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

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "bkey_buf.h"
   5 #include "alloc_background.h"
   6 #include "btree_gc.h"
   7 #include "btree_update.h"
   8 #include "btree_update_interior.h"
   9 #include "btree_io.h"
  10 #include "buckets.h"
  11 #include "dirent.h"
  12 #include "ec.h"
  13 #include "error.h"
  14 #include "fs-common.h"
  15 #include "fsck.h"
  16 #include "journal_io.h"
  17 #include "journal_reclaim.h"
  18 #include "journal_seq_blacklist.h"
  19 #include "lru.h"
  20 #include "move.h"
  21 #include "quota.h"
  22 #include "recovery.h"
  23 #include "replicas.h"
  24 #include "subvolume.h"
  25 #include "super-io.h"
  26
  27 #include <linux/sort.h>
  28 #include <linux/stat.h>
  29
  30 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  31
  32 /* for -o reconstruct_alloc: */
  33 static void drop_alloc_keys(struct journal_keys *keys)
  34 {
  35         size_t src, dst;
  36
  37         for (src = 0, dst = 0; src < keys->nr; src++)
  38                 if (keys->d[src].btree_id != BTREE_ID_alloc)
  39                         keys->d[dst++] = keys->d[src];
  40
  41         keys->nr = dst;
  42 }
  43
  44 /*
  45  * Btree node pointers have a field to stack a pointer to the in memory btree
  46  * node; we need to zero out this field when reading in btree nodes, or when
  47  * reading in keys from the journal:
  48  */
  49 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
  50 {
  51         struct journal_key *i;
  52
  53         for (i = keys->d; i < keys->d + keys->nr; i++)
  54                 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
  55                         bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
  56 }
  57
  58 /* iterate over keys read from the journal: */
  59
  60 static int __journal_key_cmp(enum btree_id      l_btree_id,
  61                              unsigned           l_level,
  62                              struct bpos        l_pos,
  63                              const struct journal_key *r)
  64 {
  65         return (cmp_int(l_btree_id,     r->btree_id) ?:
  66                 cmp_int(l_level,        r->level) ?:
  67                 bpos_cmp(l_pos, r->k->k.p));
  68 }
  69
  70 static int journal_key_cmp(const struct journal_key *l, const struct journal_key *r)
  71 {
  72         return __journal_key_cmp(l->btree_id, l->level, l->k->k.p, r);
  73 }
  74
  75 static inline size_t idx_to_pos(struct journal_keys *keys, size_t idx)
  76 {
  77         size_t gap_size = keys->size - keys->nr;
  78
  79         if (idx >= keys->gap)
  80                 idx += gap_size;
  81         return idx;
  82 }
  83
  84 static inline struct journal_key *idx_to_key(struct journal_keys *keys, size_t idx)
  85 {
  86         return keys->d + idx_to_pos(keys, idx);
  87 }
  88
  89 size_t bch2_journal_key_search(struct journal_keys *keys,
  90                                enum btree_id id, unsigned level,
  91                                struct bpos pos)
  92 {
  93         size_t l = 0, r = keys->nr, m;
  94
  95         while (l < r) {
  96                 m = l + ((r - l) >> 1);
  97                 if (__journal_key_cmp(id, level, pos, idx_to_key(keys, m)) > 0)
  98                         l = m + 1;
  99                 else
 100                         r = m;
 101         }
 102
 103         BUG_ON(l < keys->nr &&
 104                __journal_key_cmp(id, level, pos, idx_to_key(keys, l)) > 0);
 105
 106         BUG_ON(l &&
 107                __journal_key_cmp(id, level, pos, idx_to_key(keys, l - 1)) <= 0);
 108
 109         return idx_to_pos(keys, l);
 110 }
 111
 112 struct bkey_i *bch2_journal_keys_peek(struct bch_fs *c, enum btree_id btree_id,
 113                                       unsigned level, struct bpos pos)
 114 {
 115         struct journal_keys *keys = &c->journal_keys;
 116         size_t idx = bch2_journal_key_search(keys, btree_id, level, pos);
 117
 118         while (idx < keys->size &&
 119                keys->d[idx].overwritten) {
 120                 idx++;
 121                 if (idx == keys->gap)
 122                         idx += keys->size - keys->nr;
 123         }
 124
 125         if (idx < keys->size &&
 126             keys->d[idx].btree_id == btree_id &&
 127             keys->d[idx].level == level)
 128                 return keys->d[idx].k;
 129         return NULL;
 130 }
 131
 132 static void journal_iters_fix(struct bch_fs *c)
 133 {
 134         struct journal_keys *keys = &c->journal_keys;
 135         /* The key we just inserted is immediately before the gap: */
 136         struct journal_key *n = &keys->d[keys->gap - 1];
 137         size_t gap_end = keys->gap + (keys->size - keys->nr);
 138         struct btree_and_journal_iter *iter;
 139
 140         /*
 141          * If an iterator points one after the key we just inserted,
 142          * and the key we just inserted compares >= the iterator's position,
 143          * decrement the iterator so it points at the key we just inserted:
 144          */
 145         list_for_each_entry(iter, &c->journal_iters, journal.list)
 146                 if (iter->journal.idx == gap_end &&
 147                     iter->last &&
 148                     iter->b->c.btree_id == n->btree_id &&
 149                     iter->b->c.level    == n->level &&
 150                     bpos_cmp(n->k->k.p, iter->unpacked.p) >= 0)
 151                         iter->journal.idx = keys->gap - 1;
 152 }
 153
 154 static void journal_iters_move_gap(struct bch_fs *c, size_t old_gap, size_t new_gap)
 155 {
 156         struct journal_keys *keys = &c->journal_keys;
 157         struct journal_iter *iter;
 158         size_t gap_size = keys->size - keys->nr;
 159
 160         list_for_each_entry(iter, &c->journal_iters, list) {
 161                 if (iter->idx > old_gap)
 162                         iter->idx -= gap_size;
 163                 if (iter->idx >= new_gap)
 164                         iter->idx += gap_size;
 165         }
 166 }
 167
 168 int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
 169                                  unsigned level, struct bkey_i *k)
 170 {
 171         struct journal_key n = {
 172                 .btree_id       = id,
 173                 .level          = level,
 174                 .k              = k,
 175                 .allocated      = true,
 176                 /*
 177                  * Ensure these keys are done last by journal replay, to unblock
 178                  * journal reclaim:
 179                  */
 180                 .journal_seq    = U32_MAX,
 181         };
 182         struct journal_keys *keys = &c->journal_keys;
 183         size_t idx = bch2_journal_key_search(keys, id, level, k->k.p);
 184
 185         BUG_ON(test_bit(BCH_FS_RW, &c->flags));
 186
 187         if (idx < keys->size &&
 188             journal_key_cmp(&n, &keys->d[idx]) == 0) {
 189                 if (keys->d[idx].allocated)
 190                         kfree(keys->d[idx].k);
 191                 keys->d[idx] = n;
 192                 return 0;
 193         }
 194
 195         if (idx > keys->gap)
 196                 idx -= keys->size - keys->nr;
 197
 198         if (keys->nr == keys->size) {
 199                 struct journal_keys new_keys = {
 200                         .nr                     = keys->nr,
 201                         .size                   = keys->size * 2,
 202                         .journal_seq_base       = keys->journal_seq_base,
 203                 };
 204
 205                 new_keys.d = kvmalloc(sizeof(new_keys.d[0]) * new_keys.size, GFP_KERNEL);
 206                 if (!new_keys.d) {
 207                         bch_err(c, "%s: error allocating new key array (size %zu)",
 208                                 __func__, new_keys.size);
 209                         return -ENOMEM;
 210                 }
 211
 212                 /* Since @keys was full, there was no gap: */
 213                 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
 214                 kvfree(keys->d);
 215                 *keys = new_keys;
 216
 217                 /* And now the gap is at the end: */
 218                 keys->gap = keys->nr;
 219         }
 220
 221         journal_iters_move_gap(c, keys->gap, idx);
 222
 223         move_gap(keys->d, keys->nr, keys->size, keys->gap, idx);
 224         keys->gap = idx;
 225
 226         keys->nr++;
 227         keys->d[keys->gap++] = n;
 228
 229         journal_iters_fix(c);
 230
 231         return 0;
 232 }
 233
 234 /*
 235  * Can only be used from the recovery thread while we're still RO - can't be
 236  * used once we've got RW, as journal_keys is at that point used by multiple
 237  * threads:
 238  */
 239 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
 240                             unsigned level, struct bkey_i *k)
 241 {
 242         struct bkey_i *n;
 243         int ret;
 244
 245         n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
 246         if (!n)
 247                 return -ENOMEM;
 248
 249         bkey_copy(n, k);
 250         ret = bch2_journal_key_insert_take(c, id, level, n);
 251         if (ret)
 252                 kfree(n);
 253         return ret;
 254 }
 255
 256 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
 257                             unsigned level, struct bpos pos)
 258 {
 259         struct bkey_i whiteout;
 260
 261         bkey_init(&whiteout.k);
 262         whiteout.k.p = pos;
 263
 264         return bch2_journal_key_insert(c, id, level, &whiteout);
 265 }
 266
 267 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
 268                                   unsigned level, struct bpos pos)
 269 {
 270         struct journal_keys *keys = &c->journal_keys;
 271         size_t idx = bch2_journal_key_search(keys, btree, level, pos);
 272
 273         if (idx < keys->size &&
 274             keys->d[idx].btree_id       == btree &&
 275             keys->d[idx].level          == level &&
 276             !bpos_cmp(keys->d[idx].k->k.p, pos))
 277                 keys->d[idx].overwritten = true;
 278 }
 279
 280 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
 281 {
 282         struct journal_key *k = iter->keys->d + iter->idx;
 283
 284         while (k < iter->keys->d + iter->keys->nr &&
 285                k->btree_id      == iter->btree_id &&
 286                k->level         == iter->level) {
 287                 if (!k->overwritten)
 288                         return k->k;
 289
 290                 iter->idx++;
 291                 k = iter->keys->d + iter->idx;
 292         }
 293
 294         return NULL;
 295 }
 296
 297 static void bch2_journal_iter_advance(struct journal_iter *iter)
 298 {
 299         if (iter->idx < iter->keys->size) {
 300                 iter->idx++;
 301                 if (iter->idx == iter->keys->gap)
 302                         iter->idx += iter->keys->size - iter->keys->nr;
 303         }
 304 }
 305
 306 static void bch2_journal_iter_exit(struct journal_iter *iter)
 307 {
 308         list_del(&iter->list);
 309 }
 310
 311 static void bch2_journal_iter_init(struct bch_fs *c,
 312                                    struct journal_iter *iter,
 313                                    enum btree_id id, unsigned level,
 314                                    struct bpos pos)
 315 {
 316         iter->btree_id  = id;
 317         iter->level     = level;
 318         iter->keys      = &c->journal_keys;
 319         iter->idx       = bch2_journal_key_search(&c->journal_keys, id, level, pos);
 320 }
 321
 322 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 323 {
 324         return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 325                                                 iter->b, &iter->unpacked);
 326 }
 327
 328 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 329 {
 330         bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 331 }
 332
 333 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 334 {
 335         switch (iter->last) {
 336         case none:
 337                 break;
 338         case btree:
 339                 bch2_journal_iter_advance_btree(iter);
 340                 break;
 341         case journal:
 342                 bch2_journal_iter_advance(&iter->journal);
 343                 break;
 344         }
 345
 346         iter->last = none;
 347 }
 348
 349 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 350 {
 351         struct bkey_s_c ret;
 352
 353         while (1) {
 354                 struct bkey_s_c btree_k         =
 355                         bch2_journal_iter_peek_btree(iter);
 356                 struct bkey_s_c journal_k       =
 357                         bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
 358
 359                 if (btree_k.k && journal_k.k) {
 360                         int cmp = bpos_cmp(btree_k.k->p, journal_k.k->p);
 361
 362                         if (!cmp)
 363                                 bch2_journal_iter_advance_btree(iter);
 364
 365                         iter->last = cmp < 0 ? btree : journal;
 366                 } else if (btree_k.k) {
 367                         iter->last = btree;
 368                 } else if (journal_k.k) {
 369                         iter->last = journal;
 370                 } else {
 371                         iter->last = none;
 372                         return bkey_s_c_null;
 373                 }
 374
 375                 ret = iter->last == journal ? journal_k : btree_k;
 376
 377                 if (iter->b &&
 378                     bpos_cmp(ret.k->p, iter->b->data->max_key) > 0) {
 379                         iter->journal.idx = iter->journal.keys->nr;
 380                         iter->last = none;
 381                         return bkey_s_c_null;
 382                 }
 383
 384                 if (!bkey_deleted(ret.k))
 385                         break;
 386
 387                 bch2_btree_and_journal_iter_advance(iter);
 388         }
 389
 390         return ret;
 391 }
 392
 393 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 394 {
 395         bch2_btree_and_journal_iter_advance(iter);
 396
 397         return bch2_btree_and_journal_iter_peek(iter);
 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 }
 418
 419 /*
 420  * this version is used by btree_gc before filesystem has gone RW and
 421  * multithreaded, so uses the journal_iters list:
 422  */
 423 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 424                                                 struct bch_fs *c,
 425                                                 struct btree *b)
 426 {
 427         struct btree_node_iter node_iter;
 428
 429         bch2_btree_node_iter_init_from_start(&node_iter, b);
 430         __bch2_btree_and_journal_iter_init_node_iter(iter, c, b, node_iter, b->data->min_key);
 431         list_add(&iter->journal.list, &c->journal_iters);
 432 }
 433
 434 /* sort and dedup all keys in the journal: */
 435
 436 void bch2_journal_entries_free(struct list_head *list)
 437 {
 438
 439         while (!list_empty(list)) {
 440                 struct journal_replay *i =
 441                         list_first_entry(list, struct journal_replay, list);
 442                 list_del(&i->list);
 443                 kvpfree(i, offsetof(struct journal_replay, j) +
 444                         vstruct_bytes(&i->j));
 445         }
 446 }
 447
 448 /*
 449  * When keys compare equal, oldest compares first:
 450  */
 451 static int journal_sort_key_cmp(const void *_l, const void *_r)
 452 {
 453         const struct journal_key *l = _l;
 454         const struct journal_key *r = _r;
 455
 456         return  journal_key_cmp(l, r) ?:
 457                 cmp_int(l->journal_seq, r->journal_seq) ?:
 458                 cmp_int(l->journal_offset, r->journal_offset);
 459 }
 460
 461 void bch2_journal_keys_free(struct journal_keys *keys)
 462 {
 463         struct journal_key *i;
 464
 465         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 466         keys->gap = keys->nr;
 467
 468         for (i = keys->d; i < keys->d + keys->nr; i++)
 469                 if (i->allocated)
 470                         kfree(i->k);
 471
 472         kvfree(keys->d);
 473         keys->d = NULL;
 474         keys->nr = keys->gap = keys->size = 0;
 475 }
 476
 477 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 478 {
 479         struct journal_replay *i;
 480         struct jset_entry *entry;
 481         struct bkey_i *k, *_n;
 482         struct journal_keys keys = { NULL };
 483         struct journal_key *src, *dst;
 484         size_t nr_keys = 0;
 485
 486         if (list_empty(journal_entries))
 487                 return keys;
 488
 489         list_for_each_entry(i, journal_entries, list) {
 490                 if (i->ignore)
 491                         continue;
 492
 493                 if (!keys.journal_seq_base)
 494                         keys.journal_seq_base = le64_to_cpu(i->j.seq);
 495
 496                 for_each_jset_key(k, _n, entry, &i->j)
 497                         nr_keys++;
 498         }
 499
 500         keys.size = roundup_pow_of_two(nr_keys);
 501
 502         keys.d = kvmalloc(sizeof(keys.d[0]) * keys.size, GFP_KERNEL);
 503         if (!keys.d)
 504                 goto err;
 505
 506         list_for_each_entry(i, journal_entries, list) {
 507                 if (i->ignore)
 508                         continue;
 509
 510                 BUG_ON(le64_to_cpu(i->j.seq) - keys.journal_seq_base > U32_MAX);
 511
 512                 for_each_jset_key(k, _n, entry, &i->j)
 513                         keys.d[keys.nr++] = (struct journal_key) {
 514                                 .btree_id       = entry->btree_id,
 515                                 .level          = entry->level,
 516                                 .k              = k,
 517                                 .journal_seq    = le64_to_cpu(i->j.seq) -
 518                                         keys.journal_seq_base,
 519                                 .journal_offset = k->_data - i->j._data,
 520                         };
 521         }
 522
 523         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 524
 525         src = dst = keys.d;
 526         while (src < keys.d + keys.nr) {
 527                 while (src + 1 < keys.d + keys.nr &&
 528                        src[0].btree_id  == src[1].btree_id &&
 529                        src[0].level     == src[1].level &&
 530                        !bpos_cmp(src[0].k->k.p, src[1].k->k.p))
 531                         src++;
 532
 533                 *dst++ = *src++;
 534         }
 535
 536         keys.nr = dst - keys.d;
 537         keys.gap = keys.nr;
 538 err:
 539         return keys;
 540 }
 541
 542 /* journal replay: */
 543
 544 static void replay_now_at(struct journal *j, u64 seq)
 545 {
 546         BUG_ON(seq < j->replay_journal_seq);
 547         BUG_ON(seq > j->replay_journal_seq_end);
 548
 549         while (j->replay_journal_seq < seq)
 550                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 551 }
 552
 553 static int bch2_journal_replay_key(struct btree_trans *trans,
 554                                    struct journal_key *k)
 555 {
 556         struct btree_iter iter;
 557         unsigned iter_flags =
 558                 BTREE_ITER_INTENT|
 559                 BTREE_ITER_NOT_EXTENTS;
 560         int ret;
 561
 562         if (!k->level && k->btree_id == BTREE_ID_alloc)
 563                 iter_flags |= BTREE_ITER_CACHED;
 564
 565         bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 566                                   BTREE_MAX_DEPTH, k->level,
 567                                   iter_flags);
 568         ret = bch2_btree_iter_traverse(&iter);
 569         if (ret)
 570                 goto out;
 571
 572         /* Must be checked with btree locked: */
 573         if (k->overwritten)
 574                 goto out;
 575
 576         ret = bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
 577 out:
 578         bch2_trans_iter_exit(trans, &iter);
 579         return ret;
 580 }
 581
 582 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 583 {
 584         const struct journal_key *l = *((const struct journal_key **)_l);
 585         const struct journal_key *r = *((const struct journal_key **)_r);
 586
 587         return cmp_int(l->journal_seq, r->journal_seq);
 588 }
 589
 590 static int bch2_journal_replay(struct bch_fs *c)
 591 {
 592         struct journal_keys *keys = &c->journal_keys;
 593         struct journal_key **keys_sorted, *k;
 594         struct journal *j = &c->journal;
 595         size_t i;
 596         int ret;
 597
 598         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 599         keys->gap = keys->nr;
 600
 601         keys_sorted = kvmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL);
 602         if (!keys_sorted)
 603                 return -ENOMEM;
 604
 605         for (i = 0; i < keys->nr; i++)
 606                 keys_sorted[i] = &keys->d[i];
 607
 608         sort(keys_sorted, keys->nr,
 609              sizeof(keys_sorted[0]),
 610              journal_sort_seq_cmp, NULL);
 611
 612         if (keys->nr)
 613                 replay_now_at(j, keys->journal_seq_base);
 614
 615         for (i = 0; i < keys->nr; i++) {
 616                 k = keys_sorted[i];
 617
 618                 cond_resched();
 619
 620                 if (!k->allocated)
 621                         replay_now_at(j, keys->journal_seq_base + k->journal_seq);
 622
 623                 ret = bch2_trans_do(c, NULL, NULL,
 624                                     BTREE_INSERT_LAZY_RW|
 625                                     BTREE_INSERT_NOFAIL|
 626                                     (!k->allocated
 627                                      ? BTREE_INSERT_JOURNAL_REPLAY|JOURNAL_WATERMARK_reserved
 628                                      : 0),
 629                              bch2_journal_replay_key(&trans, k));
 630                 if (ret) {
 631                         bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
 632                                 ret, bch2_btree_ids[k->btree_id], k->level);
 633                         goto err;
 634                 }
 635         }
 636
 637         replay_now_at(j, j->replay_journal_seq_end);
 638         j->replay_journal_seq = 0;
 639
 640         bch2_journal_set_replay_done(j);
 641         bch2_journal_flush_all_pins(j);
 642         ret = bch2_journal_error(j);
 643
 644         if (keys->nr && !ret)
 645                 bch2_journal_log_msg(&c->journal, "journal replay finished");
 646 err:
 647         kvfree(keys_sorted);
 648         return ret;
 649 }
 650
 651 /* journal replay early: */
 652
 653 static int journal_replay_entry_early(struct bch_fs *c,
 654                                       struct jset_entry *entry)
 655 {
 656         int ret = 0;
 657
 658         switch (entry->type) {
 659         case BCH_JSET_ENTRY_btree_root: {
 660                 struct btree_root *r;
 661
 662                 if (entry->btree_id >= BTREE_ID_NR) {
 663                         bch_err(c, "filesystem has unknown btree type %u",
 664                                 entry->btree_id);
 665                         return -EINVAL;
 666                 }
 667
 668                 r = &c->btree_roots[entry->btree_id];
 669
 670                 if (entry->u64s) {
 671                         r->level = entry->level;
 672                         bkey_copy(&r->key, &entry->start[0]);
 673                         r->error = 0;
 674                 } else {
 675                         r->error = -EIO;
 676                 }
 677                 r->alive = true;
 678                 break;
 679         }
 680         case BCH_JSET_ENTRY_usage: {
 681                 struct jset_entry_usage *u =
 682                         container_of(entry, struct jset_entry_usage, entry);
 683
 684                 switch (entry->btree_id) {
 685                 case BCH_FS_USAGE_reserved:
 686                         if (entry->level < BCH_REPLICAS_MAX)
 687                                 c->usage_base->persistent_reserved[entry->level] =
 688                                         le64_to_cpu(u->v);
 689                         break;
 690                 case BCH_FS_USAGE_inodes:
 691                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 692                         break;
 693                 case BCH_FS_USAGE_key_version:
 694                         atomic64_set(&c->key_version,
 695                                      le64_to_cpu(u->v));
 696                         break;
 697                 }
 698
 699                 break;
 700         }
 701         case BCH_JSET_ENTRY_data_usage: {
 702                 struct jset_entry_data_usage *u =
 703                         container_of(entry, struct jset_entry_data_usage, entry);
 704
 705                 ret = bch2_replicas_set_usage(c, &u->r,
 706                                               le64_to_cpu(u->v));
 707                 break;
 708         }
 709         case BCH_JSET_ENTRY_dev_usage: {
 710                 struct jset_entry_dev_usage *u =
 711                         container_of(entry, struct jset_entry_dev_usage, entry);
 712                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 713                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 714
 715                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 716                 ca->usage_base->buckets_unavailable     = le64_to_cpu(u->buckets_unavailable);
 717
 718                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 719                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 720                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 721                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 722                 }
 723
 724                 break;
 725         }
 726         case BCH_JSET_ENTRY_blacklist: {
 727                 struct jset_entry_blacklist *bl_entry =
 728                         container_of(entry, struct jset_entry_blacklist, entry);
 729
 730                 ret = bch2_journal_seq_blacklist_add(c,
 731                                 le64_to_cpu(bl_entry->seq),
 732                                 le64_to_cpu(bl_entry->seq) + 1);
 733                 break;
 734         }
 735         case BCH_JSET_ENTRY_blacklist_v2: {
 736                 struct jset_entry_blacklist_v2 *bl_entry =
 737                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 738
 739                 ret = bch2_journal_seq_blacklist_add(c,
 740                                 le64_to_cpu(bl_entry->start),
 741                                 le64_to_cpu(bl_entry->end) + 1);
 742                 break;
 743         }
 744         case BCH_JSET_ENTRY_clock: {
 745                 struct jset_entry_clock *clock =
 746                         container_of(entry, struct jset_entry_clock, entry);
 747
 748                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 749         }
 750         }
 751
 752         return ret;
 753 }
 754
 755 static int journal_replay_early(struct bch_fs *c,
 756                                 struct bch_sb_field_clean *clean,
 757                                 struct list_head *journal)
 758 {
 759         struct journal_replay *i;
 760         struct jset_entry *entry;
 761         int ret;
 762
 763         if (clean) {
 764                 for (entry = clean->start;
 765                      entry != vstruct_end(&clean->field);
 766                      entry = vstruct_next(entry)) {
 767                         ret = journal_replay_entry_early(c, entry);
 768                         if (ret)
 769                                 return ret;
 770                 }
 771         } else {
 772                 list_for_each_entry(i, journal, list) {
 773                         if (i->ignore)
 774                                 continue;
 775
 776                         vstruct_for_each(&i->j, entry) {
 777                                 ret = journal_replay_entry_early(c, entry);
 778                                 if (ret)
 779                                         return ret;
 780                         }
 781                 }
 782         }
 783
 784         bch2_fs_usage_initialize(c);
 785
 786         return 0;
 787 }
 788
 789 /* sb clean section: */
 790
 791 static struct bkey_i *btree_root_find(struct bch_fs *c,
 792                                       struct bch_sb_field_clean *clean,
 793                                       struct jset *j,
 794                                       enum btree_id id, unsigned *level)
 795 {
 796         struct bkey_i *k;
 797         struct jset_entry *entry, *start, *end;
 798
 799         if (clean) {
 800                 start = clean->start;
 801                 end = vstruct_end(&clean->field);
 802         } else {
 803                 start = j->start;
 804                 end = vstruct_last(j);
 805         }
 806
 807         for (entry = start; entry < end; entry = vstruct_next(entry))
 808                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 809                     entry->btree_id == id)
 810                         goto found;
 811
 812         return NULL;
 813 found:
 814         if (!entry->u64s)
 815                 return ERR_PTR(-EINVAL);
 816
 817         k = entry->start;
 818         *level = entry->level;
 819         return k;
 820 }
 821
 822 static int verify_superblock_clean(struct bch_fs *c,
 823                                    struct bch_sb_field_clean **cleanp,
 824                                    struct jset *j)
 825 {
 826         unsigned i;
 827         struct bch_sb_field_clean *clean = *cleanp;
 828         struct printbuf buf1 = PRINTBUF;
 829         struct printbuf buf2 = PRINTBUF;
 830         int ret = 0;
 831
 832         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 833                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 834                         le64_to_cpu(clean->journal_seq),
 835                         le64_to_cpu(j->seq))) {
 836                 kfree(clean);
 837                 *cleanp = NULL;
 838                 return 0;
 839         }
 840
 841         for (i = 0; i < BTREE_ID_NR; i++) {
 842                 struct bkey_i *k1, *k2;
 843                 unsigned l1 = 0, l2 = 0;
 844
 845                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 846                 k2 = btree_root_find(c, NULL, j, i, &l2);
 847
 848                 if (!k1 && !k2)
 849                         continue;
 850
 851                 printbuf_reset(&buf1);
 852                 printbuf_reset(&buf2);
 853
 854                 if (k1)
 855                         bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(k1));
 856                 else
 857                         pr_buf(&buf1, "(none)");
 858
 859                 if (k2)
 860                         bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(k2));
 861                 else
 862                         pr_buf(&buf2, "(none)");
 863
 864                 mustfix_fsck_err_on(!k1 || !k2 ||
 865                                     IS_ERR(k1) ||
 866                                     IS_ERR(k2) ||
 867                                     k1->k.u64s != k2->k.u64s ||
 868                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 869                                     l1 != l2, c,
 870                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 871                         "sb:      l=%u %s\n"
 872                         "journal: l=%u %s\n", i,
 873                         l1, buf1.buf,
 874                         l2, buf2.buf);
 875         }
 876 fsck_err:
 877         printbuf_exit(&buf2);
 878         printbuf_exit(&buf1);
 879         return ret;
 880 }
 881
 882 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 883 {
 884         struct bch_sb_field_clean *clean, *sb_clean;
 885         int ret;
 886
 887         mutex_lock(&c->sb_lock);
 888         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 889
 890         if (fsck_err_on(!sb_clean, c,
 891                         "superblock marked clean but clean section not present")) {
 892                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 893                 c->sb.clean = false;
 894                 mutex_unlock(&c->sb_lock);
 895                 return NULL;
 896         }
 897
 898         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 899                         GFP_KERNEL);
 900         if (!clean) {
 901                 mutex_unlock(&c->sb_lock);
 902                 return ERR_PTR(-ENOMEM);
 903         }
 904
 905         ret = bch2_sb_clean_validate_late(c, clean, READ);
 906         if (ret) {
 907                 mutex_unlock(&c->sb_lock);
 908                 return ERR_PTR(ret);
 909         }
 910
 911         mutex_unlock(&c->sb_lock);
 912
 913         return clean;
 914 fsck_err:
 915         mutex_unlock(&c->sb_lock);
 916         return ERR_PTR(ret);
 917 }
 918
 919 static int read_btree_roots(struct bch_fs *c)
 920 {
 921         unsigned i;
 922         int ret = 0;
 923
 924         for (i = 0; i < BTREE_ID_NR; i++) {
 925                 struct btree_root *r = &c->btree_roots[i];
 926
 927                 if (!r->alive)
 928                         continue;
 929
 930                 if (i == BTREE_ID_alloc &&
 931                     c->opts.reconstruct_alloc) {
 932                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 933                         continue;
 934                 }
 935
 936                 if (r->error) {
 937                         __fsck_err(c, i == BTREE_ID_alloc
 938                                    ? FSCK_CAN_IGNORE : 0,
 939                                    "invalid btree root %s",
 940                                    bch2_btree_ids[i]);
 941                         if (i == BTREE_ID_alloc)
 942                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 943                 }
 944
 945                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 946                 if (ret) {
 947                         __fsck_err(c, i == BTREE_ID_alloc
 948                                    ? FSCK_CAN_IGNORE : 0,
 949                                    "error reading btree root %s",
 950                                    bch2_btree_ids[i]);
 951                         if (i == BTREE_ID_alloc)
 952                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 953                 }
 954         }
 955
 956         for (i = 0; i < BTREE_ID_NR; i++)
 957                 if (!c->btree_roots[i].b)
 958                         bch2_btree_root_alloc(c, i);
 959 fsck_err:
 960         return ret;
 961 }
 962
 963 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
 964 {
 965         struct bkey_i_snapshot  root_snapshot;
 966         struct bkey_i_subvolume root_volume;
 967         int ret;
 968
 969         bkey_snapshot_init(&root_snapshot.k_i);
 970         root_snapshot.k.p.offset = U32_MAX;
 971         root_snapshot.v.flags   = 0;
 972         root_snapshot.v.parent  = 0;
 973         root_snapshot.v.subvol  = BCACHEFS_ROOT_SUBVOL;
 974         root_snapshot.v.pad     = 0;
 975         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
 976
 977         ret = bch2_btree_insert(c, BTREE_ID_snapshots,
 978                                 &root_snapshot.k_i,
 979                                 NULL, NULL, 0);
 980         if (ret)
 981                 return ret;
 982
 983
 984         bkey_subvolume_init(&root_volume.k_i);
 985         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
 986         root_volume.v.flags     = 0;
 987         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
 988         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
 989
 990         ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
 991                                 &root_volume.k_i,
 992                                 NULL, NULL, 0);
 993         if (ret)
 994                 return ret;
 995
 996         return 0;
 997 }
 998
 999 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
1000 {
1001         struct btree_iter iter;
1002         struct bkey_s_c k;
1003         struct bch_inode_unpacked inode;
1004         int ret;
1005
1006         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
1007                              SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
1008         k = bch2_btree_iter_peek_slot(&iter);
1009         ret = bkey_err(k);
1010         if (ret)
1011                 goto err;
1012
1013         if (!bkey_is_inode(k.k)) {
1014                 bch_err(trans->c, "root inode not found");
1015                 ret = -ENOENT;
1016                 goto err;
1017         }
1018
1019         ret = bch2_inode_unpack(k, &inode);
1020         BUG_ON(ret);
1021
1022         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1023
1024         ret = bch2_inode_write(trans, &iter, &inode);
1025 err:
1026         bch2_trans_iter_exit(trans, &iter);
1027         return ret;
1028 }
1029
1030 int bch2_fs_recovery(struct bch_fs *c)
1031 {
1032         const char *err = "cannot allocate memory";
1033         struct bch_sb_field_clean *clean = NULL;
1034         struct jset *last_journal_entry = NULL;
1035         u64 blacklist_seq, journal_seq;
1036         bool write_sb = false;
1037         int ret = 0;
1038
1039         if (c->sb.clean)
1040                 clean = read_superblock_clean(c);
1041         ret = PTR_ERR_OR_ZERO(clean);
1042         if (ret)
1043                 goto err;
1044
1045         if (c->sb.clean)
1046                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
1047                          le64_to_cpu(clean->journal_seq));
1048         else
1049                 bch_info(c, "recovering from unclean shutdown");
1050
1051         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
1052                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
1053                 ret = -EINVAL;
1054                 goto err;
1055         }
1056
1057         if (!c->sb.clean &&
1058             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1059                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1060                 ret = -EINVAL;
1061                 goto err;
1062         }
1063
1064         if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1065                 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
1066                 ret = -EINVAL;
1067                 goto err;
1068         }
1069
1070         if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
1071                 bch_info(c, "alloc_v2 feature bit not set, fsck required");
1072                 c->opts.fsck = true;
1073                 c->opts.fix_errors = FSCK_OPT_YES;
1074         }
1075
1076         if (!c->replicas.entries ||
1077             c->opts.rebuild_replicas) {
1078                 bch_info(c, "building replicas info");
1079                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1080         }
1081
1082         if (!c->opts.nochanges) {
1083                 if (c->sb.version < bcachefs_metadata_version_inode_backpointers) {
1084                         bch_info(c, "version prior to inode backpointers, upgrade and fsck required");
1085                         c->opts.version_upgrade = true;
1086                         c->opts.fsck            = true;
1087                         c->opts.fix_errors      = FSCK_OPT_YES;
1088                 } else if (c->sb.version < bcachefs_metadata_version_subvol_dirent) {
1089                         bch_info(c, "filesystem version is prior to subvol_dirent - upgrading");
1090                         c->opts.version_upgrade = true;
1091                         c->opts.fsck            = true;
1092                 } else if (c->sb.version < bcachefs_metadata_version_alloc_v4) {
1093                         bch_info(c, "filesystem version is prior to alloc_v4 - upgrading");
1094                         c->opts.version_upgrade = true;
1095                 }
1096         }
1097
1098         ret = bch2_blacklist_table_initialize(c);
1099         if (ret) {
1100                 bch_err(c, "error initializing blacklist table");
1101                 goto err;
1102         }
1103
1104         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1105                 struct journal_replay *i;
1106
1107                 bch_verbose(c, "starting journal read");
1108                 ret = bch2_journal_read(c, &c->journal_entries,
1109                                         &blacklist_seq, &journal_seq);
1110                 if (ret)
1111                         goto err;
1112
1113                 list_for_each_entry_reverse(i, &c->journal_entries, list)
1114                         if (!i->ignore) {
1115                                 last_journal_entry = &i->j;
1116                                 break;
1117                         }
1118
1119                 if (mustfix_fsck_err_on(c->sb.clean &&
1120                                         last_journal_entry &&
1121                                         !journal_entry_empty(last_journal_entry), c,
1122                                 "filesystem marked clean but journal not empty")) {
1123                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1124                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1125                         c->sb.clean = false;
1126                 }
1127
1128                 if (!last_journal_entry) {
1129                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
1130                         goto use_clean;
1131                 }
1132
1133                 c->journal_keys = journal_keys_sort(&c->journal_entries);
1134                 if (!c->journal_keys.d) {
1135                         ret = -ENOMEM;
1136                         goto err;
1137                 }
1138
1139                 if (c->sb.clean && last_journal_entry) {
1140                         ret = verify_superblock_clean(c, &clean,
1141                                                       last_journal_entry);
1142                         if (ret)
1143                                 goto err;
1144                 }
1145         } else {
1146 use_clean:
1147                 if (!clean) {
1148                         bch_err(c, "no superblock clean section found");
1149                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1150                         goto err;
1151
1152                 }
1153                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1154         }
1155
1156         if (c->opts.read_journal_only)
1157                 goto out;
1158
1159         if (c->opts.reconstruct_alloc) {
1160                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1161                 drop_alloc_keys(&c->journal_keys);
1162         }
1163
1164         zero_out_btree_mem_ptr(&c->journal_keys);
1165
1166         ret = journal_replay_early(c, clean, &c->journal_entries);
1167         if (ret)
1168                 goto err;
1169
1170         /*
1171          * After an unclean shutdown, skip then next few journal sequence
1172          * numbers as they may have been referenced by btree writes that
1173          * happened before their corresponding journal writes - those btree
1174          * writes need to be ignored, by skipping and blacklisting the next few
1175          * journal sequence numbers:
1176          */
1177         if (!c->sb.clean)
1178                 journal_seq += 8;
1179
1180         if (blacklist_seq != journal_seq) {
1181                 ret = bch2_journal_seq_blacklist_add(c,
1182                                         blacklist_seq, journal_seq);
1183                 if (ret) {
1184                         bch_err(c, "error creating new journal seq blacklist entry");
1185                         goto err;
1186                 }
1187         }
1188
1189         ret = bch2_fs_journal_start(&c->journal, journal_seq,
1190                                     &c->journal_entries);
1191         if (ret)
1192                 goto err;
1193
1194         ret = read_btree_roots(c);
1195         if (ret)
1196                 goto err;
1197
1198         bch_verbose(c, "starting alloc read");
1199         err = "error reading allocation information";
1200
1201         down_read(&c->gc_lock);
1202         ret = bch2_alloc_read(c);
1203         up_read(&c->gc_lock);
1204
1205         if (ret)
1206                 goto err;
1207         bch_verbose(c, "alloc read done");
1208
1209         bch_verbose(c, "starting stripes_read");
1210         err = "error reading stripes";
1211         ret = bch2_stripes_read(c);
1212         if (ret)
1213                 goto err;
1214         bch_verbose(c, "stripes_read done");
1215
1216         /*
1217          * If we're not running fsck, this ensures bch2_fsck_err() calls are
1218          * instead interpreted as bch2_inconsistent_err() calls:
1219          */
1220         if (!c->opts.fsck)
1221                 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1222
1223         if (c->opts.fsck ||
1224             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)) ||
1225             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_metadata)) ||
1226             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1227                 bool metadata_only = c->opts.norecovery;
1228
1229                 bch_info(c, "checking allocations");
1230                 err = "error checking allocations";
1231                 ret = bch2_gc(c, true, metadata_only);
1232                 if (ret)
1233                         goto err;
1234                 bch_verbose(c, "done checking allocations");
1235         }
1236
1237         if (c->opts.fsck) {
1238                 bch_info(c, "checking need_discard and freespace btrees");
1239                 err = "error checking need_discard and freespace btrees";
1240                 ret = bch2_check_alloc_info(c);
1241                 if (ret)
1242                         goto err;
1243
1244                 ret = bch2_check_lrus(c, true);
1245                 if (ret)
1246                         goto err;
1247                 bch_verbose(c, "done checking need_discard and freespace btrees");
1248         }
1249
1250         bch2_stripes_heap_start(c);
1251
1252         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1253         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1254         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1255
1256         /*
1257          * Skip past versions that might have possibly been used (as nonces),
1258          * but hadn't had their pointers written:
1259          */
1260         if (c->sb.encryption_type && !c->sb.clean)
1261                 atomic64_add(1 << 16, &c->key_version);
1262
1263         if (c->opts.norecovery)
1264                 goto out;
1265
1266         bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1267         err = "journal replay failed";
1268         ret = bch2_journal_replay(c);
1269         if (ret)
1270                 goto err;
1271         if (c->opts.verbose || !c->sb.clean)
1272                 bch_info(c, "journal replay done");
1273
1274         err = "error initializing freespace";
1275         ret = bch2_fs_freespace_init(c);
1276         if (ret)
1277                 goto err;
1278
1279         if (c->opts.fsck) {
1280                 bch_info(c, "checking alloc to lru refs");
1281                 err = "error checking alloc to lru refs";
1282                 ret = bch2_check_alloc_to_lru_refs(c);
1283                 if (ret)
1284                         goto err;
1285
1286                 ret = bch2_check_lrus(c, true);
1287                 if (ret)
1288                         goto err;
1289                 bch_verbose(c, "done checking alloc to lru refs");
1290         }
1291
1292         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1293                 bch2_fs_lazy_rw(c);
1294
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->sb.version < bcachefs_metadata_version_snapshot_2) {
1309                 /* set bi_subvol on root inode */
1310                 err = "error upgrade root inode for subvolumes";
1311                 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1312                                     bch2_fs_upgrade_for_subvolumes(&trans));
1313                 if (ret)
1314                         goto err;
1315         }
1316
1317         if (c->opts.fsck) {
1318                 bch_info(c, "starting fsck");
1319                 err = "error in fsck";
1320                 ret = bch2_fsck_full(c);
1321                 if (ret)
1322                         goto err;
1323                 bch_verbose(c, "fsck done");
1324         } else if (!c->sb.clean) {
1325                 bch_verbose(c, "checking for deleted inodes");
1326                 err = "error in recovery";
1327                 ret = bch2_fsck_walk_inodes_only(c);
1328                 if (ret)
1329                         goto err;
1330                 bch_verbose(c, "check inodes done");
1331         }
1332
1333         if (enabled_qtypes(c)) {
1334                 bch_verbose(c, "reading quotas");
1335                 ret = bch2_fs_quota_read(c);
1336                 if (ret)
1337                         goto err;
1338                 bch_verbose(c, "quotas done");
1339         }
1340
1341         mutex_lock(&c->sb_lock);
1342         if (c->opts.version_upgrade) {
1343                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1344                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1345                 write_sb = true;
1346         }
1347
1348         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1349                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1350                 write_sb = true;
1351         }
1352
1353         if (c->opts.fsck &&
1354             !test_bit(BCH_FS_ERROR, &c->flags) &&
1355             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1356                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1357                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1358                 write_sb = true;
1359         }
1360
1361         if (write_sb)
1362                 bch2_write_super(c);
1363         mutex_unlock(&c->sb_lock);
1364
1365         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1366             !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1367             le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) {
1368                 struct bch_move_stats stats;
1369
1370                 bch_move_stats_init(&stats, "recovery");
1371
1372                 bch_info(c, "scanning for old btree nodes");
1373                 ret = bch2_fs_read_write(c);
1374                 if (ret)
1375                         goto err;
1376
1377                 ret = bch2_scan_old_btree_nodes(c, &stats);
1378                 if (ret)
1379                         goto err;
1380                 bch_info(c, "scanning for old btree nodes done");
1381         }
1382
1383         if (c->journal_seq_blacklist_table &&
1384             c->journal_seq_blacklist_table->nr > 128)
1385                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1386
1387         ret = 0;
1388 out:
1389         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1390         bch2_flush_fsck_errs(c);
1391
1392         if (!c->opts.keep_journal) {
1393                 bch2_journal_keys_free(&c->journal_keys);
1394                 bch2_journal_entries_free(&c->journal_entries);
1395         }
1396         kfree(clean);
1397         if (ret)
1398                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1399         else
1400                 bch_verbose(c, "ret %i", ret);
1401         return ret;
1402 err:
1403 fsck_err:
1404         bch2_fs_emergency_read_only(c);
1405         goto out;
1406 }
1407
1408 int bch2_fs_initialize(struct bch_fs *c)
1409 {
1410         struct bch_inode_unpacked root_inode, lostfound_inode;
1411         struct bkey_inode_buf packed_inode;
1412         struct qstr lostfound = QSTR("lost+found");
1413         const char *err = "cannot allocate memory";
1414         struct bch_dev *ca;
1415         LIST_HEAD(journal);
1416         unsigned i;
1417         int ret;
1418
1419         bch_notice(c, "initializing new filesystem");
1420
1421         mutex_lock(&c->sb_lock);
1422         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1423         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1424
1425         if (c->opts.version_upgrade) {
1426                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1427                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1428                 bch2_write_super(c);
1429         }
1430         mutex_unlock(&c->sb_lock);
1431
1432         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1433         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1434         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1435
1436         for (i = 0; i < BTREE_ID_NR; i++)
1437                 bch2_btree_root_alloc(c, i);
1438
1439         err = "unable to allocate journal buckets";
1440         for_each_online_member(ca, c, i) {
1441                 ret = bch2_dev_journal_alloc(ca);
1442                 if (ret) {
1443                         percpu_ref_put(&ca->io_ref);
1444                         goto err;
1445                 }
1446         }
1447
1448         /*
1449          * journal_res_get() will crash if called before this has
1450          * set up the journal.pin FIFO and journal.cur pointer:
1451          */
1452         bch2_fs_journal_start(&c->journal, 1, &journal);
1453         bch2_journal_set_replay_done(&c->journal);
1454
1455         err = "error going read-write";
1456         ret = bch2_fs_read_write_early(c);
1457         if (ret)
1458                 goto err;
1459
1460         /*
1461          * Write out the superblock and journal buckets, now that we can do
1462          * btree updates
1463          */
1464         bch_verbose(c, "marking superblocks");
1465         err = "error marking superblock and journal";
1466         for_each_member_device(ca, c, i) {
1467                 ret = bch2_trans_mark_dev_sb(c, ca);
1468                 if (ret) {
1469                         percpu_ref_put(&ca->ref);
1470                         goto err;
1471                 }
1472
1473                 ca->new_fs_bucket_idx = 0;
1474         }
1475
1476         bch_verbose(c, "initializing freespace");
1477         err = "error initializing freespace";
1478         ret = bch2_fs_freespace_init(c);
1479         if (ret)
1480                 goto err;
1481
1482         err = "error creating root snapshot node";
1483         ret = bch2_fs_initialize_subvolumes(c);
1484         if (ret)
1485                 goto err;
1486
1487         bch_verbose(c, "reading snapshots table");
1488         err = "error reading snapshots table";
1489         ret = bch2_fs_snapshots_start(c);
1490         if (ret)
1491                 goto err;
1492         bch_verbose(c, "reading snapshots done");
1493
1494         bch2_inode_init(c, &root_inode, 0, 0,
1495                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1496         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1497         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1498         bch2_inode_pack(c, &packed_inode, &root_inode);
1499         packed_inode.inode.k.p.snapshot = U32_MAX;
1500
1501         err = "error creating root directory";
1502         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1503                                 &packed_inode.inode.k_i,
1504                                 NULL, NULL, 0);
1505         if (ret)
1506                 goto err;
1507
1508         bch2_inode_init_early(c, &lostfound_inode);
1509
1510         err = "error creating lost+found";
1511         ret = bch2_trans_do(c, NULL, NULL, 0,
1512                 bch2_create_trans(&trans,
1513                                   BCACHEFS_ROOT_SUBVOL_INUM,
1514                                   &root_inode, &lostfound_inode,
1515                                   &lostfound,
1516                                   0, 0, S_IFDIR|0700, 0,
1517                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1518         if (ret) {
1519                 bch_err(c, "error creating lost+found");
1520                 goto err;
1521         }
1522
1523         if (enabled_qtypes(c)) {
1524                 ret = bch2_fs_quota_read(c);
1525                 if (ret)
1526                         goto err;
1527         }
1528
1529         err = "error writing first journal entry";
1530         ret = bch2_journal_flush(&c->journal);
1531         if (ret)
1532                 goto err;
1533
1534         mutex_lock(&c->sb_lock);
1535         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1536         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1537
1538         bch2_write_super(c);
1539         mutex_unlock(&c->sb_lock);
1540
1541         return 0;
1542 err:
1543         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1544         return ret;
1545 }