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 static 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                 cond_resched();
 547
 548                 for_each_jset_key(k, entry, &i->j) {
 549                         if (keys->nr == keys->size) {
 550                                 __journal_keys_sort(keys);
 551
 552                                 if (keys->nr > keys->size * 7 / 8) {
 553                                         bch_err(c, "Too many journal keys for slowpath; have %zu compacted, buf size %zu, processed %zu/%zu",
 554                                                 keys->nr, keys->size, nr_read, nr_keys);
 555                                         return -BCH_ERR_ENOMEM_journal_keys_sort;
 556                                 }
 557                         }
 558
 559                         keys->d[keys->nr++] = (struct journal_key) {
 560                                 .btree_id       = entry->btree_id,
 561                                 .level          = entry->level,
 562                                 .k              = k,
 563                                 .journal_seq    = le64_to_cpu(i->j.seq),
 564                                 .journal_offset = k->_data - i->j._data,
 565                         };
 566
 567                         nr_read++;
 568                 }
 569         }
 570
 571         __journal_keys_sort(keys);
 572         keys->gap = keys->nr;
 573
 574         bch_verbose(c, "Journal keys: %zu read, %zu after sorting and compacting", nr_keys, keys->nr);
 575         return 0;
 576 }
 577
 578 /* journal replay: */
 579
 580 static void replay_now_at(struct journal *j, u64 seq)
 581 {
 582         BUG_ON(seq < j->replay_journal_seq);
 583
 584         seq = min(seq, j->replay_journal_seq_end);
 585
 586         while (j->replay_journal_seq < seq)
 587                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 588 }
 589
 590 static int bch2_journal_replay_key(struct btree_trans *trans,
 591                                    struct journal_key *k)
 592 {
 593         struct btree_iter iter;
 594         unsigned iter_flags =
 595                 BTREE_ITER_INTENT|
 596                 BTREE_ITER_NOT_EXTENTS;
 597         int ret;
 598
 599         if (!k->level && k->btree_id == BTREE_ID_alloc)
 600                 iter_flags |= BTREE_ITER_CACHED;
 601
 602         bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 603                                   BTREE_MAX_DEPTH, k->level,
 604                                   iter_flags);
 605         ret = bch2_btree_iter_traverse(&iter);
 606         if (ret)
 607                 goto out;
 608
 609         /* Must be checked with btree locked: */
 610         if (k->overwritten)
 611                 goto out;
 612
 613         ret = bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
 614 out:
 615         bch2_trans_iter_exit(trans, &iter);
 616         return ret;
 617 }
 618
 619 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 620 {
 621         const struct journal_key *l = *((const struct journal_key **)_l);
 622         const struct journal_key *r = *((const struct journal_key **)_r);
 623
 624         return cmp_int(l->journal_seq, r->journal_seq);
 625 }
 626
 627 static int bch2_journal_replay(struct bch_fs *c)
 628 {
 629         struct journal_keys *keys = &c->journal_keys;
 630         struct journal_key **keys_sorted, *k;
 631         struct journal *j = &c->journal;
 632         u64 start_seq   = c->journal_replay_seq_start;
 633         u64 end_seq     = c->journal_replay_seq_start;
 634         size_t i;
 635         int ret;
 636
 637         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 638         keys->gap = keys->nr;
 639
 640         keys_sorted = kvmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL);
 641         if (!keys_sorted)
 642                 return -BCH_ERR_ENOMEM_journal_replay;
 643
 644         for (i = 0; i < keys->nr; i++)
 645                 keys_sorted[i] = &keys->d[i];
 646
 647         sort(keys_sorted, keys->nr,
 648              sizeof(keys_sorted[0]),
 649              journal_sort_seq_cmp, NULL);
 650
 651         if (keys->nr) {
 652                 ret = bch2_journal_log_msg(c, "Starting journal replay (%zu keys in entries %llu-%llu)",
 653                                            keys->nr, start_seq, end_seq);
 654                 if (ret)
 655                         goto err;
 656         }
 657
 658         for (i = 0; i < keys->nr; i++) {
 659                 k = keys_sorted[i];
 660
 661                 cond_resched();
 662
 663                 replay_now_at(j, k->journal_seq);
 664
 665                 ret = bch2_trans_do(c, NULL, NULL,
 666                                     BTREE_INSERT_LAZY_RW|
 667                                     BTREE_INSERT_NOFAIL|
 668                                     (!k->allocated
 669                                      ? BTREE_INSERT_JOURNAL_REPLAY|BCH_WATERMARK_reclaim
 670                                      : 0),
 671                              bch2_journal_replay_key(&trans, k));
 672                 if (ret) {
 673                         bch_err(c, "journal replay: error while replaying key at btree %s level %u: %s",
 674                                 bch2_btree_ids[k->btree_id], k->level, bch2_err_str(ret));
 675                         goto err;
 676                 }
 677         }
 678
 679         replay_now_at(j, j->replay_journal_seq_end);
 680         j->replay_journal_seq = 0;
 681
 682         bch2_journal_set_replay_done(j);
 683         bch2_journal_flush_all_pins(j);
 684         ret = bch2_journal_error(j);
 685
 686         if (keys->nr && !ret)
 687                 bch2_journal_log_msg(c, "journal replay finished");
 688 err:
 689         kvfree(keys_sorted);
 690
 691         if (ret)
 692                 bch_err_fn(c, ret);
 693         return ret;
 694 }
 695
 696 /* journal replay early: */
 697
 698 static int journal_replay_entry_early(struct bch_fs *c,
 699                                       struct jset_entry *entry)
 700 {
 701         int ret = 0;
 702
 703         switch (entry->type) {
 704         case BCH_JSET_ENTRY_btree_root: {
 705                 struct btree_root *r;
 706
 707                 while (entry->btree_id >= c->btree_roots_extra.nr + BTREE_ID_NR) {
 708                         ret = darray_push(&c->btree_roots_extra, (struct btree_root) { NULL });
 709                         if (ret)
 710                                 return ret;
 711                 }
 712
 713                 r = bch2_btree_id_root(c, entry->btree_id);
 714
 715                 if (entry->u64s) {
 716                         r->level = entry->level;
 717                         bkey_copy(&r->key, &entry->start[0]);
 718                         r->error = 0;
 719                 } else {
 720                         r->error = -EIO;
 721                 }
 722                 r->alive = true;
 723                 break;
 724         }
 725         case BCH_JSET_ENTRY_usage: {
 726                 struct jset_entry_usage *u =
 727                         container_of(entry, struct jset_entry_usage, entry);
 728
 729                 switch (entry->btree_id) {
 730                 case BCH_FS_USAGE_reserved:
 731                         if (entry->level < BCH_REPLICAS_MAX)
 732                                 c->usage_base->persistent_reserved[entry->level] =
 733                                         le64_to_cpu(u->v);
 734                         break;
 735                 case BCH_FS_USAGE_inodes:
 736                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 737                         break;
 738                 case BCH_FS_USAGE_key_version:
 739                         atomic64_set(&c->key_version,
 740                                      le64_to_cpu(u->v));
 741                         break;
 742                 }
 743
 744                 break;
 745         }
 746         case BCH_JSET_ENTRY_data_usage: {
 747                 struct jset_entry_data_usage *u =
 748                         container_of(entry, struct jset_entry_data_usage, entry);
 749
 750                 ret = bch2_replicas_set_usage(c, &u->r,
 751                                               le64_to_cpu(u->v));
 752                 break;
 753         }
 754         case BCH_JSET_ENTRY_dev_usage: {
 755                 struct jset_entry_dev_usage *u =
 756                         container_of(entry, struct jset_entry_dev_usage, entry);
 757                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 758                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 759
 760                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 761
 762                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 763                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 764                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 765                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 766                 }
 767
 768                 break;
 769         }
 770         case BCH_JSET_ENTRY_blacklist: {
 771                 struct jset_entry_blacklist *bl_entry =
 772                         container_of(entry, struct jset_entry_blacklist, entry);
 773
 774                 ret = bch2_journal_seq_blacklist_add(c,
 775                                 le64_to_cpu(bl_entry->seq),
 776                                 le64_to_cpu(bl_entry->seq) + 1);
 777                 break;
 778         }
 779         case BCH_JSET_ENTRY_blacklist_v2: {
 780                 struct jset_entry_blacklist_v2 *bl_entry =
 781                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 782
 783                 ret = bch2_journal_seq_blacklist_add(c,
 784                                 le64_to_cpu(bl_entry->start),
 785                                 le64_to_cpu(bl_entry->end) + 1);
 786                 break;
 787         }
 788         case BCH_JSET_ENTRY_clock: {
 789                 struct jset_entry_clock *clock =
 790                         container_of(entry, struct jset_entry_clock, entry);
 791
 792                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 793         }
 794         }
 795
 796         return ret;
 797 }
 798
 799 static int journal_replay_early(struct bch_fs *c,
 800                                 struct bch_sb_field_clean *clean)
 801 {
 802         struct jset_entry *entry;
 803         int ret;
 804
 805         if (clean) {
 806                 for (entry = clean->start;
 807                      entry != vstruct_end(&clean->field);
 808                      entry = vstruct_next(entry)) {
 809                         ret = journal_replay_entry_early(c, entry);
 810                         if (ret)
 811                                 return ret;
 812                 }
 813         } else {
 814                 struct genradix_iter iter;
 815                 struct journal_replay *i, **_i;
 816
 817                 genradix_for_each(&c->journal_entries, iter, _i) {
 818                         i = *_i;
 819
 820                         if (!i || i->ignore)
 821                                 continue;
 822
 823                         vstruct_for_each(&i->j, entry) {
 824                                 ret = journal_replay_entry_early(c, entry);
 825                                 if (ret)
 826                                         return ret;
 827                         }
 828                 }
 829         }
 830
 831         bch2_fs_usage_initialize(c);
 832
 833         return 0;
 834 }
 835
 836 /* sb clean section: */
 837
 838 static struct bkey_i *btree_root_find(struct bch_fs *c,
 839                                       struct bch_sb_field_clean *clean,
 840                                       struct jset *j,
 841                                       enum btree_id id, unsigned *level)
 842 {
 843         struct bkey_i *k;
 844         struct jset_entry *entry, *start, *end;
 845
 846         if (clean) {
 847                 start = clean->start;
 848                 end = vstruct_end(&clean->field);
 849         } else {
 850                 start = j->start;
 851                 end = vstruct_last(j);
 852         }
 853
 854         for (entry = start; entry < end; entry = vstruct_next(entry))
 855                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 856                     entry->btree_id == id)
 857                         goto found;
 858
 859         return NULL;
 860 found:
 861         if (!entry->u64s)
 862                 return ERR_PTR(-EINVAL);
 863
 864         k = entry->start;
 865         *level = entry->level;
 866         return k;
 867 }
 868
 869 static int verify_superblock_clean(struct bch_fs *c,
 870                                    struct bch_sb_field_clean **cleanp,
 871                                    struct jset *j)
 872 {
 873         unsigned i;
 874         struct bch_sb_field_clean *clean = *cleanp;
 875         struct printbuf buf1 = PRINTBUF;
 876         struct printbuf buf2 = PRINTBUF;
 877         int ret = 0;
 878
 879         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 880                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 881                         le64_to_cpu(clean->journal_seq),
 882                         le64_to_cpu(j->seq))) {
 883                 kfree(clean);
 884                 *cleanp = NULL;
 885                 return 0;
 886         }
 887
 888         for (i = 0; i < BTREE_ID_NR; i++) {
 889                 struct bkey_i *k1, *k2;
 890                 unsigned l1 = 0, l2 = 0;
 891
 892                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 893                 k2 = btree_root_find(c, NULL, j, i, &l2);
 894
 895                 if (!k1 && !k2)
 896                         continue;
 897
 898                 printbuf_reset(&buf1);
 899                 printbuf_reset(&buf2);
 900
 901                 if (k1)
 902                         bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(k1));
 903                 else
 904                         prt_printf(&buf1, "(none)");
 905
 906                 if (k2)
 907                         bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(k2));
 908                 else
 909                         prt_printf(&buf2, "(none)");
 910
 911                 mustfix_fsck_err_on(!k1 || !k2 ||
 912                                     IS_ERR(k1) ||
 913                                     IS_ERR(k2) ||
 914                                     k1->k.u64s != k2->k.u64s ||
 915                                     memcmp(k1, k2, bkey_bytes(&k1->k)) ||
 916                                     l1 != l2, c,
 917                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 918                         "sb:      l=%u %s\n"
 919                         "journal: l=%u %s\n", i,
 920                         l1, buf1.buf,
 921                         l2, buf2.buf);
 922         }
 923 fsck_err:
 924         printbuf_exit(&buf2);
 925         printbuf_exit(&buf1);
 926         return ret;
 927 }
 928
 929 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 930 {
 931         struct bch_sb_field_clean *clean, *sb_clean;
 932         int ret;
 933
 934         mutex_lock(&c->sb_lock);
 935         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 936
 937         if (fsck_err_on(!sb_clean, c,
 938                         "superblock marked clean but clean section not present")) {
 939                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 940                 c->sb.clean = false;
 941                 mutex_unlock(&c->sb_lock);
 942                 return NULL;
 943         }
 944
 945         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 946                         GFP_KERNEL);
 947         if (!clean) {
 948                 mutex_unlock(&c->sb_lock);
 949                 return ERR_PTR(-BCH_ERR_ENOMEM_read_superblock_clean);
 950         }
 951
 952         ret = bch2_sb_clean_validate_late(c, clean, READ);
 953         if (ret) {
 954                 mutex_unlock(&c->sb_lock);
 955                 return ERR_PTR(ret);
 956         }
 957
 958         mutex_unlock(&c->sb_lock);
 959
 960         return clean;
 961 fsck_err:
 962         mutex_unlock(&c->sb_lock);
 963         return ERR_PTR(ret);
 964 }
 965
 966 static bool btree_id_is_alloc(enum btree_id id)
 967 {
 968         switch (id) {
 969         case BTREE_ID_alloc:
 970         case BTREE_ID_backpointers:
 971         case BTREE_ID_need_discard:
 972         case BTREE_ID_freespace:
 973         case BTREE_ID_bucket_gens:
 974                 return true;
 975         default:
 976                 return false;
 977         }
 978 }
 979
 980 static int read_btree_roots(struct bch_fs *c)
 981 {
 982         unsigned i;
 983         int ret = 0;
 984
 985         for (i = 0; i < btree_id_nr_alive(c); i++) {
 986                 struct btree_root *r = bch2_btree_id_root(c, i);
 987
 988                 if (!r->alive)
 989                         continue;
 990
 991                 if (btree_id_is_alloc(i) &&
 992                     c->opts.reconstruct_alloc) {
 993                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 994                         continue;
 995                 }
 996
 997                 if (r->error) {
 998                         __fsck_err(c, btree_id_is_alloc(i)
 999                                    ? FSCK_CAN_IGNORE : 0,
1000                                    "invalid btree root %s",
1001                                    bch2_btree_ids[i]);
1002                         if (i == BTREE_ID_alloc)
1003                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1004                 }
1005
1006                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
1007                 if (ret) {
1008                         __fsck_err(c,
1009                                    btree_id_is_alloc(i)
1010                                    ? FSCK_CAN_IGNORE : 0,
1011                                    "error reading btree root %s",
1012                                    bch2_btree_ids[i]);
1013                         if (btree_id_is_alloc(i))
1014                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1015                 }
1016         }
1017
1018         for (i = 0; i < BTREE_ID_NR; i++) {
1019                 struct btree_root *r = bch2_btree_id_root(c, i);
1020
1021                 if (!r->b) {
1022                         r->alive = false;
1023                         r->level = 0;
1024                         bch2_btree_root_alloc(c, i);
1025                 }
1026         }
1027 fsck_err:
1028         return ret;
1029 }
1030
1031 static int bch2_initialize_subvolumes(struct bch_fs *c)
1032 {
1033         struct bkey_i_snapshot_tree     root_tree;
1034         struct bkey_i_snapshot          root_snapshot;
1035         struct bkey_i_subvolume         root_volume;
1036         int ret;
1037
1038         bkey_snapshot_tree_init(&root_tree.k_i);
1039         root_tree.k.p.offset            = 1;
1040         root_tree.v.master_subvol       = cpu_to_le32(1);
1041         root_tree.v.root_snapshot       = cpu_to_le32(U32_MAX);
1042
1043         bkey_snapshot_init(&root_snapshot.k_i);
1044         root_snapshot.k.p.offset = U32_MAX;
1045         root_snapshot.v.flags   = 0;
1046         root_snapshot.v.parent  = 0;
1047         root_snapshot.v.subvol  = cpu_to_le32(BCACHEFS_ROOT_SUBVOL);
1048         root_snapshot.v.tree    = cpu_to_le32(1);
1049         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
1050
1051         bkey_subvolume_init(&root_volume.k_i);
1052         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1053         root_volume.v.flags     = 0;
1054         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
1055         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
1056
1057         ret =   bch2_btree_insert(c, BTREE_ID_snapshot_trees,
1058                                   &root_tree.k_i,
1059                                   NULL, NULL, 0) ?:
1060                 bch2_btree_insert(c, BTREE_ID_snapshots,
1061                                   &root_snapshot.k_i,
1062                                   NULL, NULL, 0) ?:
1063                 bch2_btree_insert(c, BTREE_ID_subvolumes,
1064                                   &root_volume.k_i,
1065                                   NULL, NULL, 0);
1066         if (ret)
1067                 bch_err_fn(c, ret);
1068         return ret;
1069 }
1070
1071 static int __bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
1072 {
1073         struct btree_iter iter;
1074         struct bkey_s_c k;
1075         struct bch_inode_unpacked inode;
1076         int ret;
1077
1078         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
1079                                SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
1080         ret = bkey_err(k);
1081         if (ret)
1082                 return ret;
1083
1084         if (!bkey_is_inode(k.k)) {
1085                 bch_err(trans->c, "root inode not found");
1086                 ret = -BCH_ERR_ENOENT_inode;
1087                 goto err;
1088         }
1089
1090         ret = bch2_inode_unpack(k, &inode);
1091         BUG_ON(ret);
1092
1093         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1094
1095         ret = bch2_inode_write(trans, &iter, &inode);
1096 err:
1097         bch2_trans_iter_exit(trans, &iter);
1098         return ret;
1099 }
1100
1101 /* set bi_subvol on root inode */
1102 noinline_for_stack
1103 static int bch2_fs_upgrade_for_subvolumes(struct bch_fs *c)
1104 {
1105         int ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1106                                 __bch2_fs_upgrade_for_subvolumes(&trans));
1107         if (ret)
1108                 bch_err_fn(c, ret);
1109         return ret;
1110 }
1111
1112 static void check_version_upgrade(struct bch_fs *c)
1113 {
1114         unsigned version = c->sb.version_upgrade_complete ?: c->sb.version;
1115
1116         if (version < bcachefs_metadata_required_upgrade_below ||
1117             (version < bcachefs_metadata_version_current &&
1118              c->opts.version_upgrade != BCH_VERSION_UPGRADE_none)) {
1119                 struct printbuf buf = PRINTBUF;
1120
1121                 if (version != c->sb.version) {
1122                         prt_str(&buf, "version upgrade to ");
1123                         bch2_version_to_text(&buf, c->sb.version);
1124                         prt_str(&buf, " incomplete:\n");
1125                 }
1126
1127                 prt_str(&buf, "version ");
1128                 bch2_version_to_text(&buf, version);
1129                 prt_str(&buf, " prior to ");
1130                 bch2_version_to_text(&buf, bcachefs_metadata_required_upgrade_below);
1131                 prt_str(&buf, ", upgrade and fsck required");
1132
1133                 bch_info(c, "%s", buf.buf);
1134                 printbuf_exit(&buf);
1135
1136                 c->opts.fsck            = true;
1137                 c->opts.fix_errors      = FSCK_OPT_YES;
1138                 set_bit(BCH_FS_VERSION_UPGRADE, &c->flags);
1139         }
1140 }
1141
1142 static int bch2_check_allocations(struct bch_fs *c)
1143 {
1144         return bch2_gc(c, true, c->opts.norecovery);
1145 }
1146
1147 static int bch2_set_may_go_rw(struct bch_fs *c)
1148 {
1149         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1150         return 0;
1151 }
1152
1153 struct recovery_pass_fn {
1154         int             (*fn)(struct bch_fs *);
1155         const char      *name;
1156         unsigned        when;
1157 };
1158
1159 static struct recovery_pass_fn recovery_passes[] = {
1160 #define x(_fn, _when)   { .fn = bch2_##_fn, .name = #_fn, .when = _when },
1161         BCH_RECOVERY_PASSES()
1162 #undef x
1163 };
1164
1165 static bool should_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
1166 {
1167         struct recovery_pass_fn *p = recovery_passes + c->curr_recovery_pass;
1168
1169         if (c->opts.norecovery && pass > BCH_RECOVERY_PASS_snapshots_read)
1170                 return false;
1171         if ((p->when & PASS_FSCK) && c->opts.fsck)
1172                 return true;
1173         if ((p->when & PASS_UNCLEAN) && !c->sb.clean)
1174                 return true;
1175         if (p->when & PASS_ALWAYS)
1176                 return true;
1177         if (p->when >= PASS_UPGRADE(0) &&
1178             bch2_version_upgrading_to(c, p->when >> 4))
1179                 return true;
1180         return false;
1181 }
1182
1183 static int bch2_run_recovery_pass(struct bch_fs *c, enum bch_recovery_pass pass)
1184 {
1185         int ret;
1186
1187         c->curr_recovery_pass = pass;
1188
1189         if (should_run_recovery_pass(c, pass)) {
1190                 struct recovery_pass_fn *p = recovery_passes + pass;
1191
1192                 if (!(p->when & PASS_SILENT))
1193                         printk(KERN_INFO bch2_log_msg(c, "%s..."), p->name);
1194                 ret = p->fn(c);
1195                 if (ret)
1196                         return ret;
1197                 if (!(p->when & PASS_SILENT))
1198                         printk(KERN_CONT " done\n");
1199         }
1200
1201         return 0;
1202 }
1203
1204 static int bch2_run_recovery_passes(struct bch_fs *c)
1205 {
1206         int ret = 0;
1207 again:
1208         while (c->curr_recovery_pass < ARRAY_SIZE(recovery_passes)) {
1209                 ret = bch2_run_recovery_pass(c, c->curr_recovery_pass);
1210                 if (ret)
1211                         break;
1212                 c->curr_recovery_pass++;
1213         }
1214
1215         if (bch2_err_matches(ret, BCH_ERR_need_snapshot_cleanup)) {
1216                 set_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
1217                 c->curr_recovery_pass = BCH_RECOVERY_PASS_delete_dead_snapshots;
1218                 goto again;
1219         }
1220
1221         return ret;
1222 }
1223
1224 int bch2_fs_recovery(struct bch_fs *c)
1225 {
1226         struct bch_sb_field_clean *clean = NULL;
1227         struct jset *last_journal_entry = NULL;
1228         u64 last_seq, blacklist_seq, journal_seq;
1229         bool write_sb = false;
1230         int ret = 0;
1231
1232         if (c->sb.clean)
1233                 clean = read_superblock_clean(c);
1234         ret = PTR_ERR_OR_ZERO(clean);
1235         if (ret)
1236                 goto err;
1237
1238         if (c->sb.clean)
1239                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
1240                          le64_to_cpu(clean->journal_seq));
1241         else
1242                 bch_info(c, "recovering from unclean shutdown");
1243
1244         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
1245                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
1246                 ret = -EINVAL;
1247                 goto err;
1248         }
1249
1250         if (!c->sb.clean &&
1251             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1252                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1253                 ret = -EINVAL;
1254                 goto err;
1255         }
1256
1257         if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1258                 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
1259                 ret = -EINVAL;
1260                 goto err;
1261         }
1262
1263         if (!c->opts.nochanges)
1264                 check_version_upgrade(c);
1265
1266         if (c->opts.fsck && c->opts.norecovery) {
1267                 bch_err(c, "cannot select both norecovery and fsck");
1268                 ret = -EINVAL;
1269                 goto err;
1270         }
1271
1272         ret = bch2_blacklist_table_initialize(c);
1273         if (ret) {
1274                 bch_err(c, "error initializing blacklist table");
1275                 goto err;
1276         }
1277
1278         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1279                 struct genradix_iter iter;
1280                 struct journal_replay **i;
1281
1282                 bch_verbose(c, "starting journal read");
1283                 ret = bch2_journal_read(c, &last_seq, &blacklist_seq, &journal_seq);
1284                 if (ret)
1285                         goto err;
1286
1287                 /*
1288                  * note: cmd_list_journal needs the blacklist table fully up to date so
1289                  * it can asterisk ignored journal entries:
1290                  */
1291                 if (c->opts.read_journal_only)
1292                         goto out;
1293
1294                 genradix_for_each_reverse(&c->journal_entries, iter, i)
1295                         if (*i && !(*i)->ignore) {
1296                                 last_journal_entry = &(*i)->j;
1297                                 break;
1298                         }
1299
1300                 if (mustfix_fsck_err_on(c->sb.clean &&
1301                                         last_journal_entry &&
1302                                         !journal_entry_empty(last_journal_entry), c,
1303                                 "filesystem marked clean but journal not empty")) {
1304                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1305                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1306                         c->sb.clean = false;
1307                 }
1308
1309                 if (!last_journal_entry) {
1310                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
1311                         if (clean)
1312                                 goto use_clean;
1313
1314                         genradix_for_each_reverse(&c->journal_entries, iter, i)
1315                                 if (*i) {
1316                                         last_journal_entry = &(*i)->j;
1317                                         (*i)->ignore = false;
1318                                         break;
1319                                 }
1320                 }
1321
1322                 ret = journal_keys_sort(c);
1323                 if (ret)
1324                         goto err;
1325
1326                 if (c->sb.clean && last_journal_entry) {
1327                         ret = verify_superblock_clean(c, &clean,
1328                                                       last_journal_entry);
1329                         if (ret)
1330                                 goto err;
1331                 }
1332         } else {
1333 use_clean:
1334                 if (!clean) {
1335                         bch_err(c, "no superblock clean section found");
1336                         ret = -BCH_ERR_fsck_repair_impossible;
1337                         goto err;
1338
1339                 }
1340                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1341         }
1342
1343         c->journal_replay_seq_start     = last_seq;
1344         c->journal_replay_seq_end       = blacklist_seq - 1;;
1345
1346         if (c->opts.reconstruct_alloc) {
1347                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1348                 drop_alloc_keys(&c->journal_keys);
1349         }
1350
1351         zero_out_btree_mem_ptr(&c->journal_keys);
1352
1353         ret = journal_replay_early(c, clean);
1354         if (ret)
1355                 goto err;
1356
1357         /*
1358          * After an unclean shutdown, skip then next few journal sequence
1359          * numbers as they may have been referenced by btree writes that
1360          * happened before their corresponding journal writes - those btree
1361          * writes need to be ignored, by skipping and blacklisting the next few
1362          * journal sequence numbers:
1363          */
1364         if (!c->sb.clean)
1365                 journal_seq += 8;
1366
1367         if (blacklist_seq != journal_seq) {
1368                 ret =   bch2_journal_log_msg(c, "blacklisting entries %llu-%llu",
1369                                              blacklist_seq, journal_seq) ?:
1370                         bch2_journal_seq_blacklist_add(c,
1371                                         blacklist_seq, journal_seq);
1372                 if (ret) {
1373                         bch_err(c, "error creating new journal seq blacklist entry");
1374                         goto err;
1375                 }
1376         }
1377
1378         ret =   bch2_journal_log_msg(c, "starting journal at entry %llu, replaying %llu-%llu",
1379                                      journal_seq, last_seq, blacklist_seq - 1) ?:
1380                 bch2_fs_journal_start(&c->journal, journal_seq);
1381         if (ret)
1382                 goto err;
1383
1384         if (c->opts.reconstruct_alloc)
1385                 bch2_journal_log_msg(c, "dropping alloc info");
1386
1387         /*
1388          * Skip past versions that might have possibly been used (as nonces),
1389          * but hadn't had their pointers written:
1390          */
1391         if (c->sb.encryption_type && !c->sb.clean)
1392                 atomic64_add(1 << 16, &c->key_version);
1393
1394         ret = read_btree_roots(c);
1395         if (ret)
1396                 goto err;
1397
1398         ret = bch2_run_recovery_passes(c);
1399         if (ret)
1400                 goto err;
1401
1402         if (enabled_qtypes(c)) {
1403                 bch_verbose(c, "reading quotas");
1404                 ret = bch2_fs_quota_read(c);
1405                 if (ret)
1406                         goto err;
1407                 bch_verbose(c, "quotas done");
1408         }
1409
1410         mutex_lock(&c->sb_lock);
1411         if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) != c->sb.version) {
1412                 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, c->sb.version);
1413                 write_sb = true;
1414         }
1415
1416         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1417                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1418                 write_sb = true;
1419         }
1420
1421         if (c->opts.fsck &&
1422             !test_bit(BCH_FS_ERROR, &c->flags) &&
1423             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1424                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1425                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1426                 write_sb = true;
1427         }
1428
1429         if (write_sb)
1430                 bch2_write_super(c);
1431         mutex_unlock(&c->sb_lock);
1432
1433         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1434             !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1435             c->sb.version_min < bcachefs_metadata_version_btree_ptr_sectors_written) {
1436                 struct bch_move_stats stats;
1437
1438                 bch2_move_stats_init(&stats, "recovery");
1439
1440                 bch_info(c, "scanning for old btree nodes");
1441                 ret =   bch2_fs_read_write(c) ?:
1442                         bch2_scan_old_btree_nodes(c, &stats);
1443                 if (ret)
1444                         goto err;
1445                 bch_info(c, "scanning for old btree nodes done");
1446         }
1447
1448         if (c->journal_seq_blacklist_table &&
1449             c->journal_seq_blacklist_table->nr > 128)
1450                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1451
1452         ret = 0;
1453 out:
1454         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1455         bch2_flush_fsck_errs(c);
1456
1457         if (!c->opts.keep_journal &&
1458             test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)) {
1459                 bch2_journal_keys_free(&c->journal_keys);
1460                 bch2_journal_entries_free(c);
1461         }
1462         kfree(clean);
1463
1464         if (!ret && test_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags)) {
1465                 bch2_fs_read_write_early(c);
1466                 bch2_delete_dead_snapshots_async(c);
1467         }
1468
1469         if (ret)
1470                 bch_err_fn(c, ret);
1471         return ret;
1472 err:
1473 fsck_err:
1474         bch2_fs_emergency_read_only(c);
1475         goto out;
1476 }
1477
1478 int bch2_fs_initialize(struct bch_fs *c)
1479 {
1480         struct bch_inode_unpacked root_inode, lostfound_inode;
1481         struct bkey_inode_buf packed_inode;
1482         struct qstr lostfound = QSTR("lost+found");
1483         struct bch_dev *ca;
1484         unsigned i;
1485         int ret;
1486
1487         bch_notice(c, "initializing new filesystem");
1488
1489         mutex_lock(&c->sb_lock);
1490         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1491         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1492
1493         if (c->opts.version_upgrade != BCH_VERSION_UPGRADE_none) {
1494                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1495                 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
1496                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1497                 bch2_write_super(c);
1498         }
1499         mutex_unlock(&c->sb_lock);
1500
1501         c->curr_recovery_pass = ARRAY_SIZE(recovery_passes);
1502         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1503         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1504
1505         for (i = 0; i < BTREE_ID_NR; i++)
1506                 bch2_btree_root_alloc(c, i);
1507
1508         for_each_online_member(ca, c, i)
1509                 bch2_dev_usage_init(ca);
1510
1511         for_each_online_member(ca, c, i) {
1512                 ret = bch2_dev_journal_alloc(ca);
1513                 if (ret) {
1514                         percpu_ref_put(&ca->io_ref);
1515                         goto err;
1516                 }
1517         }
1518
1519         /*
1520          * journal_res_get() will crash if called before this has
1521          * set up the journal.pin FIFO and journal.cur pointer:
1522          */
1523         bch2_fs_journal_start(&c->journal, 1);
1524         bch2_journal_set_replay_done(&c->journal);
1525
1526         ret = bch2_fs_read_write_early(c);
1527         if (ret)
1528                 goto err;
1529
1530         /*
1531          * Write out the superblock and journal buckets, now that we can do
1532          * btree updates
1533          */
1534         bch_verbose(c, "marking superblocks");
1535         for_each_member_device(ca, c, i) {
1536                 ret = bch2_trans_mark_dev_sb(c, ca);
1537                 if (ret) {
1538                         percpu_ref_put(&ca->ref);
1539                         goto err;
1540                 }
1541
1542                 ca->new_fs_bucket_idx = 0;
1543         }
1544
1545         ret = bch2_fs_freespace_init(c);
1546         if (ret)
1547                 goto err;
1548
1549         ret = bch2_initialize_subvolumes(c);
1550         if (ret)
1551                 goto err;
1552
1553         bch_verbose(c, "reading snapshots table");
1554         ret = bch2_snapshots_read(c);
1555         if (ret)
1556                 goto err;
1557         bch_verbose(c, "reading snapshots done");
1558
1559         bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL);
1560         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1561         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1562         bch2_inode_pack(&packed_inode, &root_inode);
1563         packed_inode.inode.k.p.snapshot = U32_MAX;
1564
1565         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1566                                 &packed_inode.inode.k_i,
1567                                 NULL, NULL, 0);
1568         if (ret) {
1569                 bch_err_msg(c, ret, "creating root directory");
1570                 goto err;
1571         }
1572
1573         bch2_inode_init_early(c, &lostfound_inode);
1574
1575         ret = bch2_trans_do(c, NULL, NULL, 0,
1576                 bch2_create_trans(&trans,
1577                                   BCACHEFS_ROOT_SUBVOL_INUM,
1578                                   &root_inode, &lostfound_inode,
1579                                   &lostfound,
1580                                   0, 0, S_IFDIR|0700, 0,
1581                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1582         if (ret) {
1583                 bch_err_msg(c, ret, "creating lost+found");
1584                 goto err;
1585         }
1586
1587         if (enabled_qtypes(c)) {
1588                 ret = bch2_fs_quota_read(c);
1589                 if (ret)
1590                         goto err;
1591         }
1592
1593         ret = bch2_journal_flush(&c->journal);
1594         if (ret) {
1595                 bch_err_msg(c, ret, "writing first journal entry");
1596                 goto err;
1597         }
1598
1599         mutex_lock(&c->sb_lock);
1600         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1601         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1602
1603         bch2_write_super(c);
1604         mutex_unlock(&c->sb_lock);
1605
1606         return 0;
1607 err:
1608         bch_err_fn(ca, ret);
1609         return ret;
1610 }