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 "backpointers.h"
   5 #include "bkey_buf.h"
   6 #include "alloc_background.h"
   7 #include "btree_gc.h"
   8 #include "btree_update.h"
   9 #include "btree_update_interior.h"
  10 #include "btree_io.h"
  11 #include "buckets.h"
  12 #include "dirent.h"
  13 #include "ec.h"
  14 #include "errcode.h"
  15 #include "error.h"
  16 #include "fs-common.h"
  17 #include "fsck.h"
  18 #include "journal_io.h"
  19 #include "journal_reclaim.h"
  20 #include "journal_seq_blacklist.h"
  21 #include "lru.h"
  22 #include "move.h"
  23 #include "quota.h"
  24 #include "recovery.h"
  25 #include "replicas.h"
  26 #include "subvolume.h"
  27 #include "super-io.h"
  28
  29 #include <linux/sort.h>
  30 #include <linux/stat.h>
  31
  32 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  33
  34 /* for -o reconstruct_alloc: */
  35 static void drop_alloc_keys(struct journal_keys *keys)
  36 {
  37         size_t src, dst;
  38
  39         for (src = 0, dst = 0; src < keys->nr; src++)
  40                 if (keys->d[src].btree_id != BTREE_ID_alloc)
  41                         keys->d[dst++] = keys->d[src];
  42
  43         keys->nr = dst;
  44 }
  45
  46 /*
  47  * Btree node pointers have a field to stack a pointer to the in memory btree
  48  * node; we need to zero out this field when reading in btree nodes, or when
  49  * reading in keys from the journal:
  50  */
  51 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
  52 {
  53         struct journal_key *i;
  54
  55         for (i = keys->d; i < keys->d + keys->nr; i++)
  56                 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
  57                         bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
  58 }
  59
  60 /* iterate over keys read from the journal: */
  61
  62 static int __journal_key_cmp(enum btree_id      l_btree_id,
  63                              unsigned           l_level,
  64                              struct bpos        l_pos,
  65                              const struct journal_key *r)
  66 {
  67         return (cmp_int(l_btree_id,     r->btree_id) ?:
  68                 cmp_int(l_level,        r->level) ?:
  69                 bpos_cmp(l_pos, r->k->k.p));
  70 }
  71
  72 static int journal_key_cmp(const struct journal_key *l, const struct journal_key *r)
  73 {
  74         return __journal_key_cmp(l->btree_id, l->level, l->k->k.p, r);
  75 }
  76
  77 static inline size_t idx_to_pos(struct journal_keys *keys, size_t idx)
  78 {
  79         size_t gap_size = keys->size - keys->nr;
  80
  81         if (idx >= keys->gap)
  82                 idx += gap_size;
  83         return idx;
  84 }
  85
  86 static inline struct journal_key *idx_to_key(struct journal_keys *keys, size_t idx)
  87 {
  88         return keys->d + idx_to_pos(keys, idx);
  89 }
  90
  91 static size_t __bch2_journal_key_search(struct journal_keys *keys,
  92                                         enum btree_id id, unsigned level,
  93                                         struct bpos pos)
  94 {
  95         size_t l = 0, r = keys->nr, m;
  96
  97         while (l < r) {
  98                 m = l + ((r - l) >> 1);
  99                 if (__journal_key_cmp(id, level, pos, idx_to_key(keys, m)) > 0)
 100                         l = m + 1;
 101                 else
 102                         r = m;
 103         }
 104
 105         BUG_ON(l < keys->nr &&
 106                __journal_key_cmp(id, level, pos, idx_to_key(keys, l)) > 0);
 107
 108         BUG_ON(l &&
 109                __journal_key_cmp(id, level, pos, idx_to_key(keys, l - 1)) <= 0);
 110
 111         return l;
 112 }
 113
 114 static size_t bch2_journal_key_search(struct journal_keys *keys,
 115                                       enum btree_id id, unsigned level,
 116                                       struct bpos pos)
 117 {
 118         return idx_to_pos(keys, __bch2_journal_key_search(keys, id, level, pos));
 119 }
 120
 121 struct bkey_i *bch2_journal_keys_peek_upto(struct bch_fs *c, enum btree_id btree_id,
 122                                            unsigned level, struct bpos pos,
 123                                            struct bpos end_pos, size_t *idx)
 124 {
 125         struct journal_keys *keys = &c->journal_keys;
 126         unsigned iters = 0;
 127         struct journal_key *k;
 128 search:
 129         if (!*idx)
 130                 *idx = __bch2_journal_key_search(keys, btree_id, level, pos);
 131
 132         while (*idx < keys->nr &&
 133                (k = idx_to_key(keys, *idx),
 134                 k->btree_id == btree_id &&
 135                 k->level == level &&
 136                 bpos_cmp(k->k->k.p, end_pos) <= 0)) {
 137                 if (bpos_cmp(k->k->k.p, pos) >= 0 &&
 138                     !k->overwritten)
 139                         return k->k;
 140
 141                 (*idx)++;
 142                 iters++;
 143                 if (iters == 10) {
 144                         *idx = 0;
 145                         goto search;
 146                 }
 147         }
 148
 149         return NULL;
 150 }
 151
 152 struct bkey_i *bch2_journal_keys_peek_slot(struct bch_fs *c, enum btree_id btree_id,
 153                                            unsigned level, struct bpos pos)
 154 {
 155         size_t idx = 0;
 156
 157         return bch2_journal_keys_peek_upto(c, btree_id, level, pos, pos, &idx);
 158 }
 159
 160 static void journal_iters_fix(struct bch_fs *c)
 161 {
 162         struct journal_keys *keys = &c->journal_keys;
 163         /* The key we just inserted is immediately before the gap: */
 164         size_t gap_end = keys->gap + (keys->size - keys->nr);
 165         struct btree_and_journal_iter *iter;
 166
 167         /*
 168          * If an iterator points one after the key we just inserted, decrement
 169          * the iterator so it points at the key we just inserted - if the
 170          * decrement was unnecessary, bch2_btree_and_journal_iter_peek() will
 171          * handle that:
 172          */
 173         list_for_each_entry(iter, &c->journal_iters, journal.list)
 174                 if (iter->journal.idx == gap_end)
 175                         iter->journal.idx = keys->gap - 1;
 176 }
 177
 178 static void journal_iters_move_gap(struct bch_fs *c, size_t old_gap, size_t new_gap)
 179 {
 180         struct journal_keys *keys = &c->journal_keys;
 181         struct journal_iter *iter;
 182         size_t gap_size = keys->size - keys->nr;
 183
 184         list_for_each_entry(iter, &c->journal_iters, list) {
 185                 if (iter->idx > old_gap)
 186                         iter->idx -= gap_size;
 187                 if (iter->idx >= new_gap)
 188                         iter->idx += gap_size;
 189         }
 190 }
 191
 192 int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
 193                                  unsigned level, struct bkey_i *k)
 194 {
 195         struct journal_key n = {
 196                 .btree_id       = id,
 197                 .level          = level,
 198                 .k              = k,
 199                 .allocated      = true,
 200                 /*
 201                  * Ensure these keys are done last by journal replay, to unblock
 202                  * journal reclaim:
 203                  */
 204                 .journal_seq    = U32_MAX,
 205         };
 206         struct journal_keys *keys = &c->journal_keys;
 207         size_t idx = bch2_journal_key_search(keys, id, level, k->k.p);
 208
 209         BUG_ON(test_bit(BCH_FS_RW, &c->flags));
 210
 211         if (idx < keys->size &&
 212             journal_key_cmp(&n, &keys->d[idx]) == 0) {
 213                 if (keys->d[idx].allocated)
 214                         kfree(keys->d[idx].k);
 215                 keys->d[idx] = n;
 216                 return 0;
 217         }
 218
 219         if (idx > keys->gap)
 220                 idx -= keys->size - keys->nr;
 221
 222         if (keys->nr == keys->size) {
 223                 struct journal_keys new_keys = {
 224                         .nr                     = keys->nr,
 225                         .size                   = max_t(size_t, keys->size, 8) * 2,
 226                         .journal_seq_base       = keys->journal_seq_base,
 227                 };
 228
 229                 new_keys.d = kvmalloc(sizeof(new_keys.d[0]) * new_keys.size, GFP_KERNEL);
 230                 if (!new_keys.d) {
 231                         bch_err(c, "%s: error allocating new key array (size %zu)",
 232                                 __func__, new_keys.size);
 233                         return -ENOMEM;
 234                 }
 235
 236                 /* Since @keys was full, there was no gap: */
 237                 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
 238                 kvfree(keys->d);
 239                 *keys = new_keys;
 240
 241                 /* And now the gap is at the end: */
 242                 keys->gap = keys->nr;
 243         }
 244
 245         journal_iters_move_gap(c, keys->gap, idx);
 246
 247         move_gap(keys->d, keys->nr, keys->size, keys->gap, idx);
 248         keys->gap = idx;
 249
 250         keys->nr++;
 251         keys->d[keys->gap++] = n;
 252
 253         journal_iters_fix(c);
 254
 255         return 0;
 256 }
 257
 258 /*
 259  * Can only be used from the recovery thread while we're still RO - can't be
 260  * used once we've got RW, as journal_keys is at that point used by multiple
 261  * threads:
 262  */
 263 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
 264                             unsigned level, struct bkey_i *k)
 265 {
 266         struct bkey_i *n;
 267         int ret;
 268
 269         n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
 270         if (!n)
 271                 return -ENOMEM;
 272
 273         bkey_copy(n, k);
 274         ret = bch2_journal_key_insert_take(c, id, level, n);
 275         if (ret)
 276                 kfree(n);
 277         return ret;
 278 }
 279
 280 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
 281                             unsigned level, struct bpos pos)
 282 {
 283         struct bkey_i whiteout;
 284
 285         bkey_init(&whiteout.k);
 286         whiteout.k.p = pos;
 287
 288         return bch2_journal_key_insert(c, id, level, &whiteout);
 289 }
 290
 291 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
 292                                   unsigned level, struct bpos pos)
 293 {
 294         struct journal_keys *keys = &c->journal_keys;
 295         size_t idx = bch2_journal_key_search(keys, btree, level, pos);
 296
 297         if (idx < keys->size &&
 298             keys->d[idx].btree_id       == btree &&
 299             keys->d[idx].level          == level &&
 300             !bpos_cmp(keys->d[idx].k->k.p, pos))
 301                 keys->d[idx].overwritten = true;
 302 }
 303
 304 static void bch2_journal_iter_advance(struct journal_iter *iter)
 305 {
 306         if (iter->idx < iter->keys->size) {
 307                 iter->idx++;
 308                 if (iter->idx == iter->keys->gap)
 309                         iter->idx += iter->keys->size - iter->keys->nr;
 310         }
 311 }
 312
 313 struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
 314 {
 315         struct journal_key *k = iter->keys->d + iter->idx;
 316
 317         while (k < iter->keys->d + iter->keys->size &&
 318                k->btree_id      == iter->btree_id &&
 319                k->level         == iter->level) {
 320                 if (!k->overwritten)
 321                         return bkey_i_to_s_c(k->k);
 322
 323                 bch2_journal_iter_advance(iter);
 324                 k = iter->keys->d + iter->idx;
 325         }
 326
 327         return bkey_s_c_null;
 328 }
 329
 330 static void bch2_journal_iter_exit(struct journal_iter *iter)
 331 {
 332         list_del(&iter->list);
 333 }
 334
 335 static void bch2_journal_iter_init(struct bch_fs *c,
 336                                    struct journal_iter *iter,
 337                                    enum btree_id id, unsigned level,
 338                                    struct bpos pos)
 339 {
 340         iter->btree_id  = id;
 341         iter->level     = level;
 342         iter->keys      = &c->journal_keys;
 343         iter->idx       = bch2_journal_key_search(&c->journal_keys, id, level, pos);
 344 }
 345
 346 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 347 {
 348         return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 349                                                 iter->b, &iter->unpacked);
 350 }
 351
 352 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 353 {
 354         bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 355 }
 356
 357 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 358 {
 359         if (!bpos_cmp(iter->pos, SPOS_MAX))
 360                 iter->at_end = true;
 361         else
 362                 iter->pos = bpos_successor(iter->pos);
 363 }
 364
 365 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 366 {
 367         struct bkey_s_c btree_k, journal_k, ret;
 368 again:
 369         if (iter->at_end)
 370                 return bkey_s_c_null;
 371
 372         while ((btree_k = bch2_journal_iter_peek_btree(iter)).k &&
 373                bpos_cmp(btree_k.k->p, iter->pos) < 0)
 374                 bch2_journal_iter_advance_btree(iter);
 375
 376         while ((journal_k = bch2_journal_iter_peek(&iter->journal)).k &&
 377                bpos_cmp(journal_k.k->p, iter->pos) < 0)
 378                 bch2_journal_iter_advance(&iter->journal);
 379
 380         ret = journal_k.k &&
 381                 (!btree_k.k || bpos_cmp(journal_k.k->p, btree_k.k->p) <= 0)
 382                 ? journal_k
 383                 : btree_k;
 384
 385         if (ret.k && iter->b && bpos_cmp(ret.k->p, iter->b->data->max_key) > 0)
 386                 ret = bkey_s_c_null;
 387
 388         if (ret.k) {
 389                 iter->pos = ret.k->p;
 390                 if (bkey_deleted(ret.k)) {
 391                         bch2_btree_and_journal_iter_advance(iter);
 392                         goto again;
 393                 }
 394         } else {
 395                 iter->pos = SPOS_MAX;
 396                 iter->at_end = true;
 397         }
 398
 399         return ret;
 400 }
 401
 402 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
 403 {
 404         bch2_journal_iter_exit(&iter->journal);
 405 }
 406
 407 void __bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 408                                                   struct bch_fs *c,
 409                                                   struct btree *b,
 410                                                   struct btree_node_iter node_iter,
 411                                                   struct bpos pos)
 412 {
 413         memset(iter, 0, sizeof(*iter));
 414
 415         iter->b = b;
 416         iter->node_iter = node_iter;
 417         bch2_journal_iter_init(c, &iter->journal, b->c.btree_id, b->c.level, pos);
 418         INIT_LIST_HEAD(&iter->journal.list);
 419         iter->pos = b->data->min_key;
 420         iter->at_end = false;
 421 }
 422
 423 /*
 424  * this version is used by btree_gc before filesystem has gone RW and
 425  * multithreaded, so uses the journal_iters list:
 426  */
 427 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 428                                                 struct bch_fs *c,
 429                                                 struct btree *b)
 430 {
 431         struct btree_node_iter node_iter;
 432
 433         bch2_btree_node_iter_init_from_start(&node_iter, b);
 434         __bch2_btree_and_journal_iter_init_node_iter(iter, c, b, node_iter, b->data->min_key);
 435         list_add(&iter->journal.list, &c->journal_iters);
 436 }
 437
 438 /* sort and dedup all keys in the journal: */
 439
 440 void bch2_journal_entries_free(struct bch_fs *c)
 441 {
 442         struct journal_replay **i;
 443         struct genradix_iter iter;
 444
 445         genradix_for_each(&c->journal_entries, iter, i)
 446                 if (*i)
 447                         kvpfree(*i, offsetof(struct journal_replay, j) +
 448                                 vstruct_bytes(&(*i)->j));
 449         genradix_free(&c->journal_entries);
 450 }
 451
 452 /*
 453  * When keys compare equal, oldest compares first:
 454  */
 455 static int journal_sort_key_cmp(const void *_l, const void *_r)
 456 {
 457         const struct journal_key *l = _l;
 458         const struct journal_key *r = _r;
 459
 460         return  journal_key_cmp(l, r) ?:
 461                 cmp_int(l->journal_seq, r->journal_seq) ?:
 462                 cmp_int(l->journal_offset, r->journal_offset);
 463 }
 464
 465 void bch2_journal_keys_free(struct journal_keys *keys)
 466 {
 467         struct journal_key *i;
 468
 469         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 470         keys->gap = keys->nr;
 471
 472         for (i = keys->d; i < keys->d + keys->nr; i++)
 473                 if (i->allocated)
 474                         kfree(i->k);
 475
 476         kvfree(keys->d);
 477         keys->d = NULL;
 478         keys->nr = keys->gap = keys->size = 0;
 479 }
 480
 481 static int journal_keys_sort(struct bch_fs *c)
 482 {
 483         struct genradix_iter iter;
 484         struct journal_replay *i, **_i;
 485         struct jset_entry *entry;
 486         struct bkey_i *k, *_n;
 487         struct journal_keys *keys = &c->journal_keys;
 488         struct journal_key *src, *dst;
 489         size_t nr_keys = 0;
 490
 491         genradix_for_each(&c->journal_entries, iter, _i) {
 492                 i = *_i;
 493
 494                 if (!i || i->ignore)
 495                         continue;
 496
 497                 if (!keys->journal_seq_base)
 498                         keys->journal_seq_base = le64_to_cpu(i->j.seq);
 499
 500                 for_each_jset_key(k, _n, entry, &i->j)
 501                         nr_keys++;
 502         }
 503
 504         if (!nr_keys)
 505                 return 0;
 506
 507         keys->size = roundup_pow_of_two(nr_keys);
 508
 509         keys->d = kvmalloc(sizeof(keys->d[0]) * keys->size, GFP_KERNEL);
 510         if (!keys->d)
 511                 return -ENOMEM;
 512
 513         genradix_for_each(&c->journal_entries, iter, _i) {
 514                 i = *_i;
 515
 516                 if (!i || i->ignore)
 517                         continue;
 518
 519                 BUG_ON(le64_to_cpu(i->j.seq) - keys->journal_seq_base > U32_MAX);
 520
 521                 for_each_jset_key(k, _n, entry, &i->j)
 522                         keys->d[keys->nr++] = (struct journal_key) {
 523                                 .btree_id       = entry->btree_id,
 524                                 .level          = entry->level,
 525                                 .k              = k,
 526                                 .journal_seq    = le64_to_cpu(i->j.seq) -
 527                                         keys->journal_seq_base,
 528                                 .journal_offset = k->_data - i->j._data,
 529                         };
 530         }
 531
 532         sort(keys->d, keys->nr, sizeof(keys->d[0]), journal_sort_key_cmp, NULL);
 533
 534         src = dst = keys->d;
 535         while (src < keys->d + keys->nr) {
 536                 while (src + 1 < keys->d + keys->nr &&
 537                        src[0].btree_id  == src[1].btree_id &&
 538                        src[0].level     == src[1].level &&
 539                        !bpos_cmp(src[0].k->k.p, src[1].k->k.p))
 540                         src++;
 541
 542                 *dst++ = *src++;
 543         }
 544
 545         keys->nr = dst - keys->d;
 546         keys->gap = keys->nr;
 547         return 0;
 548 }
 549
 550 /* journal replay: */
 551
 552 static void replay_now_at(struct journal *j, u64 seq)
 553 {
 554         BUG_ON(seq < j->replay_journal_seq);
 555
 556         seq = min(seq, j->replay_journal_seq_end);
 557
 558         while (j->replay_journal_seq < seq)
 559                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 560 }
 561
 562 static int bch2_journal_replay_key(struct btree_trans *trans,
 563                                    struct journal_key *k)
 564 {
 565         struct btree_iter iter;
 566         unsigned iter_flags =
 567                 BTREE_ITER_INTENT|
 568                 BTREE_ITER_NOT_EXTENTS;
 569         int ret;
 570
 571         if (!k->level && k->btree_id == BTREE_ID_alloc)
 572                 iter_flags |= BTREE_ITER_CACHED;
 573
 574         bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 575                                   BTREE_MAX_DEPTH, k->level,
 576                                   iter_flags);
 577         ret = bch2_btree_iter_traverse(&iter);
 578         if (ret)
 579                 goto out;
 580
 581         /* Must be checked with btree locked: */
 582         if (k->overwritten)
 583                 goto out;
 584
 585         ret = bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
 586 out:
 587         bch2_trans_iter_exit(trans, &iter);
 588         return ret;
 589 }
 590
 591 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 592 {
 593         const struct journal_key *l = *((const struct journal_key **)_l);
 594         const struct journal_key *r = *((const struct journal_key **)_r);
 595
 596         return cmp_int(l->journal_seq, r->journal_seq);
 597 }
 598
 599 static int bch2_journal_replay(struct bch_fs *c)
 600 {
 601         struct journal_keys *keys = &c->journal_keys;
 602         struct journal_key **keys_sorted, *k;
 603         struct journal *j = &c->journal;
 604         size_t i;
 605         int ret;
 606
 607         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 608         keys->gap = keys->nr;
 609
 610         keys_sorted = kvmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL);
 611         if (!keys_sorted)
 612                 return -ENOMEM;
 613
 614         for (i = 0; i < keys->nr; i++)
 615                 keys_sorted[i] = &keys->d[i];
 616
 617         sort(keys_sorted, keys->nr,
 618              sizeof(keys_sorted[0]),
 619              journal_sort_seq_cmp, NULL);
 620
 621         if (keys->nr)
 622                 replay_now_at(j, keys->journal_seq_base);
 623
 624         for (i = 0; i < keys->nr; i++) {
 625                 k = keys_sorted[i];
 626
 627                 cond_resched();
 628
 629                 replay_now_at(j, keys->journal_seq_base + k->journal_seq);
 630
 631                 ret = bch2_trans_do(c, NULL, NULL,
 632                                     BTREE_INSERT_LAZY_RW|
 633                                     BTREE_INSERT_NOFAIL|
 634                                     (!k->allocated
 635                                      ? BTREE_INSERT_JOURNAL_REPLAY|JOURNAL_WATERMARK_reserved
 636                                      : 0),
 637                              bch2_journal_replay_key(&trans, k));
 638                 if (ret) {
 639                         bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
 640                                 ret, bch2_btree_ids[k->btree_id], k->level);
 641                         goto err;
 642                 }
 643         }
 644
 645         replay_now_at(j, j->replay_journal_seq_end);
 646         j->replay_journal_seq = 0;
 647
 648         bch2_journal_set_replay_done(j);
 649         bch2_journal_flush_all_pins(j);
 650         ret = bch2_journal_error(j);
 651
 652         if (keys->nr && !ret)
 653                 bch2_journal_log_msg(&c->journal, "journal replay finished");
 654 err:
 655         kvfree(keys_sorted);
 656         return ret;
 657 }
 658
 659 /* journal replay early: */
 660
 661 static int journal_replay_entry_early(struct bch_fs *c,
 662                                       struct jset_entry *entry)
 663 {
 664         int ret = 0;
 665
 666         switch (entry->type) {
 667         case BCH_JSET_ENTRY_btree_root: {
 668                 struct btree_root *r;
 669
 670                 if (entry->btree_id >= BTREE_ID_NR) {
 671                         bch_err(c, "filesystem has unknown btree type %u",
 672                                 entry->btree_id);
 673                         return -EINVAL;
 674                 }
 675
 676                 r = &c->btree_roots[entry->btree_id];
 677
 678                 if (entry->u64s) {
 679                         r->level = entry->level;
 680                         bkey_copy(&r->key, &entry->start[0]);
 681                         r->error = 0;
 682                 } else {
 683                         r->error = -EIO;
 684                 }
 685                 r->alive = true;
 686                 break;
 687         }
 688         case BCH_JSET_ENTRY_usage: {
 689                 struct jset_entry_usage *u =
 690                         container_of(entry, struct jset_entry_usage, entry);
 691
 692                 switch (entry->btree_id) {
 693                 case BCH_FS_USAGE_reserved:
 694                         if (entry->level < BCH_REPLICAS_MAX)
 695                                 c->usage_base->persistent_reserved[entry->level] =
 696                                         le64_to_cpu(u->v);
 697                         break;
 698                 case BCH_FS_USAGE_inodes:
 699                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 700                         break;
 701                 case BCH_FS_USAGE_key_version:
 702                         atomic64_set(&c->key_version,
 703                                      le64_to_cpu(u->v));
 704                         break;
 705                 }
 706
 707                 break;
 708         }
 709         case BCH_JSET_ENTRY_data_usage: {
 710                 struct jset_entry_data_usage *u =
 711                         container_of(entry, struct jset_entry_data_usage, entry);
 712
 713                 ret = bch2_replicas_set_usage(c, &u->r,
 714                                               le64_to_cpu(u->v));
 715                 break;
 716         }
 717         case BCH_JSET_ENTRY_dev_usage: {
 718                 struct jset_entry_dev_usage *u =
 719                         container_of(entry, struct jset_entry_dev_usage, entry);
 720                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 721                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 722
 723                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 724
 725                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 726                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 727                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 728                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 729                 }
 730
 731                 break;
 732         }
 733         case BCH_JSET_ENTRY_blacklist: {
 734                 struct jset_entry_blacklist *bl_entry =
 735                         container_of(entry, struct jset_entry_blacklist, entry);
 736
 737                 ret = bch2_journal_seq_blacklist_add(c,
 738                                 le64_to_cpu(bl_entry->seq),
 739                                 le64_to_cpu(bl_entry->seq) + 1);
 740                 break;
 741         }
 742         case BCH_JSET_ENTRY_blacklist_v2: {
 743                 struct jset_entry_blacklist_v2 *bl_entry =
 744                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 745
 746                 ret = bch2_journal_seq_blacklist_add(c,
 747                                 le64_to_cpu(bl_entry->start),
 748                                 le64_to_cpu(bl_entry->end) + 1);
 749                 break;
 750         }
 751         case BCH_JSET_ENTRY_clock: {
 752                 struct jset_entry_clock *clock =
 753                         container_of(entry, struct jset_entry_clock, entry);
 754
 755                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 756         }
 757         }
 758
 759         return ret;
 760 }
 761
 762 static int journal_replay_early(struct bch_fs *c,
 763                                 struct bch_sb_field_clean *clean)
 764 {
 765         struct jset_entry *entry;
 766         int ret;
 767
 768         if (clean) {
 769                 for (entry = clean->start;
 770                      entry != vstruct_end(&clean->field);
 771                      entry = vstruct_next(entry)) {
 772                         ret = journal_replay_entry_early(c, entry);
 773                         if (ret)
 774                                 return ret;
 775                 }
 776         } else {
 777                 struct genradix_iter iter;
 778                 struct journal_replay *i, **_i;
 779
 780                 genradix_for_each(&c->journal_entries, iter, _i) {
 781                         i = *_i;
 782
 783                         if (!i || i->ignore)
 784                                 continue;
 785
 786                         vstruct_for_each(&i->j, entry) {
 787                                 ret = journal_replay_entry_early(c, entry);
 788                                 if (ret)
 789                                         return ret;
 790                         }
 791                 }
 792         }
 793
 794         bch2_fs_usage_initialize(c);
 795
 796         return 0;
 797 }
 798
 799 /* sb clean section: */
 800
 801 static struct bkey_i *btree_root_find(struct bch_fs *c,
 802                                       struct bch_sb_field_clean *clean,
 803                                       struct jset *j,
 804                                       enum btree_id id, unsigned *level)
 805 {
 806         struct bkey_i *k;
 807         struct jset_entry *entry, *start, *end;
 808
 809         if (clean) {
 810                 start = clean->start;
 811                 end = vstruct_end(&clean->field);
 812         } else {
 813                 start = j->start;
 814                 end = vstruct_last(j);
 815         }
 816
 817         for (entry = start; entry < end; entry = vstruct_next(entry))
 818                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 819                     entry->btree_id == id)
 820                         goto found;
 821
 822         return NULL;
 823 found:
 824         if (!entry->u64s)
 825                 return ERR_PTR(-EINVAL);
 826
 827         k = entry->start;
 828         *level = entry->level;
 829         return k;
 830 }
 831
 832 static int verify_superblock_clean(struct bch_fs *c,
 833                                    struct bch_sb_field_clean **cleanp,
 834                                    struct jset *j)
 835 {
 836         unsigned i;
 837         struct bch_sb_field_clean *clean = *cleanp;
 838         struct printbuf buf1 = PRINTBUF;
 839         struct printbuf buf2 = PRINTBUF;
 840         int ret = 0;
 841
 842         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 843                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 844                         le64_to_cpu(clean->journal_seq),
 845                         le64_to_cpu(j->seq))) {
 846                 kfree(clean);
 847                 *cleanp = NULL;
 848                 return 0;
 849         }
 850
 851         for (i = 0; i < BTREE_ID_NR; i++) {
 852                 struct bkey_i *k1, *k2;
 853                 unsigned l1 = 0, l2 = 0;
 854
 855                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 856                 k2 = btree_root_find(c, NULL, j, i, &l2);
 857
 858                 if (!k1 && !k2)
 859                         continue;
 860
 861                 printbuf_reset(&buf1);
 862                 printbuf_reset(&buf2);
 863
 864                 if (k1)
 865                         bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(k1));
 866                 else
 867                         prt_printf(&buf1, "(none)");
 868
 869                 if (k2)
 870                         bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(k2));
 871                 else
 872                         prt_printf(&buf2, "(none)");
 873
 874                 mustfix_fsck_err_on(!k1 || !k2 ||
 875                                     IS_ERR(k1) ||
 876                                     IS_ERR(k2) ||
 877                                     k1->k.u64s != k2->k.u64s ||
 878                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 879                                     l1 != l2, c,
 880                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 881                         "sb:      l=%u %s\n"
 882                         "journal: l=%u %s\n", i,
 883                         l1, buf1.buf,
 884                         l2, buf2.buf);
 885         }
 886 fsck_err:
 887         printbuf_exit(&buf2);
 888         printbuf_exit(&buf1);
 889         return ret;
 890 }
 891
 892 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 893 {
 894         struct bch_sb_field_clean *clean, *sb_clean;
 895         int ret;
 896
 897         mutex_lock(&c->sb_lock);
 898         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 899
 900         if (fsck_err_on(!sb_clean, c,
 901                         "superblock marked clean but clean section not present")) {
 902                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 903                 c->sb.clean = false;
 904                 mutex_unlock(&c->sb_lock);
 905                 return NULL;
 906         }
 907
 908         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 909                         GFP_KERNEL);
 910         if (!clean) {
 911                 mutex_unlock(&c->sb_lock);
 912                 return ERR_PTR(-ENOMEM);
 913         }
 914
 915         ret = bch2_sb_clean_validate_late(c, clean, READ);
 916         if (ret) {
 917                 mutex_unlock(&c->sb_lock);
 918                 return ERR_PTR(ret);
 919         }
 920
 921         mutex_unlock(&c->sb_lock);
 922
 923         return clean;
 924 fsck_err:
 925         mutex_unlock(&c->sb_lock);
 926         return ERR_PTR(ret);
 927 }
 928
 929 static bool btree_id_is_alloc(enum btree_id id)
 930 {
 931         switch (id) {
 932         case BTREE_ID_alloc:
 933         case BTREE_ID_backpointers:
 934         case BTREE_ID_need_discard:
 935         case BTREE_ID_freespace:
 936                 return true;
 937         default:
 938                 return false;
 939         }
 940 }
 941
 942 static int read_btree_roots(struct bch_fs *c)
 943 {
 944         unsigned i;
 945         int ret = 0;
 946
 947         for (i = 0; i < BTREE_ID_NR; i++) {
 948                 struct btree_root *r = &c->btree_roots[i];
 949
 950                 if (!r->alive)
 951                         continue;
 952
 953                 if (btree_id_is_alloc(i) &&
 954                     c->opts.reconstruct_alloc) {
 955                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 956                         continue;
 957                 }
 958
 959                 if (r->error) {
 960                         __fsck_err(c, btree_id_is_alloc(i)
 961                                    ? FSCK_CAN_IGNORE : 0,
 962                                    "invalid btree root %s",
 963                                    bch2_btree_ids[i]);
 964                         if (i == BTREE_ID_alloc)
 965                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 966                 }
 967
 968                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 969                 if (ret) {
 970                         __fsck_err(c,
 971                                    btree_id_is_alloc(i)
 972                                    ? FSCK_CAN_IGNORE : 0,
 973                                    "error reading btree root %s",
 974                                    bch2_btree_ids[i]);
 975                         if (i == BTREE_ID_alloc)
 976                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 977                 }
 978         }
 979
 980         for (i = 0; i < BTREE_ID_NR; i++)
 981                 if (!c->btree_roots[i].b)
 982                         bch2_btree_root_alloc(c, i);
 983 fsck_err:
 984         return ret;
 985 }
 986
 987 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
 988 {
 989         struct bkey_i_snapshot  root_snapshot;
 990         struct bkey_i_subvolume root_volume;
 991         int ret;
 992
 993         bkey_snapshot_init(&root_snapshot.k_i);
 994         root_snapshot.k.p.offset = U32_MAX;
 995         root_snapshot.v.flags   = 0;
 996         root_snapshot.v.parent  = 0;
 997         root_snapshot.v.subvol  = BCACHEFS_ROOT_SUBVOL;
 998         root_snapshot.v.pad     = 0;
 999         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
1000
1001         ret = bch2_btree_insert(c, BTREE_ID_snapshots,
1002                                 &root_snapshot.k_i,
1003                                 NULL, NULL, 0);
1004         if (ret)
1005                 return ret;
1006
1007         bkey_subvolume_init(&root_volume.k_i);
1008         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1009         root_volume.v.flags     = 0;
1010         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
1011         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
1012
1013         ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
1014                                 &root_volume.k_i,
1015                                 NULL, NULL, 0);
1016         if (ret)
1017                 return ret;
1018
1019         return 0;
1020 }
1021
1022 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
1023 {
1024         struct btree_iter iter;
1025         struct bkey_s_c k;
1026         struct bch_inode_unpacked inode;
1027         int ret;
1028
1029         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
1030                              SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
1031         k = bch2_btree_iter_peek_slot(&iter);
1032         ret = bkey_err(k);
1033         if (ret)
1034                 goto err;
1035
1036         if (!bkey_is_inode(k.k)) {
1037                 bch_err(trans->c, "root inode not found");
1038                 ret = -ENOENT;
1039                 goto err;
1040         }
1041
1042         ret = bch2_inode_unpack(k, &inode);
1043         BUG_ON(ret);
1044
1045         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1046
1047         ret = bch2_inode_write(trans, &iter, &inode);
1048 err:
1049         bch2_trans_iter_exit(trans, &iter);
1050         return ret;
1051 }
1052
1053 int bch2_fs_recovery(struct bch_fs *c)
1054 {
1055         const char *err = "cannot allocate memory";
1056         struct bch_sb_field_clean *clean = NULL;
1057         struct jset *last_journal_entry = NULL;
1058         u64 blacklist_seq, journal_seq;
1059         bool write_sb = false;
1060         int ret = 0;
1061
1062         if (c->sb.clean)
1063                 clean = read_superblock_clean(c);
1064         ret = PTR_ERR_OR_ZERO(clean);
1065         if (ret)
1066                 goto err;
1067
1068         if (c->sb.clean)
1069                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
1070                          le64_to_cpu(clean->journal_seq));
1071         else
1072                 bch_info(c, "recovering from unclean shutdown");
1073
1074         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
1075                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
1076                 ret = -EINVAL;
1077                 goto err;
1078         }
1079
1080         if (!c->sb.clean &&
1081             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1082                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1083                 ret = -EINVAL;
1084                 goto err;
1085         }
1086
1087         if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1088                 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
1089                 ret = -EINVAL;
1090                 goto err;
1091         }
1092
1093         if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
1094                 bch_info(c, "alloc_v2 feature bit not set, fsck required");
1095                 c->opts.fsck = true;
1096                 c->opts.fix_errors = FSCK_OPT_YES;
1097         }
1098
1099         if (!c->opts.nochanges) {
1100                 if (c->sb.version < bcachefs_metadata_version_backpointers) {
1101                         bch_info(c, "version prior to backpointers, upgrade and fsck required");
1102                         c->opts.version_upgrade = true;
1103                         c->opts.fsck            = true;
1104                         c->opts.fix_errors      = FSCK_OPT_YES;
1105                 }
1106         }
1107
1108         if (c->opts.fsck && c->opts.norecovery) {
1109                 bch_err(c, "cannot select both norecovery and fsck");
1110                 ret = -EINVAL;
1111                 goto err;
1112         }
1113
1114         ret = bch2_blacklist_table_initialize(c);
1115         if (ret) {
1116                 bch_err(c, "error initializing blacklist table");
1117                 goto err;
1118         }
1119
1120         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1121                 struct genradix_iter iter;
1122                 struct journal_replay **i;
1123
1124                 bch_verbose(c, "starting journal read");
1125                 ret = bch2_journal_read(c, &blacklist_seq, &journal_seq);
1126                 if (ret)
1127                         goto err;
1128
1129                 genradix_for_each_reverse(&c->journal_entries, iter, i)
1130                         if (*i && !(*i)->ignore) {
1131                                 last_journal_entry = &(*i)->j;
1132                                 break;
1133                         }
1134
1135                 if (mustfix_fsck_err_on(c->sb.clean &&
1136                                         last_journal_entry &&
1137                                         !journal_entry_empty(last_journal_entry), c,
1138                                 "filesystem marked clean but journal not empty")) {
1139                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1140                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1141                         c->sb.clean = false;
1142                 }
1143
1144                 if (!last_journal_entry) {
1145                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
1146                         goto use_clean;
1147                 }
1148
1149                 ret = journal_keys_sort(c);
1150                 if (ret)
1151                         goto err;
1152
1153                 if (c->sb.clean && last_journal_entry) {
1154                         ret = verify_superblock_clean(c, &clean,
1155                                                       last_journal_entry);
1156                         if (ret)
1157                                 goto err;
1158                 }
1159         } else {
1160 use_clean:
1161                 if (!clean) {
1162                         bch_err(c, "no superblock clean section found");
1163                         ret = -BCH_ERR_fsck_repair_impossible;
1164                         goto err;
1165
1166                 }
1167                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1168         }
1169
1170         if (c->opts.reconstruct_alloc) {
1171                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1172                 drop_alloc_keys(&c->journal_keys);
1173         }
1174
1175         zero_out_btree_mem_ptr(&c->journal_keys);
1176
1177         ret = journal_replay_early(c, clean);
1178         if (ret)
1179                 goto err;
1180
1181         /*
1182          * After an unclean shutdown, skip then next few journal sequence
1183          * numbers as they may have been referenced by btree writes that
1184          * happened before their corresponding journal writes - those btree
1185          * writes need to be ignored, by skipping and blacklisting the next few
1186          * journal sequence numbers:
1187          */
1188         if (!c->sb.clean)
1189                 journal_seq += 8;
1190
1191         if (blacklist_seq != journal_seq) {
1192                 ret = bch2_journal_seq_blacklist_add(c,
1193                                         blacklist_seq, journal_seq);
1194                 if (ret) {
1195                         bch_err(c, "error creating new journal seq blacklist entry");
1196                         goto err;
1197                 }
1198         }
1199
1200         /*
1201          * note: cmd_list_journal needs the blacklist table fully up to date so
1202          * it can asterisk ignored journal entries:
1203          */
1204         if (c->opts.read_journal_only)
1205                 goto out;
1206
1207         ret = bch2_fs_journal_start(&c->journal, journal_seq);
1208         if (ret)
1209                 goto err;
1210
1211         /*
1212          * Skip past versions that might have possibly been used (as nonces),
1213          * but hadn't had their pointers written:
1214          */
1215         if (c->sb.encryption_type && !c->sb.clean)
1216                 atomic64_add(1 << 16, &c->key_version);
1217
1218         ret = read_btree_roots(c);
1219         if (ret)
1220                 goto err;
1221
1222         bch_verbose(c, "starting alloc read");
1223         err = "error reading allocation information";
1224
1225         down_read(&c->gc_lock);
1226         ret = bch2_alloc_read(c);
1227         up_read(&c->gc_lock);
1228
1229         if (ret)
1230                 goto err;
1231         bch_verbose(c, "alloc read done");
1232
1233         bch_verbose(c, "starting stripes_read");
1234         err = "error reading stripes";
1235         ret = bch2_stripes_read(c);
1236         if (ret)
1237                 goto err;
1238         bch_verbose(c, "stripes_read done");
1239
1240         bch2_stripes_heap_start(c);
1241
1242         if (c->opts.fsck) {
1243                 bool metadata_only = c->opts.norecovery;
1244
1245                 bch_info(c, "checking allocations");
1246                 err = "error checking allocations";
1247                 ret = bch2_gc(c, true, metadata_only);
1248                 if (ret)
1249                         goto err;
1250                 bch_verbose(c, "done checking allocations");
1251
1252                 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1253
1254                 bch_info(c, "checking need_discard and freespace btrees");
1255                 err = "error checking need_discard and freespace btrees";
1256                 ret = bch2_check_alloc_info(c);
1257                 if (ret)
1258                         goto err;
1259                 bch_verbose(c, "done checking need_discard and freespace btrees");
1260
1261                 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1262
1263                 bch_info(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1264                 err = "journal replay failed";
1265                 ret = bch2_journal_replay(c);
1266                 if (ret)
1267                         goto err;
1268                 if (c->opts.verbose || !c->sb.clean)
1269                         bch_info(c, "journal replay done");
1270
1271                 bch_info(c, "checking lrus");
1272                 err = "error checking lrus";
1273                 ret = bch2_check_lrus(c);
1274                 if (ret)
1275                         goto err;
1276                 bch_verbose(c, "done checking lrus");
1277                 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1278
1279                 bch_info(c, "checking backpointers to alloc keys");
1280                 err = "error checking backpointers to alloc keys";
1281                 ret = bch2_check_btree_backpointers(c);
1282                 if (ret)
1283                         goto err;
1284                 bch_verbose(c, "done checking backpointers to alloc keys");
1285
1286                 bch_info(c, "checking backpointers to extents");
1287                 err = "error checking backpointers to extents";
1288                 ret = bch2_check_backpointers_to_extents(c);
1289                 if (ret)
1290                         goto err;
1291                 bch_verbose(c, "done checking backpointers to extents");
1292
1293                 bch_info(c, "checking extents to backpointers");
1294                 err = "error checking extents to backpointers";
1295                 ret = bch2_check_extents_to_backpointers(c);
1296                 if (ret)
1297                         goto err;
1298                 bch_verbose(c, "done checking extents to backpointers");
1299                 set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1300
1301                 bch_info(c, "checking alloc to lru refs");
1302                 err = "error checking alloc to lru refs";
1303                 ret = bch2_check_alloc_to_lru_refs(c);
1304                 if (ret)
1305                         goto err;
1306                 bch_verbose(c, "done checking alloc to lru refs");
1307                 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1308         } else {
1309                 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1310                 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1311                 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1312                 set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1313                 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1314                 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1315
1316                 if (c->opts.norecovery)
1317                         goto out;
1318
1319                 bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1320                 err = "journal replay failed";
1321                 ret = bch2_journal_replay(c);
1322                 if (ret)
1323                         goto err;
1324                 if (c->opts.verbose || !c->sb.clean)
1325                         bch_info(c, "journal replay done");
1326         }
1327
1328         err = "error initializing freespace";
1329         ret = bch2_fs_freespace_init(c);
1330         if (ret)
1331                 goto err;
1332
1333         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1334                 bch2_fs_lazy_rw(c);
1335
1336                 err = "error creating root snapshot node";
1337                 ret = bch2_fs_initialize_subvolumes(c);
1338                 if (ret)
1339                         goto err;
1340         }
1341
1342         bch_verbose(c, "reading snapshots table");
1343         err = "error reading snapshots table";
1344         ret = bch2_fs_snapshots_start(c);
1345         if (ret)
1346                 goto err;
1347         bch_verbose(c, "reading snapshots done");
1348
1349         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1350                 /* set bi_subvol on root inode */
1351                 err = "error upgrade root inode for subvolumes";
1352                 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1353                                     bch2_fs_upgrade_for_subvolumes(&trans));
1354                 if (ret)
1355                         goto err;
1356         }
1357
1358         if (c->opts.fsck) {
1359                 bch_info(c, "starting fsck");
1360                 err = "error in fsck";
1361                 ret = bch2_fsck_full(c);
1362                 if (ret)
1363                         goto err;
1364                 bch_verbose(c, "fsck done");
1365         } else if (!c->sb.clean) {
1366                 bch_verbose(c, "checking for deleted inodes");
1367                 err = "error in recovery";
1368                 ret = bch2_fsck_walk_inodes_only(c);
1369                 if (ret)
1370                         goto err;
1371                 bch_verbose(c, "check inodes done");
1372         }
1373
1374         if (enabled_qtypes(c)) {
1375                 bch_verbose(c, "reading quotas");
1376                 ret = bch2_fs_quota_read(c);
1377                 if (ret)
1378                         goto err;
1379                 bch_verbose(c, "quotas done");
1380         }
1381
1382         mutex_lock(&c->sb_lock);
1383         if (c->opts.version_upgrade) {
1384                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1385                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1386                 write_sb = true;
1387         }
1388
1389         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1390                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1391                 write_sb = true;
1392         }
1393
1394         if (c->opts.fsck &&
1395             !test_bit(BCH_FS_ERROR, &c->flags) &&
1396             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1397                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1398                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1399                 write_sb = true;
1400         }
1401
1402         if (write_sb)
1403                 bch2_write_super(c);
1404         mutex_unlock(&c->sb_lock);
1405
1406         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1407             !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1408             le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) {
1409                 struct bch_move_stats stats;
1410
1411                 bch_move_stats_init(&stats, "recovery");
1412
1413                 bch_info(c, "scanning for old btree nodes");
1414                 ret = bch2_fs_read_write(c);
1415                 if (ret)
1416                         goto err;
1417
1418                 ret = bch2_scan_old_btree_nodes(c, &stats);
1419                 if (ret)
1420                         goto err;
1421                 bch_info(c, "scanning for old btree nodes done");
1422         }
1423
1424         if (c->journal_seq_blacklist_table &&
1425             c->journal_seq_blacklist_table->nr > 128)
1426                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1427
1428         ret = 0;
1429 out:
1430         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1431         bch2_flush_fsck_errs(c);
1432
1433         if (!c->opts.keep_journal) {
1434                 bch2_journal_keys_free(&c->journal_keys);
1435                 bch2_journal_entries_free(c);
1436         }
1437         kfree(clean);
1438
1439         if (!ret && test_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags)) {
1440                 bch2_fs_read_write_early(c);
1441                 bch2_delete_dead_snapshots_async(c);
1442         }
1443
1444         if (ret)
1445                 bch_err(c, "Error in recovery: %s (%s)", err, bch2_err_str(ret));
1446         else
1447                 bch_verbose(c, "ret %s", bch2_err_str(ret));
1448         return ret;
1449 err:
1450 fsck_err:
1451         bch2_fs_emergency_read_only(c);
1452         goto out;
1453 }
1454
1455 int bch2_fs_initialize(struct bch_fs *c)
1456 {
1457         struct bch_inode_unpacked root_inode, lostfound_inode;
1458         struct bkey_inode_buf packed_inode;
1459         struct qstr lostfound = QSTR("lost+found");
1460         const char *err = "cannot allocate memory";
1461         struct bch_dev *ca;
1462         unsigned i;
1463         int ret;
1464
1465         bch_notice(c, "initializing new filesystem");
1466
1467         mutex_lock(&c->sb_lock);
1468         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1469         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1470
1471         if (c->sb.version < bcachefs_metadata_version_backpointers)
1472                 c->opts.version_upgrade = true;
1473
1474         if (c->opts.version_upgrade) {
1475                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1476                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1477                 bch2_write_super(c);
1478         }
1479         mutex_unlock(&c->sb_lock);
1480
1481         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1482         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1483         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1484
1485         for (i = 0; i < BTREE_ID_NR; i++)
1486                 bch2_btree_root_alloc(c, i);
1487
1488         for_each_online_member(ca, c, i)
1489                 bch2_dev_usage_init(ca);
1490
1491         err = "unable to allocate journal buckets";
1492         for_each_online_member(ca, c, i) {
1493                 ret = bch2_dev_journal_alloc(ca);
1494                 if (ret) {
1495                         percpu_ref_put(&ca->io_ref);
1496                         goto err;
1497                 }
1498         }
1499
1500         /*
1501          * journal_res_get() will crash if called before this has
1502          * set up the journal.pin FIFO and journal.cur pointer:
1503          */
1504         bch2_fs_journal_start(&c->journal, 1);
1505         bch2_journal_set_replay_done(&c->journal);
1506
1507         err = "error going read-write";
1508         ret = bch2_fs_read_write_early(c);
1509         if (ret)
1510                 goto err;
1511
1512         /*
1513          * Write out the superblock and journal buckets, now that we can do
1514          * btree updates
1515          */
1516         bch_verbose(c, "marking superblocks");
1517         err = "error marking superblock and journal";
1518         for_each_member_device(ca, c, i) {
1519                 ret = bch2_trans_mark_dev_sb(c, ca);
1520                 if (ret) {
1521                         percpu_ref_put(&ca->ref);
1522                         goto err;
1523                 }
1524
1525                 ca->new_fs_bucket_idx = 0;
1526         }
1527
1528         bch_verbose(c, "initializing freespace");
1529         err = "error initializing freespace";
1530         ret = bch2_fs_freespace_init(c);
1531         if (ret)
1532                 goto err;
1533
1534         err = "error creating root snapshot node";
1535         ret = bch2_fs_initialize_subvolumes(c);
1536         if (ret)
1537                 goto err;
1538
1539         bch_verbose(c, "reading snapshots table");
1540         err = "error reading snapshots table";
1541         ret = bch2_fs_snapshots_start(c);
1542         if (ret)
1543                 goto err;
1544         bch_verbose(c, "reading snapshots done");
1545
1546         bch2_inode_init(c, &root_inode, 0, 0,
1547                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1548         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1549         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1550         bch2_inode_pack(c, &packed_inode, &root_inode);
1551         packed_inode.inode.k.p.snapshot = U32_MAX;
1552
1553         err = "error creating root directory";
1554         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1555                                 &packed_inode.inode.k_i,
1556                                 NULL, NULL, 0);
1557         if (ret)
1558                 goto err;
1559
1560         bch2_inode_init_early(c, &lostfound_inode);
1561
1562         err = "error creating lost+found";
1563         ret = bch2_trans_do(c, NULL, NULL, 0,
1564                 bch2_create_trans(&trans,
1565                                   BCACHEFS_ROOT_SUBVOL_INUM,
1566                                   &root_inode, &lostfound_inode,
1567                                   &lostfound,
1568                                   0, 0, S_IFDIR|0700, 0,
1569                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1570         if (ret) {
1571                 bch_err(c, "error creating lost+found");
1572                 goto err;
1573         }
1574
1575         if (enabled_qtypes(c)) {
1576                 ret = bch2_fs_quota_read(c);
1577                 if (ret)
1578                         goto err;
1579         }
1580
1581         err = "error writing first journal entry";
1582         ret = bch2_journal_flush(&c->journal);
1583         if (ret)
1584                 goto err;
1585
1586         mutex_lock(&c->sb_lock);
1587         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1588         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1589
1590         bch2_write_super(c);
1591         mutex_unlock(&c->sb_lock);
1592
1593         return 0;
1594 err:
1595         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1596         return ret;
1597 }