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

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "backpointers.h"
   5 #include "bkey_buf.h"
   6 #include "alloc_background.h"
   7 #include "btree_gc.h"
   8 #include "btree_update.h"
   9 #include "btree_update_interior.h"
  10 #include "btree_io.h"
  11 #include "buckets.h"
  12 #include "dirent.h"
  13 #include "ec.h"
  14 #include "errcode.h"
  15 #include "error.h"
  16 #include "fs-common.h"
  17 #include "fsck.h"
  18 #include "journal_io.h"
  19 #include "journal_reclaim.h"
  20 #include "journal_seq_blacklist.h"
  21 #include "lru.h"
  22 #include "move.h"
  23 #include "quota.h"
  24 #include "recovery.h"
  25 #include "replicas.h"
  26 #include "subvolume.h"
  27 #include "super-io.h"
  28
  29 #include <linux/sort.h>
  30 #include <linux/stat.h>
  31
  32 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  33
  34 /* for -o reconstruct_alloc: */
  35 static void drop_alloc_keys(struct journal_keys *keys)
  36 {
  37         size_t src, dst;
  38
  39         for (src = 0, dst = 0; src < keys->nr; src++)
  40                 if (keys->d[src].btree_id != BTREE_ID_alloc)
  41                         keys->d[dst++] = keys->d[src];
  42
  43         keys->nr = dst;
  44 }
  45
  46 /*
  47  * Btree node pointers have a field to stack a pointer to the in memory btree
  48  * node; we need to zero out this field when reading in btree nodes, or when
  49  * reading in keys from the journal:
  50  */
  51 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
  52 {
  53         struct journal_key *i;
  54
  55         for (i = keys->d; i < keys->d + keys->nr; i++)
  56                 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
  57                         bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
  58 }
  59
  60 /* iterate over keys read from the journal: */
  61
  62 static int __journal_key_cmp(enum btree_id      l_btree_id,
  63                              unsigned           l_level,
  64                              struct bpos        l_pos,
  65                              const struct journal_key *r)
  66 {
  67         return (cmp_int(l_btree_id,     r->btree_id) ?:
  68                 cmp_int(l_level,        r->level) ?:
  69                 bpos_cmp(l_pos, r->k->k.p));
  70 }
  71
  72 static int journal_key_cmp(const struct journal_key *l, const struct journal_key *r)
  73 {
  74         return __journal_key_cmp(l->btree_id, l->level, l->k->k.p, r);
  75 }
  76
  77 static inline size_t idx_to_pos(struct journal_keys *keys, size_t idx)
  78 {
  79         size_t gap_size = keys->size - keys->nr;
  80
  81         if (idx >= keys->gap)
  82                 idx += gap_size;
  83         return idx;
  84 }
  85
  86 static inline struct journal_key *idx_to_key(struct journal_keys *keys, size_t idx)
  87 {
  88         return keys->d + idx_to_pos(keys, idx);
  89 }
  90
  91 static size_t __bch2_journal_key_search(struct journal_keys *keys,
  92                                         enum btree_id id, unsigned level,
  93                                         struct bpos pos)
  94 {
  95         size_t l = 0, r = keys->nr, m;
  96
  97         while (l < r) {
  98                 m = l + ((r - l) >> 1);
  99                 if (__journal_key_cmp(id, level, pos, idx_to_key(keys, m)) > 0)
 100                         l = m + 1;
 101                 else
 102                         r = m;
 103         }
 104
 105         BUG_ON(l < keys->nr &&
 106                __journal_key_cmp(id, level, pos, idx_to_key(keys, l)) > 0);
 107
 108         BUG_ON(l &&
 109                __journal_key_cmp(id, level, pos, idx_to_key(keys, l - 1)) <= 0);
 110
 111         return l;
 112 }
 113
 114 static size_t bch2_journal_key_search(struct journal_keys *keys,
 115                                       enum btree_id id, unsigned level,
 116                                       struct bpos pos)
 117 {
 118         return idx_to_pos(keys, __bch2_journal_key_search(keys, id, level, pos));
 119 }
 120
 121 struct bkey_i *bch2_journal_keys_peek_upto(struct bch_fs *c, enum btree_id btree_id,
 122                                            unsigned level, struct bpos pos,
 123                                            struct bpos end_pos, size_t *idx)
 124 {
 125         struct journal_keys *keys = &c->journal_keys;
 126         unsigned iters = 0;
 127         struct journal_key *k;
 128 search:
 129         if (!*idx)
 130                 *idx = __bch2_journal_key_search(keys, btree_id, level, pos);
 131
 132         while ((k = *idx < keys->nr ? idx_to_key(keys, *idx) : NULL)) {
 133                 if (__journal_key_cmp(btree_id, level, end_pos, k) < 0)
 134                         return NULL;
 135
 136                 if (__journal_key_cmp(btree_id, level, pos, k) <= 0 &&
 137                     !k->overwritten)
 138                         return k->k;
 139
 140                 (*idx)++;
 141                 iters++;
 142                 if (iters == 10) {
 143                         *idx = 0;
 144                         goto search;
 145                 }
 146         }
 147
 148         return NULL;
 149 }
 150
 151 struct bkey_i *bch2_journal_keys_peek_slot(struct bch_fs *c, enum btree_id btree_id,
 152                                            unsigned level, struct bpos pos)
 153 {
 154         size_t idx = 0;
 155
 156         return bch2_journal_keys_peek_upto(c, btree_id, level, pos, pos, &idx);
 157 }
 158
 159 static void journal_iters_fix(struct bch_fs *c)
 160 {
 161         struct journal_keys *keys = &c->journal_keys;
 162         /* The key we just inserted is immediately before the gap: */
 163         size_t gap_end = keys->gap + (keys->size - keys->nr);
 164         struct btree_and_journal_iter *iter;
 165
 166         /*
 167          * If an iterator points one after the key we just inserted, decrement
 168          * the iterator so it points at the key we just inserted - if the
 169          * decrement was unnecessary, bch2_btree_and_journal_iter_peek() will
 170          * handle that:
 171          */
 172         list_for_each_entry(iter, &c->journal_iters, journal.list)
 173                 if (iter->journal.idx == gap_end)
 174                         iter->journal.idx = keys->gap - 1;
 175 }
 176
 177 static void journal_iters_move_gap(struct bch_fs *c, size_t old_gap, size_t new_gap)
 178 {
 179         struct journal_keys *keys = &c->journal_keys;
 180         struct journal_iter *iter;
 181         size_t gap_size = keys->size - keys->nr;
 182
 183         list_for_each_entry(iter, &c->journal_iters, list) {
 184                 if (iter->idx > old_gap)
 185                         iter->idx -= gap_size;
 186                 if (iter->idx >= new_gap)
 187                         iter->idx += gap_size;
 188         }
 189 }
 190
 191 int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
 192                                  unsigned level, struct bkey_i *k)
 193 {
 194         struct journal_key n = {
 195                 .btree_id       = id,
 196                 .level          = level,
 197                 .k              = k,
 198                 .allocated      = true,
 199                 /*
 200                  * Ensure these keys are done last by journal replay, to unblock
 201                  * journal reclaim:
 202                  */
 203                 .journal_seq    = U32_MAX,
 204         };
 205         struct journal_keys *keys = &c->journal_keys;
 206         size_t idx = bch2_journal_key_search(keys, id, level, k->k.p);
 207
 208         BUG_ON(test_bit(BCH_FS_RW, &c->flags));
 209
 210         if (idx < keys->size &&
 211             journal_key_cmp(&n, &keys->d[idx]) == 0) {
 212                 if (keys->d[idx].allocated)
 213                         kfree(keys->d[idx].k);
 214                 keys->d[idx] = n;
 215                 return 0;
 216         }
 217
 218         if (idx > keys->gap)
 219                 idx -= keys->size - keys->nr;
 220
 221         if (keys->nr == keys->size) {
 222                 struct journal_keys new_keys = {
 223                         .nr                     = keys->nr,
 224                         .size                   = max_t(size_t, keys->size, 8) * 2,
 225                 };
 226
 227                 new_keys.d = kvmalloc_array(new_keys.size, sizeof(new_keys.d[0]), GFP_KERNEL);
 228                 if (!new_keys.d) {
 229                         bch_err(c, "%s: error allocating new key array (size %zu)",
 230                                 __func__, new_keys.size);
 231                         return -BCH_ERR_ENOMEM_journal_key_insert;
 232                 }
 233
 234                 /* Since @keys was full, there was no gap: */
 235                 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
 236                 kvfree(keys->d);
 237                 *keys = new_keys;
 238
 239                 /* And now the gap is at the end: */
 240                 keys->gap = keys->nr;
 241         }
 242
 243         journal_iters_move_gap(c, keys->gap, idx);
 244
 245         move_gap(keys->d, keys->nr, keys->size, keys->gap, idx);
 246         keys->gap = idx;
 247
 248         keys->nr++;
 249         keys->d[keys->gap++] = n;
 250
 251         journal_iters_fix(c);
 252
 253         return 0;
 254 }
 255
 256 /*
 257  * Can only be used from the recovery thread while we're still RO - can't be
 258  * used once we've got RW, as journal_keys is at that point used by multiple
 259  * threads:
 260  */
 261 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
 262                             unsigned level, struct bkey_i *k)
 263 {
 264         struct bkey_i *n;
 265         int ret;
 266
 267         n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
 268         if (!n)
 269                 return -BCH_ERR_ENOMEM_journal_key_insert;
 270
 271         bkey_copy(n, k);
 272         ret = bch2_journal_key_insert_take(c, id, level, n);
 273         if (ret)
 274                 kfree(n);
 275         return ret;
 276 }
 277
 278 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
 279                             unsigned level, struct bpos pos)
 280 {
 281         struct bkey_i whiteout;
 282
 283         bkey_init(&whiteout.k);
 284         whiteout.k.p = pos;
 285
 286         return bch2_journal_key_insert(c, id, level, &whiteout);
 287 }
 288
 289 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
 290                                   unsigned level, struct bpos pos)
 291 {
 292         struct journal_keys *keys = &c->journal_keys;
 293         size_t idx = bch2_journal_key_search(keys, btree, level, pos);
 294
 295         if (idx < keys->size &&
 296             keys->d[idx].btree_id       == btree &&
 297             keys->d[idx].level          == level &&
 298             bpos_eq(keys->d[idx].k->k.p, pos))
 299                 keys->d[idx].overwritten = true;
 300 }
 301
 302 static void bch2_journal_iter_advance(struct journal_iter *iter)
 303 {
 304         if (iter->idx < iter->keys->size) {
 305                 iter->idx++;
 306                 if (iter->idx == iter->keys->gap)
 307                         iter->idx += iter->keys->size - iter->keys->nr;
 308         }
 309 }
 310
 311 struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
 312 {
 313         struct journal_key *k = iter->keys->d + iter->idx;
 314
 315         while (k < iter->keys->d + iter->keys->size &&
 316                k->btree_id      == iter->btree_id &&
 317                k->level         == iter->level) {
 318                 if (!k->overwritten)
 319                         return bkey_i_to_s_c(k->k);
 320
 321                 bch2_journal_iter_advance(iter);
 322                 k = iter->keys->d + iter->idx;
 323         }
 324
 325         return bkey_s_c_null;
 326 }
 327
 328 static void bch2_journal_iter_exit(struct journal_iter *iter)
 329 {
 330         list_del(&iter->list);
 331 }
 332
 333 static void bch2_journal_iter_init(struct bch_fs *c,
 334                                    struct journal_iter *iter,
 335                                    enum btree_id id, unsigned level,
 336                                    struct bpos pos)
 337 {
 338         iter->btree_id  = id;
 339         iter->level     = level;
 340         iter->keys      = &c->journal_keys;
 341         iter->idx       = bch2_journal_key_search(&c->journal_keys, id, level, pos);
 342 }
 343
 344 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 345 {
 346         return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 347                                                 iter->b, &iter->unpacked);
 348 }
 349
 350 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 351 {
 352         bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 353 }
 354
 355 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 356 {
 357         if (bpos_eq(iter->pos, SPOS_MAX))
 358                 iter->at_end = true;
 359         else
 360                 iter->pos = bpos_successor(iter->pos);
 361 }
 362
 363 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 364 {
 365         struct bkey_s_c btree_k, journal_k, ret;
 366 again:
 367         if (iter->at_end)
 368                 return bkey_s_c_null;
 369
 370         while ((btree_k = bch2_journal_iter_peek_btree(iter)).k &&
 371                bpos_lt(btree_k.k->p, iter->pos))
 372                 bch2_journal_iter_advance_btree(iter);
 373
 374         while ((journal_k = bch2_journal_iter_peek(&iter->journal)).k &&
 375                bpos_lt(journal_k.k->p, iter->pos))
 376                 bch2_journal_iter_advance(&iter->journal);
 377
 378         ret = journal_k.k &&
 379                 (!btree_k.k || bpos_le(journal_k.k->p, btree_k.k->p))
 380                 ? journal_k
 381                 : btree_k;
 382
 383         if (ret.k && iter->b && bpos_gt(ret.k->p, iter->b->data->max_key))
 384                 ret = bkey_s_c_null;
 385
 386         if (ret.k) {
 387                 iter->pos = ret.k->p;
 388                 if (bkey_deleted(ret.k)) {
 389                         bch2_btree_and_journal_iter_advance(iter);
 390                         goto again;
 391                 }
 392         } else {
 393                 iter->pos = SPOS_MAX;
 394                 iter->at_end = true;
 395         }
 396
 397         return ret;
 398 }
 399
 400 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
 401 {
 402         bch2_journal_iter_exit(&iter->journal);
 403 }
 404
 405 void __bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 406                                                   struct bch_fs *c,
 407                                                   struct btree *b,
 408                                                   struct btree_node_iter node_iter,
 409                                                   struct bpos pos)
 410 {
 411         memset(iter, 0, sizeof(*iter));
 412
 413         iter->b = b;
 414         iter->node_iter = node_iter;
 415         bch2_journal_iter_init(c, &iter->journal, b->c.btree_id, b->c.level, pos);
 416         INIT_LIST_HEAD(&iter->journal.list);
 417         iter->pos = b->data->min_key;
 418         iter->at_end = false;
 419 }
 420
 421 /*
 422  * this version is used by btree_gc before filesystem has gone RW and
 423  * multithreaded, so uses the journal_iters list:
 424  */
 425 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 426                                                 struct bch_fs *c,
 427                                                 struct btree *b)
 428 {
 429         struct btree_node_iter node_iter;
 430
 431         bch2_btree_node_iter_init_from_start(&node_iter, b);
 432         __bch2_btree_and_journal_iter_init_node_iter(iter, c, b, node_iter, b->data->min_key);
 433         list_add(&iter->journal.list, &c->journal_iters);
 434 }
 435
 436 /* sort and dedup all keys in the journal: */
 437
 438 void bch2_journal_entries_free(struct bch_fs *c)
 439 {
 440         struct journal_replay **i;
 441         struct genradix_iter iter;
 442
 443         genradix_for_each(&c->journal_entries, iter, i)
 444                 if (*i)
 445                         kvpfree(*i, offsetof(struct journal_replay, j) +
 446                                 vstruct_bytes(&(*i)->j));
 447         genradix_free(&c->journal_entries);
 448 }
 449
 450 /*
 451  * When keys compare equal, oldest compares first:
 452  */
 453 static int journal_sort_key_cmp(const void *_l, const void *_r)
 454 {
 455         const struct journal_key *l = _l;
 456         const struct journal_key *r = _r;
 457
 458         return  journal_key_cmp(l, r) ?:
 459                 cmp_int(l->journal_seq, r->journal_seq) ?:
 460                 cmp_int(l->journal_offset, r->journal_offset);
 461 }
 462
 463 void bch2_journal_keys_free(struct journal_keys *keys)
 464 {
 465         struct journal_key *i;
 466
 467         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 468         keys->gap = keys->nr;
 469
 470         for (i = keys->d; i < keys->d + keys->nr; i++)
 471                 if (i->allocated)
 472                         kfree(i->k);
 473
 474         kvfree(keys->d);
 475         keys->d = NULL;
 476         keys->nr = keys->gap = keys->size = 0;
 477 }
 478
 479 static void __journal_keys_sort(struct journal_keys *keys)
 480 {
 481         struct journal_key *src, *dst;
 482
 483         sort(keys->d, keys->nr, sizeof(keys->d[0]), journal_sort_key_cmp, NULL);
 484
 485         src = dst = keys->d;
 486         while (src < keys->d + keys->nr) {
 487                 while (src + 1 < keys->d + keys->nr &&
 488                        src[0].btree_id  == src[1].btree_id &&
 489                        src[0].level     == src[1].level &&
 490                        bpos_eq(src[0].k->k.p, src[1].k->k.p))
 491                         src++;
 492
 493                 *dst++ = *src++;
 494         }
 495
 496         keys->nr = dst - keys->d;
 497 }
 498
 499 static int journal_keys_sort(struct bch_fs *c)
 500 {
 501         struct genradix_iter iter;
 502         struct journal_replay *i, **_i;
 503         struct jset_entry *entry;
 504         struct bkey_i *k;
 505         struct journal_keys *keys = &c->journal_keys;
 506         size_t nr_keys = 0, nr_read = 0;
 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, entry, &i->j)
 515                         nr_keys++;
 516         }
 517
 518         if (!nr_keys)
 519                 return 0;
 520
 521         keys->size = roundup_pow_of_two(nr_keys);
 522
 523         keys->d = kvmalloc_array(keys->size, sizeof(keys->d[0]), GFP_KERNEL);
 524         if (!keys->d) {
 525                 bch_err(c, "Failed to allocate buffer for sorted journal keys (%zu keys); trying slowpath",
 526                         nr_keys);
 527
 528                 do {
 529                         keys->size >>= 1;
 530                         keys->d = kvmalloc_array(keys->size, sizeof(keys->d[0]), GFP_KERNEL);
 531                 } while (!keys->d && keys->size > nr_keys / 8);
 532
 533                 if (!keys->d) {
 534                         bch_err(c, "Failed to allocate %zu size buffer for sorted journal keys; exiting",
 535                                 keys->size);
 536                         return -BCH_ERR_ENOMEM_journal_keys_sort;
 537                 }
 538         }
 539
 540         genradix_for_each(&c->journal_entries, iter, _i) {
 541                 i = *_i;
 542
 543                 if (!i || i->ignore)
 544                         continue;
 545
 546                 for_each_jset_key(k, entry, &i->j) {
 547                         if (keys->nr == keys->size) {
 548                                 __journal_keys_sort(keys);
 549
 550                                 if (keys->nr > keys->size * 7 / 8) {
 551                                         bch_err(c, "Too many journal keys for slowpath; have %zu compacted, buf size %zu, processed %zu/%zu",
 552                                                 keys->nr, keys->size, nr_read, nr_keys);
 553                                         return -BCH_ERR_ENOMEM_journal_keys_sort;
 554                                 }
 555                         }
 556
 557                         keys->d[keys->nr++] = (struct journal_key) {
 558                                 .btree_id       = entry->btree_id,
 559                                 .level          = entry->level,
 560                                 .k              = k,
 561                                 .journal_seq    = le64_to_cpu(i->j.seq),
 562                                 .journal_offset = k->_data - i->j._data,
 563                         };
 564
 565                         nr_read++;
 566                 }
 567         }
 568
 569         __journal_keys_sort(keys);
 570         keys->gap = keys->nr;
 571
 572         bch_verbose(c, "Journal keys: %zu read, %zu after sorting and compacting", nr_keys, keys->nr);
 573         return 0;
 574 }
 575
 576 /* journal replay: */
 577
 578 static void replay_now_at(struct journal *j, u64 seq)
 579 {
 580         BUG_ON(seq < j->replay_journal_seq);
 581
 582         seq = min(seq, j->replay_journal_seq_end);
 583
 584         while (j->replay_journal_seq < seq)
 585                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 586 }
 587
 588 static int bch2_journal_replay_key(struct btree_trans *trans,
 589                                    struct journal_key *k)
 590 {
 591         struct btree_iter iter;
 592         unsigned iter_flags =
 593                 BTREE_ITER_INTENT|
 594                 BTREE_ITER_NOT_EXTENTS;
 595         int ret;
 596
 597         if (!k->level && k->btree_id == BTREE_ID_alloc)
 598                 iter_flags |= BTREE_ITER_CACHED;
 599
 600         bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 601                                   BTREE_MAX_DEPTH, k->level,
 602                                   iter_flags);
 603         ret = bch2_btree_iter_traverse(&iter);
 604         if (ret)
 605                 goto out;
 606
 607         /* Must be checked with btree locked: */
 608         if (k->overwritten)
 609                 goto out;
 610
 611         ret = bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
 612 out:
 613         bch2_trans_iter_exit(trans, &iter);
 614         return ret;
 615 }
 616
 617 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 618 {
 619         const struct journal_key *l = *((const struct journal_key **)_l);
 620         const struct journal_key *r = *((const struct journal_key **)_r);
 621
 622         return cmp_int(l->journal_seq, r->journal_seq);
 623 }
 624
 625 static int bch2_journal_replay(struct bch_fs *c, u64 start_seq, u64 end_seq)
 626 {
 627         struct journal_keys *keys = &c->journal_keys;
 628         struct journal_key **keys_sorted, *k;
 629         struct journal *j = &c->journal;
 630         size_t i;
 631         int ret;
 632
 633         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 634         keys->gap = keys->nr;
 635
 636         keys_sorted = kvmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL);
 637         if (!keys_sorted)
 638                 return -BCH_ERR_ENOMEM_journal_replay;
 639
 640         for (i = 0; i < keys->nr; i++)
 641                 keys_sorted[i] = &keys->d[i];
 642
 643         sort(keys_sorted, keys->nr,
 644              sizeof(keys_sorted[0]),
 645              journal_sort_seq_cmp, NULL);
 646
 647         if (keys->nr) {
 648                 ret = bch2_journal_log_msg(c, "Starting journal replay (%zu keys in entries %llu-%llu)",
 649                                            keys->nr, start_seq, end_seq);
 650                 if (ret)
 651                         goto err;
 652         }
 653
 654         for (i = 0; i < keys->nr; i++) {
 655                 k = keys_sorted[i];
 656
 657                 cond_resched();
 658
 659                 replay_now_at(j, k->journal_seq);
 660
 661                 ret = bch2_trans_do(c, NULL, NULL,
 662                                     BTREE_INSERT_LAZY_RW|
 663                                     BTREE_INSERT_NOFAIL|
 664                                     (!k->allocated
 665                                      ? BTREE_INSERT_JOURNAL_REPLAY|JOURNAL_WATERMARK_reserved
 666                                      : 0),
 667                              bch2_journal_replay_key(&trans, k));
 668                 if (ret) {
 669                         bch_err(c, "journal replay: error while replaying key at btree %s level %u: %s",
 670                                 bch2_btree_ids[k->btree_id], k->level, bch2_err_str(ret));
 671                         goto err;
 672                 }
 673         }
 674
 675         replay_now_at(j, j->replay_journal_seq_end);
 676         j->replay_journal_seq = 0;
 677
 678         bch2_journal_set_replay_done(j);
 679         bch2_journal_flush_all_pins(j);
 680         ret = bch2_journal_error(j);
 681
 682         if (keys->nr && !ret)
 683                 bch2_journal_log_msg(c, "journal replay finished");
 684 err:
 685         kvfree(keys_sorted);
 686         return ret;
 687 }
 688
 689 /* journal replay early: */
 690
 691 static int journal_replay_entry_early(struct bch_fs *c,
 692                                       struct jset_entry *entry)
 693 {
 694         int ret = 0;
 695
 696         switch (entry->type) {
 697         case BCH_JSET_ENTRY_btree_root: {
 698                 struct btree_root *r;
 699
 700                 if (entry->btree_id >= BTREE_ID_NR) {
 701                         bch_err(c, "filesystem has unknown btree type %u",
 702                                 entry->btree_id);
 703                         return -EINVAL;
 704                 }
 705
 706                 r = &c->btree_roots[entry->btree_id];
 707
 708                 if (entry->u64s) {
 709                         r->level = entry->level;
 710                         bkey_copy(&r->key, &entry->start[0]);
 711                         r->error = 0;
 712                 } else {
 713                         r->error = -EIO;
 714                 }
 715                 r->alive = true;
 716                 break;
 717         }
 718         case BCH_JSET_ENTRY_usage: {
 719                 struct jset_entry_usage *u =
 720                         container_of(entry, struct jset_entry_usage, entry);
 721
 722                 switch (entry->btree_id) {
 723                 case BCH_FS_USAGE_reserved:
 724                         if (entry->level < BCH_REPLICAS_MAX)
 725                                 c->usage_base->persistent_reserved[entry->level] =
 726                                         le64_to_cpu(u->v);
 727                         break;
 728                 case BCH_FS_USAGE_inodes:
 729                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 730                         break;
 731                 case BCH_FS_USAGE_key_version:
 732                         atomic64_set(&c->key_version,
 733                                      le64_to_cpu(u->v));
 734                         break;
 735                 }
 736
 737                 break;
 738         }
 739         case BCH_JSET_ENTRY_data_usage: {
 740                 struct jset_entry_data_usage *u =
 741                         container_of(entry, struct jset_entry_data_usage, entry);
 742
 743                 ret = bch2_replicas_set_usage(c, &u->r,
 744                                               le64_to_cpu(u->v));
 745                 break;
 746         }
 747         case BCH_JSET_ENTRY_dev_usage: {
 748                 struct jset_entry_dev_usage *u =
 749                         container_of(entry, struct jset_entry_dev_usage, entry);
 750                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 751                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 752
 753                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 754
 755                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 756                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 757                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 758                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 759                 }
 760
 761                 break;
 762         }
 763         case BCH_JSET_ENTRY_blacklist: {
 764                 struct jset_entry_blacklist *bl_entry =
 765                         container_of(entry, struct jset_entry_blacklist, entry);
 766
 767                 ret = bch2_journal_seq_blacklist_add(c,
 768                                 le64_to_cpu(bl_entry->seq),
 769                                 le64_to_cpu(bl_entry->seq) + 1);
 770                 break;
 771         }
 772         case BCH_JSET_ENTRY_blacklist_v2: {
 773                 struct jset_entry_blacklist_v2 *bl_entry =
 774                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 775
 776                 ret = bch2_journal_seq_blacklist_add(c,
 777                                 le64_to_cpu(bl_entry->start),
 778                                 le64_to_cpu(bl_entry->end) + 1);
 779                 break;
 780         }
 781         case BCH_JSET_ENTRY_clock: {
 782                 struct jset_entry_clock *clock =
 783                         container_of(entry, struct jset_entry_clock, entry);
 784
 785                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 786         }
 787         }
 788
 789         return ret;
 790 }
 791
 792 static int journal_replay_early(struct bch_fs *c,
 793                                 struct bch_sb_field_clean *clean)
 794 {
 795         struct jset_entry *entry;
 796         int ret;
 797
 798         if (clean) {
 799                 for (entry = clean->start;
 800                      entry != vstruct_end(&clean->field);
 801                      entry = vstruct_next(entry)) {
 802                         ret = journal_replay_entry_early(c, entry);
 803                         if (ret)
 804                                 return ret;
 805                 }
 806         } else {
 807                 struct genradix_iter iter;
 808                 struct journal_replay *i, **_i;
 809
 810                 genradix_for_each(&c->journal_entries, iter, _i) {
 811                         i = *_i;
 812
 813                         if (!i || i->ignore)
 814                                 continue;
 815
 816                         vstruct_for_each(&i->j, entry) {
 817                                 ret = journal_replay_entry_early(c, entry);
 818                                 if (ret)
 819                                         return ret;
 820                         }
 821                 }
 822         }
 823
 824         bch2_fs_usage_initialize(c);
 825
 826         return 0;
 827 }
 828
 829 /* sb clean section: */
 830
 831 static struct bkey_i *btree_root_find(struct bch_fs *c,
 832                                       struct bch_sb_field_clean *clean,
 833                                       struct jset *j,
 834                                       enum btree_id id, unsigned *level)
 835 {
 836         struct bkey_i *k;
 837         struct jset_entry *entry, *start, *end;
 838
 839         if (clean) {
 840                 start = clean->start;
 841                 end = vstruct_end(&clean->field);
 842         } else {
 843                 start = j->start;
 844                 end = vstruct_last(j);
 845         }
 846
 847         for (entry = start; entry < end; entry = vstruct_next(entry))
 848                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 849                     entry->btree_id == id)
 850                         goto found;
 851
 852         return NULL;
 853 found:
 854         if (!entry->u64s)
 855                 return ERR_PTR(-EINVAL);
 856
 857         k = entry->start;
 858         *level = entry->level;
 859         return k;
 860 }
 861
 862 static int verify_superblock_clean(struct bch_fs *c,
 863                                    struct bch_sb_field_clean **cleanp,
 864                                    struct jset *j)
 865 {
 866         unsigned i;
 867         struct bch_sb_field_clean *clean = *cleanp;
 868         struct printbuf buf1 = PRINTBUF;
 869         struct printbuf buf2 = PRINTBUF;
 870         int ret = 0;
 871
 872         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 873                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 874                         le64_to_cpu(clean->journal_seq),
 875                         le64_to_cpu(j->seq))) {
 876                 kfree(clean);
 877                 *cleanp = NULL;
 878                 return 0;
 879         }
 880
 881         for (i = 0; i < BTREE_ID_NR; i++) {
 882                 struct bkey_i *k1, *k2;
 883                 unsigned l1 = 0, l2 = 0;
 884
 885                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 886                 k2 = btree_root_find(c, NULL, j, i, &l2);
 887
 888                 if (!k1 && !k2)
 889                         continue;
 890
 891                 printbuf_reset(&buf1);
 892                 printbuf_reset(&buf2);
 893
 894                 if (k1)
 895                         bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(k1));
 896                 else
 897                         prt_printf(&buf1, "(none)");
 898
 899                 if (k2)
 900                         bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(k2));
 901                 else
 902                         prt_printf(&buf2, "(none)");
 903
 904                 mustfix_fsck_err_on(!k1 || !k2 ||
 905                                     IS_ERR(k1) ||
 906                                     IS_ERR(k2) ||
 907                                     k1->k.u64s != k2->k.u64s ||
 908                                     memcmp(k1, k2, bkey_bytes(&k1->k)) ||
 909                                     l1 != l2, c,
 910                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 911                         "sb:      l=%u %s\n"
 912                         "journal: l=%u %s\n", i,
 913                         l1, buf1.buf,
 914                         l2, buf2.buf);
 915         }
 916 fsck_err:
 917         printbuf_exit(&buf2);
 918         printbuf_exit(&buf1);
 919         return ret;
 920 }
 921
 922 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 923 {
 924         struct bch_sb_field_clean *clean, *sb_clean;
 925         int ret;
 926
 927         mutex_lock(&c->sb_lock);
 928         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 929
 930         if (fsck_err_on(!sb_clean, c,
 931                         "superblock marked clean but clean section not present")) {
 932                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 933                 c->sb.clean = false;
 934                 mutex_unlock(&c->sb_lock);
 935                 return NULL;
 936         }
 937
 938         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 939                         GFP_KERNEL);
 940         if (!clean) {
 941                 mutex_unlock(&c->sb_lock);
 942                 return ERR_PTR(-BCH_ERR_ENOMEM_read_superblock_clean);
 943         }
 944
 945         ret = bch2_sb_clean_validate_late(c, clean, READ);
 946         if (ret) {
 947                 mutex_unlock(&c->sb_lock);
 948                 return ERR_PTR(ret);
 949         }
 950
 951         mutex_unlock(&c->sb_lock);
 952
 953         return clean;
 954 fsck_err:
 955         mutex_unlock(&c->sb_lock);
 956         return ERR_PTR(ret);
 957 }
 958
 959 static bool btree_id_is_alloc(enum btree_id id)
 960 {
 961         switch (id) {
 962         case BTREE_ID_alloc:
 963         case BTREE_ID_backpointers:
 964         case BTREE_ID_need_discard:
 965         case BTREE_ID_freespace:
 966         case BTREE_ID_bucket_gens:
 967                 return true;
 968         default:
 969                 return false;
 970         }
 971 }
 972
 973 static int read_btree_roots(struct bch_fs *c)
 974 {
 975         unsigned i;
 976         int ret = 0;
 977
 978         for (i = 0; i < BTREE_ID_NR; i++) {
 979                 struct btree_root *r = &c->btree_roots[i];
 980
 981                 if (!r->alive)
 982                         continue;
 983
 984                 if (btree_id_is_alloc(i) &&
 985                     c->opts.reconstruct_alloc) {
 986                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 987                         continue;
 988                 }
 989
 990                 if (r->error) {
 991                         __fsck_err(c, btree_id_is_alloc(i)
 992                                    ? FSCK_CAN_IGNORE : 0,
 993                                    "invalid btree root %s",
 994                                    bch2_btree_ids[i]);
 995                         if (i == BTREE_ID_alloc)
 996                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 997                 }
 998
 999                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
1000                 if (ret) {
1001                         __fsck_err(c,
1002                                    btree_id_is_alloc(i)
1003                                    ? FSCK_CAN_IGNORE : 0,
1004                                    "error reading btree root %s",
1005                                    bch2_btree_ids[i]);
1006                         if (btree_id_is_alloc(i))
1007                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1008                 }
1009         }
1010
1011         for (i = 0; i < BTREE_ID_NR; i++) {
1012                 struct btree_root *r = &c->btree_roots[i];
1013
1014                 if (!r->b) {
1015                         r->alive = false;
1016                         r->level = 0;
1017                         bch2_btree_root_alloc(c, i);
1018                 }
1019         }
1020 fsck_err:
1021         return ret;
1022 }
1023
1024 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
1025 {
1026         struct bkey_i_snapshot  root_snapshot;
1027         struct bkey_i_subvolume root_volume;
1028         int ret;
1029
1030         bkey_snapshot_init(&root_snapshot.k_i);
1031         root_snapshot.k.p.offset = U32_MAX;
1032         root_snapshot.v.flags   = 0;
1033         root_snapshot.v.parent  = 0;
1034         root_snapshot.v.subvol  = BCACHEFS_ROOT_SUBVOL;
1035         root_snapshot.v.pad     = 0;
1036         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
1037
1038         ret = bch2_btree_insert(c, BTREE_ID_snapshots,
1039                                 &root_snapshot.k_i,
1040                                 NULL, NULL, 0);
1041         if (ret)
1042                 return ret;
1043
1044         bkey_subvolume_init(&root_volume.k_i);
1045         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1046         root_volume.v.flags     = 0;
1047         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
1048         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
1049
1050         ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
1051                                 &root_volume.k_i,
1052                                 NULL, NULL, 0);
1053         if (ret)
1054                 return ret;
1055
1056         return 0;
1057 }
1058
1059 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
1060 {
1061         struct btree_iter iter;
1062         struct bkey_s_c k;
1063         struct bch_inode_unpacked inode;
1064         int ret;
1065
1066         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
1067                              SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
1068         k = bch2_btree_iter_peek_slot(&iter);
1069         ret = bkey_err(k);
1070         if (ret)
1071                 goto err;
1072
1073         if (!bkey_is_inode(k.k)) {
1074                 bch_err(trans->c, "root inode not found");
1075                 ret = -ENOENT;
1076                 goto err;
1077         }
1078
1079         ret = bch2_inode_unpack(k, &inode);
1080         BUG_ON(ret);
1081
1082         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1083
1084         ret = bch2_inode_write(trans, &iter, &inode);
1085 err:
1086         bch2_trans_iter_exit(trans, &iter);
1087         return ret;
1088 }
1089
1090 int bch2_fs_recovery(struct bch_fs *c)
1091 {
1092         const char *err = "cannot allocate memory";
1093         struct bch_sb_field_clean *clean = NULL;
1094         struct jset *last_journal_entry = NULL;
1095         u64 last_seq, blacklist_seq, journal_seq;
1096         bool write_sb = false;
1097         int ret = 0;
1098
1099         if (c->sb.clean)
1100                 clean = read_superblock_clean(c);
1101         ret = PTR_ERR_OR_ZERO(clean);
1102         if (ret)
1103                 goto err;
1104
1105         if (c->sb.clean)
1106                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
1107                          le64_to_cpu(clean->journal_seq));
1108         else
1109                 bch_info(c, "recovering from unclean shutdown");
1110
1111         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
1112                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
1113                 ret = -EINVAL;
1114                 goto err;
1115         }
1116
1117         if (!c->sb.clean &&
1118             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1119                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1120                 ret = -EINVAL;
1121                 goto err;
1122         }
1123
1124         if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1125                 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
1126                 ret = -EINVAL;
1127                 goto err;
1128         }
1129
1130         if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
1131                 bch_info(c, "alloc_v2 feature bit not set, fsck required");
1132                 c->opts.fsck = true;
1133                 c->opts.fix_errors = FSCK_OPT_YES;
1134         }
1135
1136         if (!c->opts.nochanges) {
1137                 if (c->sb.version < bcachefs_metadata_version_no_bps_in_alloc_keys) {
1138                         bch_info(c, "version prior to no_bps_in_alloc_keys, upgrade and fsck required");
1139                         c->opts.version_upgrade = true;
1140                         c->opts.fsck            = true;
1141                         c->opts.fix_errors      = FSCK_OPT_YES;
1142                 }
1143         }
1144
1145         if (c->opts.fsck && c->opts.norecovery) {
1146                 bch_err(c, "cannot select both norecovery and fsck");
1147                 ret = -EINVAL;
1148                 goto err;
1149         }
1150
1151         ret = bch2_blacklist_table_initialize(c);
1152         if (ret) {
1153                 bch_err(c, "error initializing blacklist table");
1154                 goto err;
1155         }
1156
1157         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1158                 struct genradix_iter iter;
1159                 struct journal_replay **i;
1160
1161                 bch_verbose(c, "starting journal read");
1162                 ret = bch2_journal_read(c, &last_seq, &blacklist_seq, &journal_seq);
1163                 if (ret)
1164                         goto err;
1165
1166                 /*
1167                  * note: cmd_list_journal needs the blacklist table fully up to date so
1168                  * it can asterisk ignored journal entries:
1169                  */
1170                 if (c->opts.read_journal_only)
1171                         goto out;
1172
1173                 genradix_for_each_reverse(&c->journal_entries, iter, i)
1174                         if (*i && !(*i)->ignore) {
1175                                 last_journal_entry = &(*i)->j;
1176                                 break;
1177                         }
1178
1179                 if (mustfix_fsck_err_on(c->sb.clean &&
1180                                         last_journal_entry &&
1181                                         !journal_entry_empty(last_journal_entry), c,
1182                                 "filesystem marked clean but journal not empty")) {
1183                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1184                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1185                         c->sb.clean = false;
1186                 }
1187
1188                 if (!last_journal_entry) {
1189                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
1190                         if (clean)
1191                                 goto use_clean;
1192
1193                         genradix_for_each_reverse(&c->journal_entries, iter, i)
1194                                 if (*i) {
1195                                         last_journal_entry = &(*i)->j;
1196                                         (*i)->ignore = false;
1197                                         break;
1198                                 }
1199                 }
1200
1201                 ret = journal_keys_sort(c);
1202                 if (ret)
1203                         goto err;
1204
1205                 if (c->sb.clean && last_journal_entry) {
1206                         ret = verify_superblock_clean(c, &clean,
1207                                                       last_journal_entry);
1208                         if (ret)
1209                                 goto err;
1210                 }
1211         } else {
1212 use_clean:
1213                 if (!clean) {
1214                         bch_err(c, "no superblock clean section found");
1215                         ret = -BCH_ERR_fsck_repair_impossible;
1216                         goto err;
1217
1218                 }
1219                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1220         }
1221
1222         if (c->opts.reconstruct_alloc) {
1223                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1224                 drop_alloc_keys(&c->journal_keys);
1225         }
1226
1227         zero_out_btree_mem_ptr(&c->journal_keys);
1228
1229         ret = journal_replay_early(c, clean);
1230         if (ret)
1231                 goto err;
1232
1233         /*
1234          * After an unclean shutdown, skip then next few journal sequence
1235          * numbers as they may have been referenced by btree writes that
1236          * happened before their corresponding journal writes - those btree
1237          * writes need to be ignored, by skipping and blacklisting the next few
1238          * journal sequence numbers:
1239          */
1240         if (!c->sb.clean)
1241                 journal_seq += 8;
1242
1243         if (blacklist_seq != journal_seq) {
1244                 ret =   bch2_journal_log_msg(c, "blacklisting entries %llu-%llu",
1245                                              blacklist_seq, journal_seq) ?:
1246                         bch2_journal_seq_blacklist_add(c,
1247                                         blacklist_seq, journal_seq);
1248                 if (ret) {
1249                         bch_err(c, "error creating new journal seq blacklist entry");
1250                         goto err;
1251                 }
1252         }
1253
1254         ret =   bch2_journal_log_msg(c, "starting journal at entry %llu, replaying %llu-%llu",
1255                                      journal_seq, last_seq, blacklist_seq - 1) ?:
1256                 bch2_fs_journal_start(&c->journal, journal_seq);
1257         if (ret)
1258                 goto err;
1259
1260         if (c->opts.reconstruct_alloc)
1261                 bch2_journal_log_msg(c, "dropping alloc info");
1262
1263         /*
1264          * Skip past versions that might have possibly been used (as nonces),
1265          * but hadn't had their pointers written:
1266          */
1267         if (c->sb.encryption_type && !c->sb.clean)
1268                 atomic64_add(1 << 16, &c->key_version);
1269
1270         ret = read_btree_roots(c);
1271         if (ret)
1272                 goto err;
1273
1274         bch_verbose(c, "starting alloc read");
1275         err = "error reading allocation information";
1276
1277         down_read(&c->gc_lock);
1278         ret = c->sb.version < bcachefs_metadata_version_bucket_gens
1279                 ? bch2_alloc_read(c)
1280                 : bch2_bucket_gens_read(c);
1281         up_read(&c->gc_lock);
1282
1283         if (ret)
1284                 goto err;
1285         bch_verbose(c, "alloc read done");
1286
1287         bch_verbose(c, "starting stripes_read");
1288         err = "error reading stripes";
1289         ret = bch2_stripes_read(c);
1290         if (ret)
1291                 goto err;
1292         bch_verbose(c, "stripes_read done");
1293
1294         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1295                 err = "error creating root snapshot node";
1296                 ret = bch2_fs_initialize_subvolumes(c);
1297                 if (ret)
1298                         goto err;
1299         }
1300
1301         bch_verbose(c, "reading snapshots table");
1302         err = "error reading snapshots table";
1303         ret = bch2_fs_snapshots_start(c);
1304         if (ret)
1305                 goto err;
1306         bch_verbose(c, "reading snapshots done");
1307
1308         if (c->opts.fsck) {
1309                 bool metadata_only = c->opts.norecovery;
1310
1311                 bch_info(c, "checking allocations");
1312                 err = "error checking allocations";
1313                 ret = bch2_gc(c, true, metadata_only);
1314                 if (ret)
1315                         goto err;
1316                 bch_verbose(c, "done checking allocations");
1317
1318                 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1319
1320                 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1321
1322                 bch_info(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1323                 err = "journal replay failed";
1324                 ret = bch2_journal_replay(c, last_seq, blacklist_seq - 1);
1325                 if (ret)
1326                         goto err;
1327                 if (c->opts.verbose || !c->sb.clean)
1328                         bch_info(c, "journal replay done");
1329
1330                 bch_info(c, "checking need_discard and freespace btrees");
1331                 err = "error checking need_discard and freespace btrees";
1332                 ret = bch2_check_alloc_info(c);
1333                 if (ret)
1334                         goto err;
1335                 bch_verbose(c, "done checking need_discard and freespace btrees");
1336
1337                 set_bit(BCH_FS_CHECK_ALLOC_DONE, &c->flags);
1338
1339                 bch_info(c, "checking lrus");
1340                 err = "error checking lrus";
1341                 ret = bch2_check_lrus(c);
1342                 if (ret)
1343                         goto err;
1344                 bch_verbose(c, "done checking lrus");
1345                 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1346
1347                 bch_info(c, "checking backpointers to alloc keys");
1348                 err = "error checking backpointers to alloc keys";
1349                 ret = bch2_check_btree_backpointers(c);
1350                 if (ret)
1351                         goto err;
1352                 bch_verbose(c, "done checking backpointers to alloc keys");
1353
1354                 bch_info(c, "checking backpointers to extents");
1355                 err = "error checking backpointers to extents";
1356                 ret = bch2_check_backpointers_to_extents(c);
1357                 if (ret)
1358                         goto err;
1359                 bch_verbose(c, "done checking backpointers to extents");
1360
1361                 bch_info(c, "checking extents to backpointers");
1362                 err = "error checking extents to backpointers";
1363                 ret = bch2_check_extents_to_backpointers(c);
1364                 if (ret)
1365                         goto err;
1366                 bch_verbose(c, "done checking extents to backpointers");
1367                 set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1368
1369                 bch_info(c, "checking alloc to lru refs");
1370                 err = "error checking alloc to lru refs";
1371                 ret = bch2_check_alloc_to_lru_refs(c);
1372                 if (ret)
1373                         goto err;
1374                 bch_verbose(c, "done checking alloc to lru refs");
1375                 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1376         } else {
1377                 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1378                 set_bit(BCH_FS_CHECK_ALLOC_DONE, &c->flags);
1379                 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1380                 set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1381                 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1382                 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1383
1384                 if (c->opts.norecovery)
1385                         goto out;
1386
1387                 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1388
1389                 bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1390                 err = "journal replay failed";
1391                 ret = bch2_journal_replay(c, last_seq, blacklist_seq - 1);
1392                 if (ret)
1393                         goto err;
1394                 if (c->opts.verbose || !c->sb.clean)
1395                         bch_info(c, "journal replay done");
1396         }
1397
1398         err = "error initializing freespace";
1399         ret = bch2_fs_freespace_init(c);
1400         if (ret)
1401                 goto err;
1402
1403         if (c->sb.version < bcachefs_metadata_version_bucket_gens &&
1404             c->opts.version_upgrade) {
1405                 bch_info(c, "initializing bucket_gens");
1406                 err = "error initializing bucket gens";
1407                 ret = bch2_bucket_gens_init(c);
1408                 if (ret)
1409                         goto err;
1410                 bch_verbose(c, "bucket_gens init done");
1411         }
1412
1413         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1414                 /* set bi_subvol on root inode */
1415                 err = "error upgrade root inode for subvolumes";
1416                 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1417                                     bch2_fs_upgrade_for_subvolumes(&trans));
1418                 if (ret)
1419                         goto err;
1420         }
1421
1422         if (c->opts.fsck) {
1423                 bch_info(c, "starting fsck");
1424                 err = "error in fsck";
1425                 ret = bch2_fsck_full(c);
1426                 if (ret)
1427                         goto err;
1428                 bch_verbose(c, "fsck done");
1429         } else if (!c->sb.clean) {
1430                 bch_verbose(c, "checking for deleted inodes");
1431                 err = "error in recovery";
1432                 ret = bch2_fsck_walk_inodes_only(c);
1433                 if (ret)
1434                         goto err;
1435                 bch_verbose(c, "check inodes done");
1436         }
1437
1438         if (enabled_qtypes(c)) {
1439                 bch_verbose(c, "reading quotas");
1440                 ret = bch2_fs_quota_read(c);
1441                 if (ret)
1442                         goto err;
1443                 bch_verbose(c, "quotas done");
1444         }
1445
1446         mutex_lock(&c->sb_lock);
1447         if (c->opts.version_upgrade) {
1448                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1449                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1450                 write_sb = true;
1451         }
1452
1453         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1454                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1455                 write_sb = true;
1456         }
1457
1458         if (c->opts.fsck &&
1459             !test_bit(BCH_FS_ERROR, &c->flags) &&
1460             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1461                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1462                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1463                 write_sb = true;
1464         }
1465
1466         if (write_sb)
1467                 bch2_write_super(c);
1468         mutex_unlock(&c->sb_lock);
1469
1470         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1471             !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1472             le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) {
1473                 struct bch_move_stats stats;
1474
1475                 bch2_move_stats_init(&stats, "recovery");
1476
1477                 bch_info(c, "scanning for old btree nodes");
1478                 ret = bch2_fs_read_write(c);
1479                 if (ret)
1480                         goto err;
1481
1482                 ret = bch2_scan_old_btree_nodes(c, &stats);
1483                 if (ret)
1484                         goto err;
1485                 bch_info(c, "scanning for old btree nodes done");
1486         }
1487
1488         if (c->journal_seq_blacklist_table &&
1489             c->journal_seq_blacklist_table->nr > 128)
1490                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1491
1492         ret = 0;
1493 out:
1494         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1495         bch2_flush_fsck_errs(c);
1496
1497         if (!c->opts.keep_journal &&
1498             test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)) {
1499                 bch2_journal_keys_free(&c->journal_keys);
1500                 bch2_journal_entries_free(c);
1501         }
1502         kfree(clean);
1503
1504         if (!ret && test_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags)) {
1505                 bch2_fs_read_write_early(c);
1506                 bch2_delete_dead_snapshots_async(c);
1507         }
1508
1509         if (ret)
1510                 bch_err(c, "Error in recovery: %s (%s)", err, bch2_err_str(ret));
1511         else
1512                 bch_verbose(c, "ret %s", bch2_err_str(ret));
1513         return ret;
1514 err:
1515 fsck_err:
1516         bch2_fs_emergency_read_only(c);
1517         goto out;
1518 }
1519
1520 int bch2_fs_initialize(struct bch_fs *c)
1521 {
1522         struct bch_inode_unpacked root_inode, lostfound_inode;
1523         struct bkey_inode_buf packed_inode;
1524         struct qstr lostfound = QSTR("lost+found");
1525         const char *err = "cannot allocate memory";
1526         struct bch_dev *ca;
1527         unsigned i;
1528         int ret;
1529
1530         bch_notice(c, "initializing new filesystem");
1531
1532         mutex_lock(&c->sb_lock);
1533         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1534         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1535
1536         if (c->sb.version < bcachefs_metadata_version_inode_v3)
1537                 c->opts.version_upgrade = true;
1538
1539         if (c->opts.version_upgrade) {
1540                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1541                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1542                 bch2_write_super(c);
1543         }
1544         mutex_unlock(&c->sb_lock);
1545
1546         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1547         set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1548         set_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags);
1549         set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1550         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1551         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1552
1553         for (i = 0; i < BTREE_ID_NR; i++)
1554                 bch2_btree_root_alloc(c, i);
1555
1556         for_each_online_member(ca, c, i)
1557                 bch2_dev_usage_init(ca);
1558
1559         err = "unable to allocate journal buckets";
1560         for_each_online_member(ca, c, i) {
1561                 ret = bch2_dev_journal_alloc(ca);
1562                 if (ret) {
1563                         percpu_ref_put(&ca->io_ref);
1564                         goto err;
1565                 }
1566         }
1567
1568         /*
1569          * journal_res_get() will crash if called before this has
1570          * set up the journal.pin FIFO and journal.cur pointer:
1571          */
1572         bch2_fs_journal_start(&c->journal, 1);
1573         bch2_journal_set_replay_done(&c->journal);
1574
1575         err = "error going read-write";
1576         ret = bch2_fs_read_write_early(c);
1577         if (ret)
1578                 goto err;
1579
1580         /*
1581          * Write out the superblock and journal buckets, now that we can do
1582          * btree updates
1583          */
1584         bch_verbose(c, "marking superblocks");
1585         err = "error marking superblock and journal";
1586         for_each_member_device(ca, c, i) {
1587                 ret = bch2_trans_mark_dev_sb(c, ca);
1588                 if (ret) {
1589                         percpu_ref_put(&ca->ref);
1590                         goto err;
1591                 }
1592
1593                 ca->new_fs_bucket_idx = 0;
1594         }
1595
1596         bch_verbose(c, "initializing freespace");
1597         err = "error initializing freespace";
1598         ret = bch2_fs_freespace_init(c);
1599         if (ret)
1600                 goto err;
1601
1602         err = "error creating root snapshot node";
1603         ret = bch2_fs_initialize_subvolumes(c);
1604         if (ret)
1605                 goto err;
1606
1607         bch_verbose(c, "reading snapshots table");
1608         err = "error reading snapshots table";
1609         ret = bch2_fs_snapshots_start(c);
1610         if (ret)
1611                 goto err;
1612         bch_verbose(c, "reading snapshots done");
1613
1614         bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL);
1615         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1616         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1617         bch2_inode_pack(&packed_inode, &root_inode);
1618         packed_inode.inode.k.p.snapshot = U32_MAX;
1619
1620         err = "error creating root directory";
1621         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1622                                 &packed_inode.inode.k_i,
1623                                 NULL, NULL, 0);
1624         if (ret)
1625                 goto err;
1626
1627         bch2_inode_init_early(c, &lostfound_inode);
1628
1629         err = "error creating lost+found";
1630         ret = bch2_trans_do(c, NULL, NULL, 0,
1631                 bch2_create_trans(&trans,
1632                                   BCACHEFS_ROOT_SUBVOL_INUM,
1633                                   &root_inode, &lostfound_inode,
1634                                   &lostfound,
1635                                   0, 0, S_IFDIR|0700, 0,
1636                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1637         if (ret) {
1638                 bch_err(c, "error creating lost+found");
1639                 goto err;
1640         }
1641
1642         if (enabled_qtypes(c)) {
1643                 ret = bch2_fs_quota_read(c);
1644                 if (ret)
1645                         goto err;
1646         }
1647
1648         err = "error writing first journal entry";
1649         ret = bch2_journal_flush(&c->journal);
1650         if (ret)
1651                 goto err;
1652
1653         mutex_lock(&c->sb_lock);
1654         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1655         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1656
1657         bch2_write_super(c);
1658         mutex_unlock(&c->sb_lock);
1659
1660         return 0;
1661 err:
1662         pr_err("Error initializing new filesystem: %s (%s)", err, bch2_err_str(ret));
1663         return ret;
1664 }