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

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "backpointers.h"
   5 #include "bkey_buf.h"
   6 #include "alloc_background.h"
   7 #include "btree_gc.h"
   8 #include "btree_update.h"
   9 #include "btree_update_interior.h"
  10 #include "btree_io.h"
  11 #include "buckets.h"
  12 #include "dirent.h"
  13 #include "ec.h"
  14 #include "errcode.h"
  15 #include "error.h"
  16 #include "fs-common.h"
  17 #include "fsck.h"
  18 #include "journal_io.h"
  19 #include "journal_reclaim.h"
  20 #include "journal_seq_blacklist.h"
  21 #include "lru.h"
  22 #include "move.h"
  23 #include "quota.h"
  24 #include "recovery.h"
  25 #include "replicas.h"
  26 #include "subvolume.h"
  27 #include "super-io.h"
  28
  29 #include <linux/sort.h>
  30 #include <linux/stat.h>
  31
  32 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  33
  34 /* for -o reconstruct_alloc: */
  35 static void drop_alloc_keys(struct journal_keys *keys)
  36 {
  37         size_t src, dst;
  38
  39         for (src = 0, dst = 0; src < keys->nr; src++)
  40                 if (keys->d[src].btree_id != BTREE_ID_alloc)
  41                         keys->d[dst++] = keys->d[src];
  42
  43         keys->nr = dst;
  44 }
  45
  46 /*
  47  * Btree node pointers have a field to stack a pointer to the in memory btree
  48  * node; we need to zero out this field when reading in btree nodes, or when
  49  * reading in keys from the journal:
  50  */
  51 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
  52 {
  53         struct journal_key *i;
  54
  55         for (i = keys->d; i < keys->d + keys->nr; i++)
  56                 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
  57                         bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
  58 }
  59
  60 /* iterate over keys read from the journal: */
  61
  62 static int __journal_key_cmp(enum btree_id      l_btree_id,
  63                              unsigned           l_level,
  64                              struct bpos        l_pos,
  65                              const struct journal_key *r)
  66 {
  67         return (cmp_int(l_btree_id,     r->btree_id) ?:
  68                 cmp_int(l_level,        r->level) ?:
  69                 bpos_cmp(l_pos, r->k->k.p));
  70 }
  71
  72 static int journal_key_cmp(const struct journal_key *l, const struct journal_key *r)
  73 {
  74         return __journal_key_cmp(l->btree_id, l->level, l->k->k.p, r);
  75 }
  76
  77 static inline size_t idx_to_pos(struct journal_keys *keys, size_t idx)
  78 {
  79         size_t gap_size = keys->size - keys->nr;
  80
  81         if (idx >= keys->gap)
  82                 idx += gap_size;
  83         return idx;
  84 }
  85
  86 static inline struct journal_key *idx_to_key(struct journal_keys *keys, size_t idx)
  87 {
  88         return keys->d + idx_to_pos(keys, idx);
  89 }
  90
  91 static size_t __bch2_journal_key_search(struct journal_keys *keys,
  92                                         enum btree_id id, unsigned level,
  93                                         struct bpos pos)
  94 {
  95         size_t l = 0, r = keys->nr, m;
  96
  97         while (l < r) {
  98                 m = l + ((r - l) >> 1);
  99                 if (__journal_key_cmp(id, level, pos, idx_to_key(keys, m)) > 0)
 100                         l = m + 1;
 101                 else
 102                         r = m;
 103         }
 104
 105         BUG_ON(l < keys->nr &&
 106                __journal_key_cmp(id, level, pos, idx_to_key(keys, l)) > 0);
 107
 108         BUG_ON(l &&
 109                __journal_key_cmp(id, level, pos, idx_to_key(keys, l - 1)) <= 0);
 110
 111         return l;
 112 }
 113
 114 static size_t bch2_journal_key_search(struct journal_keys *keys,
 115                                       enum btree_id id, unsigned level,
 116                                       struct bpos pos)
 117 {
 118         return idx_to_pos(keys, __bch2_journal_key_search(keys, id, level, pos));
 119 }
 120
 121 struct bkey_i *bch2_journal_keys_peek_upto(struct bch_fs *c, enum btree_id btree_id,
 122                                            unsigned level, struct bpos pos,
 123                                            struct bpos end_pos, size_t *idx)
 124 {
 125         struct journal_keys *keys = &c->journal_keys;
 126         unsigned iters = 0;
 127         struct journal_key *k;
 128 search:
 129         if (!*idx)
 130                 *idx = __bch2_journal_key_search(keys, btree_id, level, pos);
 131
 132         while ((k = *idx < keys->nr ? idx_to_key(keys, *idx) : NULL)) {
 133                 if (__journal_key_cmp(btree_id, level, end_pos, k) < 0)
 134                         return NULL;
 135
 136                 if (__journal_key_cmp(btree_id, level, pos, k) <= 0 &&
 137                     !k->overwritten)
 138                         return k->k;
 139
 140                 (*idx)++;
 141                 iters++;
 142                 if (iters == 10) {
 143                         *idx = 0;
 144                         goto search;
 145                 }
 146         }
 147
 148         return NULL;
 149 }
 150
 151 struct bkey_i *bch2_journal_keys_peek_slot(struct bch_fs *c, enum btree_id btree_id,
 152                                            unsigned level, struct bpos pos)
 153 {
 154         size_t idx = 0;
 155
 156         return bch2_journal_keys_peek_upto(c, btree_id, level, pos, pos, &idx);
 157 }
 158
 159 static void journal_iters_fix(struct bch_fs *c)
 160 {
 161         struct journal_keys *keys = &c->journal_keys;
 162         /* The key we just inserted is immediately before the gap: */
 163         size_t gap_end = keys->gap + (keys->size - keys->nr);
 164         struct btree_and_journal_iter *iter;
 165
 166         /*
 167          * If an iterator points one after the key we just inserted, decrement
 168          * the iterator so it points at the key we just inserted - if the
 169          * decrement was unnecessary, bch2_btree_and_journal_iter_peek() will
 170          * handle that:
 171          */
 172         list_for_each_entry(iter, &c->journal_iters, journal.list)
 173                 if (iter->journal.idx == gap_end)
 174                         iter->journal.idx = keys->gap - 1;
 175 }
 176
 177 static void journal_iters_move_gap(struct bch_fs *c, size_t old_gap, size_t new_gap)
 178 {
 179         struct journal_keys *keys = &c->journal_keys;
 180         struct journal_iter *iter;
 181         size_t gap_size = keys->size - keys->nr;
 182
 183         list_for_each_entry(iter, &c->journal_iters, list) {
 184                 if (iter->idx > old_gap)
 185                         iter->idx -= gap_size;
 186                 if (iter->idx >= new_gap)
 187                         iter->idx += gap_size;
 188         }
 189 }
 190
 191 int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
 192                                  unsigned level, struct bkey_i *k)
 193 {
 194         struct journal_key n = {
 195                 .btree_id       = id,
 196                 .level          = level,
 197                 .k              = k,
 198                 .allocated      = true,
 199                 /*
 200                  * Ensure these keys are done last by journal replay, to unblock
 201                  * journal reclaim:
 202                  */
 203                 .journal_seq    = U32_MAX,
 204         };
 205         struct journal_keys *keys = &c->journal_keys;
 206         size_t idx = bch2_journal_key_search(keys, id, level, k->k.p);
 207
 208         BUG_ON(test_bit(BCH_FS_RW, &c->flags));
 209
 210         if (idx < keys->size &&
 211             journal_key_cmp(&n, &keys->d[idx]) == 0) {
 212                 if (keys->d[idx].allocated)
 213                         kfree(keys->d[idx].k);
 214                 keys->d[idx] = n;
 215                 return 0;
 216         }
 217
 218         if (idx > keys->gap)
 219                 idx -= keys->size - keys->nr;
 220
 221         if (keys->nr == keys->size) {
 222                 struct journal_keys new_keys = {
 223                         .nr                     = keys->nr,
 224                         .size                   = max_t(size_t, keys->size, 8) * 2,
 225                 };
 226
 227                 new_keys.d = kvmalloc_array(new_keys.size, sizeof(new_keys.d[0]), GFP_KERNEL);
 228                 if (!new_keys.d) {
 229                         bch_err(c, "%s: error allocating new key array (size %zu)",
 230                                 __func__, new_keys.size);
 231                         return -ENOMEM;
 232                 }
 233
 234                 /* Since @keys was full, there was no gap: */
 235                 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
 236                 kvfree(keys->d);
 237                 *keys = new_keys;
 238
 239                 /* And now the gap is at the end: */
 240                 keys->gap = keys->nr;
 241         }
 242
 243         journal_iters_move_gap(c, keys->gap, idx);
 244
 245         move_gap(keys->d, keys->nr, keys->size, keys->gap, idx);
 246         keys->gap = idx;
 247
 248         keys->nr++;
 249         keys->d[keys->gap++] = n;
 250
 251         journal_iters_fix(c);
 252
 253         return 0;
 254 }
 255
 256 /*
 257  * Can only be used from the recovery thread while we're still RO - can't be
 258  * used once we've got RW, as journal_keys is at that point used by multiple
 259  * threads:
 260  */
 261 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
 262                             unsigned level, struct bkey_i *k)
 263 {
 264         struct bkey_i *n;
 265         int ret;
 266
 267         n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
 268         if (!n)
 269                 return -ENOMEM;
 270
 271         bkey_copy(n, k);
 272         ret = bch2_journal_key_insert_take(c, id, level, n);
 273         if (ret)
 274                 kfree(n);
 275         return ret;
 276 }
 277
 278 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
 279                             unsigned level, struct bpos pos)
 280 {
 281         struct bkey_i whiteout;
 282
 283         bkey_init(&whiteout.k);
 284         whiteout.k.p = pos;
 285
 286         return bch2_journal_key_insert(c, id, level, &whiteout);
 287 }
 288
 289 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
 290                                   unsigned level, struct bpos pos)
 291 {
 292         struct journal_keys *keys = &c->journal_keys;
 293         size_t idx = bch2_journal_key_search(keys, btree, level, pos);
 294
 295         if (idx < keys->size &&
 296             keys->d[idx].btree_id       == btree &&
 297             keys->d[idx].level          == level &&
 298             bpos_eq(keys->d[idx].k->k.p, pos))
 299                 keys->d[idx].overwritten = true;
 300 }
 301
 302 static void bch2_journal_iter_advance(struct journal_iter *iter)
 303 {
 304         if (iter->idx < iter->keys->size) {
 305                 iter->idx++;
 306                 if (iter->idx == iter->keys->gap)
 307                         iter->idx += iter->keys->size - iter->keys->nr;
 308         }
 309 }
 310
 311 struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
 312 {
 313         struct journal_key *k = iter->keys->d + iter->idx;
 314
 315         while (k < iter->keys->d + iter->keys->size &&
 316                k->btree_id      == iter->btree_id &&
 317                k->level         == iter->level) {
 318                 if (!k->overwritten)
 319                         return bkey_i_to_s_c(k->k);
 320
 321                 bch2_journal_iter_advance(iter);
 322                 k = iter->keys->d + iter->idx;
 323         }
 324
 325         return bkey_s_c_null;
 326 }
 327
 328 static void bch2_journal_iter_exit(struct journal_iter *iter)
 329 {
 330         list_del(&iter->list);
 331 }
 332
 333 static void bch2_journal_iter_init(struct bch_fs *c,
 334                                    struct journal_iter *iter,
 335                                    enum btree_id id, unsigned level,
 336                                    struct bpos pos)
 337 {
 338         iter->btree_id  = id;
 339         iter->level     = level;
 340         iter->keys      = &c->journal_keys;
 341         iter->idx       = bch2_journal_key_search(&c->journal_keys, id, level, pos);
 342 }
 343
 344 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 345 {
 346         return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 347                                                 iter->b, &iter->unpacked);
 348 }
 349
 350 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 351 {
 352         bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 353 }
 354
 355 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 356 {
 357         if (bpos_eq(iter->pos, SPOS_MAX))
 358                 iter->at_end = true;
 359         else
 360                 iter->pos = bpos_successor(iter->pos);
 361 }
 362
 363 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 364 {
 365         struct bkey_s_c btree_k, journal_k, ret;
 366 again:
 367         if (iter->at_end)
 368                 return bkey_s_c_null;
 369
 370         while ((btree_k = bch2_journal_iter_peek_btree(iter)).k &&
 371                bpos_lt(btree_k.k->p, iter->pos))
 372                 bch2_journal_iter_advance_btree(iter);
 373
 374         while ((journal_k = bch2_journal_iter_peek(&iter->journal)).k &&
 375                bpos_lt(journal_k.k->p, iter->pos))
 376                 bch2_journal_iter_advance(&iter->journal);
 377
 378         ret = journal_k.k &&
 379                 (!btree_k.k || bpos_le(journal_k.k->p, btree_k.k->p))
 380                 ? journal_k
 381                 : btree_k;
 382
 383         if (ret.k && iter->b && bpos_gt(ret.k->p, iter->b->data->max_key))
 384                 ret = bkey_s_c_null;
 385
 386         if (ret.k) {
 387                 iter->pos = ret.k->p;
 388                 if (bkey_deleted(ret.k)) {
 389                         bch2_btree_and_journal_iter_advance(iter);
 390                         goto again;
 391                 }
 392         } else {
 393                 iter->pos = SPOS_MAX;
 394                 iter->at_end = true;
 395         }
 396
 397         return ret;
 398 }
 399
 400 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
 401 {
 402         bch2_journal_iter_exit(&iter->journal);
 403 }
 404
 405 void __bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 406                                                   struct bch_fs *c,
 407                                                   struct btree *b,
 408                                                   struct btree_node_iter node_iter,
 409                                                   struct bpos pos)
 410 {
 411         memset(iter, 0, sizeof(*iter));
 412
 413         iter->b = b;
 414         iter->node_iter = node_iter;
 415         bch2_journal_iter_init(c, &iter->journal, b->c.btree_id, b->c.level, pos);
 416         INIT_LIST_HEAD(&iter->journal.list);
 417         iter->pos = b->data->min_key;
 418         iter->at_end = false;
 419 }
 420
 421 /*
 422  * this version is used by btree_gc before filesystem has gone RW and
 423  * multithreaded, so uses the journal_iters list:
 424  */
 425 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 426                                                 struct bch_fs *c,
 427                                                 struct btree *b)
 428 {
 429         struct btree_node_iter node_iter;
 430
 431         bch2_btree_node_iter_init_from_start(&node_iter, b);
 432         __bch2_btree_and_journal_iter_init_node_iter(iter, c, b, node_iter, b->data->min_key);
 433         list_add(&iter->journal.list, &c->journal_iters);
 434 }
 435
 436 /* sort and dedup all keys in the journal: */
 437
 438 void bch2_journal_entries_free(struct bch_fs *c)
 439 {
 440         struct journal_replay **i;
 441         struct genradix_iter iter;
 442
 443         genradix_for_each(&c->journal_entries, iter, i)
 444                 if (*i)
 445                         kvpfree(*i, offsetof(struct journal_replay, j) +
 446                                 vstruct_bytes(&(*i)->j));
 447         genradix_free(&c->journal_entries);
 448 }
 449
 450 /*
 451  * When keys compare equal, oldest compares first:
 452  */
 453 static int journal_sort_key_cmp(const void *_l, const void *_r)
 454 {
 455         const struct journal_key *l = _l;
 456         const struct journal_key *r = _r;
 457
 458         return  journal_key_cmp(l, r) ?:
 459                 cmp_int(l->journal_seq, r->journal_seq) ?:
 460                 cmp_int(l->journal_offset, r->journal_offset);
 461 }
 462
 463 void bch2_journal_keys_free(struct journal_keys *keys)
 464 {
 465         struct journal_key *i;
 466
 467         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 468         keys->gap = keys->nr;
 469
 470         for (i = keys->d; i < keys->d + keys->nr; i++)
 471                 if (i->allocated)
 472                         kfree(i->k);
 473
 474         kvfree(keys->d);
 475         keys->d = NULL;
 476         keys->nr = keys->gap = keys->size = 0;
 477 }
 478
 479 static int journal_keys_sort(struct bch_fs *c)
 480 {
 481         struct genradix_iter iter;
 482         struct journal_replay *i, **_i;
 483         struct jset_entry *entry;
 484         struct bkey_i *k, *_n;
 485         struct journal_keys *keys = &c->journal_keys;
 486         struct journal_key *src, *dst;
 487         size_t nr_keys = 0;
 488
 489         genradix_for_each(&c->journal_entries, iter, _i) {
 490                 i = *_i;
 491
 492                 if (!i || i->ignore)
 493                         continue;
 494
 495                 for_each_jset_key(k, _n, entry, &i->j)
 496                         nr_keys++;
 497         }
 498
 499         if (!nr_keys)
 500                 return 0;
 501
 502         keys->size = roundup_pow_of_two(nr_keys);
 503
 504         keys->d = kvmalloc_array(keys->size, sizeof(keys->d[0]), GFP_KERNEL);
 505         if (!keys->d)
 506                 return -ENOMEM;
 507
 508         genradix_for_each(&c->journal_entries, iter, _i) {
 509                 i = *_i;
 510
 511                 if (!i || i->ignore)
 512                         continue;
 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                                 .journal_offset = k->_data - i->j._data,
 521                         };
 522         }
 523
 524         sort(keys->d, keys->nr, sizeof(keys->d[0]), journal_sort_key_cmp, NULL);
 525
 526         src = dst = keys->d;
 527         while (src < keys->d + keys->nr) {
 528                 while (src + 1 < keys->d + keys->nr &&
 529                        src[0].btree_id  == src[1].btree_id &&
 530                        src[0].level     == src[1].level &&
 531                        bpos_eq(src[0].k->k.p, src[1].k->k.p))
 532                         src++;
 533
 534                 *dst++ = *src++;
 535         }
 536
 537         keys->nr = dst - keys->d;
 538         keys->gap = keys->nr;
 539         return 0;
 540 }
 541
 542 /* journal replay: */
 543
 544 static void replay_now_at(struct journal *j, u64 seq)
 545 {
 546         BUG_ON(seq < j->replay_journal_seq);
 547
 548         seq = min(seq, j->replay_journal_seq_end);
 549
 550         while (j->replay_journal_seq < seq)
 551                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 552 }
 553
 554 static int bch2_journal_replay_key(struct btree_trans *trans,
 555                                    struct journal_key *k)
 556 {
 557         struct btree_iter iter;
 558         unsigned iter_flags =
 559                 BTREE_ITER_INTENT|
 560                 BTREE_ITER_NOT_EXTENTS;
 561         int ret;
 562
 563         if (!k->level && k->btree_id == BTREE_ID_alloc)
 564                 iter_flags |= BTREE_ITER_CACHED;
 565
 566         bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 567                                   BTREE_MAX_DEPTH, k->level,
 568                                   iter_flags);
 569         ret = bch2_btree_iter_traverse(&iter);
 570         if (ret)
 571                 goto out;
 572
 573         /* Must be checked with btree locked: */
 574         if (k->overwritten)
 575                 goto out;
 576
 577         ret = bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
 578 out:
 579         bch2_trans_iter_exit(trans, &iter);
 580         return ret;
 581 }
 582
 583 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 584 {
 585         const struct journal_key *l = *((const struct journal_key **)_l);
 586         const struct journal_key *r = *((const struct journal_key **)_r);
 587
 588         return cmp_int(l->journal_seq, r->journal_seq);
 589 }
 590
 591 static int bch2_journal_replay(struct bch_fs *c, u64 start_seq, u64 end_seq)
 592 {
 593         struct journal_keys *keys = &c->journal_keys;
 594         struct journal_key **keys_sorted, *k;
 595         struct journal *j = &c->journal;
 596         size_t i;
 597         int ret;
 598
 599         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 600         keys->gap = keys->nr;
 601
 602         keys_sorted = kvmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL);
 603         if (!keys_sorted)
 604                 return -ENOMEM;
 605
 606         for (i = 0; i < keys->nr; i++)
 607                 keys_sorted[i] = &keys->d[i];
 608
 609         sort(keys_sorted, keys->nr,
 610              sizeof(keys_sorted[0]),
 611              journal_sort_seq_cmp, NULL);
 612
 613         if (keys->nr) {
 614                 ret = bch2_fs_log_msg(c, "Starting journal replay (%zu keys in entries %llu-%llu)",
 615                                       keys->nr, start_seq, end_seq);
 616                 if (ret)
 617                         goto err;
 618         }
 619
 620         for (i = 0; i < keys->nr; i++) {
 621                 k = keys_sorted[i];
 622
 623                 cond_resched();
 624
 625                 replay_now_at(j, k->journal_seq);
 626
 627                 ret = bch2_trans_do(c, NULL, NULL,
 628                                     BTREE_INSERT_LAZY_RW|
 629                                     BTREE_INSERT_NOFAIL|
 630                                     (!k->allocated
 631                                      ? BTREE_INSERT_JOURNAL_REPLAY|JOURNAL_WATERMARK_reserved
 632                                      : 0),
 633                              bch2_journal_replay_key(&trans, k));
 634                 if (ret) {
 635                         bch_err(c, "journal replay: error while replaying key at btree %s level %u: %s",
 636                                 bch2_btree_ids[k->btree_id], k->level, bch2_err_str(ret));
 637                         goto err;
 638                 }
 639         }
 640
 641         replay_now_at(j, j->replay_journal_seq_end);
 642         j->replay_journal_seq = 0;
 643
 644         bch2_journal_set_replay_done(j);
 645         bch2_journal_flush_all_pins(j);
 646         ret = bch2_journal_error(j);
 647
 648         if (keys->nr && !ret)
 649                 bch2_fs_log_msg(c, "journal replay finished");
 650 err:
 651         kvfree(keys_sorted);
 652         return ret;
 653 }
 654
 655 /* journal replay early: */
 656
 657 static int journal_replay_entry_early(struct bch_fs *c,
 658                                       struct jset_entry *entry)
 659 {
 660         int ret = 0;
 661
 662         switch (entry->type) {
 663         case BCH_JSET_ENTRY_btree_root: {
 664                 struct btree_root *r;
 665
 666                 if (entry->btree_id >= BTREE_ID_NR) {
 667                         bch_err(c, "filesystem has unknown btree type %u",
 668                                 entry->btree_id);
 669                         return -EINVAL;
 670                 }
 671
 672                 r = &c->btree_roots[entry->btree_id];
 673
 674                 if (entry->u64s) {
 675                         r->level = entry->level;
 676                         bkey_copy(&r->key, &entry->start[0]);
 677                         r->error = 0;
 678                 } else {
 679                         r->error = -EIO;
 680                 }
 681                 r->alive = true;
 682                 break;
 683         }
 684         case BCH_JSET_ENTRY_usage: {
 685                 struct jset_entry_usage *u =
 686                         container_of(entry, struct jset_entry_usage, entry);
 687
 688                 switch (entry->btree_id) {
 689                 case BCH_FS_USAGE_reserved:
 690                         if (entry->level < BCH_REPLICAS_MAX)
 691                                 c->usage_base->persistent_reserved[entry->level] =
 692                                         le64_to_cpu(u->v);
 693                         break;
 694                 case BCH_FS_USAGE_inodes:
 695                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 696                         break;
 697                 case BCH_FS_USAGE_key_version:
 698                         atomic64_set(&c->key_version,
 699                                      le64_to_cpu(u->v));
 700                         break;
 701                 }
 702
 703                 break;
 704         }
 705         case BCH_JSET_ENTRY_data_usage: {
 706                 struct jset_entry_data_usage *u =
 707                         container_of(entry, struct jset_entry_data_usage, entry);
 708
 709                 ret = bch2_replicas_set_usage(c, &u->r,
 710                                               le64_to_cpu(u->v));
 711                 break;
 712         }
 713         case BCH_JSET_ENTRY_dev_usage: {
 714                 struct jset_entry_dev_usage *u =
 715                         container_of(entry, struct jset_entry_dev_usage, entry);
 716                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 717                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 718
 719                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 720
 721                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 722                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 723                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 724                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 725                 }
 726
 727                 break;
 728         }
 729         case BCH_JSET_ENTRY_blacklist: {
 730                 struct jset_entry_blacklist *bl_entry =
 731                         container_of(entry, struct jset_entry_blacklist, entry);
 732
 733                 ret = bch2_journal_seq_blacklist_add(c,
 734                                 le64_to_cpu(bl_entry->seq),
 735                                 le64_to_cpu(bl_entry->seq) + 1);
 736                 break;
 737         }
 738         case BCH_JSET_ENTRY_blacklist_v2: {
 739                 struct jset_entry_blacklist_v2 *bl_entry =
 740                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 741
 742                 ret = bch2_journal_seq_blacklist_add(c,
 743                                 le64_to_cpu(bl_entry->start),
 744                                 le64_to_cpu(bl_entry->end) + 1);
 745                 break;
 746         }
 747         case BCH_JSET_ENTRY_clock: {
 748                 struct jset_entry_clock *clock =
 749                         container_of(entry, struct jset_entry_clock, entry);
 750
 751                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 752         }
 753         }
 754
 755         return ret;
 756 }
 757
 758 static int journal_replay_early(struct bch_fs *c,
 759                                 struct bch_sb_field_clean *clean)
 760 {
 761         struct jset_entry *entry;
 762         int ret;
 763
 764         if (clean) {
 765                 for (entry = clean->start;
 766                      entry != vstruct_end(&clean->field);
 767                      entry = vstruct_next(entry)) {
 768                         ret = journal_replay_entry_early(c, entry);
 769                         if (ret)
 770                                 return ret;
 771                 }
 772         } else {
 773                 struct genradix_iter iter;
 774                 struct journal_replay *i, **_i;
 775
 776                 genradix_for_each(&c->journal_entries, iter, _i) {
 777                         i = *_i;
 778
 779                         if (!i || i->ignore)
 780                                 continue;
 781
 782                         vstruct_for_each(&i->j, entry) {
 783                                 ret = journal_replay_entry_early(c, entry);
 784                                 if (ret)
 785                                         return ret;
 786                         }
 787                 }
 788         }
 789
 790         bch2_fs_usage_initialize(c);
 791
 792         return 0;
 793 }
 794
 795 /* sb clean section: */
 796
 797 static struct bkey_i *btree_root_find(struct bch_fs *c,
 798                                       struct bch_sb_field_clean *clean,
 799                                       struct jset *j,
 800                                       enum btree_id id, unsigned *level)
 801 {
 802         struct bkey_i *k;
 803         struct jset_entry *entry, *start, *end;
 804
 805         if (clean) {
 806                 start = clean->start;
 807                 end = vstruct_end(&clean->field);
 808         } else {
 809                 start = j->start;
 810                 end = vstruct_last(j);
 811         }
 812
 813         for (entry = start; entry < end; entry = vstruct_next(entry))
 814                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 815                     entry->btree_id == id)
 816                         goto found;
 817
 818         return NULL;
 819 found:
 820         if (!entry->u64s)
 821                 return ERR_PTR(-EINVAL);
 822
 823         k = entry->start;
 824         *level = entry->level;
 825         return k;
 826 }
 827
 828 static int verify_superblock_clean(struct bch_fs *c,
 829                                    struct bch_sb_field_clean **cleanp,
 830                                    struct jset *j)
 831 {
 832         unsigned i;
 833         struct bch_sb_field_clean *clean = *cleanp;
 834         struct printbuf buf1 = PRINTBUF;
 835         struct printbuf buf2 = PRINTBUF;
 836         int ret = 0;
 837
 838         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 839                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 840                         le64_to_cpu(clean->journal_seq),
 841                         le64_to_cpu(j->seq))) {
 842                 kfree(clean);
 843                 *cleanp = NULL;
 844                 return 0;
 845         }
 846
 847         for (i = 0; i < BTREE_ID_NR; i++) {
 848                 struct bkey_i *k1, *k2;
 849                 unsigned l1 = 0, l2 = 0;
 850
 851                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 852                 k2 = btree_root_find(c, NULL, j, i, &l2);
 853
 854                 if (!k1 && !k2)
 855                         continue;
 856
 857                 printbuf_reset(&buf1);
 858                 printbuf_reset(&buf2);
 859
 860                 if (k1)
 861                         bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(k1));
 862                 else
 863                         prt_printf(&buf1, "(none)");
 864
 865                 if (k2)
 866                         bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(k2));
 867                 else
 868                         prt_printf(&buf2, "(none)");
 869
 870                 mustfix_fsck_err_on(!k1 || !k2 ||
 871                                     IS_ERR(k1) ||
 872                                     IS_ERR(k2) ||
 873                                     k1->k.u64s != k2->k.u64s ||
 874                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 875                                     l1 != l2, c,
 876                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 877                         "sb:      l=%u %s\n"
 878                         "journal: l=%u %s\n", i,
 879                         l1, buf1.buf,
 880                         l2, buf2.buf);
 881         }
 882 fsck_err:
 883         printbuf_exit(&buf2);
 884         printbuf_exit(&buf1);
 885         return ret;
 886 }
 887
 888 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 889 {
 890         struct bch_sb_field_clean *clean, *sb_clean;
 891         int ret;
 892
 893         mutex_lock(&c->sb_lock);
 894         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 895
 896         if (fsck_err_on(!sb_clean, c,
 897                         "superblock marked clean but clean section not present")) {
 898                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 899                 c->sb.clean = false;
 900                 mutex_unlock(&c->sb_lock);
 901                 return NULL;
 902         }
 903
 904         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 905                         GFP_KERNEL);
 906         if (!clean) {
 907                 mutex_unlock(&c->sb_lock);
 908                 return ERR_PTR(-ENOMEM);
 909         }
 910
 911         ret = bch2_sb_clean_validate_late(c, clean, READ);
 912         if (ret) {
 913                 mutex_unlock(&c->sb_lock);
 914                 return ERR_PTR(ret);
 915         }
 916
 917         mutex_unlock(&c->sb_lock);
 918
 919         return clean;
 920 fsck_err:
 921         mutex_unlock(&c->sb_lock);
 922         return ERR_PTR(ret);
 923 }
 924
 925 static bool btree_id_is_alloc(enum btree_id id)
 926 {
 927         switch (id) {
 928         case BTREE_ID_alloc:
 929         case BTREE_ID_backpointers:
 930         case BTREE_ID_need_discard:
 931         case BTREE_ID_freespace:
 932         case BTREE_ID_bucket_gens:
 933                 return true;
 934         default:
 935                 return false;
 936         }
 937 }
 938
 939 static int read_btree_roots(struct bch_fs *c)
 940 {
 941         unsigned i;
 942         int ret = 0;
 943
 944         for (i = 0; i < BTREE_ID_NR; i++) {
 945                 struct btree_root *r = &c->btree_roots[i];
 946
 947                 if (!r->alive)
 948                         continue;
 949
 950                 if (btree_id_is_alloc(i) &&
 951                     c->opts.reconstruct_alloc) {
 952                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 953                         continue;
 954                 }
 955
 956                 if (r->error) {
 957                         __fsck_err(c, btree_id_is_alloc(i)
 958                                    ? FSCK_CAN_IGNORE : 0,
 959                                    "invalid btree root %s",
 960                                    bch2_btree_ids[i]);
 961                         if (i == BTREE_ID_alloc)
 962                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 963                 }
 964
 965                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 966                 if (ret) {
 967                         __fsck_err(c,
 968                                    btree_id_is_alloc(i)
 969                                    ? FSCK_CAN_IGNORE : 0,
 970                                    "error reading btree root %s",
 971                                    bch2_btree_ids[i]);
 972                         if (i == BTREE_ID_alloc)
 973                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 974                 }
 975         }
 976
 977         for (i = 0; i < BTREE_ID_NR; i++)
 978                 if (!c->btree_roots[i].b)
 979                         bch2_btree_root_alloc(c, i);
 980 fsck_err:
 981         return ret;
 982 }
 983
 984 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
 985 {
 986         struct bkey_i_snapshot  root_snapshot;
 987         struct bkey_i_subvolume root_volume;
 988         int ret;
 989
 990         bkey_snapshot_init(&root_snapshot.k_i);
 991         root_snapshot.k.p.offset = U32_MAX;
 992         root_snapshot.v.flags   = 0;
 993         root_snapshot.v.parent  = 0;
 994         root_snapshot.v.subvol  = BCACHEFS_ROOT_SUBVOL;
 995         root_snapshot.v.pad     = 0;
 996         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
 997
 998         ret = bch2_btree_insert(c, BTREE_ID_snapshots,
 999                                 &root_snapshot.k_i,
1000                                 NULL, NULL, 0);
1001         if (ret)
1002                 return ret;
1003
1004         bkey_subvolume_init(&root_volume.k_i);
1005         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1006         root_volume.v.flags     = 0;
1007         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
1008         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
1009
1010         ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
1011                                 &root_volume.k_i,
1012                                 NULL, NULL, 0);
1013         if (ret)
1014                 return ret;
1015
1016         return 0;
1017 }
1018
1019 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
1020 {
1021         struct btree_iter iter;
1022         struct bkey_s_c k;
1023         struct bch_inode_unpacked inode;
1024         int ret;
1025
1026         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
1027                              SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
1028         k = bch2_btree_iter_peek_slot(&iter);
1029         ret = bkey_err(k);
1030         if (ret)
1031                 goto err;
1032
1033         if (!bkey_is_inode(k.k)) {
1034                 bch_err(trans->c, "root inode not found");
1035                 ret = -ENOENT;
1036                 goto err;
1037         }
1038
1039         ret = bch2_inode_unpack(k, &inode);
1040         BUG_ON(ret);
1041
1042         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1043
1044         ret = bch2_inode_write(trans, &iter, &inode);
1045 err:
1046         bch2_trans_iter_exit(trans, &iter);
1047         return ret;
1048 }
1049
1050 int bch2_fs_recovery(struct bch_fs *c)
1051 {
1052         const char *err = "cannot allocate memory";
1053         struct bch_sb_field_clean *clean = NULL;
1054         struct jset *last_journal_entry = NULL;
1055         u64 last_seq, blacklist_seq, journal_seq;
1056         bool write_sb = false;
1057         int ret = 0;
1058
1059         if (c->sb.clean)
1060                 clean = read_superblock_clean(c);
1061         ret = PTR_ERR_OR_ZERO(clean);
1062         if (ret)
1063                 goto err;
1064
1065         if (c->sb.clean)
1066                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
1067                          le64_to_cpu(clean->journal_seq));
1068         else
1069                 bch_info(c, "recovering from unclean shutdown");
1070
1071         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
1072                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
1073                 ret = -EINVAL;
1074                 goto err;
1075         }
1076
1077         if (!c->sb.clean &&
1078             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1079                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1080                 ret = -EINVAL;
1081                 goto err;
1082         }
1083
1084         if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1085                 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
1086                 ret = -EINVAL;
1087                 goto err;
1088         }
1089
1090         if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
1091                 bch_info(c, "alloc_v2 feature bit not set, fsck required");
1092                 c->opts.fsck = true;
1093                 c->opts.fix_errors = FSCK_OPT_YES;
1094         }
1095
1096         if (!c->opts.nochanges) {
1097                 if (c->sb.version < bcachefs_metadata_version_lru_v2) {
1098                         bch_info(c, "version prior to backpointers, upgrade and fsck required");
1099                         c->opts.version_upgrade = true;
1100                         c->opts.fsck            = true;
1101                         c->opts.fix_errors      = FSCK_OPT_YES;
1102                 }
1103         }
1104
1105         if (c->opts.fsck && c->opts.norecovery) {
1106                 bch_err(c, "cannot select both norecovery and fsck");
1107                 ret = -EINVAL;
1108                 goto err;
1109         }
1110
1111         ret = bch2_blacklist_table_initialize(c);
1112         if (ret) {
1113                 bch_err(c, "error initializing blacklist table");
1114                 goto err;
1115         }
1116
1117         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1118                 struct genradix_iter iter;
1119                 struct journal_replay **i;
1120
1121                 bch_verbose(c, "starting journal read");
1122                 ret = bch2_journal_read(c, &last_seq, &blacklist_seq, &journal_seq);
1123                 if (ret)
1124                         goto err;
1125
1126                 /*
1127                  * note: cmd_list_journal needs the blacklist table fully up to date so
1128                  * it can asterisk ignored journal entries:
1129                  */
1130                 if (c->opts.read_journal_only)
1131                         goto out;
1132
1133                 genradix_for_each_reverse(&c->journal_entries, iter, i)
1134                         if (*i && !(*i)->ignore) {
1135                                 last_journal_entry = &(*i)->j;
1136                                 break;
1137                         }
1138
1139                 if (mustfix_fsck_err_on(c->sb.clean &&
1140                                         last_journal_entry &&
1141                                         !journal_entry_empty(last_journal_entry), c,
1142                                 "filesystem marked clean but journal not empty")) {
1143                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1144                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1145                         c->sb.clean = false;
1146                 }
1147
1148                 if (!last_journal_entry) {
1149                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
1150                         if (clean)
1151                                 goto use_clean;
1152
1153                         genradix_for_each_reverse(&c->journal_entries, iter, i)
1154                                 if (*i) {
1155                                         last_journal_entry = &(*i)->j;
1156                                         (*i)->ignore = false;
1157                                         break;
1158                                 }
1159                 }
1160
1161                 ret = journal_keys_sort(c);
1162                 if (ret)
1163                         goto err;
1164
1165                 if (c->sb.clean && last_journal_entry) {
1166                         ret = verify_superblock_clean(c, &clean,
1167                                                       last_journal_entry);
1168                         if (ret)
1169                                 goto err;
1170                 }
1171         } else {
1172 use_clean:
1173                 if (!clean) {
1174                         bch_err(c, "no superblock clean section found");
1175                         ret = -BCH_ERR_fsck_repair_impossible;
1176                         goto err;
1177
1178                 }
1179                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1180         }
1181
1182         if (c->opts.reconstruct_alloc) {
1183                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1184                 drop_alloc_keys(&c->journal_keys);
1185         }
1186
1187         zero_out_btree_mem_ptr(&c->journal_keys);
1188
1189         ret = journal_replay_early(c, clean);
1190         if (ret)
1191                 goto err;
1192
1193         /*
1194          * After an unclean shutdown, skip then next few journal sequence
1195          * numbers as they may have been referenced by btree writes that
1196          * happened before their corresponding journal writes - those btree
1197          * writes need to be ignored, by skipping and blacklisting the next few
1198          * journal sequence numbers:
1199          */
1200         if (!c->sb.clean)
1201                 journal_seq += 8;
1202
1203         if (blacklist_seq != journal_seq) {
1204                 ret =   bch2_fs_log_msg(c, "blacklisting entries %llu-%llu",
1205                                         blacklist_seq, journal_seq) ?:
1206                         bch2_journal_seq_blacklist_add(c,
1207                                         blacklist_seq, journal_seq);
1208                 if (ret) {
1209                         bch_err(c, "error creating new journal seq blacklist entry");
1210                         goto err;
1211                 }
1212         }
1213
1214         ret =   bch2_fs_log_msg(c, "starting journal at entry %llu, replaying %llu-%llu",
1215                                 journal_seq, last_seq, blacklist_seq - 1) ?:
1216                 bch2_fs_journal_start(&c->journal, journal_seq);
1217         if (ret)
1218                 goto err;
1219
1220         /*
1221          * Skip past versions that might have possibly been used (as nonces),
1222          * but hadn't had their pointers written:
1223          */
1224         if (c->sb.encryption_type && !c->sb.clean)
1225                 atomic64_add(1 << 16, &c->key_version);
1226
1227         ret = read_btree_roots(c);
1228         if (ret)
1229                 goto err;
1230
1231         bch_verbose(c, "starting alloc read");
1232         err = "error reading allocation information";
1233
1234         down_read(&c->gc_lock);
1235         ret = c->sb.version < bcachefs_metadata_version_bucket_gens
1236                 ? bch2_alloc_read(c)
1237                 : bch2_bucket_gens_read(c);
1238         up_read(&c->gc_lock);
1239
1240         if (ret)
1241                 goto err;
1242         bch_verbose(c, "alloc read done");
1243
1244         bch_verbose(c, "starting stripes_read");
1245         err = "error reading stripes";
1246         ret = bch2_stripes_read(c);
1247         if (ret)
1248                 goto err;
1249         bch_verbose(c, "stripes_read done");
1250
1251         bch2_stripes_heap_start(c);
1252
1253         if (c->opts.fsck) {
1254                 bool metadata_only = c->opts.norecovery;
1255
1256                 bch_info(c, "checking allocations");
1257                 err = "error checking allocations";
1258                 ret = bch2_gc(c, true, metadata_only);
1259                 if (ret)
1260                         goto err;
1261                 bch_verbose(c, "done checking allocations");
1262
1263                 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1264
1265                 if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1266                         err = "error creating root snapshot node";
1267                         ret = bch2_fs_initialize_subvolumes(c);
1268                         if (ret)
1269                                 goto err;
1270                 }
1271
1272                 bch_verbose(c, "reading snapshots table");
1273                 err = "error reading snapshots table";
1274                 ret = bch2_fs_snapshots_start(c);
1275                 if (ret)
1276                         goto err;
1277                 bch_verbose(c, "reading snapshots done");
1278
1279                 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1280
1281                 bch_info(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1282                 err = "journal replay failed";
1283                 ret = bch2_journal_replay(c, last_seq, blacklist_seq - 1);
1284                 if (ret)
1285                         goto err;
1286                 if (c->opts.verbose || !c->sb.clean)
1287                         bch_info(c, "journal replay done");
1288
1289                 bch_info(c, "checking need_discard and freespace btrees");
1290                 err = "error checking need_discard and freespace btrees";
1291                 ret = bch2_check_alloc_info(c);
1292                 if (ret)
1293                         goto err;
1294                 bch_verbose(c, "done checking need_discard and freespace btrees");
1295
1296                 set_bit(BCH_FS_CHECK_ALLOC_DONE, &c->flags);
1297
1298                 bch_info(c, "checking lrus");
1299                 err = "error checking lrus";
1300                 ret = bch2_check_lrus(c);
1301                 if (ret)
1302                         goto err;
1303                 bch_verbose(c, "done checking lrus");
1304                 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1305
1306                 bch_info(c, "checking backpointers to alloc keys");
1307                 err = "error checking backpointers to alloc keys";
1308                 ret = bch2_check_btree_backpointers(c);
1309                 if (ret)
1310                         goto err;
1311                 bch_verbose(c, "done checking backpointers to alloc keys");
1312
1313                 bch_info(c, "checking backpointers to extents");
1314                 err = "error checking backpointers to extents";
1315                 ret = bch2_check_backpointers_to_extents(c);
1316                 if (ret)
1317                         goto err;
1318                 bch_verbose(c, "done checking backpointers to extents");
1319
1320                 bch_info(c, "checking extents to backpointers");
1321                 err = "error checking extents to backpointers";
1322                 ret = bch2_check_extents_to_backpointers(c);
1323                 if (ret)
1324                         goto err;
1325                 bch_verbose(c, "done checking extents to backpointers");
1326                 set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1327
1328                 bch_info(c, "checking alloc to lru refs");
1329                 err = "error checking alloc to lru refs";
1330                 ret = bch2_check_alloc_to_lru_refs(c);
1331                 if (ret)
1332                         goto err;
1333                 bch_verbose(c, "done checking alloc to lru refs");
1334                 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1335         } else {
1336                 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1337                 set_bit(BCH_FS_CHECK_ALLOC_DONE, &c->flags);
1338                 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1339                 set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1340                 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1341                 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1342
1343                 if (c->opts.norecovery)
1344                         goto out;
1345
1346                 if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1347                         err = "error creating root snapshot node";
1348                         ret = bch2_fs_initialize_subvolumes(c);
1349                         if (ret)
1350                                 goto err;
1351                 }
1352
1353                 bch_verbose(c, "reading snapshots table");
1354                 err = "error reading snapshots table";
1355                 ret = bch2_fs_snapshots_start(c);
1356                 if (ret)
1357                         goto err;
1358                 bch_verbose(c, "reading snapshots done");
1359
1360                 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1361
1362                 bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1363                 err = "journal replay failed";
1364                 ret = bch2_journal_replay(c, last_seq, blacklist_seq - 1);
1365                 if (ret)
1366                         goto err;
1367                 if (c->opts.verbose || !c->sb.clean)
1368                         bch_info(c, "journal replay done");
1369         }
1370
1371         err = "error initializing freespace";
1372         ret = bch2_fs_freespace_init(c);
1373         if (ret)
1374                 goto err;
1375
1376         if (c->sb.version < bcachefs_metadata_version_bucket_gens &&
1377             c->opts.version_upgrade) {
1378                 bch_info(c, "initializing bucket_gens");
1379                 err = "error initializing bucket gens";
1380                 ret = bch2_bucket_gens_init(c);
1381                 if (ret)
1382                         goto err;
1383                 bch_verbose(c, "bucket_gens init done");
1384         }
1385
1386         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1387                 /* set bi_subvol on root inode */
1388                 err = "error upgrade root inode for subvolumes";
1389                 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1390                                     bch2_fs_upgrade_for_subvolumes(&trans));
1391                 if (ret)
1392                         goto err;
1393         }
1394
1395         if (c->opts.fsck) {
1396                 bch_info(c, "starting fsck");
1397                 err = "error in fsck";
1398                 ret = bch2_fsck_full(c);
1399                 if (ret)
1400                         goto err;
1401                 bch_verbose(c, "fsck done");
1402         } else if (!c->sb.clean) {
1403                 bch_verbose(c, "checking for deleted inodes");
1404                 err = "error in recovery";
1405                 ret = bch2_fsck_walk_inodes_only(c);
1406                 if (ret)
1407                         goto err;
1408                 bch_verbose(c, "check inodes done");
1409         }
1410
1411         if (enabled_qtypes(c)) {
1412                 bch_verbose(c, "reading quotas");
1413                 ret = bch2_fs_quota_read(c);
1414                 if (ret)
1415                         goto err;
1416                 bch_verbose(c, "quotas done");
1417         }
1418
1419         mutex_lock(&c->sb_lock);
1420         if (c->opts.version_upgrade) {
1421                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1422                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1423                 write_sb = true;
1424         }
1425
1426         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1427                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1428                 write_sb = true;
1429         }
1430
1431         if (c->opts.fsck &&
1432             !test_bit(BCH_FS_ERROR, &c->flags) &&
1433             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1434                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1435                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1436                 write_sb = true;
1437         }
1438
1439         if (write_sb)
1440                 bch2_write_super(c);
1441         mutex_unlock(&c->sb_lock);
1442
1443         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1444             !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1445             le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) {
1446                 struct bch_move_stats stats;
1447
1448                 bch2_move_stats_init(&stats, "recovery");
1449
1450                 bch_info(c, "scanning for old btree nodes");
1451                 ret = bch2_fs_read_write(c);
1452                 if (ret)
1453                         goto err;
1454
1455                 ret = bch2_scan_old_btree_nodes(c, &stats);
1456                 if (ret)
1457                         goto err;
1458                 bch_info(c, "scanning for old btree nodes done");
1459         }
1460
1461         if (c->journal_seq_blacklist_table &&
1462             c->journal_seq_blacklist_table->nr > 128)
1463                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1464
1465         ret = 0;
1466 out:
1467         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1468         bch2_flush_fsck_errs(c);
1469
1470         if (!c->opts.keep_journal &&
1471             test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)) {
1472                 bch2_journal_keys_free(&c->journal_keys);
1473                 bch2_journal_entries_free(c);
1474         }
1475         kfree(clean);
1476
1477         if (!ret && test_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags)) {
1478                 bch2_fs_read_write_early(c);
1479                 bch2_delete_dead_snapshots_async(c);
1480         }
1481
1482         if (ret)
1483                 bch_err(c, "Error in recovery: %s (%s)", err, bch2_err_str(ret));
1484         else
1485                 bch_verbose(c, "ret %s", bch2_err_str(ret));
1486         return ret;
1487 err:
1488 fsck_err:
1489         bch2_fs_emergency_read_only(c);
1490         goto out;
1491 }
1492
1493 int bch2_fs_initialize(struct bch_fs *c)
1494 {
1495         struct bch_inode_unpacked root_inode, lostfound_inode;
1496         struct bkey_inode_buf packed_inode;
1497         struct qstr lostfound = QSTR("lost+found");
1498         const char *err = "cannot allocate memory";
1499         struct bch_dev *ca;
1500         unsigned i;
1501         int ret;
1502
1503         bch_notice(c, "initializing new filesystem");
1504
1505         mutex_lock(&c->sb_lock);
1506         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1507         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1508
1509         if (c->sb.version < bcachefs_metadata_version_inode_v3)
1510                 c->opts.version_upgrade = true;
1511
1512         if (c->opts.version_upgrade) {
1513                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1514                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1515                 bch2_write_super(c);
1516         }
1517         mutex_unlock(&c->sb_lock);
1518
1519         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1520         set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1521         set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1522         set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1523         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1524         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1525
1526         for (i = 0; i < BTREE_ID_NR; i++)
1527                 bch2_btree_root_alloc(c, i);
1528
1529         for_each_online_member(ca, c, i)
1530                 bch2_dev_usage_init(ca);
1531
1532         err = "unable to allocate journal buckets";
1533         for_each_online_member(ca, c, i) {
1534                 ret = bch2_dev_journal_alloc(ca);
1535                 if (ret) {
1536                         percpu_ref_put(&ca->io_ref);
1537                         goto err;
1538                 }
1539         }
1540
1541         /*
1542          * journal_res_get() will crash if called before this has
1543          * set up the journal.pin FIFO and journal.cur pointer:
1544          */
1545         bch2_fs_journal_start(&c->journal, 1);
1546         bch2_journal_set_replay_done(&c->journal);
1547
1548         err = "error going read-write";
1549         ret = bch2_fs_read_write_early(c);
1550         if (ret)
1551                 goto err;
1552
1553         /*
1554          * Write out the superblock and journal buckets, now that we can do
1555          * btree updates
1556          */
1557         bch_verbose(c, "marking superblocks");
1558         err = "error marking superblock and journal";
1559         for_each_member_device(ca, c, i) {
1560                 ret = bch2_trans_mark_dev_sb(c, ca);
1561                 if (ret) {
1562                         percpu_ref_put(&ca->ref);
1563                         goto err;
1564                 }
1565
1566                 ca->new_fs_bucket_idx = 0;
1567         }
1568
1569         bch_verbose(c, "initializing freespace");
1570         err = "error initializing freespace";
1571         ret = bch2_fs_freespace_init(c);
1572         if (ret)
1573                 goto err;
1574
1575         err = "error creating root snapshot node";
1576         ret = bch2_fs_initialize_subvolumes(c);
1577         if (ret)
1578                 goto err;
1579
1580         bch_verbose(c, "reading snapshots table");
1581         err = "error reading snapshots table";
1582         ret = bch2_fs_snapshots_start(c);
1583         if (ret)
1584                 goto err;
1585         bch_verbose(c, "reading snapshots done");
1586
1587         bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL);
1588         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1589         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1590         bch2_inode_pack(&packed_inode, &root_inode);
1591         packed_inode.inode.k.p.snapshot = U32_MAX;
1592
1593         err = "error creating root directory";
1594         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1595                                 &packed_inode.inode.k_i,
1596                                 NULL, NULL, 0);
1597         if (ret)
1598                 goto err;
1599
1600         bch2_inode_init_early(c, &lostfound_inode);
1601
1602         err = "error creating lost+found";
1603         ret = bch2_trans_do(c, NULL, NULL, 0,
1604                 bch2_create_trans(&trans,
1605                                   BCACHEFS_ROOT_SUBVOL_INUM,
1606                                   &root_inode, &lostfound_inode,
1607                                   &lostfound,
1608                                   0, 0, S_IFDIR|0700, 0,
1609                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1610         if (ret) {
1611                 bch_err(c, "error creating lost+found");
1612                 goto err;
1613         }
1614
1615         if (enabled_qtypes(c)) {
1616                 ret = bch2_fs_quota_read(c);
1617                 if (ret)
1618                         goto err;
1619         }
1620
1621         err = "error writing first journal entry";
1622         ret = bch2_journal_flush(&c->journal);
1623         if (ret)
1624                 goto err;
1625
1626         mutex_lock(&c->sb_lock);
1627         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1628         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1629
1630         bch2_write_super(c);
1631         mutex_unlock(&c->sb_lock);
1632
1633         return 0;
1634 err:
1635         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1636         return ret;
1637 }