git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/recovery.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "bkey_buf.h"
   5 #include "alloc_background.h"
   6 #include "btree_gc.h"
   7 #include "btree_update.h"
   8 #include "btree_update_interior.h"
   9 #include "btree_io.h"
  10 #include "buckets.h"
  11 #include "dirent.h"
  12 #include "ec.h"
  13 #include "error.h"
  14 #include "fs-common.h"
  15 #include "fsck.h"
  16 #include "journal_io.h"
  17 #include "journal_reclaim.h"
  18 #include "journal_seq_blacklist.h"
  19 #include "lru.h"
  20 #include "move.h"
  21 #include "quota.h"
  22 #include "recovery.h"
  23 #include "replicas.h"
  24 #include "subvolume.h"
  25 #include "super-io.h"
  26
  27 #include <linux/sort.h>
  28 #include <linux/stat.h>
  29
  30 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  31
  32 /* for -o reconstruct_alloc: */
  33 static void drop_alloc_keys(struct journal_keys *keys)
  34 {
  35         size_t src, dst;
  36
  37         for (src = 0, dst = 0; src < keys->nr; src++)
  38                 if (keys->d[src].btree_id != BTREE_ID_alloc)
  39                         keys->d[dst++] = keys->d[src];
  40
  41         keys->nr = dst;
  42 }
  43
  44 /*
  45  * Btree node pointers have a field to stack a pointer to the in memory btree
  46  * node; we need to zero out this field when reading in btree nodes, or when
  47  * reading in keys from the journal:
  48  */
  49 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
  50 {
  51         struct journal_key *i;
  52
  53         for (i = keys->d; i < keys->d + keys->nr; i++)
  54                 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
  55                         bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
  56 }
  57
  58 /* iterate over keys read from the journal: */
  59
  60 static int __journal_key_cmp(enum btree_id      l_btree_id,
  61                              unsigned           l_level,
  62                              struct bpos        l_pos,
  63                              const struct journal_key *r)
  64 {
  65         return (cmp_int(l_btree_id,     r->btree_id) ?:
  66                 cmp_int(l_level,        r->level) ?:
  67                 bpos_cmp(l_pos, r->k->k.p));
  68 }
  69
  70 static int journal_key_cmp(const struct journal_key *l, const struct journal_key *r)
  71 {
  72         return __journal_key_cmp(l->btree_id, l->level, l->k->k.p, r);
  73 }
  74
  75 static inline size_t idx_to_pos(struct journal_keys *keys, size_t idx)
  76 {
  77         size_t gap_size = keys->size - keys->nr;
  78
  79         if (idx >= keys->gap)
  80                 idx += gap_size;
  81         return idx;
  82 }
  83
  84 static inline struct journal_key *idx_to_key(struct journal_keys *keys, size_t idx)
  85 {
  86         return keys->d + idx_to_pos(keys, idx);
  87 }
  88
  89 size_t bch2_journal_key_search(struct journal_keys *keys,
  90                                enum btree_id id, unsigned level,
  91                                struct bpos pos)
  92 {
  93         size_t l = 0, r = keys->nr, m;
  94
  95         while (l < r) {
  96                 m = l + ((r - l) >> 1);
  97                 if (__journal_key_cmp(id, level, pos, idx_to_key(keys, m)) > 0)
  98                         l = m + 1;
  99                 else
 100                         r = m;
 101         }
 102
 103         BUG_ON(l < keys->nr &&
 104                __journal_key_cmp(id, level, pos, idx_to_key(keys, l)) > 0);
 105
 106         BUG_ON(l &&
 107                __journal_key_cmp(id, level, pos, idx_to_key(keys, l - 1)) <= 0);
 108
 109         return idx_to_pos(keys, l);
 110 }
 111
 112 struct bkey_i *bch2_journal_keys_peek_upto(struct bch_fs *c, enum btree_id btree_id,
 113                                            unsigned level, struct bpos pos,
 114                                            struct bpos end_pos)
 115 {
 116         struct journal_keys *keys = &c->journal_keys;
 117         size_t idx = bch2_journal_key_search(keys, btree_id, level, pos);
 118
 119         while (idx < keys->size &&
 120                keys->d[idx].btree_id == btree_id &&
 121                keys->d[idx].level == level &&
 122                bpos_cmp(keys->d[idx].k->k.p, end_pos) <= 0) {
 123                 if (!keys->d[idx].overwritten)
 124                         return keys->d[idx].k;
 125
 126                 idx++;
 127                 if (idx == keys->gap)
 128                         idx += keys->size - keys->nr;
 129         }
 130
 131         return NULL;
 132 }
 133
 134 struct bkey_i *bch2_journal_keys_peek_slot(struct bch_fs *c, enum btree_id btree_id,
 135                                            unsigned level, struct bpos pos)
 136 {
 137         return bch2_journal_keys_peek_upto(c, btree_id, level, pos, pos);
 138 }
 139
 140 static void journal_iters_fix(struct bch_fs *c)
 141 {
 142         struct journal_keys *keys = &c->journal_keys;
 143         /* The key we just inserted is immediately before the gap: */
 144         struct journal_key *n = &keys->d[keys->gap - 1];
 145         size_t gap_end = keys->gap + (keys->size - keys->nr);
 146         struct btree_and_journal_iter *iter;
 147
 148         /*
 149          * If an iterator points one after the key we just inserted,
 150          * and the key we just inserted compares > the iterator's position,
 151          * decrement the iterator so it points at the key we just inserted:
 152          */
 153         list_for_each_entry(iter, &c->journal_iters, journal.list)
 154                 if (iter->journal.idx == gap_end &&
 155                     iter->last &&
 156                     iter->b->c.btree_id == n->btree_id &&
 157                     iter->b->c.level    == n->level &&
 158                     bpos_cmp(n->k->k.p, iter->unpacked.p) > 0)
 159                         iter->journal.idx = keys->gap - 1;
 160 }
 161
 162 static void journal_iters_move_gap(struct bch_fs *c, size_t old_gap, size_t new_gap)
 163 {
 164         struct journal_keys *keys = &c->journal_keys;
 165         struct journal_iter *iter;
 166         size_t gap_size = keys->size - keys->nr;
 167
 168         list_for_each_entry(iter, &c->journal_iters, list) {
 169                 if (iter->idx > old_gap)
 170                         iter->idx -= gap_size;
 171                 if (iter->idx >= new_gap)
 172                         iter->idx += gap_size;
 173         }
 174 }
 175
 176 int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
 177                                  unsigned level, struct bkey_i *k)
 178 {
 179         struct journal_key n = {
 180                 .btree_id       = id,
 181                 .level          = level,
 182                 .k              = k,
 183                 .allocated      = true,
 184                 /*
 185                  * Ensure these keys are done last by journal replay, to unblock
 186                  * journal reclaim:
 187                  */
 188                 .journal_seq    = U32_MAX,
 189         };
 190         struct journal_keys *keys = &c->journal_keys;
 191         size_t idx = bch2_journal_key_search(keys, id, level, k->k.p);
 192
 193         BUG_ON(test_bit(BCH_FS_RW, &c->flags));
 194
 195         if (idx < keys->size &&
 196             journal_key_cmp(&n, &keys->d[idx]) == 0) {
 197                 if (keys->d[idx].allocated)
 198                         kfree(keys->d[idx].k);
 199                 keys->d[idx] = n;
 200                 return 0;
 201         }
 202
 203         if (idx > keys->gap)
 204                 idx -= keys->size - keys->nr;
 205
 206         if (keys->nr == keys->size) {
 207                 struct journal_keys new_keys = {
 208                         .nr                     = keys->nr,
 209                         .size                   = max_t(size_t, keys->size, 8) * 2,
 210                         .journal_seq_base       = keys->journal_seq_base,
 211                 };
 212
 213                 new_keys.d = kvmalloc(sizeof(new_keys.d[0]) * new_keys.size, GFP_KERNEL);
 214                 if (!new_keys.d) {
 215                         bch_err(c, "%s: error allocating new key array (size %zu)",
 216                                 __func__, new_keys.size);
 217                         return -ENOMEM;
 218                 }
 219
 220                 /* Since @keys was full, there was no gap: */
 221                 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
 222                 kvfree(keys->d);
 223                 *keys = new_keys;
 224
 225                 /* And now the gap is at the end: */
 226                 keys->gap = keys->nr;
 227         }
 228
 229         journal_iters_move_gap(c, keys->gap, idx);
 230
 231         move_gap(keys->d, keys->nr, keys->size, keys->gap, idx);
 232         keys->gap = idx;
 233
 234         keys->nr++;
 235         keys->d[keys->gap++] = n;
 236
 237         journal_iters_fix(c);
 238
 239         return 0;
 240 }
 241
 242 /*
 243  * Can only be used from the recovery thread while we're still RO - can't be
 244  * used once we've got RW, as journal_keys is at that point used by multiple
 245  * threads:
 246  */
 247 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
 248                             unsigned level, struct bkey_i *k)
 249 {
 250         struct bkey_i *n;
 251         int ret;
 252
 253         n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
 254         if (!n)
 255                 return -ENOMEM;
 256
 257         bkey_copy(n, k);
 258         ret = bch2_journal_key_insert_take(c, id, level, n);
 259         if (ret)
 260                 kfree(n);
 261         return ret;
 262 }
 263
 264 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
 265                             unsigned level, struct bpos pos)
 266 {
 267         struct bkey_i whiteout;
 268
 269         bkey_init(&whiteout.k);
 270         whiteout.k.p = pos;
 271
 272         return bch2_journal_key_insert(c, id, level, &whiteout);
 273 }
 274
 275 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
 276                                   unsigned level, struct bpos pos)
 277 {
 278         struct journal_keys *keys = &c->journal_keys;
 279         size_t idx = bch2_journal_key_search(keys, btree, level, pos);
 280
 281         if (idx < keys->size &&
 282             keys->d[idx].btree_id       == btree &&
 283             keys->d[idx].level          == level &&
 284             !bpos_cmp(keys->d[idx].k->k.p, pos))
 285                 keys->d[idx].overwritten = true;
 286 }
 287
 288 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
 289 {
 290         struct journal_key *k = iter->keys->d + iter->idx;
 291
 292         while (k < iter->keys->d + iter->keys->nr &&
 293                k->btree_id      == iter->btree_id &&
 294                k->level         == iter->level) {
 295                 if (!k->overwritten)
 296                         return k->k;
 297
 298                 iter->idx++;
 299                 k = iter->keys->d + iter->idx;
 300         }
 301
 302         return NULL;
 303 }
 304
 305 static void bch2_journal_iter_advance(struct journal_iter *iter)
 306 {
 307         if (iter->idx < iter->keys->size) {
 308                 iter->idx++;
 309                 if (iter->idx == iter->keys->gap)
 310                         iter->idx += iter->keys->size - iter->keys->nr;
 311         }
 312 }
 313
 314 static void bch2_journal_iter_exit(struct journal_iter *iter)
 315 {
 316         list_del(&iter->list);
 317 }
 318
 319 static void bch2_journal_iter_init(struct bch_fs *c,
 320                                    struct journal_iter *iter,
 321                                    enum btree_id id, unsigned level,
 322                                    struct bpos pos)
 323 {
 324         iter->btree_id  = id;
 325         iter->level     = level;
 326         iter->keys      = &c->journal_keys;
 327         iter->idx       = bch2_journal_key_search(&c->journal_keys, id, level, pos);
 328 }
 329
 330 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 331 {
 332         return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 333                                                 iter->b, &iter->unpacked);
 334 }
 335
 336 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 337 {
 338         bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 339 }
 340
 341 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 342 {
 343         switch (iter->last) {
 344         case none:
 345                 break;
 346         case btree:
 347                 bch2_journal_iter_advance_btree(iter);
 348                 break;
 349         case journal:
 350                 bch2_journal_iter_advance(&iter->journal);
 351                 break;
 352         }
 353
 354         iter->last = none;
 355 }
 356
 357 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 358 {
 359         struct bkey_s_c ret;
 360
 361         while (1) {
 362                 struct bkey_s_c btree_k         =
 363                         bch2_journal_iter_peek_btree(iter);
 364                 struct bkey_s_c journal_k       =
 365                         bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
 366
 367                 if (btree_k.k && journal_k.k) {
 368                         int cmp = bpos_cmp(btree_k.k->p, journal_k.k->p);
 369
 370                         if (!cmp)
 371                                 bch2_journal_iter_advance_btree(iter);
 372
 373                         iter->last = cmp < 0 ? btree : journal;
 374                 } else if (btree_k.k) {
 375                         iter->last = btree;
 376                 } else if (journal_k.k) {
 377                         iter->last = journal;
 378                 } else {
 379                         iter->last = none;
 380                         return bkey_s_c_null;
 381                 }
 382
 383                 ret = iter->last == journal ? journal_k : btree_k;
 384
 385                 if (iter->b &&
 386                     bpos_cmp(ret.k->p, iter->b->data->max_key) > 0) {
 387                         iter->journal.idx = iter->journal.keys->nr;
 388                         iter->last = none;
 389                         return bkey_s_c_null;
 390                 }
 391
 392                 if (!bkey_deleted(ret.k))
 393                         break;
 394
 395                 bch2_btree_and_journal_iter_advance(iter);
 396         }
 397
 398         return ret;
 399 }
 400
 401 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 402 {
 403         bch2_btree_and_journal_iter_advance(iter);
 404
 405         return bch2_btree_and_journal_iter_peek(iter);
 406 }
 407
 408 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
 409 {
 410         bch2_journal_iter_exit(&iter->journal);
 411 }
 412
 413 void __bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 414                                                   struct bch_fs *c,
 415                                                   struct btree *b,
 416                                                   struct btree_node_iter node_iter,
 417                                                   struct bpos pos)
 418 {
 419         memset(iter, 0, sizeof(*iter));
 420
 421         iter->b = b;
 422         iter->node_iter = node_iter;
 423         bch2_journal_iter_init(c, &iter->journal, b->c.btree_id, b->c.level, pos);
 424         INIT_LIST_HEAD(&iter->journal.list);
 425 }
 426
 427 /*
 428  * this version is used by btree_gc before filesystem has gone RW and
 429  * multithreaded, so uses the journal_iters list:
 430  */
 431 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 432                                                 struct bch_fs *c,
 433                                                 struct btree *b)
 434 {
 435         struct btree_node_iter node_iter;
 436
 437         bch2_btree_node_iter_init_from_start(&node_iter, b);
 438         __bch2_btree_and_journal_iter_init_node_iter(iter, c, b, node_iter, b->data->min_key);
 439         list_add(&iter->journal.list, &c->journal_iters);
 440 }
 441
 442 /* sort and dedup all keys in the journal: */
 443
 444 void bch2_journal_entries_free(struct bch_fs *c)
 445 {
 446         struct journal_replay **i;
 447         struct genradix_iter iter;
 448
 449         genradix_for_each(&c->journal_entries, iter, i)
 450                 if (*i)
 451                         kvpfree(*i, offsetof(struct journal_replay, j) +
 452                                 vstruct_bytes(&(*i)->j));
 453         genradix_free(&c->journal_entries);
 454 }
 455
 456 /*
 457  * When keys compare equal, oldest compares first:
 458  */
 459 static int journal_sort_key_cmp(const void *_l, const void *_r)
 460 {
 461         const struct journal_key *l = _l;
 462         const struct journal_key *r = _r;
 463
 464         return  journal_key_cmp(l, r) ?:
 465                 cmp_int(l->journal_seq, r->journal_seq) ?:
 466                 cmp_int(l->journal_offset, r->journal_offset);
 467 }
 468
 469 void bch2_journal_keys_free(struct journal_keys *keys)
 470 {
 471         struct journal_key *i;
 472
 473         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 474         keys->gap = keys->nr;
 475
 476         for (i = keys->d; i < keys->d + keys->nr; i++)
 477                 if (i->allocated)
 478                         kfree(i->k);
 479
 480         kvfree(keys->d);
 481         keys->d = NULL;
 482         keys->nr = keys->gap = keys->size = 0;
 483 }
 484
 485 static int journal_keys_sort(struct bch_fs *c)
 486 {
 487         struct genradix_iter iter;
 488         struct journal_replay *i, **_i;
 489         struct jset_entry *entry;
 490         struct bkey_i *k, *_n;
 491         struct journal_keys *keys = &c->journal_keys;
 492         struct journal_key *src, *dst;
 493         size_t nr_keys = 0;
 494
 495         genradix_for_each(&c->journal_entries, iter, _i) {
 496                 i = *_i;
 497
 498                 if (!i || i->ignore)
 499                         continue;
 500
 501                 if (!keys->journal_seq_base)
 502                         keys->journal_seq_base = le64_to_cpu(i->j.seq);
 503
 504                 for_each_jset_key(k, _n, entry, &i->j)
 505                         nr_keys++;
 506         }
 507
 508         if (!nr_keys)
 509                 return 0;
 510
 511         keys->size = roundup_pow_of_two(nr_keys);
 512
 513         keys->d = kvmalloc(sizeof(keys->d[0]) * keys->size, GFP_KERNEL);
 514         if (!keys->d)
 515                 return -ENOMEM;
 516
 517         genradix_for_each(&c->journal_entries, iter, _i) {
 518                 i = *_i;
 519
 520                 if (!i || i->ignore)
 521                         continue;
 522
 523                 BUG_ON(le64_to_cpu(i->j.seq) - keys->journal_seq_base > U32_MAX);
 524
 525                 for_each_jset_key(k, _n, entry, &i->j)
 526                         keys->d[keys->nr++] = (struct journal_key) {
 527                                 .btree_id       = entry->btree_id,
 528                                 .level          = entry->level,
 529                                 .k              = k,
 530                                 .journal_seq    = le64_to_cpu(i->j.seq) -
 531                                         keys->journal_seq_base,
 532                                 .journal_offset = k->_data - i->j._data,
 533                         };
 534         }
 535
 536         sort(keys->d, keys->nr, sizeof(keys->d[0]), journal_sort_key_cmp, NULL);
 537
 538         src = dst = keys->d;
 539         while (src < keys->d + keys->nr) {
 540                 while (src + 1 < keys->d + keys->nr &&
 541                        src[0].btree_id  == src[1].btree_id &&
 542                        src[0].level     == src[1].level &&
 543                        !bpos_cmp(src[0].k->k.p, src[1].k->k.p))
 544                         src++;
 545
 546                 *dst++ = *src++;
 547         }
 548
 549         keys->nr = dst - keys->d;
 550         keys->gap = keys->nr;
 551         return 0;
 552 }
 553
 554 /* journal replay: */
 555
 556 static void replay_now_at(struct journal *j, u64 seq)
 557 {
 558         BUG_ON(seq < j->replay_journal_seq);
 559
 560         seq = min(seq, j->replay_journal_seq_end);
 561
 562         while (j->replay_journal_seq < seq)
 563                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 564 }
 565
 566 static int bch2_journal_replay_key(struct btree_trans *trans,
 567                                    struct journal_key *k)
 568 {
 569         struct btree_iter iter;
 570         unsigned iter_flags =
 571                 BTREE_ITER_INTENT|
 572                 BTREE_ITER_NOT_EXTENTS;
 573         int ret;
 574
 575         if (!k->level && k->btree_id == BTREE_ID_alloc)
 576                 iter_flags |= BTREE_ITER_CACHED;
 577
 578         bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 579                                   BTREE_MAX_DEPTH, k->level,
 580                                   iter_flags);
 581         ret = bch2_btree_iter_traverse(&iter);
 582         if (ret)
 583                 goto out;
 584
 585         /* Must be checked with btree locked: */
 586         if (k->overwritten)
 587                 goto out;
 588
 589         ret = bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
 590 out:
 591         bch2_trans_iter_exit(trans, &iter);
 592         return ret;
 593 }
 594
 595 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 596 {
 597         const struct journal_key *l = *((const struct journal_key **)_l);
 598         const struct journal_key *r = *((const struct journal_key **)_r);
 599
 600         return cmp_int(l->journal_seq, r->journal_seq);
 601 }
 602
 603 static int bch2_journal_replay(struct bch_fs *c)
 604 {
 605         struct journal_keys *keys = &c->journal_keys;
 606         struct journal_key **keys_sorted, *k;
 607         struct journal *j = &c->journal;
 608         size_t i;
 609         int ret;
 610
 611         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 612         keys->gap = keys->nr;
 613
 614         keys_sorted = kvmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL);
 615         if (!keys_sorted)
 616                 return -ENOMEM;
 617
 618         for (i = 0; i < keys->nr; i++)
 619                 keys_sorted[i] = &keys->d[i];
 620
 621         sort(keys_sorted, keys->nr,
 622              sizeof(keys_sorted[0]),
 623              journal_sort_seq_cmp, NULL);
 624
 625         if (keys->nr)
 626                 replay_now_at(j, keys->journal_seq_base);
 627
 628         for (i = 0; i < keys->nr; i++) {
 629                 k = keys_sorted[i];
 630
 631                 cond_resched();
 632
 633                 replay_now_at(j, keys->journal_seq_base + k->journal_seq);
 634
 635                 ret = bch2_trans_do(c, NULL, NULL,
 636                                     BTREE_INSERT_LAZY_RW|
 637                                     BTREE_INSERT_NOFAIL|
 638                                     (!k->allocated
 639                                      ? BTREE_INSERT_JOURNAL_REPLAY|JOURNAL_WATERMARK_reserved
 640                                      : 0),
 641                              bch2_journal_replay_key(&trans, k));
 642                 if (ret) {
 643                         bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
 644                                 ret, bch2_btree_ids[k->btree_id], k->level);
 645                         goto err;
 646                 }
 647         }
 648
 649         replay_now_at(j, j->replay_journal_seq_end);
 650         j->replay_journal_seq = 0;
 651
 652         bch2_journal_set_replay_done(j);
 653         bch2_journal_flush_all_pins(j);
 654         ret = bch2_journal_error(j);
 655
 656         if (keys->nr && !ret)
 657                 bch2_journal_log_msg(&c->journal, "journal replay finished");
 658 err:
 659         kvfree(keys_sorted);
 660         return ret;
 661 }
 662
 663 /* journal replay early: */
 664
 665 static int journal_replay_entry_early(struct bch_fs *c,
 666                                       struct jset_entry *entry)
 667 {
 668         int ret = 0;
 669
 670         switch (entry->type) {
 671         case BCH_JSET_ENTRY_btree_root: {
 672                 struct btree_root *r;
 673
 674                 if (entry->btree_id >= BTREE_ID_NR) {
 675                         bch_err(c, "filesystem has unknown btree type %u",
 676                                 entry->btree_id);
 677                         return -EINVAL;
 678                 }
 679
 680                 r = &c->btree_roots[entry->btree_id];
 681
 682                 if (entry->u64s) {
 683                         r->level = entry->level;
 684                         bkey_copy(&r->key, &entry->start[0]);
 685                         r->error = 0;
 686                 } else {
 687                         r->error = -EIO;
 688                 }
 689                 r->alive = true;
 690                 break;
 691         }
 692         case BCH_JSET_ENTRY_usage: {
 693                 struct jset_entry_usage *u =
 694                         container_of(entry, struct jset_entry_usage, entry);
 695
 696                 switch (entry->btree_id) {
 697                 case BCH_FS_USAGE_reserved:
 698                         if (entry->level < BCH_REPLICAS_MAX)
 699                                 c->usage_base->persistent_reserved[entry->level] =
 700                                         le64_to_cpu(u->v);
 701                         break;
 702                 case BCH_FS_USAGE_inodes:
 703                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 704                         break;
 705                 case BCH_FS_USAGE_key_version:
 706                         atomic64_set(&c->key_version,
 707                                      le64_to_cpu(u->v));
 708                         break;
 709                 }
 710
 711                 break;
 712         }
 713         case BCH_JSET_ENTRY_data_usage: {
 714                 struct jset_entry_data_usage *u =
 715                         container_of(entry, struct jset_entry_data_usage, entry);
 716
 717                 ret = bch2_replicas_set_usage(c, &u->r,
 718                                               le64_to_cpu(u->v));
 719                 break;
 720         }
 721         case BCH_JSET_ENTRY_dev_usage: {
 722                 struct jset_entry_dev_usage *u =
 723                         container_of(entry, struct jset_entry_dev_usage, entry);
 724                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 725                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 726
 727                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 728
 729                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 730                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 731                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 732                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 733                 }
 734
 735                 break;
 736         }
 737         case BCH_JSET_ENTRY_blacklist: {
 738                 struct jset_entry_blacklist *bl_entry =
 739                         container_of(entry, struct jset_entry_blacklist, entry);
 740
 741                 ret = bch2_journal_seq_blacklist_add(c,
 742                                 le64_to_cpu(bl_entry->seq),
 743                                 le64_to_cpu(bl_entry->seq) + 1);
 744                 break;
 745         }
 746         case BCH_JSET_ENTRY_blacklist_v2: {
 747                 struct jset_entry_blacklist_v2 *bl_entry =
 748                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 749
 750                 ret = bch2_journal_seq_blacklist_add(c,
 751                                 le64_to_cpu(bl_entry->start),
 752                                 le64_to_cpu(bl_entry->end) + 1);
 753                 break;
 754         }
 755         case BCH_JSET_ENTRY_clock: {
 756                 struct jset_entry_clock *clock =
 757                         container_of(entry, struct jset_entry_clock, entry);
 758
 759                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 760         }
 761         }
 762
 763         return ret;
 764 }
 765
 766 static int journal_replay_early(struct bch_fs *c,
 767                                 struct bch_sb_field_clean *clean)
 768 {
 769         struct jset_entry *entry;
 770         int ret;
 771
 772         if (clean) {
 773                 for (entry = clean->start;
 774                      entry != vstruct_end(&clean->field);
 775                      entry = vstruct_next(entry)) {
 776                         ret = journal_replay_entry_early(c, entry);
 777                         if (ret)
 778                                 return ret;
 779                 }
 780         } else {
 781                 struct genradix_iter iter;
 782                 struct journal_replay *i, **_i;
 783
 784                 genradix_for_each(&c->journal_entries, iter, _i) {
 785                         i = *_i;
 786
 787                         if (!i || i->ignore)
 788                                 continue;
 789
 790                         vstruct_for_each(&i->j, entry) {
 791                                 ret = journal_replay_entry_early(c, entry);
 792                                 if (ret)
 793                                         return ret;
 794                         }
 795                 }
 796         }
 797
 798         bch2_fs_usage_initialize(c);
 799
 800         return 0;
 801 }
 802
 803 /* sb clean section: */
 804
 805 static struct bkey_i *btree_root_find(struct bch_fs *c,
 806                                       struct bch_sb_field_clean *clean,
 807                                       struct jset *j,
 808                                       enum btree_id id, unsigned *level)
 809 {
 810         struct bkey_i *k;
 811         struct jset_entry *entry, *start, *end;
 812
 813         if (clean) {
 814                 start = clean->start;
 815                 end = vstruct_end(&clean->field);
 816         } else {
 817                 start = j->start;
 818                 end = vstruct_last(j);
 819         }
 820
 821         for (entry = start; entry < end; entry = vstruct_next(entry))
 822                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 823                     entry->btree_id == id)
 824                         goto found;
 825
 826         return NULL;
 827 found:
 828         if (!entry->u64s)
 829                 return ERR_PTR(-EINVAL);
 830
 831         k = entry->start;
 832         *level = entry->level;
 833         return k;
 834 }
 835
 836 static int verify_superblock_clean(struct bch_fs *c,
 837                                    struct bch_sb_field_clean **cleanp,
 838                                    struct jset *j)
 839 {
 840         unsigned i;
 841         struct bch_sb_field_clean *clean = *cleanp;
 842         struct printbuf buf1 = PRINTBUF;
 843         struct printbuf buf2 = PRINTBUF;
 844         int ret = 0;
 845
 846         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 847                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 848                         le64_to_cpu(clean->journal_seq),
 849                         le64_to_cpu(j->seq))) {
 850                 kfree(clean);
 851                 *cleanp = NULL;
 852                 return 0;
 853         }
 854
 855         for (i = 0; i < BTREE_ID_NR; i++) {
 856                 struct bkey_i *k1, *k2;
 857                 unsigned l1 = 0, l2 = 0;
 858
 859                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 860                 k2 = btree_root_find(c, NULL, j, i, &l2);
 861
 862                 if (!k1 && !k2)
 863                         continue;
 864
 865                 printbuf_reset(&buf1);
 866                 printbuf_reset(&buf2);
 867
 868                 if (k1)
 869                         bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(k1));
 870                 else
 871                         pr_buf(&buf1, "(none)");
 872
 873                 if (k2)
 874                         bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(k2));
 875                 else
 876                         pr_buf(&buf2, "(none)");
 877
 878                 mustfix_fsck_err_on(!k1 || !k2 ||
 879                                     IS_ERR(k1) ||
 880                                     IS_ERR(k2) ||
 881                                     k1->k.u64s != k2->k.u64s ||
 882                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 883                                     l1 != l2, c,
 884                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 885                         "sb:      l=%u %s\n"
 886                         "journal: l=%u %s\n", i,
 887                         l1, buf1.buf,
 888                         l2, buf2.buf);
 889         }
 890 fsck_err:
 891         printbuf_exit(&buf2);
 892         printbuf_exit(&buf1);
 893         return ret;
 894 }
 895
 896 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 897 {
 898         struct bch_sb_field_clean *clean, *sb_clean;
 899         int ret;
 900
 901         mutex_lock(&c->sb_lock);
 902         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 903
 904         if (fsck_err_on(!sb_clean, c,
 905                         "superblock marked clean but clean section not present")) {
 906                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 907                 c->sb.clean = false;
 908                 mutex_unlock(&c->sb_lock);
 909                 return NULL;
 910         }
 911
 912         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 913                         GFP_KERNEL);
 914         if (!clean) {
 915                 mutex_unlock(&c->sb_lock);
 916                 return ERR_PTR(-ENOMEM);
 917         }
 918
 919         ret = bch2_sb_clean_validate_late(c, clean, READ);
 920         if (ret) {
 921                 mutex_unlock(&c->sb_lock);
 922                 return ERR_PTR(ret);
 923         }
 924
 925         mutex_unlock(&c->sb_lock);
 926
 927         return clean;
 928 fsck_err:
 929         mutex_unlock(&c->sb_lock);
 930         return ERR_PTR(ret);
 931 }
 932
 933 static int read_btree_roots(struct bch_fs *c)
 934 {
 935         unsigned i;
 936         int ret = 0;
 937
 938         for (i = 0; i < BTREE_ID_NR; i++) {
 939                 struct btree_root *r = &c->btree_roots[i];
 940
 941                 if (!r->alive)
 942                         continue;
 943
 944                 if (i == BTREE_ID_alloc &&
 945                     c->opts.reconstruct_alloc) {
 946                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 947                         continue;
 948                 }
 949
 950                 if (r->error) {
 951                         __fsck_err(c, i == BTREE_ID_alloc
 952                                    ? FSCK_CAN_IGNORE : 0,
 953                                    "invalid btree root %s",
 954                                    bch2_btree_ids[i]);
 955                         if (i == BTREE_ID_alloc)
 956                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 957                 }
 958
 959                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 960                 if (ret) {
 961                         __fsck_err(c, i == BTREE_ID_alloc
 962                                    ? FSCK_CAN_IGNORE : 0,
 963                                    "error reading btree root %s",
 964                                    bch2_btree_ids[i]);
 965                         if (i == BTREE_ID_alloc)
 966                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 967                 }
 968         }
 969
 970         for (i = 0; i < BTREE_ID_NR; i++)
 971                 if (!c->btree_roots[i].b)
 972                         bch2_btree_root_alloc(c, i);
 973 fsck_err:
 974         return ret;
 975 }
 976
 977 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
 978 {
 979         struct bkey_i_snapshot  root_snapshot;
 980         struct bkey_i_subvolume root_volume;
 981         int ret;
 982
 983         bkey_snapshot_init(&root_snapshot.k_i);
 984         root_snapshot.k.p.offset = U32_MAX;
 985         root_snapshot.v.flags   = 0;
 986         root_snapshot.v.parent  = 0;
 987         root_snapshot.v.subvol  = BCACHEFS_ROOT_SUBVOL;
 988         root_snapshot.v.pad     = 0;
 989         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
 990
 991         ret = bch2_btree_insert(c, BTREE_ID_snapshots,
 992                                 &root_snapshot.k_i,
 993                                 NULL, NULL, 0);
 994         if (ret)
 995                 return ret;
 996
 997         bkey_subvolume_init(&root_volume.k_i);
 998         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
 999         root_volume.v.flags     = 0;
1000         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
1001         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
1002
1003         ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
1004                                 &root_volume.k_i,
1005                                 NULL, NULL, 0);
1006         if (ret)
1007                 return ret;
1008
1009         return 0;
1010 }
1011
1012 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
1013 {
1014         struct btree_iter iter;
1015         struct bkey_s_c k;
1016         struct bch_inode_unpacked inode;
1017         int ret;
1018
1019         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
1020                              SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
1021         k = bch2_btree_iter_peek_slot(&iter);
1022         ret = bkey_err(k);
1023         if (ret)
1024                 goto err;
1025
1026         if (!bkey_is_inode(k.k)) {
1027                 bch_err(trans->c, "root inode not found");
1028                 ret = -ENOENT;
1029                 goto err;
1030         }
1031
1032         ret = bch2_inode_unpack(k, &inode);
1033         BUG_ON(ret);
1034
1035         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1036
1037         ret = bch2_inode_write(trans, &iter, &inode);
1038 err:
1039         bch2_trans_iter_exit(trans, &iter);
1040         return ret;
1041 }
1042
1043 int bch2_fs_recovery(struct bch_fs *c)
1044 {
1045         const char *err = "cannot allocate memory";
1046         struct bch_sb_field_clean *clean = NULL;
1047         struct jset *last_journal_entry = NULL;
1048         u64 blacklist_seq, journal_seq;
1049         bool write_sb = false;
1050         int ret = 0;
1051
1052         if (c->sb.clean)
1053                 clean = read_superblock_clean(c);
1054         ret = PTR_ERR_OR_ZERO(clean);
1055         if (ret)
1056                 goto err;
1057
1058         if (c->sb.clean)
1059                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
1060                          le64_to_cpu(clean->journal_seq));
1061         else
1062                 bch_info(c, "recovering from unclean shutdown");
1063
1064         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
1065                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
1066                 ret = -EINVAL;
1067                 goto err;
1068         }
1069
1070         if (!c->sb.clean &&
1071             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1072                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1073                 ret = -EINVAL;
1074                 goto err;
1075         }
1076
1077         if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1078                 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
1079                 ret = -EINVAL;
1080                 goto err;
1081         }
1082
1083         if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
1084                 bch_info(c, "alloc_v2 feature bit not set, fsck required");
1085                 c->opts.fsck = true;
1086                 c->opts.fix_errors = FSCK_OPT_YES;
1087         }
1088
1089         if (!c->opts.nochanges) {
1090                 if (c->sb.version < bcachefs_metadata_version_new_data_types) {
1091                         bch_info(c, "version prior to new_data_types, upgrade and fsck required");
1092                         c->opts.version_upgrade = true;
1093                         c->opts.fsck            = true;
1094                         c->opts.fix_errors      = FSCK_OPT_YES;
1095                 }
1096         }
1097
1098         if (c->opts.fsck && c->opts.norecovery) {
1099                 bch_err(c, "cannot select both norecovery and fsck");
1100                 ret = -EINVAL;
1101                 goto err;
1102         }
1103
1104         ret = bch2_blacklist_table_initialize(c);
1105         if (ret) {
1106                 bch_err(c, "error initializing blacklist table");
1107                 goto err;
1108         }
1109
1110         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1111                 struct genradix_iter iter;
1112                 struct journal_replay **i;
1113
1114                 bch_verbose(c, "starting journal read");
1115                 ret = bch2_journal_read(c, &blacklist_seq, &journal_seq);
1116                 if (ret)
1117                         goto err;
1118
1119                 genradix_for_each_reverse(&c->journal_entries, iter, i)
1120                         if (*i && !(*i)->ignore) {
1121                                 last_journal_entry = &(*i)->j;
1122                                 break;
1123                         }
1124
1125                 if (mustfix_fsck_err_on(c->sb.clean &&
1126                                         last_journal_entry &&
1127                                         !journal_entry_empty(last_journal_entry), c,
1128                                 "filesystem marked clean but journal not empty")) {
1129                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1130                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1131                         c->sb.clean = false;
1132                 }
1133
1134                 if (!last_journal_entry) {
1135                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
1136                         goto use_clean;
1137                 }
1138
1139                 ret = journal_keys_sort(c);
1140                 if (ret)
1141                         goto err;
1142
1143                 if (c->sb.clean && last_journal_entry) {
1144                         ret = verify_superblock_clean(c, &clean,
1145                                                       last_journal_entry);
1146                         if (ret)
1147                                 goto err;
1148                 }
1149         } else {
1150 use_clean:
1151                 if (!clean) {
1152                         bch_err(c, "no superblock clean section found");
1153                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1154                         goto err;
1155
1156                 }
1157                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1158         }
1159
1160         if (c->opts.read_journal_only)
1161                 goto out;
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         ret = bch2_fs_journal_start(&c->journal, journal_seq);
1194         if (ret)
1195                 goto err;
1196
1197         /*
1198          * Skip past versions that might have possibly been used (as nonces),
1199          * but hadn't had their pointers written:
1200          */
1201         if (c->sb.encryption_type && !c->sb.clean)
1202                 atomic64_add(1 << 16, &c->key_version);
1203
1204         ret = read_btree_roots(c);
1205         if (ret)
1206                 goto err;
1207
1208         bch_verbose(c, "starting alloc read");
1209         err = "error reading allocation information";
1210
1211         down_read(&c->gc_lock);
1212         ret = bch2_alloc_read(c);
1213         up_read(&c->gc_lock);
1214
1215         if (ret)
1216                 goto err;
1217         bch_verbose(c, "alloc read done");
1218
1219         bch_verbose(c, "starting stripes_read");
1220         err = "error reading stripes";
1221         ret = bch2_stripes_read(c);
1222         if (ret)
1223                 goto err;
1224         bch_verbose(c, "stripes_read done");
1225
1226         bch2_stripes_heap_start(c);
1227
1228         if (c->opts.fsck) {
1229                 bool metadata_only = c->opts.norecovery;
1230
1231                 bch_info(c, "checking allocations");
1232                 err = "error checking allocations";
1233                 ret = bch2_gc(c, true, metadata_only);
1234                 if (ret)
1235                         goto err;
1236                 bch_verbose(c, "done checking allocations");
1237
1238                 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1239
1240                 bch_info(c, "checking need_discard and freespace btrees");
1241                 err = "error checking need_discard and freespace btrees";
1242                 ret = bch2_check_alloc_info(c);
1243                 if (ret)
1244                         goto err;
1245                 bch_verbose(c, "done checking need_discard and freespace btrees");
1246
1247                 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1248
1249                 bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1250                 err = "journal replay failed";
1251                 ret = bch2_journal_replay(c);
1252                 if (ret)
1253                         goto err;
1254                 if (c->opts.verbose || !c->sb.clean)
1255                         bch_info(c, "journal replay done");
1256
1257                 bch_info(c, "checking lrus");
1258                 err = "error checking lrus";
1259                 ret = bch2_check_lrus(c, true);
1260                 if (ret)
1261                         goto err;
1262                 bch_verbose(c, "done checking lrus");
1263
1264                 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1265
1266                 bch_info(c, "checking alloc to lru refs");
1267                 err = "error checking alloc to lru refs";
1268                 ret = bch2_check_alloc_to_lru_refs(c);
1269                 if (ret)
1270                         goto err;
1271                 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1272
1273                 ret = bch2_check_lrus(c, true);
1274                 if (ret)
1275                         goto err;
1276                 bch_verbose(c, "done checking alloc to lru refs");
1277         } else {
1278                 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1279                 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1280                 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1281                 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1282                 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1283
1284                 if (c->opts.norecovery)
1285                         goto out;
1286
1287                 bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1288                 err = "journal replay failed";
1289                 ret = bch2_journal_replay(c);
1290                 if (ret)
1291                         goto err;
1292                 if (c->opts.verbose || !c->sb.clean)
1293                         bch_info(c, "journal replay done");
1294         }
1295
1296         err = "error initializing freespace";
1297         ret = bch2_fs_freespace_init(c);
1298         if (ret)
1299                 goto err;
1300
1301         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1302                 bch2_fs_lazy_rw(c);
1303
1304                 err = "error creating root snapshot node";
1305                 ret = bch2_fs_initialize_subvolumes(c);
1306                 if (ret)
1307                         goto err;
1308         }
1309
1310         bch_verbose(c, "reading snapshots table");
1311         err = "error reading snapshots table";
1312         ret = bch2_fs_snapshots_start(c);
1313         if (ret)
1314                 goto err;
1315         bch_verbose(c, "reading snapshots done");
1316
1317         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1318                 /* set bi_subvol on root inode */
1319                 err = "error upgrade root inode for subvolumes";
1320                 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1321                                     bch2_fs_upgrade_for_subvolumes(&trans));
1322                 if (ret)
1323                         goto err;
1324         }
1325
1326         if (c->opts.fsck) {
1327                 bch_info(c, "starting fsck");
1328                 err = "error in fsck";
1329                 ret = bch2_fsck_full(c);
1330                 if (ret)
1331                         goto err;
1332                 bch_verbose(c, "fsck done");
1333         } else if (!c->sb.clean) {
1334                 bch_verbose(c, "checking for deleted inodes");
1335                 err = "error in recovery";
1336                 ret = bch2_fsck_walk_inodes_only(c);
1337                 if (ret)
1338                         goto err;
1339                 bch_verbose(c, "check inodes done");
1340         }
1341
1342         if (enabled_qtypes(c)) {
1343                 bch_verbose(c, "reading quotas");
1344                 ret = bch2_fs_quota_read(c);
1345                 if (ret)
1346                         goto err;
1347                 bch_verbose(c, "quotas done");
1348         }
1349
1350         mutex_lock(&c->sb_lock);
1351         if (c->opts.version_upgrade) {
1352                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1353                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1354                 write_sb = true;
1355         }
1356
1357         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1358                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1359                 write_sb = true;
1360         }
1361
1362         if (c->opts.fsck &&
1363             !test_bit(BCH_FS_ERROR, &c->flags) &&
1364             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1365                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1366                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1367                 write_sb = true;
1368         }
1369
1370         if (write_sb)
1371                 bch2_write_super(c);
1372         mutex_unlock(&c->sb_lock);
1373
1374         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1375             !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1376             le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) {
1377                 struct bch_move_stats stats;
1378
1379                 bch_move_stats_init(&stats, "recovery");
1380
1381                 bch_info(c, "scanning for old btree nodes");
1382                 ret = bch2_fs_read_write(c);
1383                 if (ret)
1384                         goto err;
1385
1386                 ret = bch2_scan_old_btree_nodes(c, &stats);
1387                 if (ret)
1388                         goto err;
1389                 bch_info(c, "scanning for old btree nodes done");
1390         }
1391
1392         if (c->journal_seq_blacklist_table &&
1393             c->journal_seq_blacklist_table->nr > 128)
1394                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1395
1396         ret = 0;
1397 out:
1398         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1399         bch2_flush_fsck_errs(c);
1400
1401         if (!c->opts.keep_journal) {
1402                 bch2_journal_keys_free(&c->journal_keys);
1403                 bch2_journal_entries_free(c);
1404         }
1405         kfree(clean);
1406         if (ret)
1407                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1408         else
1409                 bch_verbose(c, "ret %i", ret);
1410         return ret;
1411 err:
1412 fsck_err:
1413         bch2_fs_emergency_read_only(c);
1414         goto out;
1415 }
1416
1417 int bch2_fs_initialize(struct bch_fs *c)
1418 {
1419         struct bch_inode_unpacked root_inode, lostfound_inode;
1420         struct bkey_inode_buf packed_inode;
1421         struct qstr lostfound = QSTR("lost+found");
1422         const char *err = "cannot allocate memory";
1423         struct bch_dev *ca;
1424         unsigned i;
1425         int ret;
1426
1427         bch_notice(c, "initializing new filesystem");
1428
1429         mutex_lock(&c->sb_lock);
1430         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1431         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1432
1433         if (c->opts.version_upgrade) {
1434                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1435                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1436                 bch2_write_super(c);
1437         }
1438         mutex_unlock(&c->sb_lock);
1439
1440         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1441         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1442         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1443
1444         for (i = 0; i < BTREE_ID_NR; i++)
1445                 bch2_btree_root_alloc(c, i);
1446
1447         for_each_online_member(ca, c, i)
1448                 bch2_dev_usage_init(ca);
1449
1450         err = "unable to allocate journal buckets";
1451         for_each_online_member(ca, c, i) {
1452                 ret = bch2_dev_journal_alloc(ca);
1453                 if (ret) {
1454                         percpu_ref_put(&ca->io_ref);
1455                         goto err;
1456                 }
1457         }
1458
1459         /*
1460          * journal_res_get() will crash if called before this has
1461          * set up the journal.pin FIFO and journal.cur pointer:
1462          */
1463         bch2_fs_journal_start(&c->journal, 1);
1464         bch2_journal_set_replay_done(&c->journal);
1465
1466         err = "error going read-write";
1467         ret = bch2_fs_read_write_early(c);
1468         if (ret)
1469                 goto err;
1470
1471         /*
1472          * Write out the superblock and journal buckets, now that we can do
1473          * btree updates
1474          */
1475         bch_verbose(c, "marking superblocks");
1476         err = "error marking superblock and journal";
1477         for_each_member_device(ca, c, i) {
1478                 ret = bch2_trans_mark_dev_sb(c, ca);
1479                 if (ret) {
1480                         percpu_ref_put(&ca->ref);
1481                         goto err;
1482                 }
1483
1484                 ca->new_fs_bucket_idx = 0;
1485         }
1486
1487         bch_verbose(c, "initializing freespace");
1488         err = "error initializing freespace";
1489         ret = bch2_fs_freespace_init(c);
1490         if (ret)
1491                 goto err;
1492
1493         err = "error creating root snapshot node";
1494         ret = bch2_fs_initialize_subvolumes(c);
1495         if (ret)
1496                 goto err;
1497
1498         bch_verbose(c, "reading snapshots table");
1499         err = "error reading snapshots table";
1500         ret = bch2_fs_snapshots_start(c);
1501         if (ret)
1502                 goto err;
1503         bch_verbose(c, "reading snapshots done");
1504
1505         bch2_inode_init(c, &root_inode, 0, 0,
1506                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1507         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1508         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1509         bch2_inode_pack(c, &packed_inode, &root_inode);
1510         packed_inode.inode.k.p.snapshot = U32_MAX;
1511
1512         err = "error creating root directory";
1513         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1514                                 &packed_inode.inode.k_i,
1515                                 NULL, NULL, 0);
1516         if (ret)
1517                 goto err;
1518
1519         bch2_inode_init_early(c, &lostfound_inode);
1520
1521         err = "error creating lost+found";
1522         ret = bch2_trans_do(c, NULL, NULL, 0,
1523                 bch2_create_trans(&trans,
1524                                   BCACHEFS_ROOT_SUBVOL_INUM,
1525                                   &root_inode, &lostfound_inode,
1526                                   &lostfound,
1527                                   0, 0, S_IFDIR|0700, 0,
1528                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1529         if (ret) {
1530                 bch_err(c, "error creating lost+found");
1531                 goto err;
1532         }
1533
1534         if (enabled_qtypes(c)) {
1535                 ret = bch2_fs_quota_read(c);
1536                 if (ret)
1537                         goto err;
1538         }
1539
1540         err = "error writing first journal entry";
1541         ret = bch2_journal_flush(&c->journal);
1542         if (ret)
1543                 goto err;
1544
1545         mutex_lock(&c->sb_lock);
1546         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1547         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1548
1549         bch2_write_super(c);
1550         mutex_unlock(&c->sb_lock);
1551
1552         return 0;
1553 err:
1554         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1555         return ret;
1556 }