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                              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(struct journal_key *l, struct journal_key *r)
  70 {
  71         return (cmp_int(l->btree_id,    r->btree_id) ?:
  72                 cmp_int(l->level,       r->level) ?:
  73                 bpos_cmp(l->k->k.p,     r->k->k.p));
  74 }
  75
  76 static size_t journal_key_search(struct journal_keys *journal_keys,
  77                                  enum btree_id id, unsigned level,
  78                                  struct bpos pos)
  79 {
  80         size_t l = 0, r = journal_keys->nr, m;
  81
  82         while (l < r) {
  83                 m = l + ((r - l) >> 1);
  84                 if (__journal_key_cmp(id, level, pos, &journal_keys->d[m]) > 0)
  85                         l = m + 1;
  86                 else
  87                         r = m;
  88         }
  89
  90         BUG_ON(l < journal_keys->nr &&
  91                __journal_key_cmp(id, level, pos, &journal_keys->d[l]) > 0);
  92
  93         BUG_ON(l &&
  94                __journal_key_cmp(id, level, pos, &journal_keys->d[l - 1]) <= 0);
  95
  96         return l;
  97 }
  98
  99 static void journal_iter_fix(struct bch_fs *c, struct journal_iter *iter, unsigned idx)
 100 {
 101         struct bkey_i *n = iter->keys->d[idx].k;
 102         struct btree_and_journal_iter *biter =
 103                 container_of(iter, struct btree_and_journal_iter, journal);
 104
 105         if (iter->idx > idx ||
 106             (iter->idx == idx &&
 107              biter->last &&
 108              bpos_cmp(n->k.p, biter->unpacked.p) <= 0))
 109                 iter->idx++;
 110 }
 111
 112 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
 113                             unsigned level, struct bkey_i *k)
 114 {
 115         struct journal_key n = {
 116                 .btree_id       = id,
 117                 .level          = level,
 118                 .allocated      = true
 119         };
 120         struct journal_keys *keys = &c->journal_keys;
 121         struct journal_iter *iter;
 122         unsigned idx = journal_key_search(keys, id, level, k->k.p);
 123
 124         if (keys->nr == keys->size) {
 125                 struct journal_keys new_keys = {
 126                         .nr                     = keys->nr,
 127                         .size                   = keys->size * 2,
 128                         .journal_seq_base       = keys->journal_seq_base,
 129                 };
 130
 131                 new_keys.d = kvmalloc(sizeof(new_keys.d[0]) * new_keys.size, GFP_KERNEL);
 132                 if (!new_keys.d) {
 133                         bch_err(c, "%s: error allocating new key array (size %zu)",
 134                                 __func__, new_keys.size);
 135                         return -ENOMEM;
 136                 }
 137
 138                 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
 139                 kvfree(keys->d);
 140                 *keys = new_keys;
 141         }
 142
 143         n.k = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
 144         if (!n.k)
 145                 return -ENOMEM;
 146
 147         bkey_copy(n.k, k);
 148
 149         if (idx < keys->nr &&
 150             journal_key_cmp(&n, &keys->d[idx]) == 0) {
 151                 if (keys->d[idx].allocated)
 152                         kfree(keys->d[idx].k);
 153                 keys->d[idx] = n;
 154         } else {
 155                 array_insert_item(keys->d, keys->nr, idx, n);
 156
 157                 list_for_each_entry(iter, &c->journal_iters, list)
 158                         journal_iter_fix(c, iter, idx);
 159         }
 160
 161         return 0;
 162 }
 163
 164 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
 165                             unsigned level, struct bpos pos)
 166 {
 167         struct bkey_i whiteout;
 168
 169         bkey_init(&whiteout.k);
 170         whiteout.k.p = pos;
 171
 172         return bch2_journal_key_insert(c, id, level, &whiteout);
 173 }
 174
 175 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
 176 {
 177         struct journal_key *k = iter->idx - iter->keys->nr
 178                 ? iter->keys->d + iter->idx : NULL;
 179
 180         if (k &&
 181             k->btree_id == iter->btree_id &&
 182             k->level    == iter->level)
 183                 return k->k;
 184
 185         iter->idx = iter->keys->nr;
 186         return NULL;
 187 }
 188
 189 static void bch2_journal_iter_advance(struct journal_iter *iter)
 190 {
 191         if (iter->idx < iter->keys->nr)
 192                 iter->idx++;
 193 }
 194
 195 static void bch2_journal_iter_exit(struct journal_iter *iter)
 196 {
 197         list_del(&iter->list);
 198 }
 199
 200 static void bch2_journal_iter_init(struct bch_fs *c,
 201                                    struct journal_iter *iter,
 202                                    enum btree_id id, unsigned level,
 203                                    struct bpos pos)
 204 {
 205         iter->btree_id  = id;
 206         iter->level     = level;
 207         iter->keys      = &c->journal_keys;
 208         iter->idx       = journal_key_search(&c->journal_keys, id, level, pos);
 209         list_add(&iter->list, &c->journal_iters);
 210 }
 211
 212 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 213 {
 214         return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 215                                                 iter->b, &iter->unpacked);
 216 }
 217
 218 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 219 {
 220         bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 221 }
 222
 223 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 224 {
 225         switch (iter->last) {
 226         case none:
 227                 break;
 228         case btree:
 229                 bch2_journal_iter_advance_btree(iter);
 230                 break;
 231         case journal:
 232                 bch2_journal_iter_advance(&iter->journal);
 233                 break;
 234         }
 235
 236         iter->last = none;
 237 }
 238
 239 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 240 {
 241         struct bkey_s_c ret;
 242
 243         while (1) {
 244                 struct bkey_s_c btree_k         =
 245                         bch2_journal_iter_peek_btree(iter);
 246                 struct bkey_s_c journal_k       =
 247                         bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
 248
 249                 if (btree_k.k && journal_k.k) {
 250                         int cmp = bpos_cmp(btree_k.k->p, journal_k.k->p);
 251
 252                         if (!cmp)
 253                                 bch2_journal_iter_advance_btree(iter);
 254
 255                         iter->last = cmp < 0 ? btree : journal;
 256                 } else if (btree_k.k) {
 257                         iter->last = btree;
 258                 } else if (journal_k.k) {
 259                         iter->last = journal;
 260                 } else {
 261                         iter->last = none;
 262                         return bkey_s_c_null;
 263                 }
 264
 265                 ret = iter->last == journal ? journal_k : btree_k;
 266
 267                 if (iter->b &&
 268                     bpos_cmp(ret.k->p, iter->b->data->max_key) > 0) {
 269                         iter->journal.idx = iter->journal.keys->nr;
 270                         iter->last = none;
 271                         return bkey_s_c_null;
 272                 }
 273
 274                 if (!bkey_deleted(ret.k))
 275                         break;
 276
 277                 bch2_btree_and_journal_iter_advance(iter);
 278         }
 279
 280         return ret;
 281 }
 282
 283 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 284 {
 285         bch2_btree_and_journal_iter_advance(iter);
 286
 287         return bch2_btree_and_journal_iter_peek(iter);
 288 }
 289
 290 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
 291 {
 292         bch2_journal_iter_exit(&iter->journal);
 293 }
 294
 295 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 296                                                 struct bch_fs *c,
 297                                                 struct btree *b)
 298 {
 299         memset(iter, 0, sizeof(*iter));
 300
 301         iter->b = b;
 302         bch2_btree_node_iter_init_from_start(&iter->node_iter, iter->b);
 303         bch2_journal_iter_init(c, &iter->journal,
 304                                b->c.btree_id, b->c.level, b->data->min_key);
 305 }
 306
 307 /* Walk btree, overlaying keys from the journal: */
 308
 309 static void btree_and_journal_iter_prefetch(struct bch_fs *c, struct btree *b,
 310                                            struct btree_and_journal_iter iter)
 311 {
 312         unsigned i = 0, nr = b->c.level > 1 ? 2 : 16;
 313         struct bkey_s_c k;
 314         struct bkey_buf tmp;
 315
 316         BUG_ON(!b->c.level);
 317
 318         bch2_bkey_buf_init(&tmp);
 319
 320         while (i < nr &&
 321                (k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 322                 bch2_bkey_buf_reassemble(&tmp, c, k);
 323
 324                 bch2_btree_node_prefetch(c, NULL, NULL, tmp.k,
 325                                         b->c.btree_id, b->c.level - 1);
 326
 327                 bch2_btree_and_journal_iter_advance(&iter);
 328                 i++;
 329         }
 330
 331         bch2_bkey_buf_exit(&tmp, c);
 332 }
 333
 334 static int bch2_btree_and_journal_walk_recurse(struct btree_trans *trans, struct btree *b,
 335                                 enum btree_id btree_id,
 336                                 btree_walk_key_fn key_fn)
 337 {
 338         struct bch_fs *c = trans->c;
 339         struct btree_and_journal_iter iter;
 340         struct bkey_s_c k;
 341         struct bkey_buf tmp;
 342         struct btree *child;
 343         int ret = 0;
 344
 345         bch2_bkey_buf_init(&tmp);
 346         bch2_btree_and_journal_iter_init_node_iter(&iter, c, b);
 347
 348         while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 349                 if (b->c.level) {
 350                         bch2_bkey_buf_reassemble(&tmp, c, k);
 351
 352                         child = bch2_btree_node_get_noiter(c, tmp.k,
 353                                                 b->c.btree_id, b->c.level - 1,
 354                                                 false);
 355
 356                         ret = PTR_ERR_OR_ZERO(child);
 357                         if (ret)
 358                                 break;
 359
 360                         btree_and_journal_iter_prefetch(c, b, iter);
 361
 362                         ret = bch2_btree_and_journal_walk_recurse(trans, child,
 363                                         btree_id, key_fn);
 364                         six_unlock_read(&child->c.lock);
 365                 } else {
 366                         ret = key_fn(trans, k);
 367                 }
 368
 369                 if (ret)
 370                         break;
 371
 372                 bch2_btree_and_journal_iter_advance(&iter);
 373         }
 374
 375         bch2_btree_and_journal_iter_exit(&iter);
 376         bch2_bkey_buf_exit(&tmp, c);
 377         return ret;
 378 }
 379
 380 int bch2_btree_and_journal_walk(struct btree_trans *trans, enum btree_id btree_id,
 381                                 btree_walk_key_fn key_fn)
 382 {
 383         struct bch_fs *c = trans->c;
 384         struct btree *b = c->btree_roots[btree_id].b;
 385         int ret = 0;
 386
 387         if (btree_node_fake(b))
 388                 return 0;
 389
 390         six_lock_read(&b->c.lock, NULL, NULL);
 391         ret = bch2_btree_and_journal_walk_recurse(trans, b, btree_id, key_fn);
 392         six_unlock_read(&b->c.lock);
 393
 394         return ret;
 395 }
 396
 397 /* sort and dedup all keys in the journal: */
 398
 399 void bch2_journal_entries_free(struct list_head *list)
 400 {
 401
 402         while (!list_empty(list)) {
 403                 struct journal_replay *i =
 404                         list_first_entry(list, struct journal_replay, list);
 405                 list_del(&i->list);
 406                 kvpfree(i, offsetof(struct journal_replay, j) +
 407                         vstruct_bytes(&i->j));
 408         }
 409 }
 410
 411 /*
 412  * When keys compare equal, oldest compares first:
 413  */
 414 static int journal_sort_key_cmp(const void *_l, const void *_r)
 415 {
 416         const struct journal_key *l = _l;
 417         const struct journal_key *r = _r;
 418
 419         return  cmp_int(l->btree_id,    r->btree_id) ?:
 420                 cmp_int(l->level,       r->level) ?:
 421                 bpos_cmp(l->k->k.p, r->k->k.p) ?:
 422                 cmp_int(l->journal_seq, r->journal_seq) ?:
 423                 cmp_int(l->journal_offset, r->journal_offset);
 424 }
 425
 426 void bch2_journal_keys_free(struct journal_keys *keys)
 427 {
 428         struct journal_key *i;
 429
 430         for (i = keys->d; i < keys->d + keys->nr; i++)
 431                 if (i->allocated)
 432                         kfree(i->k);
 433
 434         kvfree(keys->d);
 435         keys->d = NULL;
 436         keys->nr = 0;
 437 }
 438
 439 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 440 {
 441         struct journal_replay *i;
 442         struct jset_entry *entry;
 443         struct bkey_i *k, *_n;
 444         struct journal_keys keys = { NULL };
 445         struct journal_key *src, *dst;
 446         size_t nr_keys = 0;
 447
 448         if (list_empty(journal_entries))
 449                 return keys;
 450
 451         list_for_each_entry(i, journal_entries, list) {
 452                 if (i->ignore)
 453                         continue;
 454
 455                 if (!keys.journal_seq_base)
 456                         keys.journal_seq_base = le64_to_cpu(i->j.seq);
 457
 458                 for_each_jset_key(k, _n, entry, &i->j)
 459                         nr_keys++;
 460         }
 461
 462         keys.size = roundup_pow_of_two(nr_keys);
 463
 464         keys.d = kvmalloc(sizeof(keys.d[0]) * keys.size, GFP_KERNEL);
 465         if (!keys.d)
 466                 goto err;
 467
 468         list_for_each_entry(i, journal_entries, list) {
 469                 if (i->ignore)
 470                         continue;
 471
 472                 BUG_ON(le64_to_cpu(i->j.seq) - keys.journal_seq_base > U32_MAX);
 473
 474                 for_each_jset_key(k, _n, entry, &i->j)
 475                         keys.d[keys.nr++] = (struct journal_key) {
 476                                 .btree_id       = entry->btree_id,
 477                                 .level          = entry->level,
 478                                 .k              = k,
 479                                 .journal_seq    = le64_to_cpu(i->j.seq) -
 480                                         keys.journal_seq_base,
 481                                 .journal_offset = k->_data - i->j._data,
 482                         };
 483         }
 484
 485         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 486
 487         src = dst = keys.d;
 488         while (src < keys.d + keys.nr) {
 489                 while (src + 1 < keys.d + keys.nr &&
 490                        src[0].btree_id  == src[1].btree_id &&
 491                        src[0].level     == src[1].level &&
 492                        !bpos_cmp(src[0].k->k.p, src[1].k->k.p))
 493                         src++;
 494
 495                 *dst++ = *src++;
 496         }
 497
 498         keys.nr = dst - keys.d;
 499 err:
 500         return keys;
 501 }
 502
 503 /* journal replay: */
 504
 505 static void replay_now_at(struct journal *j, u64 seq)
 506 {
 507         BUG_ON(seq < j->replay_journal_seq);
 508         BUG_ON(seq > j->replay_journal_seq_end);
 509
 510         while (j->replay_journal_seq < seq)
 511                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 512 }
 513
 514 static int __bch2_journal_replay_key(struct btree_trans *trans,
 515                                      struct journal_key *k)
 516 {
 517         struct btree_iter iter;
 518         unsigned iter_flags =
 519                 BTREE_ITER_INTENT|
 520                 BTREE_ITER_NOT_EXTENTS;
 521         int ret;
 522
 523         if (!k->level && k->btree_id == BTREE_ID_alloc)
 524                 iter_flags |= BTREE_ITER_CACHED|BTREE_ITER_CACHED_NOFILL;
 525
 526         bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 527                                   BTREE_MAX_DEPTH, k->level,
 528                                   iter_flags);
 529         ret   = bch2_btree_iter_traverse(&iter) ?:
 530                 bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
 531         bch2_trans_iter_exit(trans, &iter);
 532         return ret;
 533 }
 534
 535 static int bch2_journal_replay_key(struct bch_fs *c, struct journal_key *k)
 536 {
 537         unsigned commit_flags =
 538                 BTREE_INSERT_LAZY_RW|
 539                 BTREE_INSERT_NOFAIL|
 540                 BTREE_INSERT_JOURNAL_RESERVED;
 541
 542         if (!k->allocated)
 543                 commit_flags |= BTREE_INSERT_JOURNAL_REPLAY;
 544
 545         return bch2_trans_do(c, NULL, NULL, commit_flags,
 546                              __bch2_journal_replay_key(&trans, k));
 547 }
 548
 549 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 550 {
 551         const struct journal_key *l = _l;
 552         const struct journal_key *r = _r;
 553
 554         return  cmp_int(r->level,       l->level) ?:
 555                 cmp_int(l->journal_seq, r->journal_seq) ?:
 556                 cmp_int(l->btree_id,    r->btree_id) ?:
 557                 bpos_cmp(l->k->k.p,     r->k->k.p);
 558 }
 559
 560 static int bch2_journal_replay(struct bch_fs *c,
 561                                struct journal_keys keys)
 562 {
 563         struct journal *j = &c->journal;
 564         struct journal_key *i;
 565         u64 seq;
 566         int ret;
 567
 568         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
 569
 570         if (keys.nr)
 571                 replay_now_at(j, keys.journal_seq_base);
 572
 573         seq = j->replay_journal_seq;
 574
 575         /*
 576          * First replay updates to the alloc btree - these will only update the
 577          * btree key cache:
 578          */
 579         for_each_journal_key(keys, i) {
 580                 cond_resched();
 581
 582                 if (!i->level && i->btree_id == BTREE_ID_alloc) {
 583                         j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
 584                         ret = bch2_journal_replay_key(c, i);
 585                         if (ret)
 586                                 goto err;
 587                 }
 588         }
 589
 590         /*
 591          * Next replay updates to interior btree nodes:
 592          */
 593         for_each_journal_key(keys, i) {
 594                 cond_resched();
 595
 596                 if (i->level) {
 597                         j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
 598                         ret = bch2_journal_replay_key(c, i);
 599                         if (ret)
 600                                 goto err;
 601                 }
 602         }
 603
 604         /*
 605          * Now that the btree is in a consistent state, we can start journal
 606          * reclaim (which will be flushing entries from the btree key cache back
 607          * to the btree:
 608          */
 609         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
 610         set_bit(JOURNAL_RECLAIM_STARTED, &j->flags);
 611         journal_reclaim_kick(j);
 612
 613         j->replay_journal_seq = seq;
 614
 615         /*
 616          * Now replay leaf node updates:
 617          */
 618         for_each_journal_key(keys, i) {
 619                 cond_resched();
 620
 621                 if (i->level || i->btree_id == BTREE_ID_alloc)
 622                         continue;
 623
 624                 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 625
 626                 ret = bch2_journal_replay_key(c, i);
 627                 if (ret)
 628                         goto err;
 629         }
 630
 631         replay_now_at(j, j->replay_journal_seq_end);
 632         j->replay_journal_seq = 0;
 633
 634         bch2_journal_set_replay_done(j);
 635         bch2_journal_flush_all_pins(j);
 636         return bch2_journal_error(j);
 637 err:
 638         bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
 639                 ret, bch2_btree_ids[i->btree_id], i->level);
 640         return ret;
 641 }
 642
 643 /* journal replay early: */
 644
 645 static int journal_replay_entry_early(struct bch_fs *c,
 646                                       struct jset_entry *entry)
 647 {
 648         int ret = 0;
 649
 650         switch (entry->type) {
 651         case BCH_JSET_ENTRY_btree_root: {
 652                 struct btree_root *r;
 653
 654                 if (entry->btree_id >= BTREE_ID_NR) {
 655                         bch_err(c, "filesystem has unknown btree type %u",
 656                                 entry->btree_id);
 657                         return -EINVAL;
 658                 }
 659
 660                 r = &c->btree_roots[entry->btree_id];
 661
 662                 if (entry->u64s) {
 663                         r->level = entry->level;
 664                         bkey_copy(&r->key, &entry->start[0]);
 665                         r->error = 0;
 666                 } else {
 667                         r->error = -EIO;
 668                 }
 669                 r->alive = true;
 670                 break;
 671         }
 672         case BCH_JSET_ENTRY_usage: {
 673                 struct jset_entry_usage *u =
 674                         container_of(entry, struct jset_entry_usage, entry);
 675
 676                 switch (entry->btree_id) {
 677                 case FS_USAGE_RESERVED:
 678                         if (entry->level < BCH_REPLICAS_MAX)
 679                                 c->usage_base->persistent_reserved[entry->level] =
 680                                         le64_to_cpu(u->v);
 681                         break;
 682                 case FS_USAGE_INODES:
 683                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 684                         break;
 685                 case FS_USAGE_KEY_VERSION:
 686                         atomic64_set(&c->key_version,
 687                                      le64_to_cpu(u->v));
 688                         break;
 689                 }
 690
 691                 break;
 692         }
 693         case BCH_JSET_ENTRY_data_usage: {
 694                 struct jset_entry_data_usage *u =
 695                         container_of(entry, struct jset_entry_data_usage, entry);
 696
 697                 ret = bch2_replicas_set_usage(c, &u->r,
 698                                               le64_to_cpu(u->v));
 699                 break;
 700         }
 701         case BCH_JSET_ENTRY_dev_usage: {
 702                 struct jset_entry_dev_usage *u =
 703                         container_of(entry, struct jset_entry_dev_usage, entry);
 704                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 705                 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
 706                 unsigned nr_types = (bytes - sizeof(struct jset_entry_dev_usage)) /
 707                         sizeof(struct jset_entry_dev_usage_type);
 708                 unsigned i;
 709
 710                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 711                 ca->usage_base->buckets_unavailable     = le64_to_cpu(u->buckets_unavailable);
 712
 713                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 714                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 715                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 716                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 717                 }
 718
 719                 break;
 720         }
 721         case BCH_JSET_ENTRY_blacklist: {
 722                 struct jset_entry_blacklist *bl_entry =
 723                         container_of(entry, struct jset_entry_blacklist, entry);
 724
 725                 ret = bch2_journal_seq_blacklist_add(c,
 726                                 le64_to_cpu(bl_entry->seq),
 727                                 le64_to_cpu(bl_entry->seq) + 1);
 728                 break;
 729         }
 730         case BCH_JSET_ENTRY_blacklist_v2: {
 731                 struct jset_entry_blacklist_v2 *bl_entry =
 732                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 733
 734                 ret = bch2_journal_seq_blacklist_add(c,
 735                                 le64_to_cpu(bl_entry->start),
 736                                 le64_to_cpu(bl_entry->end) + 1);
 737                 break;
 738         }
 739         case BCH_JSET_ENTRY_clock: {
 740                 struct jset_entry_clock *clock =
 741                         container_of(entry, struct jset_entry_clock, entry);
 742
 743                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 744         }
 745         }
 746
 747         return ret;
 748 }
 749
 750 static int journal_replay_early(struct bch_fs *c,
 751                                 struct bch_sb_field_clean *clean,
 752                                 struct list_head *journal)
 753 {
 754         struct journal_replay *i;
 755         struct jset_entry *entry;
 756         int ret;
 757
 758         if (clean) {
 759                 for (entry = clean->start;
 760                      entry != vstruct_end(&clean->field);
 761                      entry = vstruct_next(entry)) {
 762                         ret = journal_replay_entry_early(c, entry);
 763                         if (ret)
 764                                 return ret;
 765                 }
 766         } else {
 767                 list_for_each_entry(i, journal, list) {
 768                         if (i->ignore)
 769                                 continue;
 770
 771                         vstruct_for_each(&i->j, entry) {
 772                                 ret = journal_replay_entry_early(c, entry);
 773                                 if (ret)
 774                                         return ret;
 775                         }
 776                 }
 777         }
 778
 779         bch2_fs_usage_initialize(c);
 780
 781         return 0;
 782 }
 783
 784 /* sb clean section: */
 785
 786 static struct bkey_i *btree_root_find(struct bch_fs *c,
 787                                       struct bch_sb_field_clean *clean,
 788                                       struct jset *j,
 789                                       enum btree_id id, unsigned *level)
 790 {
 791         struct bkey_i *k;
 792         struct jset_entry *entry, *start, *end;
 793
 794         if (clean) {
 795                 start = clean->start;
 796                 end = vstruct_end(&clean->field);
 797         } else {
 798                 start = j->start;
 799                 end = vstruct_last(j);
 800         }
 801
 802         for (entry = start; entry < end; entry = vstruct_next(entry))
 803                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 804                     entry->btree_id == id)
 805                         goto found;
 806
 807         return NULL;
 808 found:
 809         if (!entry->u64s)
 810                 return ERR_PTR(-EINVAL);
 811
 812         k = entry->start;
 813         *level = entry->level;
 814         return k;
 815 }
 816
 817 static int verify_superblock_clean(struct bch_fs *c,
 818                                    struct bch_sb_field_clean **cleanp,
 819                                    struct jset *j)
 820 {
 821         unsigned i;
 822         struct bch_sb_field_clean *clean = *cleanp;
 823         int ret = 0;
 824
 825         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 826                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 827                         le64_to_cpu(clean->journal_seq),
 828                         le64_to_cpu(j->seq))) {
 829                 kfree(clean);
 830                 *cleanp = NULL;
 831                 return 0;
 832         }
 833
 834         for (i = 0; i < BTREE_ID_NR; i++) {
 835                 char buf1[200], buf2[200];
 836                 struct bkey_i *k1, *k2;
 837                 unsigned l1 = 0, l2 = 0;
 838
 839                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 840                 k2 = btree_root_find(c, NULL, j, i, &l2);
 841
 842                 if (!k1 && !k2)
 843                         continue;
 844
 845                 mustfix_fsck_err_on(!k1 || !k2 ||
 846                                     IS_ERR(k1) ||
 847                                     IS_ERR(k2) ||
 848                                     k1->k.u64s != k2->k.u64s ||
 849                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 850                                     l1 != l2, c,
 851                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 852                         "sb:      l=%u %s\n"
 853                         "journal: l=%u %s\n", i,
 854                         l1, (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(k1)), buf1),
 855                         l2, (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(k2)), buf2));
 856         }
 857 fsck_err:
 858         return ret;
 859 }
 860
 861 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 862 {
 863         struct bch_sb_field_clean *clean, *sb_clean;
 864         int ret;
 865
 866         mutex_lock(&c->sb_lock);
 867         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 868
 869         if (fsck_err_on(!sb_clean, c,
 870                         "superblock marked clean but clean section not present")) {
 871                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 872                 c->sb.clean = false;
 873                 mutex_unlock(&c->sb_lock);
 874                 return NULL;
 875         }
 876
 877         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 878                         GFP_KERNEL);
 879         if (!clean) {
 880                 mutex_unlock(&c->sb_lock);
 881                 return ERR_PTR(-ENOMEM);
 882         }
 883
 884         ret = bch2_sb_clean_validate(c, clean, READ);
 885         if (ret) {
 886                 mutex_unlock(&c->sb_lock);
 887                 return ERR_PTR(ret);
 888         }
 889
 890         mutex_unlock(&c->sb_lock);
 891
 892         return clean;
 893 fsck_err:
 894         mutex_unlock(&c->sb_lock);
 895         return ERR_PTR(ret);
 896 }
 897
 898 static int read_btree_roots(struct bch_fs *c)
 899 {
 900         unsigned i;
 901         int ret = 0;
 902
 903         for (i = 0; i < BTREE_ID_NR; i++) {
 904                 struct btree_root *r = &c->btree_roots[i];
 905
 906                 if (!r->alive)
 907                         continue;
 908
 909                 if (i == BTREE_ID_alloc &&
 910                     c->opts.reconstruct_alloc) {
 911                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 912                         continue;
 913                 }
 914
 915                 if (r->error) {
 916                         __fsck_err(c, i == BTREE_ID_alloc
 917                                    ? FSCK_CAN_IGNORE : 0,
 918                                    "invalid btree root %s",
 919                                    bch2_btree_ids[i]);
 920                         if (i == BTREE_ID_alloc)
 921                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 922                 }
 923
 924                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 925                 if (ret) {
 926                         __fsck_err(c, i == BTREE_ID_alloc
 927                                    ? FSCK_CAN_IGNORE : 0,
 928                                    "error reading btree root %s",
 929                                    bch2_btree_ids[i]);
 930                         if (i == BTREE_ID_alloc)
 931                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 932                 }
 933         }
 934
 935         for (i = 0; i < BTREE_ID_NR; i++)
 936                 if (!c->btree_roots[i].b)
 937                         bch2_btree_root_alloc(c, i);
 938 fsck_err:
 939         return ret;
 940 }
 941
 942 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
 943 {
 944         struct bkey_i_snapshot  root_snapshot;
 945         struct bkey_i_subvolume root_volume;
 946         int ret;
 947
 948         bkey_snapshot_init(&root_snapshot.k_i);
 949         root_snapshot.k.p.offset = U32_MAX;
 950         root_snapshot.v.flags   = 0;
 951         root_snapshot.v.parent  = 0;
 952         root_snapshot.v.subvol  = BCACHEFS_ROOT_SUBVOL;
 953         root_snapshot.v.pad     = 0;
 954         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
 955
 956         ret = bch2_btree_insert(c, BTREE_ID_snapshots,
 957                                 &root_snapshot.k_i,
 958                                 NULL, NULL, 0);
 959         if (ret)
 960                 return ret;
 961
 962
 963         bkey_subvolume_init(&root_volume.k_i);
 964         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
 965         root_volume.v.flags     = 0;
 966         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
 967         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
 968
 969         ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
 970                                 &root_volume.k_i,
 971                                 NULL, NULL, 0);
 972         if (ret)
 973                 return ret;
 974
 975         return 0;
 976 }
 977
 978 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
 979 {
 980         struct bch_fs *c = trans->c;
 981         struct btree_iter iter;
 982         struct bkey_s_c k;
 983         struct bch_inode_unpacked inode;
 984         int ret;
 985
 986         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
 987                              SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
 988         k = bch2_btree_iter_peek_slot(&iter);
 989         ret = bkey_err(k);
 990         if (ret)
 991                 goto err;
 992
 993         if (!bkey_is_inode(k.k)) {
 994                 bch_err(c, "root inode not found");
 995                 ret = -ENOENT;
 996                 goto err;
 997         }
 998
 999         ret = bch2_inode_unpack(k, &inode);
1000         BUG_ON(ret);
1001
1002         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1003
1004         ret = bch2_inode_write(trans, &iter, &inode);
1005 err:
1006         bch2_trans_iter_exit(trans, &iter);
1007         return ret;
1008 }
1009
1010 int bch2_fs_recovery(struct bch_fs *c)
1011 {
1012         const char *err = "cannot allocate memory";
1013         struct bch_sb_field_clean *clean = NULL;
1014         struct jset *last_journal_entry = NULL;
1015         u64 blacklist_seq, journal_seq;
1016         bool write_sb = false;
1017         int ret = 0;
1018
1019         if (c->sb.clean)
1020                 clean = read_superblock_clean(c);
1021         ret = PTR_ERR_OR_ZERO(clean);
1022         if (ret)
1023                 goto err;
1024
1025         if (c->sb.clean)
1026                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
1027                          le64_to_cpu(clean->journal_seq));
1028         else
1029                 bch_info(c, "recovering from unclean shutdown");
1030
1031         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
1032                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
1033                 ret = -EINVAL;
1034                 goto err;
1035         }
1036
1037         if (!c->sb.clean &&
1038             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1039                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1040                 ret = -EINVAL;
1041                 goto err;
1042         }
1043
1044         if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1045                 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
1046                 ret = -EINVAL;
1047                 goto err;
1048         }
1049
1050         if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
1051                 bch_info(c, "alloc_v2 feature bit not set, fsck required");
1052                 c->opts.fsck = true;
1053                 c->opts.fix_errors = FSCK_OPT_YES;
1054         }
1055
1056         if (!c->replicas.entries ||
1057             c->opts.rebuild_replicas) {
1058                 bch_info(c, "building replicas info");
1059                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1060         }
1061
1062         if (!c->opts.nochanges) {
1063                 if (c->sb.version < bcachefs_metadata_version_inode_backpointers) {
1064                         bch_info(c, "version prior to inode backpointers, upgrade and fsck required");
1065                         c->opts.version_upgrade = true;
1066                         c->opts.fsck            = true;
1067                         c->opts.fix_errors      = FSCK_OPT_YES;
1068                 } else if (c->sb.version < bcachefs_metadata_version_subvol_dirent) {
1069                         bch_info(c, "filesystem version is prior to subvol_dirent - upgrading");
1070                         c->opts.version_upgrade = true;
1071                         c->opts.fsck            = true;
1072                 } else if (c->sb.version < bcachefs_metadata_version_inode_v2) {
1073                         bch_info(c, "filesystem version is prior to inode_v2 - upgrading");
1074                         c->opts.version_upgrade = true;
1075                 }
1076         }
1077
1078         ret = bch2_blacklist_table_initialize(c);
1079         if (ret) {
1080                 bch_err(c, "error initializing blacklist table");
1081                 goto err;
1082         }
1083
1084         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1085                 struct journal_replay *i;
1086
1087                 ret = bch2_journal_read(c, &c->journal_entries,
1088                                         &blacklist_seq, &journal_seq);
1089                 if (ret)
1090                         goto err;
1091
1092                 list_for_each_entry_reverse(i, &c->journal_entries, list)
1093                         if (!i->ignore) {
1094                                 last_journal_entry = &i->j;
1095                                 break;
1096                         }
1097
1098                 if (mustfix_fsck_err_on(c->sb.clean &&
1099                                         last_journal_entry &&
1100                                         !journal_entry_empty(last_journal_entry), c,
1101                                 "filesystem marked clean but journal not empty")) {
1102                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1103                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1104                         c->sb.clean = false;
1105                 }
1106
1107                 if (!last_journal_entry) {
1108                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
1109                         goto use_clean;
1110                 }
1111
1112                 c->journal_keys = journal_keys_sort(&c->journal_entries);
1113                 if (!c->journal_keys.d) {
1114                         ret = -ENOMEM;
1115                         goto err;
1116                 }
1117
1118                 if (c->sb.clean && last_journal_entry) {
1119                         ret = verify_superblock_clean(c, &clean,
1120                                                       last_journal_entry);
1121                         if (ret)
1122                                 goto err;
1123                 }
1124         } else {
1125 use_clean:
1126                 if (!clean) {
1127                         bch_err(c, "no superblock clean section found");
1128                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1129                         goto err;
1130
1131                 }
1132                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1133         }
1134
1135         if (c->opts.reconstruct_alloc) {
1136                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1137                 drop_alloc_keys(&c->journal_keys);
1138         }
1139
1140         zero_out_btree_mem_ptr(&c->journal_keys);
1141
1142         ret = journal_replay_early(c, clean, &c->journal_entries);
1143         if (ret)
1144                 goto err;
1145
1146         if (blacklist_seq != journal_seq) {
1147                 ret = bch2_journal_seq_blacklist_add(c,
1148                                         blacklist_seq, journal_seq);
1149                 if (ret) {
1150                         bch_err(c, "error creating new journal seq blacklist entry");
1151                         goto err;
1152                 }
1153         }
1154
1155         ret = bch2_fs_journal_start(&c->journal, journal_seq,
1156                                     &c->journal_entries);
1157         if (ret)
1158                 goto err;
1159
1160         ret = read_btree_roots(c);
1161         if (ret)
1162                 goto err;
1163
1164         bch_verbose(c, "starting alloc read");
1165         err = "error reading allocation information";
1166         ret = bch2_alloc_read(c);
1167         if (ret)
1168                 goto err;
1169         bch_verbose(c, "alloc read done");
1170
1171         bch_verbose(c, "starting stripes_read");
1172         err = "error reading stripes";
1173         ret = bch2_stripes_read(c);
1174         if (ret)
1175                 goto err;
1176         bch_verbose(c, "stripes_read done");
1177
1178         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1179
1180         if (c->opts.fsck ||
1181             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)) ||
1182             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_metadata)) ||
1183             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1184                 bool metadata_only = c->opts.norecovery;
1185
1186                 bch_info(c, "starting mark and sweep");
1187                 err = "error in mark and sweep";
1188                 ret = bch2_gc(c, true, metadata_only);
1189                 if (ret)
1190                         goto err;
1191                 bch_verbose(c, "mark and sweep done");
1192         }
1193
1194         bch2_stripes_heap_start(c);
1195
1196         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1197         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1198
1199         /*
1200          * Skip past versions that might have possibly been used (as nonces),
1201          * but hadn't had their pointers written:
1202          */
1203         if (c->sb.encryption_type && !c->sb.clean)
1204                 atomic64_add(1 << 16, &c->key_version);
1205
1206         if (c->opts.norecovery)
1207                 goto out;
1208
1209         bch_verbose(c, "starting journal replay");
1210         err = "journal replay failed";
1211         ret = bch2_journal_replay(c, c->journal_keys);
1212         if (ret)
1213                 goto err;
1214         bch_verbose(c, "journal replay done");
1215
1216         if (test_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags) &&
1217             !c->opts.nochanges) {
1218                 /*
1219                  * note that even when filesystem was clean there might be work
1220                  * to do here, if we ran gc (because of fsck) which recalculated
1221                  * oldest_gen:
1222                  */
1223                 bch_verbose(c, "writing allocation info");
1224                 err = "error writing out alloc info";
1225                 ret = bch2_alloc_write_all(c, BTREE_INSERT_LAZY_RW);
1226                 if (ret) {
1227                         bch_err(c, "error writing alloc info");
1228                         goto err;
1229                 }
1230                 bch_verbose(c, "alloc write done");
1231         }
1232
1233         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1234                 bch2_fs_lazy_rw(c);
1235
1236                 err = "error creating root snapshot node";
1237                 ret = bch2_fs_initialize_subvolumes(c);
1238                 if (ret)
1239                         goto err;
1240         }
1241
1242         bch_verbose(c, "reading snapshots table");
1243         err = "error reading snapshots table";
1244         ret = bch2_fs_snapshots_start(c);
1245         if (ret)
1246                 goto err;
1247         bch_verbose(c, "reading snapshots done");
1248
1249         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1250                 /* set bi_subvol on root inode */
1251                 err = "error upgrade root inode for subvolumes";
1252                 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1253                                     bch2_fs_upgrade_for_subvolumes(&trans));
1254                 if (ret)
1255                         goto err;
1256         }
1257
1258         if (c->opts.fsck) {
1259                 bch_info(c, "starting fsck");
1260                 err = "error in fsck";
1261                 ret = bch2_fsck_full(c);
1262                 if (ret)
1263                         goto err;
1264                 bch_verbose(c, "fsck done");
1265         } else if (!c->sb.clean) {
1266                 bch_verbose(c, "checking for deleted inodes");
1267                 err = "error in recovery";
1268                 ret = bch2_fsck_walk_inodes_only(c);
1269                 if (ret)
1270                         goto err;
1271                 bch_verbose(c, "check inodes done");
1272         }
1273
1274         if (enabled_qtypes(c)) {
1275                 bch_verbose(c, "reading quotas");
1276                 ret = bch2_fs_quota_read(c);
1277                 if (ret)
1278                         goto err;
1279                 bch_verbose(c, "quotas done");
1280         }
1281
1282         mutex_lock(&c->sb_lock);
1283         /*
1284          * With journal replay done, we can clear the journal seq blacklist
1285          * table:
1286          */
1287         BUG_ON(!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags));
1288         if (le16_to_cpu(c->sb.version_min) >= bcachefs_metadata_version_btree_ptr_sectors_written)
1289                 bch2_sb_resize_journal_seq_blacklist(&c->disk_sb, 0);
1290
1291         if (c->opts.version_upgrade) {
1292                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1293                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1294                 write_sb = true;
1295         }
1296
1297         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1298                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1299                 write_sb = true;
1300         }
1301
1302         if (c->opts.fsck &&
1303             !test_bit(BCH_FS_ERROR, &c->flags) &&
1304             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1305                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1306                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1307                 write_sb = true;
1308         }
1309
1310         if (write_sb)
1311                 bch2_write_super(c);
1312         mutex_unlock(&c->sb_lock);
1313
1314         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1315             !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1316             le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) {
1317                 struct bch_move_stats stats;
1318
1319                 bch_move_stats_init(&stats, "recovery");
1320
1321                 bch_info(c, "scanning for old btree nodes");
1322                 ret = bch2_fs_read_write(c);
1323                 if (ret)
1324                         goto err;
1325
1326                 ret = bch2_scan_old_btree_nodes(c, &stats);
1327                 if (ret)
1328                         goto err;
1329                 bch_info(c, "scanning for old btree nodes done");
1330         }
1331
1332         ret = 0;
1333 out:
1334         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1335         bch2_flush_fsck_errs(c);
1336
1337         if (!c->opts.keep_journal) {
1338                 bch2_journal_keys_free(&c->journal_keys);
1339                 bch2_journal_entries_free(&c->journal_entries);
1340         }
1341         kfree(clean);
1342         if (ret)
1343                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1344         else
1345                 bch_verbose(c, "ret %i", ret);
1346         return ret;
1347 err:
1348 fsck_err:
1349         bch2_fs_emergency_read_only(c);
1350         goto out;
1351 }
1352
1353 int bch2_fs_initialize(struct bch_fs *c)
1354 {
1355         struct bch_inode_unpacked root_inode, lostfound_inode;
1356         struct bkey_inode_buf packed_inode;
1357         struct qstr lostfound = QSTR("lost+found");
1358         const char *err = "cannot allocate memory";
1359         struct bch_dev *ca;
1360         LIST_HEAD(journal);
1361         unsigned i;
1362         int ret;
1363
1364         bch_notice(c, "initializing new filesystem");
1365
1366         mutex_lock(&c->sb_lock);
1367         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1368         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1369
1370         if (c->opts.version_upgrade) {
1371                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1372                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1373                 bch2_write_super(c);
1374         }
1375         mutex_unlock(&c->sb_lock);
1376
1377         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1378         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1379
1380         for (i = 0; i < BTREE_ID_NR; i++)
1381                 bch2_btree_root_alloc(c, i);
1382
1383         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
1384         set_bit(JOURNAL_RECLAIM_STARTED, &c->journal.flags);
1385
1386         err = "unable to allocate journal buckets";
1387         for_each_online_member(ca, c, i) {
1388                 ret = bch2_dev_journal_alloc(ca);
1389                 if (ret) {
1390                         percpu_ref_put(&ca->io_ref);
1391                         goto err;
1392                 }
1393         }
1394
1395         /*
1396          * journal_res_get() will crash if called before this has
1397          * set up the journal.pin FIFO and journal.cur pointer:
1398          */
1399         bch2_fs_journal_start(&c->journal, 1, &journal);
1400         bch2_journal_set_replay_done(&c->journal);
1401
1402         err = "error going read-write";
1403         ret = bch2_fs_read_write_early(c);
1404         if (ret)
1405                 goto err;
1406
1407         /*
1408          * Write out the superblock and journal buckets, now that we can do
1409          * btree updates
1410          */
1411         err = "error marking superblock and journal";
1412         for_each_member_device(ca, c, i) {
1413                 ret = bch2_trans_mark_dev_sb(c, ca);
1414                 if (ret) {
1415                         percpu_ref_put(&ca->ref);
1416                         goto err;
1417                 }
1418
1419                 ca->new_fs_bucket_idx = 0;
1420         }
1421
1422         err = "error creating root snapshot node";
1423         ret = bch2_fs_initialize_subvolumes(c);
1424         if (ret)
1425                 goto err;
1426
1427         bch_verbose(c, "reading snapshots table");
1428         err = "error reading snapshots table";
1429         ret = bch2_fs_snapshots_start(c);
1430         if (ret)
1431                 goto err;
1432         bch_verbose(c, "reading snapshots done");
1433
1434         bch2_inode_init(c, &root_inode, 0, 0,
1435                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1436         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1437         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1438         bch2_inode_pack(c, &packed_inode, &root_inode);
1439         packed_inode.inode.k.p.snapshot = U32_MAX;
1440
1441         err = "error creating root directory";
1442         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1443                                 &packed_inode.inode.k_i,
1444                                 NULL, NULL, 0);
1445         if (ret)
1446                 goto err;
1447
1448         bch2_inode_init_early(c, &lostfound_inode);
1449
1450         err = "error creating lost+found";
1451         ret = bch2_trans_do(c, NULL, NULL, 0,
1452                 bch2_create_trans(&trans,
1453                                   BCACHEFS_ROOT_SUBVOL_INUM,
1454                                   &root_inode, &lostfound_inode,
1455                                   &lostfound,
1456                                   0, 0, S_IFDIR|0700, 0,
1457                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1458         if (ret) {
1459                 bch_err(c, "error creating lost+found");
1460                 goto err;
1461         }
1462
1463         if (enabled_qtypes(c)) {
1464                 ret = bch2_fs_quota_read(c);
1465                 if (ret)
1466                         goto err;
1467         }
1468
1469         err = "error writing first journal entry";
1470         ret = bch2_journal_flush(&c->journal);
1471         if (ret)
1472                 goto err;
1473
1474         mutex_lock(&c->sb_lock);
1475         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1476         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1477
1478         bch2_write_super(c);
1479         mutex_unlock(&c->sb_lock);
1480
1481         return 0;
1482 err:
1483         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1484         return ret;
1485 }