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 = kvmalloc_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                 replay_now_at(j, keys->journal_seq_base);
 535
 536         for (i = 0; i < keys->nr; i++) {
 537                 k = keys_sorted[i];
 538
 539                 cond_resched();
 540
 541                 if (!k->allocated)
 542                         replay_now_at(j, keys->journal_seq_base + k->journal_seq);
 543
 544                 ret = bch2_trans_do(c, NULL, NULL,
 545                                     BTREE_INSERT_LAZY_RW|
 546                                     BTREE_INSERT_NOFAIL|
 547                                     BTREE_INSERT_JOURNAL_RESERVED|
 548                                     (!k->allocated ? BTREE_INSERT_JOURNAL_REPLAY : 0),
 549                              bch2_journal_replay_key(&trans, k));
 550                 if (ret) {
 551                         bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
 552                                 ret, bch2_btree_ids[k->btree_id], k->level);
 553                         goto err;
 554                 }
 555         }
 556
 557         replay_now_at(j, j->replay_journal_seq_end);
 558         j->replay_journal_seq = 0;
 559
 560         bch2_journal_set_replay_done(j);
 561         bch2_journal_flush_all_pins(j);
 562         ret = bch2_journal_error(j);
 563 err:
 564         kvfree(keys_sorted);
 565         return ret;
 566 }
 567
 568 /* journal replay early: */
 569
 570 static int journal_replay_entry_early(struct bch_fs *c,
 571                                       struct jset_entry *entry)
 572 {
 573         int ret = 0;
 574
 575         switch (entry->type) {
 576         case BCH_JSET_ENTRY_btree_root: {
 577                 struct btree_root *r;
 578
 579                 if (entry->btree_id >= BTREE_ID_NR) {
 580                         bch_err(c, "filesystem has unknown btree type %u",
 581                                 entry->btree_id);
 582                         return -EINVAL;
 583                 }
 584
 585                 r = &c->btree_roots[entry->btree_id];
 586
 587                 if (entry->u64s) {
 588                         r->level = entry->level;
 589                         bkey_copy(&r->key, &entry->start[0]);
 590                         r->error = 0;
 591                 } else {
 592                         r->error = -EIO;
 593                 }
 594                 r->alive = true;
 595                 break;
 596         }
 597         case BCH_JSET_ENTRY_usage: {
 598                 struct jset_entry_usage *u =
 599                         container_of(entry, struct jset_entry_usage, entry);
 600
 601                 switch (entry->btree_id) {
 602                 case BCH_FS_USAGE_reserved:
 603                         if (entry->level < BCH_REPLICAS_MAX)
 604                                 c->usage_base->persistent_reserved[entry->level] =
 605                                         le64_to_cpu(u->v);
 606                         break;
 607                 case BCH_FS_USAGE_inodes:
 608                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 609                         break;
 610                 case BCH_FS_USAGE_key_version:
 611                         atomic64_set(&c->key_version,
 612                                      le64_to_cpu(u->v));
 613                         break;
 614                 }
 615
 616                 break;
 617         }
 618         case BCH_JSET_ENTRY_data_usage: {
 619                 struct jset_entry_data_usage *u =
 620                         container_of(entry, struct jset_entry_data_usage, entry);
 621
 622                 ret = bch2_replicas_set_usage(c, &u->r,
 623                                               le64_to_cpu(u->v));
 624                 break;
 625         }
 626         case BCH_JSET_ENTRY_dev_usage: {
 627                 struct jset_entry_dev_usage *u =
 628                         container_of(entry, struct jset_entry_dev_usage, entry);
 629                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 630                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 631
 632                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 633                 ca->usage_base->buckets_unavailable     = le64_to_cpu(u->buckets_unavailable);
 634
 635                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 636                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 637                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 638                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 639                 }
 640
 641                 break;
 642         }
 643         case BCH_JSET_ENTRY_blacklist: {
 644                 struct jset_entry_blacklist *bl_entry =
 645                         container_of(entry, struct jset_entry_blacklist, entry);
 646
 647                 ret = bch2_journal_seq_blacklist_add(c,
 648                                 le64_to_cpu(bl_entry->seq),
 649                                 le64_to_cpu(bl_entry->seq) + 1);
 650                 break;
 651         }
 652         case BCH_JSET_ENTRY_blacklist_v2: {
 653                 struct jset_entry_blacklist_v2 *bl_entry =
 654                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 655
 656                 ret = bch2_journal_seq_blacklist_add(c,
 657                                 le64_to_cpu(bl_entry->start),
 658                                 le64_to_cpu(bl_entry->end) + 1);
 659                 break;
 660         }
 661         case BCH_JSET_ENTRY_clock: {
 662                 struct jset_entry_clock *clock =
 663                         container_of(entry, struct jset_entry_clock, entry);
 664
 665                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 666         }
 667         }
 668
 669         return ret;
 670 }
 671
 672 static int journal_replay_early(struct bch_fs *c,
 673                                 struct bch_sb_field_clean *clean,
 674                                 struct list_head *journal)
 675 {
 676         struct journal_replay *i;
 677         struct jset_entry *entry;
 678         int ret;
 679
 680         if (clean) {
 681                 for (entry = clean->start;
 682                      entry != vstruct_end(&clean->field);
 683                      entry = vstruct_next(entry)) {
 684                         ret = journal_replay_entry_early(c, entry);
 685                         if (ret)
 686                                 return ret;
 687                 }
 688         } else {
 689                 list_for_each_entry(i, journal, list) {
 690                         if (i->ignore)
 691                                 continue;
 692
 693                         vstruct_for_each(&i->j, entry) {
 694                                 ret = journal_replay_entry_early(c, entry);
 695                                 if (ret)
 696                                         return ret;
 697                         }
 698                 }
 699         }
 700
 701         bch2_fs_usage_initialize(c);
 702
 703         return 0;
 704 }
 705
 706 /* sb clean section: */
 707
 708 static struct bkey_i *btree_root_find(struct bch_fs *c,
 709                                       struct bch_sb_field_clean *clean,
 710                                       struct jset *j,
 711                                       enum btree_id id, unsigned *level)
 712 {
 713         struct bkey_i *k;
 714         struct jset_entry *entry, *start, *end;
 715
 716         if (clean) {
 717                 start = clean->start;
 718                 end = vstruct_end(&clean->field);
 719         } else {
 720                 start = j->start;
 721                 end = vstruct_last(j);
 722         }
 723
 724         for (entry = start; entry < end; entry = vstruct_next(entry))
 725                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 726                     entry->btree_id == id)
 727                         goto found;
 728
 729         return NULL;
 730 found:
 731         if (!entry->u64s)
 732                 return ERR_PTR(-EINVAL);
 733
 734         k = entry->start;
 735         *level = entry->level;
 736         return k;
 737 }
 738
 739 static int verify_superblock_clean(struct bch_fs *c,
 740                                    struct bch_sb_field_clean **cleanp,
 741                                    struct jset *j)
 742 {
 743         unsigned i;
 744         struct bch_sb_field_clean *clean = *cleanp;
 745         int ret = 0;
 746
 747         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 748                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 749                         le64_to_cpu(clean->journal_seq),
 750                         le64_to_cpu(j->seq))) {
 751                 kfree(clean);
 752                 *cleanp = NULL;
 753                 return 0;
 754         }
 755
 756         for (i = 0; i < BTREE_ID_NR; i++) {
 757                 char buf1[200], buf2[200];
 758                 struct bkey_i *k1, *k2;
 759                 unsigned l1 = 0, l2 = 0;
 760
 761                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 762                 k2 = btree_root_find(c, NULL, j, i, &l2);
 763
 764                 if (!k1 && !k2)
 765                         continue;
 766
 767                 mustfix_fsck_err_on(!k1 || !k2 ||
 768                                     IS_ERR(k1) ||
 769                                     IS_ERR(k2) ||
 770                                     k1->k.u64s != k2->k.u64s ||
 771                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 772                                     l1 != l2, c,
 773                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 774                         "sb:      l=%u %s\n"
 775                         "journal: l=%u %s\n", i,
 776                         l1, (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(k1)), buf1),
 777                         l2, (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(k2)), buf2));
 778         }
 779 fsck_err:
 780         return ret;
 781 }
 782
 783 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 784 {
 785         struct bch_sb_field_clean *clean, *sb_clean;
 786         int ret;
 787
 788         mutex_lock(&c->sb_lock);
 789         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 790
 791         if (fsck_err_on(!sb_clean, c,
 792                         "superblock marked clean but clean section not present")) {
 793                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 794                 c->sb.clean = false;
 795                 mutex_unlock(&c->sb_lock);
 796                 return NULL;
 797         }
 798
 799         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 800                         GFP_KERNEL);
 801         if (!clean) {
 802                 mutex_unlock(&c->sb_lock);
 803                 return ERR_PTR(-ENOMEM);
 804         }
 805
 806         ret = bch2_sb_clean_validate(c, clean, READ);
 807         if (ret) {
 808                 mutex_unlock(&c->sb_lock);
 809                 return ERR_PTR(ret);
 810         }
 811
 812         mutex_unlock(&c->sb_lock);
 813
 814         return clean;
 815 fsck_err:
 816         mutex_unlock(&c->sb_lock);
 817         return ERR_PTR(ret);
 818 }
 819
 820 static int read_btree_roots(struct bch_fs *c)
 821 {
 822         unsigned i;
 823         int ret = 0;
 824
 825         for (i = 0; i < BTREE_ID_NR; i++) {
 826                 struct btree_root *r = &c->btree_roots[i];
 827
 828                 if (!r->alive)
 829                         continue;
 830
 831                 if (i == BTREE_ID_alloc &&
 832                     c->opts.reconstruct_alloc) {
 833                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 834                         continue;
 835                 }
 836
 837                 if (r->error) {
 838                         __fsck_err(c, i == BTREE_ID_alloc
 839                                    ? FSCK_CAN_IGNORE : 0,
 840                                    "invalid btree root %s",
 841                                    bch2_btree_ids[i]);
 842                         if (i == BTREE_ID_alloc)
 843                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 844                 }
 845
 846                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 847                 if (ret) {
 848                         __fsck_err(c, i == BTREE_ID_alloc
 849                                    ? FSCK_CAN_IGNORE : 0,
 850                                    "error reading btree root %s",
 851                                    bch2_btree_ids[i]);
 852                         if (i == BTREE_ID_alloc)
 853                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 854                 }
 855         }
 856
 857         for (i = 0; i < BTREE_ID_NR; i++)
 858                 if (!c->btree_roots[i].b)
 859                         bch2_btree_root_alloc(c, i);
 860 fsck_err:
 861         return ret;
 862 }
 863
 864 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
 865 {
 866         struct bkey_i_snapshot  root_snapshot;
 867         struct bkey_i_subvolume root_volume;
 868         int ret;
 869
 870         bkey_snapshot_init(&root_snapshot.k_i);
 871         root_snapshot.k.p.offset = U32_MAX;
 872         root_snapshot.v.flags   = 0;
 873         root_snapshot.v.parent  = 0;
 874         root_snapshot.v.subvol  = BCACHEFS_ROOT_SUBVOL;
 875         root_snapshot.v.pad     = 0;
 876         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
 877
 878         ret = bch2_btree_insert(c, BTREE_ID_snapshots,
 879                                 &root_snapshot.k_i,
 880                                 NULL, NULL, 0);
 881         if (ret)
 882                 return ret;
 883
 884
 885         bkey_subvolume_init(&root_volume.k_i);
 886         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
 887         root_volume.v.flags     = 0;
 888         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
 889         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
 890
 891         ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
 892                                 &root_volume.k_i,
 893                                 NULL, NULL, 0);
 894         if (ret)
 895                 return ret;
 896
 897         return 0;
 898 }
 899
 900 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
 901 {
 902         struct btree_iter iter;
 903         struct bkey_s_c k;
 904         struct bch_inode_unpacked inode;
 905         int ret;
 906
 907         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
 908                              SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
 909         k = bch2_btree_iter_peek_slot(&iter);
 910         ret = bkey_err(k);
 911         if (ret)
 912                 goto err;
 913
 914         if (!bkey_is_inode(k.k)) {
 915                 bch_err(trans->c, "root inode not found");
 916                 ret = -ENOENT;
 917                 goto err;
 918         }
 919
 920         ret = bch2_inode_unpack(k, &inode);
 921         BUG_ON(ret);
 922
 923         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
 924
 925         ret = bch2_inode_write(trans, &iter, &inode);
 926 err:
 927         bch2_trans_iter_exit(trans, &iter);
 928         return ret;
 929 }
 930
 931 int bch2_fs_recovery(struct bch_fs *c)
 932 {
 933         const char *err = "cannot allocate memory";
 934         struct bch_sb_field_clean *clean = NULL;
 935         struct jset *last_journal_entry = NULL;
 936         u64 blacklist_seq, journal_seq;
 937         bool write_sb = false;
 938         int ret = 0;
 939
 940         if (c->sb.clean)
 941                 clean = read_superblock_clean(c);
 942         ret = PTR_ERR_OR_ZERO(clean);
 943         if (ret)
 944                 goto err;
 945
 946         if (c->sb.clean)
 947                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 948                          le64_to_cpu(clean->journal_seq));
 949         else
 950                 bch_info(c, "recovering from unclean shutdown");
 951
 952         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
 953                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
 954                 ret = -EINVAL;
 955                 goto err;
 956         }
 957
 958         if (!c->sb.clean &&
 959             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
 960                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
 961                 ret = -EINVAL;
 962                 goto err;
 963         }
 964
 965         if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
 966                 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
 967                 ret = -EINVAL;
 968                 goto err;
 969         }
 970
 971         if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
 972                 bch_info(c, "alloc_v2 feature bit not set, fsck required");
 973                 c->opts.fsck = true;
 974                 c->opts.fix_errors = FSCK_OPT_YES;
 975         }
 976
 977         if (!c->replicas.entries ||
 978             c->opts.rebuild_replicas) {
 979                 bch_info(c, "building replicas info");
 980                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 981         }
 982
 983         if (!c->opts.nochanges) {
 984                 if (c->sb.version < bcachefs_metadata_version_inode_backpointers) {
 985                         bch_info(c, "version prior to inode backpointers, upgrade and fsck required");
 986                         c->opts.version_upgrade = true;
 987                         c->opts.fsck            = true;
 988                         c->opts.fix_errors      = FSCK_OPT_YES;
 989                 } else if (c->sb.version < bcachefs_metadata_version_subvol_dirent) {
 990                         bch_info(c, "filesystem version is prior to subvol_dirent - upgrading");
 991                         c->opts.version_upgrade = true;
 992                         c->opts.fsck            = true;
 993                 } else if (c->sb.version < bcachefs_metadata_version_inode_v2) {
 994                         bch_info(c, "filesystem version is prior to inode_v2 - upgrading");
 995                         c->opts.version_upgrade = true;
 996                 }
 997         }
 998
 999         ret = bch2_blacklist_table_initialize(c);
1000         if (ret) {
1001                 bch_err(c, "error initializing blacklist table");
1002                 goto err;
1003         }
1004
1005         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1006                 struct journal_replay *i;
1007
1008                 bch_verbose(c, "starting journal read");
1009                 ret = bch2_journal_read(c, &c->journal_entries,
1010                                         &blacklist_seq, &journal_seq);
1011                 if (ret)
1012                         goto err;
1013
1014                 list_for_each_entry_reverse(i, &c->journal_entries, list)
1015                         if (!i->ignore) {
1016                                 last_journal_entry = &i->j;
1017                                 break;
1018                         }
1019
1020                 if (mustfix_fsck_err_on(c->sb.clean &&
1021                                         last_journal_entry &&
1022                                         !journal_entry_empty(last_journal_entry), c,
1023                                 "filesystem marked clean but journal not empty")) {
1024                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1025                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1026                         c->sb.clean = false;
1027                 }
1028
1029                 if (!last_journal_entry) {
1030                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
1031                         goto use_clean;
1032                 }
1033
1034                 c->journal_keys = journal_keys_sort(&c->journal_entries);
1035                 if (!c->journal_keys.d) {
1036                         ret = -ENOMEM;
1037                         goto err;
1038                 }
1039
1040                 if (c->sb.clean && last_journal_entry) {
1041                         ret = verify_superblock_clean(c, &clean,
1042                                                       last_journal_entry);
1043                         if (ret)
1044                                 goto err;
1045                 }
1046         } else {
1047 use_clean:
1048                 if (!clean) {
1049                         bch_err(c, "no superblock clean section found");
1050                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1051                         goto err;
1052
1053                 }
1054                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1055         }
1056
1057         if (c->opts.reconstruct_alloc) {
1058                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1059                 drop_alloc_keys(&c->journal_keys);
1060         }
1061
1062         zero_out_btree_mem_ptr(&c->journal_keys);
1063
1064         ret = journal_replay_early(c, clean, &c->journal_entries);
1065         if (ret)
1066                 goto err;
1067
1068         /*
1069          * After an unclean shutdown, skip then next few journal sequence
1070          * numbers as they may have been referenced by btree writes that
1071          * happened before their corresponding journal writes - those btree
1072          * writes need to be ignored, by skipping and blacklisting the next few
1073          * journal sequence numbers:
1074          */
1075         if (!c->sb.clean)
1076                 journal_seq += 8;
1077
1078         if (blacklist_seq != journal_seq) {
1079                 ret = bch2_journal_seq_blacklist_add(c,
1080                                         blacklist_seq, journal_seq);
1081                 if (ret) {
1082                         bch_err(c, "error creating new journal seq blacklist entry");
1083                         goto err;
1084                 }
1085         }
1086
1087         ret = bch2_fs_journal_start(&c->journal, journal_seq,
1088                                     &c->journal_entries);
1089         if (ret)
1090                 goto err;
1091
1092         ret = read_btree_roots(c);
1093         if (ret)
1094                 goto err;
1095
1096         bch_verbose(c, "starting alloc read");
1097         err = "error reading allocation information";
1098         ret = bch2_alloc_read(c);
1099         if (ret)
1100                 goto err;
1101         bch_verbose(c, "alloc read done");
1102
1103         bch_verbose(c, "starting stripes_read");
1104         err = "error reading stripes";
1105         ret = bch2_stripes_read(c);
1106         if (ret)
1107                 goto err;
1108         bch_verbose(c, "stripes_read done");
1109
1110         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1111
1112         /*
1113          * If we're not running fsck, this ensures bch2_fsck_err() calls are
1114          * instead interpreted as bch2_inconsistent_err() calls:
1115          */
1116         if (!c->opts.fsck)
1117                 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1118
1119         if (c->opts.fsck ||
1120             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)) ||
1121             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_metadata)) ||
1122             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1123                 bool metadata_only = c->opts.norecovery;
1124
1125                 bch_info(c, "starting mark and sweep");
1126                 err = "error in mark and sweep";
1127                 ret = bch2_gc(c, true, metadata_only);
1128                 if (ret)
1129                         goto err;
1130                 bch_verbose(c, "mark and sweep done");
1131         }
1132
1133         bch2_stripes_heap_start(c);
1134
1135         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1136         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1137
1138         /*
1139          * Skip past versions that might have possibly been used (as nonces),
1140          * but hadn't had their pointers written:
1141          */
1142         if (c->sb.encryption_type && !c->sb.clean)
1143                 atomic64_add(1 << 16, &c->key_version);
1144
1145         if (c->opts.norecovery)
1146                 goto out;
1147
1148         bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1149         err = "journal replay failed";
1150         ret = bch2_journal_replay(c);
1151         if (ret)
1152                 goto err;
1153         if (c->opts.verbose || !c->sb.clean)
1154                 bch_info(c, "journal replay done");
1155
1156         if (test_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags) &&
1157             !c->opts.nochanges) {
1158                 /*
1159                  * note that even when filesystem was clean there might be work
1160                  * to do here, if we ran gc (because of fsck) which recalculated
1161                  * oldest_gen:
1162                  */
1163                 bch_verbose(c, "writing allocation info");
1164                 err = "error writing out alloc info";
1165                 ret = bch2_alloc_write_all(c, BTREE_INSERT_LAZY_RW);
1166                 if (ret) {
1167                         bch_err(c, "error writing alloc info");
1168                         goto err;
1169                 }
1170                 bch_verbose(c, "alloc write done");
1171         }
1172
1173         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1174                 bch2_fs_lazy_rw(c);
1175
1176                 err = "error creating root snapshot node";
1177                 ret = bch2_fs_initialize_subvolumes(c);
1178                 if (ret)
1179                         goto err;
1180         }
1181
1182         bch_verbose(c, "reading snapshots table");
1183         err = "error reading snapshots table";
1184         ret = bch2_fs_snapshots_start(c);
1185         if (ret)
1186                 goto err;
1187         bch_verbose(c, "reading snapshots done");
1188
1189         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1190                 /* set bi_subvol on root inode */
1191                 err = "error upgrade root inode for subvolumes";
1192                 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1193                                     bch2_fs_upgrade_for_subvolumes(&trans));
1194                 if (ret)
1195                         goto err;
1196         }
1197
1198         if (c->opts.fsck) {
1199                 bch_info(c, "starting fsck");
1200                 err = "error in fsck";
1201                 ret = bch2_fsck_full(c);
1202                 if (ret)
1203                         goto err;
1204                 bch_verbose(c, "fsck done");
1205         } else if (!c->sb.clean) {
1206                 bch_verbose(c, "checking for deleted inodes");
1207                 err = "error in recovery";
1208                 ret = bch2_fsck_walk_inodes_only(c);
1209                 if (ret)
1210                         goto err;
1211                 bch_verbose(c, "check inodes done");
1212         }
1213
1214         if (enabled_qtypes(c)) {
1215                 bch_verbose(c, "reading quotas");
1216                 ret = bch2_fs_quota_read(c);
1217                 if (ret)
1218                         goto err;
1219                 bch_verbose(c, "quotas done");
1220         }
1221
1222         mutex_lock(&c->sb_lock);
1223         if (c->opts.version_upgrade) {
1224                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1225                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1226                 write_sb = true;
1227         }
1228
1229         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1230                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1231                 write_sb = true;
1232         }
1233
1234         if (c->opts.fsck &&
1235             !test_bit(BCH_FS_ERROR, &c->flags) &&
1236             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1237                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1238                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1239                 write_sb = true;
1240         }
1241
1242         if (write_sb)
1243                 bch2_write_super(c);
1244         mutex_unlock(&c->sb_lock);
1245
1246         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1247             !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1248             le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) {
1249                 struct bch_move_stats stats;
1250
1251                 bch_move_stats_init(&stats, "recovery");
1252
1253                 bch_info(c, "scanning for old btree nodes");
1254                 ret = bch2_fs_read_write(c);
1255                 if (ret)
1256                         goto err;
1257
1258                 ret = bch2_scan_old_btree_nodes(c, &stats);
1259                 if (ret)
1260                         goto err;
1261                 bch_info(c, "scanning for old btree nodes done");
1262         }
1263
1264         if (c->journal_seq_blacklist_table &&
1265             c->journal_seq_blacklist_table->nr > 128)
1266                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1267
1268         ret = 0;
1269 out:
1270         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1271         bch2_flush_fsck_errs(c);
1272
1273         if (!c->opts.keep_journal) {
1274                 bch2_journal_keys_free(&c->journal_keys);
1275                 bch2_journal_entries_free(&c->journal_entries);
1276         }
1277         kfree(clean);
1278         if (ret)
1279                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1280         else
1281                 bch_verbose(c, "ret %i", ret);
1282         return ret;
1283 err:
1284 fsck_err:
1285         bch2_fs_emergency_read_only(c);
1286         goto out;
1287 }
1288
1289 int bch2_fs_initialize(struct bch_fs *c)
1290 {
1291         struct bch_inode_unpacked root_inode, lostfound_inode;
1292         struct bkey_inode_buf packed_inode;
1293         struct qstr lostfound = QSTR("lost+found");
1294         const char *err = "cannot allocate memory";
1295         struct bch_dev *ca;
1296         LIST_HEAD(journal);
1297         unsigned i;
1298         int ret;
1299
1300         bch_notice(c, "initializing new filesystem");
1301
1302         mutex_lock(&c->sb_lock);
1303         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1304         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1305
1306         if (c->opts.version_upgrade) {
1307                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1308                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1309                 bch2_write_super(c);
1310         }
1311         mutex_unlock(&c->sb_lock);
1312
1313         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1314         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1315         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1316
1317         for (i = 0; i < BTREE_ID_NR; i++)
1318                 bch2_btree_root_alloc(c, i);
1319
1320         err = "unable to allocate journal buckets";
1321         for_each_online_member(ca, c, i) {
1322                 ret = bch2_dev_journal_alloc(ca);
1323                 if (ret) {
1324                         percpu_ref_put(&ca->io_ref);
1325                         goto err;
1326                 }
1327         }
1328
1329         /*
1330          * journal_res_get() will crash if called before this has
1331          * set up the journal.pin FIFO and journal.cur pointer:
1332          */
1333         bch2_fs_journal_start(&c->journal, 1, &journal);
1334         bch2_journal_set_replay_done(&c->journal);
1335
1336         err = "error going read-write";
1337         ret = bch2_fs_read_write_early(c);
1338         if (ret)
1339                 goto err;
1340
1341         /*
1342          * Write out the superblock and journal buckets, now that we can do
1343          * btree updates
1344          */
1345         err = "error marking superblock and journal";
1346         for_each_member_device(ca, c, i) {
1347                 ret = bch2_trans_mark_dev_sb(c, ca);
1348                 if (ret) {
1349                         percpu_ref_put(&ca->ref);
1350                         goto err;
1351                 }
1352
1353                 ca->new_fs_bucket_idx = 0;
1354         }
1355
1356         err = "error creating root snapshot node";
1357         ret = bch2_fs_initialize_subvolumes(c);
1358         if (ret)
1359                 goto err;
1360
1361         bch_verbose(c, "reading snapshots table");
1362         err = "error reading snapshots table";
1363         ret = bch2_fs_snapshots_start(c);
1364         if (ret)
1365                 goto err;
1366         bch_verbose(c, "reading snapshots done");
1367
1368         bch2_inode_init(c, &root_inode, 0, 0,
1369                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1370         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1371         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1372         bch2_inode_pack(c, &packed_inode, &root_inode);
1373         packed_inode.inode.k.p.snapshot = U32_MAX;
1374
1375         err = "error creating root directory";
1376         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1377                                 &packed_inode.inode.k_i,
1378                                 NULL, NULL, 0);
1379         if (ret)
1380                 goto err;
1381
1382         bch2_inode_init_early(c, &lostfound_inode);
1383
1384         err = "error creating lost+found";
1385         ret = bch2_trans_do(c, NULL, NULL, 0,
1386                 bch2_create_trans(&trans,
1387                                   BCACHEFS_ROOT_SUBVOL_INUM,
1388                                   &root_inode, &lostfound_inode,
1389                                   &lostfound,
1390                                   0, 0, S_IFDIR|0700, 0,
1391                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1392         if (ret) {
1393                 bch_err(c, "error creating lost+found");
1394                 goto err;
1395         }
1396
1397         if (enabled_qtypes(c)) {
1398                 ret = bch2_fs_quota_read(c);
1399                 if (ret)
1400                         goto err;
1401         }
1402
1403         err = "error writing first journal entry";
1404         ret = bch2_journal_flush(&c->journal);
1405         if (ret)
1406                 goto err;
1407
1408         mutex_lock(&c->sb_lock);
1409         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1410         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1411
1412         bch2_write_super(c);
1413         mutex_unlock(&c->sb_lock);
1414
1415         return 0;
1416 err:
1417         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1418         return ret;
1419 }