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 ((k = *idx < keys->nr ? idx_to_key(keys, *idx) : NULL)) {
 133                 if (__journal_key_cmp(btree_id, level, end_pos, k) < 0)
 134                         return NULL;
 135
 136                 if (__journal_key_cmp(btree_id, level, pos, k) <= 0 &&
 137                     !k->overwritten)
 138                         return k->k;
 139
 140                 (*idx)++;
 141                 iters++;
 142                 if (iters == 10) {
 143                         *idx = 0;
 144                         goto search;
 145                 }
 146         }
 147
 148         return NULL;
 149 }
 150
 151 struct bkey_i *bch2_journal_keys_peek_slot(struct bch_fs *c, enum btree_id btree_id,
 152                                            unsigned level, struct bpos pos)
 153 {
 154         size_t idx = 0;
 155
 156         return bch2_journal_keys_peek_upto(c, btree_id, level, pos, pos, &idx);
 157 }
 158
 159 static void journal_iters_fix(struct bch_fs *c)
 160 {
 161         struct journal_keys *keys = &c->journal_keys;
 162         /* The key we just inserted is immediately before the gap: */
 163         size_t gap_end = keys->gap + (keys->size - keys->nr);
 164         struct btree_and_journal_iter *iter;
 165
 166         /*
 167          * If an iterator points one after the key we just inserted, decrement
 168          * the iterator so it points at the key we just inserted - if the
 169          * decrement was unnecessary, bch2_btree_and_journal_iter_peek() will
 170          * handle that:
 171          */
 172         list_for_each_entry(iter, &c->journal_iters, journal.list)
 173                 if (iter->journal.idx == gap_end)
 174                         iter->journal.idx = keys->gap - 1;
 175 }
 176
 177 static void journal_iters_move_gap(struct bch_fs *c, size_t old_gap, size_t new_gap)
 178 {
 179         struct journal_keys *keys = &c->journal_keys;
 180         struct journal_iter *iter;
 181         size_t gap_size = keys->size - keys->nr;
 182
 183         list_for_each_entry(iter, &c->journal_iters, list) {
 184                 if (iter->idx > old_gap)
 185                         iter->idx -= gap_size;
 186                 if (iter->idx >= new_gap)
 187                         iter->idx += gap_size;
 188         }
 189 }
 190
 191 int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
 192                                  unsigned level, struct bkey_i *k)
 193 {
 194         struct journal_key n = {
 195                 .btree_id       = id,
 196                 .level          = level,
 197                 .k              = k,
 198                 .allocated      = true,
 199                 /*
 200                  * Ensure these keys are done last by journal replay, to unblock
 201                  * journal reclaim:
 202                  */
 203                 .journal_seq    = U32_MAX,
 204         };
 205         struct journal_keys *keys = &c->journal_keys;
 206         size_t idx = bch2_journal_key_search(keys, id, level, k->k.p);
 207
 208         BUG_ON(test_bit(BCH_FS_RW, &c->flags));
 209
 210         if (idx < keys->size &&
 211             journal_key_cmp(&n, &keys->d[idx]) == 0) {
 212                 if (keys->d[idx].allocated)
 213                         kfree(keys->d[idx].k);
 214                 keys->d[idx] = n;
 215                 return 0;
 216         }
 217
 218         if (idx > keys->gap)
 219                 idx -= keys->size - keys->nr;
 220
 221         if (keys->nr == keys->size) {
 222                 struct journal_keys new_keys = {
 223                         .nr                     = keys->nr,
 224                         .size                   = max_t(size_t, keys->size, 8) * 2,
 225                 };
 226
 227                 new_keys.d = kvmalloc_array(new_keys.size, sizeof(new_keys.d[0]), GFP_KERNEL);
 228                 if (!new_keys.d) {
 229                         bch_err(c, "%s: error allocating new key array (size %zu)",
 230                                 __func__, new_keys.size);
 231                         return -BCH_ERR_ENOMEM_journal_key_insert;
 232                 }
 233
 234                 /* Since @keys was full, there was no gap: */
 235                 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
 236                 kvfree(keys->d);
 237                 *keys = new_keys;
 238
 239                 /* And now the gap is at the end: */
 240                 keys->gap = keys->nr;
 241         }
 242
 243         journal_iters_move_gap(c, keys->gap, idx);
 244
 245         move_gap(keys->d, keys->nr, keys->size, keys->gap, idx);
 246         keys->gap = idx;
 247
 248         keys->nr++;
 249         keys->d[keys->gap++] = n;
 250
 251         journal_iters_fix(c);
 252
 253         return 0;
 254 }
 255
 256 /*
 257  * Can only be used from the recovery thread while we're still RO - can't be
 258  * used once we've got RW, as journal_keys is at that point used by multiple
 259  * threads:
 260  */
 261 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
 262                             unsigned level, struct bkey_i *k)
 263 {
 264         struct bkey_i *n;
 265         int ret;
 266
 267         n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
 268         if (!n)
 269                 return -BCH_ERR_ENOMEM_journal_key_insert;
 270
 271         bkey_copy(n, k);
 272         ret = bch2_journal_key_insert_take(c, id, level, n);
 273         if (ret)
 274                 kfree(n);
 275         return ret;
 276 }
 277
 278 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
 279                             unsigned level, struct bpos pos)
 280 {
 281         struct bkey_i whiteout;
 282
 283         bkey_init(&whiteout.k);
 284         whiteout.k.p = pos;
 285
 286         return bch2_journal_key_insert(c, id, level, &whiteout);
 287 }
 288
 289 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
 290                                   unsigned level, struct bpos pos)
 291 {
 292         struct journal_keys *keys = &c->journal_keys;
 293         size_t idx = bch2_journal_key_search(keys, btree, level, pos);
 294
 295         if (idx < keys->size &&
 296             keys->d[idx].btree_id       == btree &&
 297             keys->d[idx].level          == level &&
 298             bpos_eq(keys->d[idx].k->k.p, pos))
 299                 keys->d[idx].overwritten = true;
 300 }
 301
 302 static void bch2_journal_iter_advance(struct journal_iter *iter)
 303 {
 304         if (iter->idx < iter->keys->size) {
 305                 iter->idx++;
 306                 if (iter->idx == iter->keys->gap)
 307                         iter->idx += iter->keys->size - iter->keys->nr;
 308         }
 309 }
 310
 311 struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
 312 {
 313         struct journal_key *k = iter->keys->d + iter->idx;
 314
 315         while (k < iter->keys->d + iter->keys->size &&
 316                k->btree_id      == iter->btree_id &&
 317                k->level         == iter->level) {
 318                 if (!k->overwritten)
 319                         return bkey_i_to_s_c(k->k);
 320
 321                 bch2_journal_iter_advance(iter);
 322                 k = iter->keys->d + iter->idx;
 323         }
 324
 325         return bkey_s_c_null;
 326 }
 327
 328 static void bch2_journal_iter_exit(struct journal_iter *iter)
 329 {
 330         list_del(&iter->list);
 331 }
 332
 333 static void bch2_journal_iter_init(struct bch_fs *c,
 334                                    struct journal_iter *iter,
 335                                    enum btree_id id, unsigned level,
 336                                    struct bpos pos)
 337 {
 338         iter->btree_id  = id;
 339         iter->level     = level;
 340         iter->keys      = &c->journal_keys;
 341         iter->idx       = bch2_journal_key_search(&c->journal_keys, id, level, pos);
 342 }
 343
 344 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 345 {
 346         return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 347                                                 iter->b, &iter->unpacked);
 348 }
 349
 350 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 351 {
 352         bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 353 }
 354
 355 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 356 {
 357         if (bpos_eq(iter->pos, SPOS_MAX))
 358                 iter->at_end = true;
 359         else
 360                 iter->pos = bpos_successor(iter->pos);
 361 }
 362
 363 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 364 {
 365         struct bkey_s_c btree_k, journal_k, ret;
 366 again:
 367         if (iter->at_end)
 368                 return bkey_s_c_null;
 369
 370         while ((btree_k = bch2_journal_iter_peek_btree(iter)).k &&
 371                bpos_lt(btree_k.k->p, iter->pos))
 372                 bch2_journal_iter_advance_btree(iter);
 373
 374         while ((journal_k = bch2_journal_iter_peek(&iter->journal)).k &&
 375                bpos_lt(journal_k.k->p, iter->pos))
 376                 bch2_journal_iter_advance(&iter->journal);
 377
 378         ret = journal_k.k &&
 379                 (!btree_k.k || bpos_le(journal_k.k->p, btree_k.k->p))
 380                 ? journal_k
 381                 : btree_k;
 382
 383         if (ret.k && iter->b && bpos_gt(ret.k->p, iter->b->data->max_key))
 384                 ret = bkey_s_c_null;
 385
 386         if (ret.k) {
 387                 iter->pos = ret.k->p;
 388                 if (bkey_deleted(ret.k)) {
 389                         bch2_btree_and_journal_iter_advance(iter);
 390                         goto again;
 391                 }
 392         } else {
 393                 iter->pos = SPOS_MAX;
 394                 iter->at_end = true;
 395         }
 396
 397         return ret;
 398 }
 399
 400 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
 401 {
 402         bch2_journal_iter_exit(&iter->journal);
 403 }
 404
 405 void __bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 406                                                   struct bch_fs *c,
 407                                                   struct btree *b,
 408                                                   struct btree_node_iter node_iter,
 409                                                   struct bpos pos)
 410 {
 411         memset(iter, 0, sizeof(*iter));
 412
 413         iter->b = b;
 414         iter->node_iter = node_iter;
 415         bch2_journal_iter_init(c, &iter->journal, b->c.btree_id, b->c.level, pos);
 416         INIT_LIST_HEAD(&iter->journal.list);
 417         iter->pos = b->data->min_key;
 418         iter->at_end = false;
 419 }
 420
 421 /*
 422  * this version is used by btree_gc before filesystem has gone RW and
 423  * multithreaded, so uses the journal_iters list:
 424  */
 425 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 426                                                 struct bch_fs *c,
 427                                                 struct btree *b)
 428 {
 429         struct btree_node_iter node_iter;
 430
 431         bch2_btree_node_iter_init_from_start(&node_iter, b);
 432         __bch2_btree_and_journal_iter_init_node_iter(iter, c, b, node_iter, b->data->min_key);
 433         list_add(&iter->journal.list, &c->journal_iters);
 434 }
 435
 436 /* sort and dedup all keys in the journal: */
 437
 438 void bch2_journal_entries_free(struct bch_fs *c)
 439 {
 440         struct journal_replay **i;
 441         struct genradix_iter iter;
 442
 443         genradix_for_each(&c->journal_entries, iter, i)
 444                 if (*i)
 445                         kvpfree(*i, offsetof(struct journal_replay, j) +
 446                                 vstruct_bytes(&(*i)->j));
 447         genradix_free(&c->journal_entries);
 448 }
 449
 450 /*
 451  * When keys compare equal, oldest compares first:
 452  */
 453 static int journal_sort_key_cmp(const void *_l, const void *_r)
 454 {
 455         const struct journal_key *l = _l;
 456         const struct journal_key *r = _r;
 457
 458         return  journal_key_cmp(l, r) ?:
 459                 cmp_int(l->journal_seq, r->journal_seq) ?:
 460                 cmp_int(l->journal_offset, r->journal_offset);
 461 }
 462
 463 void bch2_journal_keys_free(struct journal_keys *keys)
 464 {
 465         struct journal_key *i;
 466
 467         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 468         keys->gap = keys->nr;
 469
 470         for (i = keys->d; i < keys->d + keys->nr; i++)
 471                 if (i->allocated)
 472                         kfree(i->k);
 473
 474         kvfree(keys->d);
 475         keys->d = NULL;
 476         keys->nr = keys->gap = keys->size = 0;
 477 }
 478
 479 static int journal_keys_sort(struct bch_fs *c)
 480 {
 481         struct genradix_iter iter;
 482         struct journal_replay *i, **_i;
 483         struct jset_entry *entry;
 484         struct bkey_i *k;
 485         struct journal_keys *keys = &c->journal_keys;
 486         struct journal_key *src, *dst;
 487         size_t nr_keys = 0;
 488
 489         genradix_for_each(&c->journal_entries, iter, _i) {
 490                 i = *_i;
 491
 492                 if (!i || i->ignore)
 493                         continue;
 494
 495                 for_each_jset_key(k, entry, &i->j)
 496                         nr_keys++;
 497         }
 498
 499         if (!nr_keys)
 500                 return 0;
 501
 502         keys->size = roundup_pow_of_two(nr_keys);
 503
 504         keys->d = kvmalloc_array(keys->size, sizeof(keys->d[0]), GFP_KERNEL);
 505         if (!keys->d) {
 506                 bch_err(c, "Failed to allocate buffer for sorted journal keys (%zu keys)",
 507                         nr_keys);
 508                 return -BCH_ERR_ENOMEM_journal_keys_sort;
 509         }
 510
 511         genradix_for_each(&c->journal_entries, iter, _i) {
 512                 i = *_i;
 513
 514                 if (!i || i->ignore)
 515                         continue;
 516
 517                 for_each_jset_key(k, entry, &i->j)
 518                         keys->d[keys->nr++] = (struct journal_key) {
 519                                 .btree_id       = entry->btree_id,
 520                                 .level          = entry->level,
 521                                 .k              = k,
 522                                 .journal_seq    = le64_to_cpu(i->j.seq),
 523                                 .journal_offset = k->_data - i->j._data,
 524                         };
 525         }
 526
 527         sort(keys->d, keys->nr, sizeof(keys->d[0]), journal_sort_key_cmp, NULL);
 528
 529         src = dst = keys->d;
 530         while (src < keys->d + keys->nr) {
 531                 while (src + 1 < keys->d + keys->nr &&
 532                        src[0].btree_id  == src[1].btree_id &&
 533                        src[0].level     == src[1].level &&
 534                        bpos_eq(src[0].k->k.p, src[1].k->k.p))
 535                         src++;
 536
 537                 *dst++ = *src++;
 538         }
 539
 540         keys->nr = dst - keys->d;
 541         keys->gap = keys->nr;
 542         return 0;
 543 }
 544
 545 /* journal replay: */
 546
 547 static void replay_now_at(struct journal *j, u64 seq)
 548 {
 549         BUG_ON(seq < j->replay_journal_seq);
 550
 551         seq = min(seq, j->replay_journal_seq_end);
 552
 553         while (j->replay_journal_seq < seq)
 554                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 555 }
 556
 557 static int bch2_journal_replay_key(struct btree_trans *trans,
 558                                    struct journal_key *k)
 559 {
 560         struct btree_iter iter;
 561         unsigned iter_flags =
 562                 BTREE_ITER_INTENT|
 563                 BTREE_ITER_NOT_EXTENTS;
 564         int ret;
 565
 566         if (!k->level && k->btree_id == BTREE_ID_alloc)
 567                 iter_flags |= BTREE_ITER_CACHED;
 568
 569         bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 570                                   BTREE_MAX_DEPTH, k->level,
 571                                   iter_flags);
 572         ret = bch2_btree_iter_traverse(&iter);
 573         if (ret)
 574                 goto out;
 575
 576         /* Must be checked with btree locked: */
 577         if (k->overwritten)
 578                 goto out;
 579
 580         ret = bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
 581 out:
 582         bch2_trans_iter_exit(trans, &iter);
 583         return ret;
 584 }
 585
 586 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 587 {
 588         const struct journal_key *l = *((const struct journal_key **)_l);
 589         const struct journal_key *r = *((const struct journal_key **)_r);
 590
 591         return cmp_int(l->journal_seq, r->journal_seq);
 592 }
 593
 594 static int bch2_journal_replay(struct bch_fs *c, u64 start_seq, u64 end_seq)
 595 {
 596         struct journal_keys *keys = &c->journal_keys;
 597         struct journal_key **keys_sorted, *k;
 598         struct journal *j = &c->journal;
 599         size_t i;
 600         int ret;
 601
 602         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 603         keys->gap = keys->nr;
 604
 605         keys_sorted = kvmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL);
 606         if (!keys_sorted)
 607                 return -BCH_ERR_ENOMEM_journal_replay;
 608
 609         for (i = 0; i < keys->nr; i++)
 610                 keys_sorted[i] = &keys->d[i];
 611
 612         sort(keys_sorted, keys->nr,
 613              sizeof(keys_sorted[0]),
 614              journal_sort_seq_cmp, NULL);
 615
 616         if (keys->nr) {
 617                 ret = bch2_fs_log_msg(c, "Starting journal replay (%zu keys in entries %llu-%llu)",
 618                                       keys->nr, start_seq, end_seq);
 619                 if (ret)
 620                         goto err;
 621         }
 622
 623         for (i = 0; i < keys->nr; i++) {
 624                 k = keys_sorted[i];
 625
 626                 cond_resched();
 627
 628                 replay_now_at(j, k->journal_seq);
 629
 630                 ret = bch2_trans_do(c, NULL, NULL,
 631                                     BTREE_INSERT_LAZY_RW|
 632                                     BTREE_INSERT_NOFAIL|
 633                                     (!k->allocated
 634                                      ? BTREE_INSERT_JOURNAL_REPLAY|JOURNAL_WATERMARK_reserved
 635                                      : 0),
 636                              bch2_journal_replay_key(&trans, k));
 637                 if (ret) {
 638                         bch_err(c, "journal replay: error while replaying key at btree %s level %u: %s",
 639                                 bch2_btree_ids[k->btree_id], k->level, bch2_err_str(ret));
 640                         goto err;
 641                 }
 642         }
 643
 644         replay_now_at(j, j->replay_journal_seq_end);
 645         j->replay_journal_seq = 0;
 646
 647         bch2_journal_set_replay_done(j);
 648         bch2_journal_flush_all_pins(j);
 649         ret = bch2_journal_error(j);
 650
 651         if (keys->nr && !ret)
 652                 bch2_fs_log_msg(c, "journal replay finished");
 653 err:
 654         kvfree(keys_sorted);
 655         return ret;
 656 }
 657
 658 /* journal replay early: */
 659
 660 static int journal_replay_entry_early(struct bch_fs *c,
 661                                       struct jset_entry *entry)
 662 {
 663         int ret = 0;
 664
 665         switch (entry->type) {
 666         case BCH_JSET_ENTRY_btree_root: {
 667                 struct btree_root *r;
 668
 669                 if (entry->btree_id >= BTREE_ID_NR) {
 670                         bch_err(c, "filesystem has unknown btree type %u",
 671                                 entry->btree_id);
 672                         return -EINVAL;
 673                 }
 674
 675                 r = &c->btree_roots[entry->btree_id];
 676
 677                 if (entry->u64s) {
 678                         r->level = entry->level;
 679                         bkey_copy(&r->key, &entry->start[0]);
 680                         r->error = 0;
 681                 } else {
 682                         r->error = -EIO;
 683                 }
 684                 r->alive = true;
 685                 break;
 686         }
 687         case BCH_JSET_ENTRY_usage: {
 688                 struct jset_entry_usage *u =
 689                         container_of(entry, struct jset_entry_usage, entry);
 690
 691                 switch (entry->btree_id) {
 692                 case BCH_FS_USAGE_reserved:
 693                         if (entry->level < BCH_REPLICAS_MAX)
 694                                 c->usage_base->persistent_reserved[entry->level] =
 695                                         le64_to_cpu(u->v);
 696                         break;
 697                 case BCH_FS_USAGE_inodes:
 698                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 699                         break;
 700                 case BCH_FS_USAGE_key_version:
 701                         atomic64_set(&c->key_version,
 702                                      le64_to_cpu(u->v));
 703                         break;
 704                 }
 705
 706                 break;
 707         }
 708         case BCH_JSET_ENTRY_data_usage: {
 709                 struct jset_entry_data_usage *u =
 710                         container_of(entry, struct jset_entry_data_usage, entry);
 711
 712                 ret = bch2_replicas_set_usage(c, &u->r,
 713                                               le64_to_cpu(u->v));
 714                 break;
 715         }
 716         case BCH_JSET_ENTRY_dev_usage: {
 717                 struct jset_entry_dev_usage *u =
 718                         container_of(entry, struct jset_entry_dev_usage, entry);
 719                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 720                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 721
 722                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 723
 724                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 725                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 726                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 727                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 728                 }
 729
 730                 break;
 731         }
 732         case BCH_JSET_ENTRY_blacklist: {
 733                 struct jset_entry_blacklist *bl_entry =
 734                         container_of(entry, struct jset_entry_blacklist, entry);
 735
 736                 ret = bch2_journal_seq_blacklist_add(c,
 737                                 le64_to_cpu(bl_entry->seq),
 738                                 le64_to_cpu(bl_entry->seq) + 1);
 739                 break;
 740         }
 741         case BCH_JSET_ENTRY_blacklist_v2: {
 742                 struct jset_entry_blacklist_v2 *bl_entry =
 743                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 744
 745                 ret = bch2_journal_seq_blacklist_add(c,
 746                                 le64_to_cpu(bl_entry->start),
 747                                 le64_to_cpu(bl_entry->end) + 1);
 748                 break;
 749         }
 750         case BCH_JSET_ENTRY_clock: {
 751                 struct jset_entry_clock *clock =
 752                         container_of(entry, struct jset_entry_clock, entry);
 753
 754                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 755         }
 756         }
 757
 758         return ret;
 759 }
 760
 761 static int journal_replay_early(struct bch_fs *c,
 762                                 struct bch_sb_field_clean *clean)
 763 {
 764         struct jset_entry *entry;
 765         int ret;
 766
 767         if (clean) {
 768                 for (entry = clean->start;
 769                      entry != vstruct_end(&clean->field);
 770                      entry = vstruct_next(entry)) {
 771                         ret = journal_replay_entry_early(c, entry);
 772                         if (ret)
 773                                 return ret;
 774                 }
 775         } else {
 776                 struct genradix_iter iter;
 777                 struct journal_replay *i, **_i;
 778
 779                 genradix_for_each(&c->journal_entries, iter, _i) {
 780                         i = *_i;
 781
 782                         if (!i || i->ignore)
 783                                 continue;
 784
 785                         vstruct_for_each(&i->j, entry) {
 786                                 ret = journal_replay_entry_early(c, entry);
 787                                 if (ret)
 788                                         return ret;
 789                         }
 790                 }
 791         }
 792
 793         bch2_fs_usage_initialize(c);
 794
 795         return 0;
 796 }
 797
 798 /* sb clean section: */
 799
 800 static struct bkey_i *btree_root_find(struct bch_fs *c,
 801                                       struct bch_sb_field_clean *clean,
 802                                       struct jset *j,
 803                                       enum btree_id id, unsigned *level)
 804 {
 805         struct bkey_i *k;
 806         struct jset_entry *entry, *start, *end;
 807
 808         if (clean) {
 809                 start = clean->start;
 810                 end = vstruct_end(&clean->field);
 811         } else {
 812                 start = j->start;
 813                 end = vstruct_last(j);
 814         }
 815
 816         for (entry = start; entry < end; entry = vstruct_next(entry))
 817                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 818                     entry->btree_id == id)
 819                         goto found;
 820
 821         return NULL;
 822 found:
 823         if (!entry->u64s)
 824                 return ERR_PTR(-EINVAL);
 825
 826         k = entry->start;
 827         *level = entry->level;
 828         return k;
 829 }
 830
 831 static int verify_superblock_clean(struct bch_fs *c,
 832                                    struct bch_sb_field_clean **cleanp,
 833                                    struct jset *j)
 834 {
 835         unsigned i;
 836         struct bch_sb_field_clean *clean = *cleanp;
 837         struct printbuf buf1 = PRINTBUF;
 838         struct printbuf buf2 = PRINTBUF;
 839         int ret = 0;
 840
 841         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 842                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 843                         le64_to_cpu(clean->journal_seq),
 844                         le64_to_cpu(j->seq))) {
 845                 kfree(clean);
 846                 *cleanp = NULL;
 847                 return 0;
 848         }
 849
 850         for (i = 0; i < BTREE_ID_NR; i++) {
 851                 struct bkey_i *k1, *k2;
 852                 unsigned l1 = 0, l2 = 0;
 853
 854                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 855                 k2 = btree_root_find(c, NULL, j, i, &l2);
 856
 857                 if (!k1 && !k2)
 858                         continue;
 859
 860                 printbuf_reset(&buf1);
 861                 printbuf_reset(&buf2);
 862
 863                 if (k1)
 864                         bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(k1));
 865                 else
 866                         prt_printf(&buf1, "(none)");
 867
 868                 if (k2)
 869                         bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(k2));
 870                 else
 871                         prt_printf(&buf2, "(none)");
 872
 873                 mustfix_fsck_err_on(!k1 || !k2 ||
 874                                     IS_ERR(k1) ||
 875                                     IS_ERR(k2) ||
 876                                     k1->k.u64s != k2->k.u64s ||
 877                                     memcmp(k1, k2, bkey_bytes(&k1->k)) ||
 878                                     l1 != l2, c,
 879                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 880                         "sb:      l=%u %s\n"
 881                         "journal: l=%u %s\n", i,
 882                         l1, buf1.buf,
 883                         l2, buf2.buf);
 884         }
 885 fsck_err:
 886         printbuf_exit(&buf2);
 887         printbuf_exit(&buf1);
 888         return ret;
 889 }
 890
 891 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 892 {
 893         struct bch_sb_field_clean *clean, *sb_clean;
 894         int ret;
 895
 896         mutex_lock(&c->sb_lock);
 897         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 898
 899         if (fsck_err_on(!sb_clean, c,
 900                         "superblock marked clean but clean section not present")) {
 901                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 902                 c->sb.clean = false;
 903                 mutex_unlock(&c->sb_lock);
 904                 return NULL;
 905         }
 906
 907         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 908                         GFP_KERNEL);
 909         if (!clean) {
 910                 mutex_unlock(&c->sb_lock);
 911                 return ERR_PTR(-BCH_ERR_ENOMEM_read_superblock_clean);
 912         }
 913
 914         ret = bch2_sb_clean_validate_late(c, clean, READ);
 915         if (ret) {
 916                 mutex_unlock(&c->sb_lock);
 917                 return ERR_PTR(ret);
 918         }
 919
 920         mutex_unlock(&c->sb_lock);
 921
 922         return clean;
 923 fsck_err:
 924         mutex_unlock(&c->sb_lock);
 925         return ERR_PTR(ret);
 926 }
 927
 928 static bool btree_id_is_alloc(enum btree_id id)
 929 {
 930         switch (id) {
 931         case BTREE_ID_alloc:
 932         case BTREE_ID_backpointers:
 933         case BTREE_ID_need_discard:
 934         case BTREE_ID_freespace:
 935         case BTREE_ID_bucket_gens:
 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 (btree_id_is_alloc(i))
 976                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 977                 }
 978         }
 979
 980         for (i = 0; i < BTREE_ID_NR; i++) {
 981                 struct btree_root *r = &c->btree_roots[i];
 982
 983                 if (!r->b) {
 984                         r->alive = false;
 985                         r->level = 0;
 986                         bch2_btree_root_alloc(c, i);
 987                 }
 988         }
 989 fsck_err:
 990         return ret;
 991 }
 992
 993 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
 994 {
 995         struct bkey_i_snapshot  root_snapshot;
 996         struct bkey_i_subvolume root_volume;
 997         int ret;
 998
 999         bkey_snapshot_init(&root_snapshot.k_i);
1000         root_snapshot.k.p.offset = U32_MAX;
1001         root_snapshot.v.flags   = 0;
1002         root_snapshot.v.parent  = 0;
1003         root_snapshot.v.subvol  = BCACHEFS_ROOT_SUBVOL;
1004         root_snapshot.v.pad     = 0;
1005         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
1006
1007         ret = bch2_btree_insert(c, BTREE_ID_snapshots,
1008                                 &root_snapshot.k_i,
1009                                 NULL, NULL, 0);
1010         if (ret)
1011                 return ret;
1012
1013         bkey_subvolume_init(&root_volume.k_i);
1014         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1015         root_volume.v.flags     = 0;
1016         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
1017         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
1018
1019         ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
1020                                 &root_volume.k_i,
1021                                 NULL, NULL, 0);
1022         if (ret)
1023                 return ret;
1024
1025         return 0;
1026 }
1027
1028 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
1029 {
1030         struct btree_iter iter;
1031         struct bkey_s_c k;
1032         struct bch_inode_unpacked inode;
1033         int ret;
1034
1035         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
1036                              SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
1037         k = bch2_btree_iter_peek_slot(&iter);
1038         ret = bkey_err(k);
1039         if (ret)
1040                 goto err;
1041
1042         if (!bkey_is_inode(k.k)) {
1043                 bch_err(trans->c, "root inode not found");
1044                 ret = -ENOENT;
1045                 goto err;
1046         }
1047
1048         ret = bch2_inode_unpack(k, &inode);
1049         BUG_ON(ret);
1050
1051         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1052
1053         ret = bch2_inode_write(trans, &iter, &inode);
1054 err:
1055         bch2_trans_iter_exit(trans, &iter);
1056         return ret;
1057 }
1058
1059 int bch2_fs_recovery(struct bch_fs *c)
1060 {
1061         const char *err = "cannot allocate memory";
1062         struct bch_sb_field_clean *clean = NULL;
1063         struct jset *last_journal_entry = NULL;
1064         u64 last_seq, blacklist_seq, journal_seq;
1065         bool write_sb = false;
1066         int ret = 0;
1067
1068         if (c->sb.clean)
1069                 clean = read_superblock_clean(c);
1070         ret = PTR_ERR_OR_ZERO(clean);
1071         if (ret)
1072                 goto err;
1073
1074         if (c->sb.clean)
1075                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
1076                          le64_to_cpu(clean->journal_seq));
1077         else
1078                 bch_info(c, "recovering from unclean shutdown");
1079
1080         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
1081                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
1082                 ret = -EINVAL;
1083                 goto err;
1084         }
1085
1086         if (!c->sb.clean &&
1087             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1088                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1089                 ret = -EINVAL;
1090                 goto err;
1091         }
1092
1093         if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1094                 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
1095                 ret = -EINVAL;
1096                 goto err;
1097         }
1098
1099         if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
1100                 bch_info(c, "alloc_v2 feature bit not set, fsck required");
1101                 c->opts.fsck = true;
1102                 c->opts.fix_errors = FSCK_OPT_YES;
1103         }
1104
1105         if (!c->opts.nochanges) {
1106                 if (c->sb.version < bcachefs_metadata_version_lru_v2) {
1107                         bch_info(c, "version prior to backpointers, upgrade and fsck required");
1108                         c->opts.version_upgrade = true;
1109                         c->opts.fsck            = true;
1110                         c->opts.fix_errors      = FSCK_OPT_YES;
1111                 } else if (c->sb.version < bcachefs_metadata_version_fragmentation_lru) {
1112                         bch_info(c, "version prior to backpointers, upgrade required");
1113                         c->opts.version_upgrade = true;
1114                 }
1115         }
1116
1117         if (c->opts.fsck && c->opts.norecovery) {
1118                 bch_err(c, "cannot select both norecovery and fsck");
1119                 ret = -EINVAL;
1120                 goto err;
1121         }
1122
1123         ret = bch2_blacklist_table_initialize(c);
1124         if (ret) {
1125                 bch_err(c, "error initializing blacklist table");
1126                 goto err;
1127         }
1128
1129         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1130                 struct genradix_iter iter;
1131                 struct journal_replay **i;
1132
1133                 bch_verbose(c, "starting journal read");
1134                 ret = bch2_journal_read(c, &last_seq, &blacklist_seq, &journal_seq);
1135                 if (ret)
1136                         goto err;
1137
1138                 /*
1139                  * note: cmd_list_journal needs the blacklist table fully up to date so
1140                  * it can asterisk ignored journal entries:
1141                  */
1142                 if (c->opts.read_journal_only)
1143                         goto out;
1144
1145                 genradix_for_each_reverse(&c->journal_entries, iter, i)
1146                         if (*i && !(*i)->ignore) {
1147                                 last_journal_entry = &(*i)->j;
1148                                 break;
1149                         }
1150
1151                 if (mustfix_fsck_err_on(c->sb.clean &&
1152                                         last_journal_entry &&
1153                                         !journal_entry_empty(last_journal_entry), c,
1154                                 "filesystem marked clean but journal not empty")) {
1155                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1156                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1157                         c->sb.clean = false;
1158                 }
1159
1160                 if (!last_journal_entry) {
1161                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
1162                         if (clean)
1163                                 goto use_clean;
1164
1165                         genradix_for_each_reverse(&c->journal_entries, iter, i)
1166                                 if (*i) {
1167                                         last_journal_entry = &(*i)->j;
1168                                         (*i)->ignore = false;
1169                                         break;
1170                                 }
1171                 }
1172
1173                 ret = journal_keys_sort(c);
1174                 if (ret)
1175                         goto err;
1176
1177                 if (c->sb.clean && last_journal_entry) {
1178                         ret = verify_superblock_clean(c, &clean,
1179                                                       last_journal_entry);
1180                         if (ret)
1181                                 goto err;
1182                 }
1183         } else {
1184 use_clean:
1185                 if (!clean) {
1186                         bch_err(c, "no superblock clean section found");
1187                         ret = -BCH_ERR_fsck_repair_impossible;
1188                         goto err;
1189
1190                 }
1191                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1192         }
1193
1194         if (c->opts.reconstruct_alloc) {
1195                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1196                 drop_alloc_keys(&c->journal_keys);
1197         }
1198
1199         zero_out_btree_mem_ptr(&c->journal_keys);
1200
1201         ret = journal_replay_early(c, clean);
1202         if (ret)
1203                 goto err;
1204
1205         /*
1206          * After an unclean shutdown, skip then next few journal sequence
1207          * numbers as they may have been referenced by btree writes that
1208          * happened before their corresponding journal writes - those btree
1209          * writes need to be ignored, by skipping and blacklisting the next few
1210          * journal sequence numbers:
1211          */
1212         if (!c->sb.clean)
1213                 journal_seq += 8;
1214
1215         if (blacklist_seq != journal_seq) {
1216                 ret =   bch2_fs_log_msg(c, "blacklisting entries %llu-%llu",
1217                                         blacklist_seq, journal_seq) ?:
1218                         bch2_journal_seq_blacklist_add(c,
1219                                         blacklist_seq, journal_seq);
1220                 if (ret) {
1221                         bch_err(c, "error creating new journal seq blacklist entry");
1222                         goto err;
1223                 }
1224         }
1225
1226         ret =   bch2_fs_log_msg(c, "starting journal at entry %llu, replaying %llu-%llu",
1227                                 journal_seq, last_seq, blacklist_seq - 1) ?:
1228                 bch2_fs_journal_start(&c->journal, journal_seq);
1229         if (ret)
1230                 goto err;
1231
1232         if (c->opts.reconstruct_alloc)
1233                 bch2_fs_log_msg(c, "dropping alloc info");
1234
1235         /*
1236          * Skip past versions that might have possibly been used (as nonces),
1237          * but hadn't had their pointers written:
1238          */
1239         if (c->sb.encryption_type && !c->sb.clean)
1240                 atomic64_add(1 << 16, &c->key_version);
1241
1242         ret = read_btree_roots(c);
1243         if (ret)
1244                 goto err;
1245
1246         bch_verbose(c, "starting alloc read");
1247         err = "error reading allocation information";
1248
1249         down_read(&c->gc_lock);
1250         ret = c->sb.version < bcachefs_metadata_version_bucket_gens
1251                 ? bch2_alloc_read(c)
1252                 : bch2_bucket_gens_read(c);
1253         up_read(&c->gc_lock);
1254
1255         if (ret)
1256                 goto err;
1257         bch_verbose(c, "alloc read done");
1258
1259         bch_verbose(c, "starting stripes_read");
1260         err = "error reading stripes";
1261         ret = bch2_stripes_read(c);
1262         if (ret)
1263                 goto err;
1264         bch_verbose(c, "stripes_read done");
1265
1266         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1267                 err = "error creating root snapshot node";
1268                 ret = bch2_fs_initialize_subvolumes(c);
1269                 if (ret)
1270                         goto err;
1271         }
1272
1273         bch_verbose(c, "reading snapshots table");
1274         err = "error reading snapshots table";
1275         ret = bch2_fs_snapshots_start(c);
1276         if (ret)
1277                 goto err;
1278         bch_verbose(c, "reading snapshots done");
1279
1280         if (c->opts.fsck) {
1281                 bool metadata_only = c->opts.norecovery;
1282
1283                 bch_info(c, "checking allocations");
1284                 err = "error checking allocations";
1285                 ret = bch2_gc(c, true, metadata_only);
1286                 if (ret)
1287                         goto err;
1288                 bch_verbose(c, "done checking allocations");
1289
1290                 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1291
1292                 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1293
1294                 bch_info(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1295                 err = "journal replay failed";
1296                 ret = bch2_journal_replay(c, last_seq, blacklist_seq - 1);
1297                 if (ret)
1298                         goto err;
1299                 if (c->opts.verbose || !c->sb.clean)
1300                         bch_info(c, "journal replay done");
1301
1302                 bch_info(c, "checking need_discard and freespace btrees");
1303                 err = "error checking need_discard and freespace btrees";
1304                 ret = bch2_check_alloc_info(c);
1305                 if (ret)
1306                         goto err;
1307                 bch_verbose(c, "done checking need_discard and freespace btrees");
1308
1309                 set_bit(BCH_FS_CHECK_ALLOC_DONE, &c->flags);
1310
1311                 bch_info(c, "checking lrus");
1312                 err = "error checking lrus";
1313                 ret = bch2_check_lrus(c);
1314                 if (ret)
1315                         goto err;
1316                 bch_verbose(c, "done checking lrus");
1317                 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1318
1319                 bch_info(c, "checking backpointers to alloc keys");
1320                 err = "error checking backpointers to alloc keys";
1321                 ret = bch2_check_btree_backpointers(c);
1322                 if (ret)
1323                         goto err;
1324                 bch_verbose(c, "done checking backpointers to alloc keys");
1325
1326                 bch_info(c, "checking backpointers to extents");
1327                 err = "error checking backpointers to extents";
1328                 ret = bch2_check_backpointers_to_extents(c);
1329                 if (ret)
1330                         goto err;
1331                 bch_verbose(c, "done checking backpointers to extents");
1332
1333                 bch_info(c, "checking extents to backpointers");
1334                 err = "error checking extents to backpointers";
1335                 ret = bch2_check_extents_to_backpointers(c);
1336                 if (ret)
1337                         goto err;
1338                 bch_verbose(c, "done checking extents to backpointers");
1339                 set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1340
1341                 bch_info(c, "checking alloc to lru refs");
1342                 err = "error checking alloc to lru refs";
1343                 ret = bch2_check_alloc_to_lru_refs(c);
1344                 if (ret)
1345                         goto err;
1346                 bch_verbose(c, "done checking alloc to lru refs");
1347                 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1348         } else {
1349                 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1350                 set_bit(BCH_FS_CHECK_ALLOC_DONE, &c->flags);
1351                 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1352                 set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1353                 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1354                 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1355
1356                 if (c->opts.norecovery)
1357                         goto out;
1358
1359                 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1360
1361                 bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1362                 err = "journal replay failed";
1363                 ret = bch2_journal_replay(c, last_seq, blacklist_seq - 1);
1364                 if (ret)
1365                         goto err;
1366                 if (c->opts.verbose || !c->sb.clean)
1367                         bch_info(c, "journal replay done");
1368         }
1369
1370         err = "error initializing freespace";
1371         ret = bch2_fs_freespace_init(c);
1372         if (ret)
1373                 goto err;
1374
1375         if (c->sb.version < bcachefs_metadata_version_bucket_gens &&
1376             c->opts.version_upgrade) {
1377                 bch_info(c, "initializing bucket_gens");
1378                 err = "error initializing bucket gens";
1379                 ret = bch2_bucket_gens_init(c);
1380                 if (ret)
1381                         goto err;
1382                 bch_verbose(c, "bucket_gens init done");
1383         }
1384
1385         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1386                 /* set bi_subvol on root inode */
1387                 err = "error upgrade root inode for subvolumes";
1388                 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1389                                     bch2_fs_upgrade_for_subvolumes(&trans));
1390                 if (ret)
1391                         goto err;
1392         }
1393
1394         if (c->opts.fsck) {
1395                 bch_info(c, "starting fsck");
1396                 err = "error in fsck";
1397                 ret = bch2_fsck_full(c);
1398                 if (ret)
1399                         goto err;
1400                 bch_verbose(c, "fsck done");
1401         } else if (!c->sb.clean) {
1402                 bch_verbose(c, "checking for deleted inodes");
1403                 err = "error in recovery";
1404                 ret = bch2_fsck_walk_inodes_only(c);
1405                 if (ret)
1406                         goto err;
1407                 bch_verbose(c, "check inodes done");
1408         }
1409
1410         if (enabled_qtypes(c)) {
1411                 bch_verbose(c, "reading quotas");
1412                 ret = bch2_fs_quota_read(c);
1413                 if (ret)
1414                         goto err;
1415                 bch_verbose(c, "quotas done");
1416         }
1417
1418         mutex_lock(&c->sb_lock);
1419         if (c->opts.version_upgrade) {
1420                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1421                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1422                 write_sb = true;
1423         }
1424
1425         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1426                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1427                 write_sb = true;
1428         }
1429
1430         if (c->opts.fsck &&
1431             !test_bit(BCH_FS_ERROR, &c->flags) &&
1432             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1433                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1434                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1435                 write_sb = true;
1436         }
1437
1438         if (write_sb)
1439                 bch2_write_super(c);
1440         mutex_unlock(&c->sb_lock);
1441
1442         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1443             !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1444             le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) {
1445                 struct bch_move_stats stats;
1446
1447                 bch2_move_stats_init(&stats, "recovery");
1448
1449                 bch_info(c, "scanning for old btree nodes");
1450                 ret = bch2_fs_read_write(c);
1451                 if (ret)
1452                         goto err;
1453
1454                 ret = bch2_scan_old_btree_nodes(c, &stats);
1455                 if (ret)
1456                         goto err;
1457                 bch_info(c, "scanning for old btree nodes done");
1458         }
1459
1460         if (c->journal_seq_blacklist_table &&
1461             c->journal_seq_blacklist_table->nr > 128)
1462                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1463
1464         ret = 0;
1465 out:
1466         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1467         bch2_flush_fsck_errs(c);
1468
1469         if (!c->opts.keep_journal &&
1470             test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)) {
1471                 bch2_journal_keys_free(&c->journal_keys);
1472                 bch2_journal_entries_free(c);
1473         }
1474         kfree(clean);
1475
1476         if (!ret && test_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags)) {
1477                 bch2_fs_read_write_early(c);
1478                 bch2_delete_dead_snapshots_async(c);
1479         }
1480
1481         if (ret)
1482                 bch_err(c, "Error in recovery: %s (%s)", err, bch2_err_str(ret));
1483         else
1484                 bch_verbose(c, "ret %s", bch2_err_str(ret));
1485         return ret;
1486 err:
1487 fsck_err:
1488         bch2_fs_emergency_read_only(c);
1489         goto out;
1490 }
1491
1492 int bch2_fs_initialize(struct bch_fs *c)
1493 {
1494         struct bch_inode_unpacked root_inode, lostfound_inode;
1495         struct bkey_inode_buf packed_inode;
1496         struct qstr lostfound = QSTR("lost+found");
1497         const char *err = "cannot allocate memory";
1498         struct bch_dev *ca;
1499         unsigned i;
1500         int ret;
1501
1502         bch_notice(c, "initializing new filesystem");
1503
1504         mutex_lock(&c->sb_lock);
1505         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1506         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1507
1508         if (c->sb.version < bcachefs_metadata_version_inode_v3)
1509                 c->opts.version_upgrade = true;
1510
1511         if (c->opts.version_upgrade) {
1512                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1513                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1514                 bch2_write_super(c);
1515         }
1516         mutex_unlock(&c->sb_lock);
1517
1518         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1519         set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1520         set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1521         set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1522         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1523         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1524
1525         for (i = 0; i < BTREE_ID_NR; i++)
1526                 bch2_btree_root_alloc(c, i);
1527
1528         for_each_online_member(ca, c, i)
1529                 bch2_dev_usage_init(ca);
1530
1531         err = "unable to allocate journal buckets";
1532         for_each_online_member(ca, c, i) {
1533                 ret = bch2_dev_journal_alloc(ca);
1534                 if (ret) {
1535                         percpu_ref_put(&ca->io_ref);
1536                         goto err;
1537                 }
1538         }
1539
1540         /*
1541          * journal_res_get() will crash if called before this has
1542          * set up the journal.pin FIFO and journal.cur pointer:
1543          */
1544         bch2_fs_journal_start(&c->journal, 1);
1545         bch2_journal_set_replay_done(&c->journal);
1546
1547         err = "error going read-write";
1548         ret = bch2_fs_read_write_early(c);
1549         if (ret)
1550                 goto err;
1551
1552         /*
1553          * Write out the superblock and journal buckets, now that we can do
1554          * btree updates
1555          */
1556         bch_verbose(c, "marking superblocks");
1557         err = "error marking superblock and journal";
1558         for_each_member_device(ca, c, i) {
1559                 ret = bch2_trans_mark_dev_sb(c, ca);
1560                 if (ret) {
1561                         percpu_ref_put(&ca->ref);
1562                         goto err;
1563                 }
1564
1565                 ca->new_fs_bucket_idx = 0;
1566         }
1567
1568         bch_verbose(c, "initializing freespace");
1569         err = "error initializing freespace";
1570         ret = bch2_fs_freespace_init(c);
1571         if (ret)
1572                 goto err;
1573
1574         err = "error creating root snapshot node";
1575         ret = bch2_fs_initialize_subvolumes(c);
1576         if (ret)
1577                 goto err;
1578
1579         bch_verbose(c, "reading snapshots table");
1580         err = "error reading snapshots table";
1581         ret = bch2_fs_snapshots_start(c);
1582         if (ret)
1583                 goto err;
1584         bch_verbose(c, "reading snapshots done");
1585
1586         bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL);
1587         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1588         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1589         bch2_inode_pack(&packed_inode, &root_inode);
1590         packed_inode.inode.k.p.snapshot = U32_MAX;
1591
1592         err = "error creating root directory";
1593         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1594                                 &packed_inode.inode.k_i,
1595                                 NULL, NULL, 0);
1596         if (ret)
1597                 goto err;
1598
1599         bch2_inode_init_early(c, &lostfound_inode);
1600
1601         err = "error creating lost+found";
1602         ret = bch2_trans_do(c, NULL, NULL, 0,
1603                 bch2_create_trans(&trans,
1604                                   BCACHEFS_ROOT_SUBVOL_INUM,
1605                                   &root_inode, &lostfound_inode,
1606                                   &lostfound,
1607                                   0, 0, S_IFDIR|0700, 0,
1608                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1609         if (ret) {
1610                 bch_err(c, "error creating lost+found");
1611                 goto err;
1612         }
1613
1614         if (enabled_qtypes(c)) {
1615                 ret = bch2_fs_quota_read(c);
1616                 if (ret)
1617                         goto err;
1618         }
1619
1620         err = "error writing first journal entry";
1621         ret = bch2_journal_flush(&c->journal);
1622         if (ret)
1623                 goto err;
1624
1625         mutex_lock(&c->sb_lock);
1626         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1627         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1628
1629         bch2_write_super(c);
1630         mutex_unlock(&c->sb_lock);
1631
1632         return 0;
1633 err:
1634         pr_err("Error initializing new filesystem: %s (%s)", err, bch2_err_str(ret));
1635         return ret;
1636 }