
1156                 ret = bch2_fs_initialize_subvolumes(c);
1157                 if (ret)
1158                         goto err;
1159         }
1160
1161         bch_verbose(c, "reading snapshots table");
1162         err = "error reading snapshots table";
1163         ret = bch2_fs_snapshots_start(c);
1164         if (ret)
1165                 goto err;
1166         bch_verbose(c, "reading snapshots done");
1167
1168         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1169                 /* set bi_subvol on root inode */
1170                 err = "error upgrade root inode for subvolumes";
1171                 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1172                                     bch2_fs_upgrade_for_subvolumes(&trans));
1173                 if (ret)
1174                         goto err;
1175         }
1176
1177         if (c->opts.fsck) {
1178                 bch_info(c, "starting fsck");
1179                 err = "error in fsck";
1180                 ret = bch2_fsck_full(c);
1181                 if (ret)
1182                         goto err;
1183                 bch_verbose(c, "fsck done");
1184         } else if (!c->sb.clean) {
1185                 bch_verbose(c, "checking for deleted inodes");
1186                 err = "error in recovery";
1187                 ret = bch2_fsck_walk_inodes_only(c);
1188                 if (ret)
1189                         goto err;
1190                 bch_verbose(c, "check inodes done");
1191         }
1192
1193         if (enabled_qtypes(c)) {
1194                 bch_verbose(c, "reading quotas");
1195                 ret = bch2_fs_quota_read(c);
1196                 if (ret)
1197                         goto err;
1198                 bch_verbose(c, "quotas done");
1199         }
1200
1201         mutex_lock(&c->sb_lock);
1202         /*
1203          * With journal replay done, we can clear the journal seq blacklist
1204          * table:
1205          */
1206         BUG_ON(!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags));
1207         if (le16_to_cpu(c->sb.version_min) >= bcachefs_metadata_version_btree_ptr_sectors_written)
1208                 bch2_sb_resize_journal_seq_blacklist(&c->disk_sb, 0);
1209
1210         if (c->opts.version_upgrade) {
1211                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1212                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1213                 write_sb = true;
1214         }
1215
1216         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1217                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1218                 write_sb = true;
1219         }
1220
1221         if (c->opts.fsck &&
1222             !test_bit(BCH_FS_ERROR, &c->flags) &&
1223             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1224                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1225                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1226                 write_sb = true;
1227         }
1228
1229         if (write_sb)
1230                 bch2_write_super(c);
1231         mutex_unlock(&c->sb_lock);
1232
1233         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1234             !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1235             le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) {
1236                 struct bch_move_stats stats;
1237
1238                 bch_move_stats_init(&stats, "recovery");
1239
1240                 bch_info(c, "scanning for old btree nodes");
1241                 ret = bch2_fs_read_write(c);
1242                 if (ret)
1243                         goto err;
1244
1245                 ret = bch2_scan_old_btree_nodes(c, &stats);
1246                 if (ret)
1247                         goto err;
1248                 bch_info(c, "scanning for old btree nodes done");
1249         }
1250
1251         ret = 0;
1252 out:
1253         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1254         bch2_flush_fsck_errs(c);
1255
1256         if (!c->opts.keep_journal) {
1257                 bch2_journal_keys_free(&c->journal_keys);
1258                 bch2_journal_entries_free(&c->journal_entries);
1259         }
1260         kfree(clean);
1261         if (ret)
1262                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1263         else
1264                 bch_verbose(c, "ret %i", ret);
1265         return ret;
1266 err:
1267 fsck_err:
1268         bch2_fs_emergency_read_only(c);
1269         goto out;
1270 }
1271
1272 int bch2_fs_initialize(struct bch_fs *c)
1273 {
1274         struct bch_inode_unpacked root_inode, lostfound_inode;
1275         struct bkey_inode_buf packed_inode;
1276         struct qstr lostfound = QSTR("lost+found");
1277         const char *err = "cannot allocate memory";
1278         struct bch_dev *ca;
1279         LIST_HEAD(journal);
1280         unsigned i;
1281         int ret;
1282
1283         bch_notice(c, "initializing new filesystem");
1284
1285         mutex_lock(&c->sb_lock);
1286         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1287         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1288
1289         if (c->opts.version_upgrade) {
1290                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1291                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1292                 bch2_write_super(c);
1293         }
1294         mutex_unlock(&c->sb_lock);
1295
1296         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1297         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1298         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1299
1300         for (i = 0; i < BTREE_ID_NR; i++)
1301                 bch2_btree_root_alloc(c, i);
1302
1303         err = "unable to allocate journal buckets";
1304         for_each_online_member(ca, c, i) {
1305                 ret = bch2_dev_journal_alloc(ca);
1306                 if (ret) {
1307                         percpu_ref_put(&ca->io_ref);
1308                         goto err;
1309                 }
1310         }
1311
1312         /*
1313          * journal_res_get() will crash if called before this has
1314          * set up the journal.pin FIFO and journal.cur pointer:
1315          */
1316         bch2_fs_journal_start(&c->journal, 1, &journal);
1317         bch2_journal_set_replay_done(&c->journal);
1318
1319         err = "error going read-write";
1320         ret = bch2_fs_read_write_early(c);
1321         if (ret)
1322                 goto err;
1323
1324         /*
1325          * Write out the superblock and journal buckets, now that we can do
1326          * btree updates
1327          */
1328         err = "error marking superblock and journal";
1329         for_each_member_device(ca, c, i) {
1330                 ret = bch2_trans_mark_dev_sb(c, ca);
1331                 if (ret) {
1332                         percpu_ref_put(&ca->ref);
1333                         goto err;
1334                 }
1335
1336                 ca->new_fs_bucket_idx = 0;
1337         }
1338
1339         err = "error creating root snapshot node";
1340         ret = bch2_fs_initialize_subvolumes(c);
1341         if (ret)
1342                 goto err;
1343
1344         bch_verbose(c, "reading snapshots table");
1345         err = "error reading snapshots table";
1346         ret = bch2_fs_snapshots_start(c);
1347         if (ret)
1348                 goto err;
1349         bch_verbose(c, "reading snapshots done");
1350
1351         bch2_inode_init(c, &root_inode, 0, 0,
1352                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1353         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1354         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1355         bch2_inode_pack(c, &packed_inode, &root_inode);
1356         packed_inode.inode.k.p.snapshot = U32_MAX;
1357
1358         err = "error creating root directory";
1359         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1360                                 &packed_inode.inode.k_i,
1361                                 NULL, NULL, 0);
1362         if (ret)
1363                 goto err;
1364
1365         bch2_inode_init_early(c, &lostfound_inode);
1366
1367         err = "error creating lost+found";
1368         ret = bch2_trans_do(c, NULL, NULL, 0,
1369                 bch2_create_trans(&trans,
1370                                   BCACHEFS_ROOT_SUBVOL_INUM,
1371                                   &root_inode, &lostfound_inode,
1372                                   &lostfound,
1373                                   0, 0, S_IFDIR|0700, 0,
1374                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1375         if (ret) {
1376                 bch_err(c, "error creating lost+found");
1377                 goto err;
1378         }
1379
1380         if (enabled_qtypes(c)) {
1381                 ret = bch2_fs_quota_read(c);
1382                 if (ret)
1383                         goto err;
1384         }
1385
1386         err = "error writing first journal entry";
1387         ret = bch2_journal_flush(&c->journal);
1388         if (ret)
1389                 goto err;
1390
1391         mutex_lock(&c->sb_lock);
1392         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1393         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1394
1395         bch2_write_super(c);
1396         mutex_unlock(&c->sb_lock);
1397
1398         return 0;
1399 err:
1400         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1401         return ret;
1402 }