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)
 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         for (i = 0; i < keys->nr; i++) {
 614                 k = keys_sorted[i];
 615
 616                 cond_resched();
 617
 618                 replay_now_at(j, k->journal_seq);
 619
 620                 ret = bch2_trans_do(c, NULL, NULL,
 621                                     BTREE_INSERT_LAZY_RW|
 622                                     BTREE_INSERT_NOFAIL|
 623                                     (!k->allocated
 624                                      ? BTREE_INSERT_JOURNAL_REPLAY|JOURNAL_WATERMARK_reserved
 625                                      : 0),
 626                              bch2_journal_replay_key(&trans, k));
 627                 if (ret) {
 628                         bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
 629                                 ret, bch2_btree_ids[k->btree_id], k->level);
 630                         goto err;
 631                 }
 632         }
 633
 634         replay_now_at(j, j->replay_journal_seq_end);
 635         j->replay_journal_seq = 0;
 636
 637         bch2_journal_set_replay_done(j);
 638         bch2_journal_flush_all_pins(j);
 639         ret = bch2_journal_error(j);
 640
 641         if (keys->nr && !ret)
 642                 bch2_journal_log_msg(&c->journal, "journal replay finished");
 643 err:
 644         kvfree(keys_sorted);
 645         return ret;
 646 }
 647
 648 /* journal replay early: */
 649
 650 static int journal_replay_entry_early(struct bch_fs *c,
 651                                       struct jset_entry *entry)
 652 {
 653         int ret = 0;
 654
 655         switch (entry->type) {
 656         case BCH_JSET_ENTRY_btree_root: {
 657                 struct btree_root *r;
 658
 659                 if (entry->btree_id >= BTREE_ID_NR) {
 660                         bch_err(c, "filesystem has unknown btree type %u",
 661                                 entry->btree_id);
 662                         return -EINVAL;
 663                 }
 664
 665                 r = &c->btree_roots[entry->btree_id];
 666
 667                 if (entry->u64s) {
 668                         r->level = entry->level;
 669                         bkey_copy(&r->key, &entry->start[0]);
 670                         r->error = 0;
 671                 } else {
 672                         r->error = -EIO;
 673                 }
 674                 r->alive = true;
 675                 break;
 676         }
 677         case BCH_JSET_ENTRY_usage: {
 678                 struct jset_entry_usage *u =
 679                         container_of(entry, struct jset_entry_usage, entry);
 680
 681                 switch (entry->btree_id) {
 682                 case BCH_FS_USAGE_reserved:
 683                         if (entry->level < BCH_REPLICAS_MAX)
 684                                 c->usage_base->persistent_reserved[entry->level] =
 685                                         le64_to_cpu(u->v);
 686                         break;
 687                 case BCH_FS_USAGE_inodes:
 688                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 689                         break;
 690                 case BCH_FS_USAGE_key_version:
 691                         atomic64_set(&c->key_version,
 692                                      le64_to_cpu(u->v));
 693                         break;
 694                 }
 695
 696                 break;
 697         }
 698         case BCH_JSET_ENTRY_data_usage: {
 699                 struct jset_entry_data_usage *u =
 700                         container_of(entry, struct jset_entry_data_usage, entry);
 701
 702                 ret = bch2_replicas_set_usage(c, &u->r,
 703                                               le64_to_cpu(u->v));
 704                 break;
 705         }
 706         case BCH_JSET_ENTRY_dev_usage: {
 707                 struct jset_entry_dev_usage *u =
 708                         container_of(entry, struct jset_entry_dev_usage, entry);
 709                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 710                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 711
 712                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 713
 714                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 715                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 716                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 717                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 718                 }
 719
 720                 break;
 721         }
 722         case BCH_JSET_ENTRY_blacklist: {
 723                 struct jset_entry_blacklist *bl_entry =
 724                         container_of(entry, struct jset_entry_blacklist, entry);
 725
 726                 ret = bch2_journal_seq_blacklist_add(c,
 727                                 le64_to_cpu(bl_entry->seq),
 728                                 le64_to_cpu(bl_entry->seq) + 1);
 729                 break;
 730         }
 731         case BCH_JSET_ENTRY_blacklist_v2: {
 732                 struct jset_entry_blacklist_v2 *bl_entry =
 733                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 734
 735                 ret = bch2_journal_seq_blacklist_add(c,
 736                                 le64_to_cpu(bl_entry->start),
 737                                 le64_to_cpu(bl_entry->end) + 1);
 738                 break;
 739         }
 740         case BCH_JSET_ENTRY_clock: {
 741                 struct jset_entry_clock *clock =
 742                         container_of(entry, struct jset_entry_clock, entry);
 743
 744                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 745         }
 746         }
 747
 748         return ret;
 749 }
 750
 751 static int journal_replay_early(struct bch_fs *c,
 752                                 struct bch_sb_field_clean *clean)
 753 {
 754         struct jset_entry *entry;
 755         int ret;
 756
 757         if (clean) {
 758                 for (entry = clean->start;
 759                      entry != vstruct_end(&clean->field);
 760                      entry = vstruct_next(entry)) {
 761                         ret = journal_replay_entry_early(c, entry);
 762                         if (ret)
 763                                 return ret;
 764                 }
 765         } else {
 766                 struct genradix_iter iter;
 767                 struct journal_replay *i, **_i;
 768
 769                 genradix_for_each(&c->journal_entries, iter, _i) {
 770                         i = *_i;
 771
 772                         if (!i || i->ignore)
 773                                 continue;
 774
 775                         vstruct_for_each(&i->j, entry) {
 776                                 ret = journal_replay_entry_early(c, entry);
 777                                 if (ret)
 778                                         return ret;
 779                         }
 780                 }
 781         }
 782
 783         bch2_fs_usage_initialize(c);
 784
 785         return 0;
 786 }
 787
 788 /* sb clean section: */
 789
 790 static struct bkey_i *btree_root_find(struct bch_fs *c,
 791                                       struct bch_sb_field_clean *clean,
 792                                       struct jset *j,
 793                                       enum btree_id id, unsigned *level)
 794 {
 795         struct bkey_i *k;
 796         struct jset_entry *entry, *start, *end;
 797
 798         if (clean) {
 799                 start = clean->start;
 800                 end = vstruct_end(&clean->field);
 801         } else {
 802                 start = j->start;
 803                 end = vstruct_last(j);
 804         }
 805
 806         for (entry = start; entry < end; entry = vstruct_next(entry))
 807                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 808                     entry->btree_id == id)
 809                         goto found;
 810
 811         return NULL;
 812 found:
 813         if (!entry->u64s)
 814                 return ERR_PTR(-EINVAL);
 815
 816         k = entry->start;
 817         *level = entry->level;
 818         return k;
 819 }
 820
 821 static int verify_superblock_clean(struct bch_fs *c,
 822                                    struct bch_sb_field_clean **cleanp,
 823                                    struct jset *j)
 824 {
 825         unsigned i;
 826         struct bch_sb_field_clean *clean = *cleanp;
 827         struct printbuf buf1 = PRINTBUF;
 828         struct printbuf buf2 = PRINTBUF;
 829         int ret = 0;
 830
 831         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 832                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 833                         le64_to_cpu(clean->journal_seq),
 834                         le64_to_cpu(j->seq))) {
 835                 kfree(clean);
 836                 *cleanp = NULL;
 837                 return 0;
 838         }
 839
 840         for (i = 0; i < BTREE_ID_NR; i++) {
 841                 struct bkey_i *k1, *k2;
 842                 unsigned l1 = 0, l2 = 0;
 843
 844                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 845                 k2 = btree_root_find(c, NULL, j, i, &l2);
 846
 847                 if (!k1 && !k2)
 848                         continue;
 849
 850                 printbuf_reset(&buf1);
 851                 printbuf_reset(&buf2);
 852
 853                 if (k1)
 854                         bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(k1));
 855                 else
 856                         prt_printf(&buf1, "(none)");
 857
 858                 if (k2)
 859                         bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(k2));
 860                 else
 861                         prt_printf(&buf2, "(none)");
 862
 863                 mustfix_fsck_err_on(!k1 || !k2 ||
 864                                     IS_ERR(k1) ||
 865                                     IS_ERR(k2) ||
 866                                     k1->k.u64s != k2->k.u64s ||
 867                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 868                                     l1 != l2, c,
 869                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 870                         "sb:      l=%u %s\n"
 871                         "journal: l=%u %s\n", i,
 872                         l1, buf1.buf,
 873                         l2, buf2.buf);
 874         }
 875 fsck_err:
 876         printbuf_exit(&buf2);
 877         printbuf_exit(&buf1);
 878         return ret;
 879 }
 880
 881 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 882 {
 883         struct bch_sb_field_clean *clean, *sb_clean;
 884         int ret;
 885
 886         mutex_lock(&c->sb_lock);
 887         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 888
 889         if (fsck_err_on(!sb_clean, c,
 890                         "superblock marked clean but clean section not present")) {
 891                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 892                 c->sb.clean = false;
 893                 mutex_unlock(&c->sb_lock);
 894                 return NULL;
 895         }
 896
 897         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 898                         GFP_KERNEL);
 899         if (!clean) {
 900                 mutex_unlock(&c->sb_lock);
 901                 return ERR_PTR(-ENOMEM);
 902         }
 903
 904         ret = bch2_sb_clean_validate_late(c, clean, READ);
 905         if (ret) {
 906                 mutex_unlock(&c->sb_lock);
 907                 return ERR_PTR(ret);
 908         }
 909
 910         mutex_unlock(&c->sb_lock);
 911
 912         return clean;
 913 fsck_err:
 914         mutex_unlock(&c->sb_lock);
 915         return ERR_PTR(ret);
 916 }
 917
 918 static bool btree_id_is_alloc(enum btree_id id)
 919 {
 920         switch (id) {
 921         case BTREE_ID_alloc:
 922         case BTREE_ID_backpointers:
 923         case BTREE_ID_need_discard:
 924         case BTREE_ID_freespace:
 925         case BTREE_ID_bucket_gens:
 926                 return true;
 927         default:
 928                 return false;
 929         }
 930 }
 931
 932 static int read_btree_roots(struct bch_fs *c)
 933 {
 934         unsigned i;
 935         int ret = 0;
 936
 937         for (i = 0; i < BTREE_ID_NR; i++) {
 938                 struct btree_root *r = &c->btree_roots[i];
 939
 940                 if (!r->alive)
 941                         continue;
 942
 943                 if (btree_id_is_alloc(i) &&
 944                     c->opts.reconstruct_alloc) {
 945                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 946                         continue;
 947                 }
 948
 949                 if (r->error) {
 950                         __fsck_err(c, btree_id_is_alloc(i)
 951                                    ? FSCK_CAN_IGNORE : 0,
 952                                    "invalid btree root %s",
 953                                    bch2_btree_ids[i]);
 954                         if (i == BTREE_ID_alloc)
 955                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 956                 }
 957
 958                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 959                 if (ret) {
 960                         __fsck_err(c,
 961                                    btree_id_is_alloc(i)
 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_backpointers) {
1091                         bch_info(c, "version prior to backpointers, upgrade and fsck required");
1092                         c->opts.version_upgrade = true;
1093                         c->opts.fsck            = true;
1094                         c->opts.fix_errors      = FSCK_OPT_YES;
1095                 } else if (c->sb.version < bcachefs_metadata_version_inode_v3) {
1096                         bch_info(c, "version prior to inode_v3, upgrade required");
1097                         c->opts.version_upgrade = true;
1098                 }
1099         }
1100
1101         if (c->opts.fsck && c->opts.norecovery) {
1102                 bch_err(c, "cannot select both norecovery and fsck");
1103                 ret = -EINVAL;
1104                 goto err;
1105         }
1106
1107         ret = bch2_blacklist_table_initialize(c);
1108         if (ret) {
1109                 bch_err(c, "error initializing blacklist table");
1110                 goto err;
1111         }
1112
1113         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1114                 struct genradix_iter iter;
1115                 struct journal_replay **i;
1116
1117                 bch_verbose(c, "starting journal read");
1118                 ret = bch2_journal_read(c, &blacklist_seq, &journal_seq);
1119                 if (ret)
1120                         goto err;
1121
1122                 /*
1123                  * note: cmd_list_journal needs the blacklist table fully up to date so
1124                  * it can asterisk ignored journal entries:
1125                  */
1126                 if (c->opts.read_journal_only)
1127                         goto out;
1128
1129                 genradix_for_each_reverse(&c->journal_entries, iter, i)
1130                         if (*i && !(*i)->ignore) {
1131                                 last_journal_entry = &(*i)->j;
1132                                 break;
1133                         }
1134
1135                 if (mustfix_fsck_err_on(c->sb.clean &&
1136                                         last_journal_entry &&
1137                                         !journal_entry_empty(last_journal_entry), c,
1138                                 "filesystem marked clean but journal not empty")) {
1139                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1140                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1141                         c->sb.clean = false;
1142                 }
1143
1144                 if (!last_journal_entry) {
1145                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
1146                         goto use_clean;
1147                 }
1148
1149                 ret = journal_keys_sort(c);
1150                 if (ret)
1151                         goto err;
1152
1153                 if (c->sb.clean && last_journal_entry) {
1154                         ret = verify_superblock_clean(c, &clean,
1155                                                       last_journal_entry);
1156                         if (ret)
1157                                 goto err;
1158                 }
1159         } else {
1160 use_clean:
1161                 if (!clean) {
1162                         bch_err(c, "no superblock clean section found");
1163                         ret = -BCH_ERR_fsck_repair_impossible;
1164                         goto err;
1165
1166                 }
1167                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1168         }
1169
1170         if (c->opts.reconstruct_alloc) {
1171                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1172                 drop_alloc_keys(&c->journal_keys);
1173         }
1174
1175         zero_out_btree_mem_ptr(&c->journal_keys);
1176
1177         ret = journal_replay_early(c, clean);
1178         if (ret)
1179                 goto err;
1180
1181         /*
1182          * After an unclean shutdown, skip then next few journal sequence
1183          * numbers as they may have been referenced by btree writes that
1184          * happened before their corresponding journal writes - those btree
1185          * writes need to be ignored, by skipping and blacklisting the next few
1186          * journal sequence numbers:
1187          */
1188         if (!c->sb.clean)
1189                 journal_seq += 8;
1190
1191         if (blacklist_seq != journal_seq) {
1192                 ret = bch2_journal_seq_blacklist_add(c,
1193                                         blacklist_seq, journal_seq);
1194                 if (ret) {
1195                         bch_err(c, "error creating new journal seq blacklist entry");
1196                         goto err;
1197                 }
1198         }
1199
1200         ret = bch2_fs_journal_start(&c->journal, journal_seq);
1201         if (ret)
1202                 goto err;
1203
1204         /*
1205          * Skip past versions that might have possibly been used (as nonces),
1206          * but hadn't had their pointers written:
1207          */
1208         if (c->sb.encryption_type && !c->sb.clean)
1209                 atomic64_add(1 << 16, &c->key_version);
1210
1211         ret = read_btree_roots(c);
1212         if (ret)
1213                 goto err;
1214
1215         bch_verbose(c, "starting alloc read");
1216         err = "error reading allocation information";
1217
1218         down_read(&c->gc_lock);
1219         ret = c->sb.version < bcachefs_metadata_version_bucket_gens
1220                 ? bch2_alloc_read(c)
1221                 : bch2_bucket_gens_read(c);
1222         up_read(&c->gc_lock);
1223
1224         if (ret)
1225                 goto err;
1226         bch_verbose(c, "alloc read done");
1227
1228         bch_verbose(c, "starting stripes_read");
1229         err = "error reading stripes";
1230         ret = bch2_stripes_read(c);
1231         if (ret)
1232                 goto err;
1233         bch_verbose(c, "stripes_read done");
1234
1235         bch2_stripes_heap_start(c);
1236
1237         if (c->opts.fsck) {
1238                 bool metadata_only = c->opts.norecovery;
1239
1240                 bch_info(c, "checking allocations");
1241                 err = "error checking allocations";
1242                 ret = bch2_gc(c, true, metadata_only);
1243                 if (ret)
1244                         goto err;
1245                 bch_verbose(c, "done checking allocations");
1246
1247                 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1248
1249                 bch_info(c, "checking need_discard and freespace btrees");
1250                 err = "error checking need_discard and freespace btrees";
1251                 ret = bch2_check_alloc_info(c);
1252                 if (ret)
1253                         goto err;
1254                 bch_verbose(c, "done checking need_discard and freespace btrees");
1255
1256                 if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1257                         err = "error creating root snapshot node";
1258                         ret = bch2_fs_initialize_subvolumes(c);
1259                         if (ret)
1260                                 goto err;
1261                 }
1262
1263                 bch_verbose(c, "reading snapshots table");
1264                 err = "error reading snapshots table";
1265                 ret = bch2_fs_snapshots_start(c);
1266                 if (ret)
1267                         goto err;
1268                 bch_verbose(c, "reading snapshots done");
1269
1270                 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1271
1272                 bch_info(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1273                 err = "journal replay failed";
1274                 ret = bch2_journal_replay(c);
1275                 if (ret)
1276                         goto err;
1277                 if (c->opts.verbose || !c->sb.clean)
1278                         bch_info(c, "journal replay done");
1279
1280                 bch_info(c, "checking lrus");
1281                 err = "error checking lrus";
1282                 ret = bch2_check_lrus(c);
1283                 if (ret)
1284                         goto err;
1285                 bch_verbose(c, "done checking lrus");
1286                 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1287
1288                 bch_info(c, "checking backpointers to alloc keys");
1289                 err = "error checking backpointers to alloc keys";
1290                 ret = bch2_check_btree_backpointers(c);
1291                 if (ret)
1292                         goto err;
1293                 bch_verbose(c, "done checking backpointers to alloc keys");
1294
1295                 bch_info(c, "checking backpointers to extents");
1296                 err = "error checking backpointers to extents";
1297                 ret = bch2_check_backpointers_to_extents(c);
1298                 if (ret)
1299                         goto err;
1300                 bch_verbose(c, "done checking backpointers to extents");
1301
1302                 bch_info(c, "checking extents to backpointers");
1303                 err = "error checking extents to backpointers";
1304                 ret = bch2_check_extents_to_backpointers(c);
1305                 if (ret)
1306                         goto err;
1307                 bch_verbose(c, "done checking extents to backpointers");
1308                 set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1309
1310                 bch_info(c, "checking alloc to lru refs");
1311                 err = "error checking alloc to lru refs";
1312                 ret = bch2_check_alloc_to_lru_refs(c);
1313                 if (ret)
1314                         goto err;
1315                 bch_verbose(c, "done checking alloc to lru refs");
1316                 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1317         } else {
1318                 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1319                 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1320                 set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1321                 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1322                 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1323
1324                 if (c->opts.norecovery)
1325                         goto out;
1326
1327                 if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1328                         err = "error creating root snapshot node";
1329                         ret = bch2_fs_initialize_subvolumes(c);
1330                         if (ret)
1331                                 goto err;
1332                 }
1333
1334                 bch_verbose(c, "reading snapshots table");
1335                 err = "error reading snapshots table";
1336                 ret = bch2_fs_snapshots_start(c);
1337                 if (ret)
1338                         goto err;
1339                 bch_verbose(c, "reading snapshots done");
1340
1341                 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1342
1343                 bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1344                 err = "journal replay failed";
1345                 ret = bch2_journal_replay(c);
1346                 if (ret)
1347                         goto err;
1348                 if (c->opts.verbose || !c->sb.clean)
1349                         bch_info(c, "journal replay done");
1350         }
1351
1352         err = "error initializing freespace";
1353         ret = bch2_fs_freespace_init(c);
1354         if (ret)
1355                 goto err;
1356
1357         if (c->sb.version < bcachefs_metadata_version_bucket_gens &&
1358             c->opts.version_upgrade) {
1359                 bch_info(c, "initializing bucket_gens");
1360                 err = "error initializing bucket gens";
1361                 ret = bch2_bucket_gens_init(c);
1362                 if (ret)
1363                         goto err;
1364                 bch_verbose(c, "bucket_gens init done");
1365         }
1366
1367         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1368                 /* set bi_subvol on root inode */
1369                 err = "error upgrade root inode for subvolumes";
1370                 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1371                                     bch2_fs_upgrade_for_subvolumes(&trans));
1372                 if (ret)
1373                         goto err;
1374         }
1375
1376         if (c->opts.fsck) {
1377                 bch_info(c, "starting fsck");
1378                 err = "error in fsck";
1379                 ret = bch2_fsck_full(c);
1380                 if (ret)
1381                         goto err;
1382                 bch_verbose(c, "fsck done");
1383         } else if (!c->sb.clean) {
1384                 bch_verbose(c, "checking for deleted inodes");
1385                 err = "error in recovery";
1386                 ret = bch2_fsck_walk_inodes_only(c);
1387                 if (ret)
1388                         goto err;
1389                 bch_verbose(c, "check inodes done");
1390         }
1391
1392         if (enabled_qtypes(c)) {
1393                 bch_verbose(c, "reading quotas");
1394                 ret = bch2_fs_quota_read(c);
1395                 if (ret)
1396                         goto err;
1397                 bch_verbose(c, "quotas done");
1398         }
1399
1400         mutex_lock(&c->sb_lock);
1401         if (c->opts.version_upgrade) {
1402                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1403                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1404                 write_sb = true;
1405         }
1406
1407         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1408                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1409                 write_sb = true;
1410         }
1411
1412         if (c->opts.fsck &&
1413             !test_bit(BCH_FS_ERROR, &c->flags) &&
1414             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1415                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1416                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1417                 write_sb = true;
1418         }
1419
1420         if (write_sb)
1421                 bch2_write_super(c);
1422         mutex_unlock(&c->sb_lock);
1423
1424         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1425             !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1426             le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) {
1427                 struct bch_move_stats stats;
1428
1429                 bch2_move_stats_init(&stats, "recovery");
1430
1431                 bch_info(c, "scanning for old btree nodes");
1432                 ret = bch2_fs_read_write(c);
1433                 if (ret)
1434                         goto err;
1435
1436                 ret = bch2_scan_old_btree_nodes(c, &stats);
1437                 if (ret)
1438                         goto err;
1439                 bch_info(c, "scanning for old btree nodes done");
1440         }
1441
1442         if (c->journal_seq_blacklist_table &&
1443             c->journal_seq_blacklist_table->nr > 128)
1444                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1445
1446         ret = 0;
1447 out:
1448         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1449         bch2_flush_fsck_errs(c);
1450
1451         if (!c->opts.keep_journal &&
1452             test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)) {
1453                 bch2_journal_keys_free(&c->journal_keys);
1454                 bch2_journal_entries_free(c);
1455         }
1456         kfree(clean);
1457
1458         if (!ret && test_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags)) {
1459                 bch2_fs_read_write_early(c);
1460                 bch2_delete_dead_snapshots_async(c);
1461         }
1462
1463         if (ret)
1464                 bch_err(c, "Error in recovery: %s (%s)", err, bch2_err_str(ret));
1465         else
1466                 bch_verbose(c, "ret %s", bch2_err_str(ret));
1467         return ret;
1468 err:
1469 fsck_err:
1470         bch2_fs_emergency_read_only(c);
1471         goto out;
1472 }
1473
1474 int bch2_fs_initialize(struct bch_fs *c)
1475 {
1476         struct bch_inode_unpacked root_inode, lostfound_inode;
1477         struct bkey_inode_buf packed_inode;
1478         struct qstr lostfound = QSTR("lost+found");
1479         const char *err = "cannot allocate memory";
1480         struct bch_dev *ca;
1481         unsigned i;
1482         int ret;
1483
1484         bch_notice(c, "initializing new filesystem");
1485
1486         mutex_lock(&c->sb_lock);
1487         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1488         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1489
1490         if (c->sb.version < bcachefs_metadata_version_inode_v3)
1491                 c->opts.version_upgrade = true;
1492
1493         if (c->opts.version_upgrade) {
1494                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1495                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1496                 bch2_write_super(c);
1497         }
1498         mutex_unlock(&c->sb_lock);
1499
1500         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1501         set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1502         set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1503         set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1504         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1505         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1506
1507         for (i = 0; i < BTREE_ID_NR; i++)
1508                 bch2_btree_root_alloc(c, i);
1509
1510         for_each_online_member(ca, c, i)
1511                 bch2_dev_usage_init(ca);
1512
1513         err = "unable to allocate journal buckets";
1514         for_each_online_member(ca, c, i) {
1515                 ret = bch2_dev_journal_alloc(ca);
1516                 if (ret) {
1517                         percpu_ref_put(&ca->io_ref);
1518                         goto err;
1519                 }
1520         }
1521
1522         /*
1523          * journal_res_get() will crash if called before this has
1524          * set up the journal.pin FIFO and journal.cur pointer:
1525          */
1526         bch2_fs_journal_start(&c->journal, 1);
1527         bch2_journal_set_replay_done(&c->journal);
1528
1529         err = "error going read-write";
1530         ret = bch2_fs_read_write_early(c);
1531         if (ret)
1532                 goto err;
1533
1534         /*
1535          * Write out the superblock and journal buckets, now that we can do
1536          * btree updates
1537          */
1538         bch_verbose(c, "marking superblocks");
1539         err = "error marking superblock and journal";
1540         for_each_member_device(ca, c, i) {
1541                 ret = bch2_trans_mark_dev_sb(c, ca);
1542                 if (ret) {
1543                         percpu_ref_put(&ca->ref);
1544                         goto err;
1545                 }
1546
1547                 ca->new_fs_bucket_idx = 0;
1548         }
1549
1550         bch_verbose(c, "initializing freespace");
1551         err = "error initializing freespace";
1552         ret = bch2_fs_freespace_init(c);
1553         if (ret)
1554                 goto err;
1555
1556         err = "error creating root snapshot node";
1557         ret = bch2_fs_initialize_subvolumes(c);
1558         if (ret)
1559                 goto err;
1560
1561         bch_verbose(c, "reading snapshots table");
1562         err = "error reading snapshots table";
1563         ret = bch2_fs_snapshots_start(c);
1564         if (ret)
1565                 goto err;
1566         bch_verbose(c, "reading snapshots done");
1567
1568         bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL);
1569         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1570         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1571         bch2_inode_pack(&packed_inode, &root_inode);
1572         packed_inode.inode.k.p.snapshot = U32_MAX;
1573
1574         err = "error creating root directory";
1575         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1576                                 &packed_inode.inode.k_i,
1577                                 NULL, NULL, 0);
1578         if (ret)
1579                 goto err;
1580
1581         bch2_inode_init_early(c, &lostfound_inode);
1582
1583         err = "error creating lost+found";
1584         ret = bch2_trans_do(c, NULL, NULL, 0,
1585                 bch2_create_trans(&trans,
1586                                   BCACHEFS_ROOT_SUBVOL_INUM,
1587                                   &root_inode, &lostfound_inode,
1588                                   &lostfound,
1589                                   0, 0, S_IFDIR|0700, 0,
1590                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1591         if (ret) {
1592                 bch_err(c, "error creating lost+found");
1593                 goto err;
1594         }
1595
1596         if (enabled_qtypes(c)) {
1597                 ret = bch2_fs_quota_read(c);
1598                 if (ret)
1599                         goto err;
1600         }
1601
1602         err = "error writing first journal entry";
1603         ret = bch2_journal_flush(&c->journal);
1604         if (ret)
1605                 goto err;
1606
1607         mutex_lock(&c->sb_lock);
1608         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1609         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1610
1611         bch2_write_super(c);
1612         mutex_unlock(&c->sb_lock);
1613
1614         return 0;
1615 err:
1616         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1617         return ret;
1618 }