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