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 "move.h"
  20 #include "quota.h"
  21 #include "recovery.h"
  22 #include "replicas.h"
  23 #include "subvolume.h"
  24 #include "super-io.h"
  25
  26 #include <linux/sort.h>
  27 #include <linux/stat.h>
  28
  29 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  30
  31 /* for -o reconstruct_alloc: */
  32 static void drop_alloc_keys(struct journal_keys *keys)
  33 {
  34         size_t src, dst;
  35
  36         for (src = 0, dst = 0; src < keys->nr; src++)
  37                 if (keys->d[src].btree_id != BTREE_ID_alloc)
  38                         keys->d[dst++] = keys->d[src];
  39
  40         keys->nr = dst;
  41 }
  42
  43 /*
  44  * Btree node pointers have a field to stack a pointer to the in memory btree
  45  * node; we need to zero out this field when reading in btree nodes, or when
  46  * reading in keys from the journal:
  47  */
  48 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
  49 {
  50         struct journal_key *i;
  51
  52         for (i = keys->d; i < keys->d + keys->nr; i++)
  53                 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
  54                         bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
  55 }
  56
  57 /* iterate over keys read from the journal: */
  58
  59 static int __journal_key_cmp(enum btree_id      l_btree_id,
  60                              unsigned           l_level,
  61                              struct bpos        l_pos,
  62                              struct journal_key *r)
  63 {
  64         return (cmp_int(l_btree_id,     r->btree_id) ?:
  65                 cmp_int(l_level,        r->level) ?:
  66                 bpos_cmp(l_pos, r->k->k.p));
  67 }
  68
  69 static int journal_key_cmp(struct journal_key *l, struct journal_key *r)
  70 {
  71         return (cmp_int(l->btree_id,    r->btree_id) ?:
  72                 cmp_int(l->level,       r->level) ?:
  73                 bpos_cmp(l->k->k.p,     r->k->k.p));
  74 }
  75
  76 static size_t journal_key_search(struct journal_keys *journal_keys,
  77                                  enum btree_id id, unsigned level,
  78                                  struct bpos pos)
  79 {
  80         size_t l = 0, r = journal_keys->nr, m;
  81
  82         while (l < r) {
  83                 m = l + ((r - l) >> 1);
  84                 if (__journal_key_cmp(id, level, pos, &journal_keys->d[m]) > 0)
  85                         l = m + 1;
  86                 else
  87                         r = m;
  88         }
  89
  90         BUG_ON(l < journal_keys->nr &&
  91                __journal_key_cmp(id, level, pos, &journal_keys->d[l]) > 0);
  92
  93         BUG_ON(l &&
  94                __journal_key_cmp(id, level, pos, &journal_keys->d[l - 1]) <= 0);
  95
  96         return l;
  97 }
  98
  99 static void journal_iter_fix(struct bch_fs *c, struct journal_iter *iter, unsigned idx)
 100 {
 101         struct bkey_i *n = iter->keys->d[idx].k;
 102         struct btree_and_journal_iter *biter =
 103                 container_of(iter, struct btree_and_journal_iter, journal);
 104
 105         if (iter->idx > idx ||
 106             (iter->idx == idx &&
 107              biter->last &&
 108              bpos_cmp(n->k.p, biter->unpacked.p) <= 0))
 109                 iter->idx++;
 110 }
 111
 112 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
 113                             unsigned level, struct bkey_i *k)
 114 {
 115         struct journal_key n = {
 116                 .btree_id       = id,
 117                 .level          = level,
 118                 .k              = k,
 119                 .allocated      = true
 120         };
 121         struct journal_keys *keys = &c->journal_keys;
 122         struct journal_iter *iter;
 123         unsigned idx = journal_key_search(keys, id, level, k->k.p);
 124
 125         if (idx < keys->nr &&
 126             journal_key_cmp(&n, &keys->d[idx]) == 0) {
 127                 if (keys->d[idx].allocated)
 128                         kfree(keys->d[idx].k);
 129                 keys->d[idx] = n;
 130                 return 0;
 131         }
 132
 133         if (keys->nr == keys->size) {
 134                 struct journal_keys new_keys = {
 135                         .nr                     = keys->nr,
 136                         .size                   = keys->size * 2,
 137                         .journal_seq_base       = keys->journal_seq_base,
 138                 };
 139
 140                 new_keys.d = kvmalloc(sizeof(new_keys.d[0]) * new_keys.size, GFP_KERNEL);
 141                 if (!new_keys.d) {
 142                         bch_err(c, "%s: error allocating new key array (size %zu)",
 143                                 __func__, new_keys.size);
 144                         return -ENOMEM;
 145                 }
 146
 147                 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
 148                 kvfree(keys->d);
 149                 *keys = new_keys;
 150         }
 151
 152         array_insert_item(keys->d, keys->nr, idx, n);
 153
 154         list_for_each_entry(iter, &c->journal_iters, list)
 155                 journal_iter_fix(c, iter, idx);
 156
 157         return 0;
 158 }
 159
 160 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
 161                             unsigned level, struct bpos pos)
 162 {
 163         struct bkey_i *whiteout =
 164                 kmalloc(sizeof(struct bkey), GFP_KERNEL);
 165         int ret;
 166
 167         if (!whiteout) {
 168                 bch_err(c, "%s: error allocating new key", __func__);
 169                 return -ENOMEM;
 170         }
 171
 172         bkey_init(&whiteout->k);
 173         whiteout->k.p = pos;
 174
 175         ret = bch2_journal_key_insert(c, id, level, whiteout);
 176         if (ret)
 177                 kfree(whiteout);
 178         return ret;
 179 }
 180
 181 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
 182 {
 183         struct journal_key *k = iter->idx - iter->keys->nr
 184                 ? iter->keys->d + iter->idx : NULL;
 185
 186         if (k &&
 187             k->btree_id == iter->btree_id &&
 188             k->level    == iter->level)
 189                 return k->k;
 190
 191         iter->idx = iter->keys->nr;
 192         return NULL;
 193 }
 194
 195 static void bch2_journal_iter_advance(struct journal_iter *iter)
 196 {
 197         if (iter->idx < iter->keys->nr)
 198                 iter->idx++;
 199 }
 200
 201 static void bch2_journal_iter_exit(struct journal_iter *iter)
 202 {
 203         list_del(&iter->list);
 204 }
 205
 206 static void bch2_journal_iter_init(struct bch_fs *c,
 207                                    struct journal_iter *iter,
 208                                    enum btree_id id, unsigned level,
 209                                    struct bpos pos)
 210 {
 211         iter->btree_id  = id;
 212         iter->level     = level;
 213         iter->keys      = &c->journal_keys;
 214         iter->idx       = journal_key_search(&c->journal_keys, id, level, pos);
 215         list_add(&iter->list, &c->journal_iters);
 216 }
 217
 218 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 219 {
 220         return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 221                                                 iter->b, &iter->unpacked);
 222 }
 223
 224 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 225 {
 226         bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 227 }
 228
 229 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 230 {
 231         switch (iter->last) {
 232         case none:
 233                 break;
 234         case btree:
 235                 bch2_journal_iter_advance_btree(iter);
 236                 break;
 237         case journal:
 238                 bch2_journal_iter_advance(&iter->journal);
 239                 break;
 240         }
 241
 242         iter->last = none;
 243 }
 244
 245 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 246 {
 247         struct bkey_s_c ret;
 248
 249         while (1) {
 250                 struct bkey_s_c btree_k         =
 251                         bch2_journal_iter_peek_btree(iter);
 252                 struct bkey_s_c journal_k       =
 253                         bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
 254
 255                 if (btree_k.k && journal_k.k) {
 256                         int cmp = bpos_cmp(btree_k.k->p, journal_k.k->p);
 257
 258                         if (!cmp)
 259                                 bch2_journal_iter_advance_btree(iter);
 260
 261                         iter->last = cmp < 0 ? btree : journal;
 262                 } else if (btree_k.k) {
 263                         iter->last = btree;
 264                 } else if (journal_k.k) {
 265                         iter->last = journal;
 266                 } else {
 267                         iter->last = none;
 268                         return bkey_s_c_null;
 269                 }
 270
 271                 ret = iter->last == journal ? journal_k : btree_k;
 272
 273                 if (iter->b &&
 274                     bpos_cmp(ret.k->p, iter->b->data->max_key) > 0) {
 275                         iter->journal.idx = iter->journal.keys->nr;
 276                         iter->last = none;
 277                         return bkey_s_c_null;
 278                 }
 279
 280                 if (!bkey_deleted(ret.k))
 281                         break;
 282
 283                 bch2_btree_and_journal_iter_advance(iter);
 284         }
 285
 286         return ret;
 287 }
 288
 289 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 290 {
 291         bch2_btree_and_journal_iter_advance(iter);
 292
 293         return bch2_btree_and_journal_iter_peek(iter);
 294 }
 295
 296 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
 297 {
 298         bch2_journal_iter_exit(&iter->journal);
 299 }
 300
 301 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 302                                                 struct bch_fs *c,
 303                                                 struct btree *b)
 304 {
 305         memset(iter, 0, sizeof(*iter));
 306
 307         iter->b = b;
 308         bch2_btree_node_iter_init_from_start(&iter->node_iter, iter->b);
 309         bch2_journal_iter_init(c, &iter->journal,
 310                                b->c.btree_id, b->c.level, b->data->min_key);
 311 }
 312
 313 /* Walk btree, overlaying keys from the journal: */
 314
 315 static void btree_and_journal_iter_prefetch(struct bch_fs *c, struct btree *b,
 316                                            struct btree_and_journal_iter iter)
 317 {
 318         unsigned i = 0, nr = b->c.level > 1 ? 2 : 16;
 319         struct bkey_s_c k;
 320         struct bkey_buf tmp;
 321
 322         BUG_ON(!b->c.level);
 323
 324         bch2_bkey_buf_init(&tmp);
 325
 326         while (i < nr &&
 327                (k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 328                 bch2_bkey_buf_reassemble(&tmp, c, k);
 329
 330                 bch2_btree_node_prefetch(c, NULL, NULL, tmp.k,
 331                                         b->c.btree_id, b->c.level - 1);
 332
 333                 bch2_btree_and_journal_iter_advance(&iter);
 334                 i++;
 335         }
 336
 337         bch2_bkey_buf_exit(&tmp, c);
 338 }
 339
 340 static int bch2_btree_and_journal_walk_recurse(struct bch_fs *c, struct btree *b,
 341                                 enum btree_id btree_id,
 342                                 btree_walk_key_fn key_fn)
 343 {
 344         struct btree_and_journal_iter iter;
 345         struct bkey_s_c k;
 346         struct bkey_buf tmp;
 347         struct btree *child;
 348         int ret = 0;
 349
 350         bch2_bkey_buf_init(&tmp);
 351         bch2_btree_and_journal_iter_init_node_iter(&iter, c, b);
 352
 353         while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 354                 if (b->c.level) {
 355                         bch2_bkey_buf_reassemble(&tmp, c, k);
 356
 357                         child = bch2_btree_node_get_noiter(c, tmp.k,
 358                                                 b->c.btree_id, b->c.level - 1,
 359                                                 false);
 360
 361                         ret = PTR_ERR_OR_ZERO(child);
 362                         if (ret)
 363                                 break;
 364
 365                         btree_and_journal_iter_prefetch(c, b, iter);
 366
 367                         ret = bch2_btree_and_journal_walk_recurse(c, child,
 368                                         btree_id, key_fn);
 369                         six_unlock_read(&child->c.lock);
 370                 } else {
 371                         ret = key_fn(c, k);
 372                 }
 373
 374                 if (ret)
 375                         break;
 376
 377                 bch2_btree_and_journal_iter_advance(&iter);
 378         }
 379
 380         bch2_btree_and_journal_iter_exit(&iter);
 381         bch2_bkey_buf_exit(&tmp, c);
 382         return ret;
 383 }
 384
 385 int bch2_btree_and_journal_walk(struct bch_fs *c, enum btree_id btree_id,
 386                                 btree_walk_key_fn key_fn)
 387 {
 388         struct btree *b = c->btree_roots[btree_id].b;
 389         int ret = 0;
 390
 391         if (btree_node_fake(b))
 392                 return 0;
 393
 394         six_lock_read(&b->c.lock, NULL, NULL);
 395         ret = bch2_btree_and_journal_walk_recurse(c, b, btree_id, key_fn);
 396         six_unlock_read(&b->c.lock);
 397
 398         return ret;
 399 }
 400
 401 /* sort and dedup all keys in the journal: */
 402
 403 void bch2_journal_entries_free(struct list_head *list)
 404 {
 405
 406         while (!list_empty(list)) {
 407                 struct journal_replay *i =
 408                         list_first_entry(list, struct journal_replay, list);
 409                 list_del(&i->list);
 410                 kvpfree(i, offsetof(struct journal_replay, j) +
 411                         vstruct_bytes(&i->j));
 412         }
 413 }
 414
 415 /*
 416  * When keys compare equal, oldest compares first:
 417  */
 418 static int journal_sort_key_cmp(const void *_l, const void *_r)
 419 {
 420         const struct journal_key *l = _l;
 421         const struct journal_key *r = _r;
 422
 423         return  cmp_int(l->btree_id,    r->btree_id) ?:
 424                 cmp_int(l->level,       r->level) ?:
 425                 bpos_cmp(l->k->k.p, r->k->k.p) ?:
 426                 cmp_int(l->journal_seq, r->journal_seq) ?:
 427                 cmp_int(l->journal_offset, r->journal_offset);
 428 }
 429
 430 void bch2_journal_keys_free(struct journal_keys *keys)
 431 {
 432         struct journal_key *i;
 433
 434         for (i = keys->d; i < keys->d + keys->nr; i++)
 435                 if (i->allocated)
 436                         kfree(i->k);
 437
 438         kvfree(keys->d);
 439         keys->d = NULL;
 440         keys->nr = 0;
 441 }
 442
 443 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 444 {
 445         struct journal_replay *i;
 446         struct jset_entry *entry;
 447         struct bkey_i *k, *_n;
 448         struct journal_keys keys = { NULL };
 449         struct journal_key *src, *dst;
 450         size_t nr_keys = 0;
 451
 452         if (list_empty(journal_entries))
 453                 return keys;
 454
 455         list_for_each_entry(i, journal_entries, list) {
 456                 if (i->ignore)
 457                         continue;
 458
 459                 if (!keys.journal_seq_base)
 460                         keys.journal_seq_base = le64_to_cpu(i->j.seq);
 461
 462                 for_each_jset_key(k, _n, entry, &i->j)
 463                         nr_keys++;
 464         }
 465
 466         keys.size = roundup_pow_of_two(nr_keys);
 467
 468         keys.d = kvmalloc(sizeof(keys.d[0]) * keys.size, GFP_KERNEL);
 469         if (!keys.d)
 470                 goto err;
 471
 472         list_for_each_entry(i, journal_entries, list) {
 473                 if (i->ignore)
 474                         continue;
 475
 476                 BUG_ON(le64_to_cpu(i->j.seq) - keys.journal_seq_base > U32_MAX);
 477
 478                 for_each_jset_key(k, _n, entry, &i->j)
 479                         keys.d[keys.nr++] = (struct journal_key) {
 480                                 .btree_id       = entry->btree_id,
 481                                 .level          = entry->level,
 482                                 .k              = k,
 483                                 .journal_seq    = le64_to_cpu(i->j.seq) -
 484                                         keys.journal_seq_base,
 485                                 .journal_offset = k->_data - i->j._data,
 486                         };
 487         }
 488
 489         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 490
 491         src = dst = keys.d;
 492         while (src < keys.d + keys.nr) {
 493                 while (src + 1 < keys.d + keys.nr &&
 494                        src[0].btree_id  == src[1].btree_id &&
 495                        src[0].level     == src[1].level &&
 496                        !bpos_cmp(src[0].k->k.p, src[1].k->k.p))
 497                         src++;
 498
 499                 *dst++ = *src++;
 500         }
 501
 502         keys.nr = dst - keys.d;
 503 err:
 504         return keys;
 505 }
 506
 507 /* journal replay: */
 508
 509 static void replay_now_at(struct journal *j, u64 seq)
 510 {
 511         BUG_ON(seq < j->replay_journal_seq);
 512         BUG_ON(seq > j->replay_journal_seq_end);
 513
 514         while (j->replay_journal_seq < seq)
 515                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 516 }
 517
 518 static int __bch2_journal_replay_key(struct btree_trans *trans,
 519                                      enum btree_id id, unsigned level,
 520                                      struct bkey_i *k)
 521 {
 522         struct btree_iter iter;
 523         int ret;
 524
 525         bch2_trans_node_iter_init(trans, &iter, id, k->k.p,
 526                                   BTREE_MAX_DEPTH, level,
 527                                   BTREE_ITER_INTENT|
 528                                   BTREE_ITER_NOT_EXTENTS);
 529         ret   = bch2_btree_iter_traverse(&iter) ?:
 530                 bch2_trans_update(trans, &iter, k, BTREE_TRIGGER_NORUN);
 531         bch2_trans_iter_exit(trans, &iter);
 532         return ret;
 533 }
 534
 535 static int bch2_journal_replay_key(struct bch_fs *c, struct journal_key *k)
 536 {
 537         unsigned commit_flags = BTREE_INSERT_NOFAIL|
 538                 BTREE_INSERT_LAZY_RW;
 539
 540         if (!k->allocated)
 541                 commit_flags |= BTREE_INSERT_JOURNAL_REPLAY;
 542
 543         return bch2_trans_do(c, NULL, NULL, commit_flags,
 544                              __bch2_journal_replay_key(&trans, k->btree_id, k->level, k->k));
 545 }
 546
 547 static int __bch2_alloc_replay_key(struct btree_trans *trans, struct bkey_i *k)
 548 {
 549         struct btree_iter iter;
 550         int ret;
 551
 552         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, k->k.p,
 553                              BTREE_ITER_CACHED|
 554                              BTREE_ITER_CACHED_NOFILL|
 555                              BTREE_ITER_INTENT);
 556         ret   = bch2_btree_iter_traverse(&iter) ?:
 557                 bch2_trans_update(trans, &iter, k, BTREE_TRIGGER_NORUN);
 558         bch2_trans_iter_exit(trans, &iter);
 559         return ret;
 560 }
 561
 562 static int bch2_alloc_replay_key(struct bch_fs *c, struct bkey_i *k)
 563 {
 564         return bch2_trans_do(c, NULL, NULL,
 565                              BTREE_INSERT_NOFAIL|
 566                              BTREE_INSERT_USE_RESERVE|
 567                              BTREE_INSERT_LAZY_RW|
 568                              BTREE_INSERT_JOURNAL_REPLAY,
 569                         __bch2_alloc_replay_key(&trans, k));
 570 }
 571
 572 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 573 {
 574         const struct journal_key *l = _l;
 575         const struct journal_key *r = _r;
 576
 577         return  cmp_int(r->level,       l->level) ?:
 578                 cmp_int(l->journal_seq, r->journal_seq) ?:
 579                 cmp_int(l->btree_id,    r->btree_id) ?:
 580                 bpos_cmp(l->k->k.p,     r->k->k.p);
 581 }
 582
 583 static int bch2_journal_replay(struct bch_fs *c,
 584                                struct journal_keys keys)
 585 {
 586         struct journal *j = &c->journal;
 587         struct journal_key *i;
 588         u64 seq;
 589         int ret;
 590
 591         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
 592
 593         if (keys.nr)
 594                 replay_now_at(j, keys.journal_seq_base);
 595
 596         seq = j->replay_journal_seq;
 597
 598         /*
 599          * First replay updates to the alloc btree - these will only update the
 600          * btree key cache:
 601          */
 602         for_each_journal_key(keys, i) {
 603                 cond_resched();
 604
 605                 if (!i->level && i->btree_id == BTREE_ID_alloc) {
 606                         j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
 607                         ret = bch2_alloc_replay_key(c, i->k);
 608                         if (ret)
 609                                 goto err;
 610                 }
 611         }
 612
 613         /*
 614          * Next replay updates to interior btree nodes:
 615          */
 616         for_each_journal_key(keys, i) {
 617                 cond_resched();
 618
 619                 if (i->level) {
 620                         j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
 621                         ret = bch2_journal_replay_key(c, i);
 622                         if (ret)
 623                                 goto err;
 624                 }
 625         }
 626
 627         /*
 628          * Now that the btree is in a consistent state, we can start journal
 629          * reclaim (which will be flushing entries from the btree key cache back
 630          * to the btree:
 631          */
 632         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
 633         set_bit(JOURNAL_RECLAIM_STARTED, &j->flags);
 634         journal_reclaim_kick(j);
 635
 636         j->replay_journal_seq = seq;
 637
 638         /*
 639          * Now replay leaf node updates:
 640          */
 641         for_each_journal_key(keys, i) {
 642                 cond_resched();
 643
 644                 if (i->level || i->btree_id == BTREE_ID_alloc)
 645                         continue;
 646
 647                 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 648
 649                 ret = bch2_journal_replay_key(c, i);
 650                 if (ret)
 651                         goto err;
 652         }
 653
 654         replay_now_at(j, j->replay_journal_seq_end);
 655         j->replay_journal_seq = 0;
 656
 657         bch2_journal_set_replay_done(j);
 658         bch2_journal_flush_all_pins(j);
 659         return bch2_journal_error(j);
 660 err:
 661         bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
 662                 ret, bch2_btree_ids[i->btree_id], i->level);
 663         return ret;
 664 }
 665
 666 /* journal replay early: */
 667
 668 static int journal_replay_entry_early(struct bch_fs *c,
 669                                       struct jset_entry *entry)
 670 {
 671         int ret = 0;
 672
 673         switch (entry->type) {
 674         case BCH_JSET_ENTRY_btree_root: {
 675                 struct btree_root *r;
 676
 677                 if (entry->btree_id >= BTREE_ID_NR) {
 678                         bch_err(c, "filesystem has unknown btree type %u",
 679                                 entry->btree_id);
 680                         return -EINVAL;
 681                 }
 682
 683                 r = &c->btree_roots[entry->btree_id];
 684
 685                 if (entry->u64s) {
 686                         r->level = entry->level;
 687                         bkey_copy(&r->key, &entry->start[0]);
 688                         r->error = 0;
 689                 } else {
 690                         r->error = -EIO;
 691                 }
 692                 r->alive = true;
 693                 break;
 694         }
 695         case BCH_JSET_ENTRY_usage: {
 696                 struct jset_entry_usage *u =
 697                         container_of(entry, struct jset_entry_usage, entry);
 698
 699                 switch (entry->btree_id) {
 700                 case FS_USAGE_RESERVED:
 701                         if (entry->level < BCH_REPLICAS_MAX)
 702                                 c->usage_base->persistent_reserved[entry->level] =
 703                                         le64_to_cpu(u->v);
 704                         break;
 705                 case FS_USAGE_INODES:
 706                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 707                         break;
 708                 case FS_USAGE_KEY_VERSION:
 709                         atomic64_set(&c->key_version,
 710                                      le64_to_cpu(u->v));
 711                         break;
 712                 }
 713
 714                 break;
 715         }
 716         case BCH_JSET_ENTRY_data_usage: {
 717                 struct jset_entry_data_usage *u =
 718                         container_of(entry, struct jset_entry_data_usage, entry);
 719
 720                 ret = bch2_replicas_set_usage(c, &u->r,
 721                                               le64_to_cpu(u->v));
 722                 break;
 723         }
 724         case BCH_JSET_ENTRY_dev_usage: {
 725                 struct jset_entry_dev_usage *u =
 726                         container_of(entry, struct jset_entry_dev_usage, entry);
 727                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 728                 unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
 729                 unsigned nr_types = (bytes - sizeof(struct jset_entry_dev_usage)) /
 730                         sizeof(struct jset_entry_dev_usage_type);
 731                 unsigned i;
 732
 733                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 734                 ca->usage_base->buckets_unavailable     = le64_to_cpu(u->buckets_unavailable);
 735
 736                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 737                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 738                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 739                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 740                 }
 741
 742                 break;
 743         }
 744         case BCH_JSET_ENTRY_blacklist: {
 745                 struct jset_entry_blacklist *bl_entry =
 746                         container_of(entry, struct jset_entry_blacklist, entry);
 747
 748                 ret = bch2_journal_seq_blacklist_add(c,
 749                                 le64_to_cpu(bl_entry->seq),
 750                                 le64_to_cpu(bl_entry->seq) + 1);
 751                 break;
 752         }
 753         case BCH_JSET_ENTRY_blacklist_v2: {
 754                 struct jset_entry_blacklist_v2 *bl_entry =
 755                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 756
 757                 ret = bch2_journal_seq_blacklist_add(c,
 758                                 le64_to_cpu(bl_entry->start),
 759                                 le64_to_cpu(bl_entry->end) + 1);
 760                 break;
 761         }
 762         case BCH_JSET_ENTRY_clock: {
 763                 struct jset_entry_clock *clock =
 764                         container_of(entry, struct jset_entry_clock, entry);
 765
 766                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 767         }
 768         }
 769
 770         return ret;
 771 }
 772
 773 static int journal_replay_early(struct bch_fs *c,
 774                                 struct bch_sb_field_clean *clean,
 775                                 struct list_head *journal)
 776 {
 777         struct journal_replay *i;
 778         struct jset_entry *entry;
 779         int ret;
 780
 781         if (clean) {
 782                 for (entry = clean->start;
 783                      entry != vstruct_end(&clean->field);
 784                      entry = vstruct_next(entry)) {
 785                         ret = journal_replay_entry_early(c, entry);
 786                         if (ret)
 787                                 return ret;
 788                 }
 789         } else {
 790                 list_for_each_entry(i, journal, list) {
 791                         if (i->ignore)
 792                                 continue;
 793
 794                         vstruct_for_each(&i->j, entry) {
 795                                 ret = journal_replay_entry_early(c, entry);
 796                                 if (ret)
 797                                         return ret;
 798                         }
 799                 }
 800         }
 801
 802         bch2_fs_usage_initialize(c);
 803
 804         return 0;
 805 }
 806
 807 /* sb clean section: */
 808
 809 static struct bkey_i *btree_root_find(struct bch_fs *c,
 810                                       struct bch_sb_field_clean *clean,
 811                                       struct jset *j,
 812                                       enum btree_id id, unsigned *level)
 813 {
 814         struct bkey_i *k;
 815         struct jset_entry *entry, *start, *end;
 816
 817         if (clean) {
 818                 start = clean->start;
 819                 end = vstruct_end(&clean->field);
 820         } else {
 821                 start = j->start;
 822                 end = vstruct_last(j);
 823         }
 824
 825         for (entry = start; entry < end; entry = vstruct_next(entry))
 826                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 827                     entry->btree_id == id)
 828                         goto found;
 829
 830         return NULL;
 831 found:
 832         if (!entry->u64s)
 833                 return ERR_PTR(-EINVAL);
 834
 835         k = entry->start;
 836         *level = entry->level;
 837         return k;
 838 }
 839
 840 static int verify_superblock_clean(struct bch_fs *c,
 841                                    struct bch_sb_field_clean **cleanp,
 842                                    struct jset *j)
 843 {
 844         unsigned i;
 845         struct bch_sb_field_clean *clean = *cleanp;
 846         int ret = 0;
 847
 848         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 849                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 850                         le64_to_cpu(clean->journal_seq),
 851                         le64_to_cpu(j->seq))) {
 852                 kfree(clean);
 853                 *cleanp = NULL;
 854                 return 0;
 855         }
 856
 857         for (i = 0; i < BTREE_ID_NR; i++) {
 858                 char buf1[200], buf2[200];
 859                 struct bkey_i *k1, *k2;
 860                 unsigned l1 = 0, l2 = 0;
 861
 862                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 863                 k2 = btree_root_find(c, NULL, j, i, &l2);
 864
 865                 if (!k1 && !k2)
 866                         continue;
 867
 868                 mustfix_fsck_err_on(!k1 || !k2 ||
 869                                     IS_ERR(k1) ||
 870                                     IS_ERR(k2) ||
 871                                     k1->k.u64s != k2->k.u64s ||
 872                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 873                                     l1 != l2, c,
 874                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 875                         "sb:      l=%u %s\n"
 876                         "journal: l=%u %s\n", i,
 877                         l1, (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(k1)), buf1),
 878                         l2, (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(k2)), buf2));
 879         }
 880 fsck_err:
 881         return ret;
 882 }
 883
 884 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 885 {
 886         struct bch_sb_field_clean *clean, *sb_clean;
 887         int ret;
 888
 889         mutex_lock(&c->sb_lock);
 890         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 891
 892         if (fsck_err_on(!sb_clean, c,
 893                         "superblock marked clean but clean section not present")) {
 894                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 895                 c->sb.clean = false;
 896                 mutex_unlock(&c->sb_lock);
 897                 return NULL;
 898         }
 899
 900         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 901                         GFP_KERNEL);
 902         if (!clean) {
 903                 mutex_unlock(&c->sb_lock);
 904                 return ERR_PTR(-ENOMEM);
 905         }
 906
 907         ret = bch2_sb_clean_validate(c, clean, READ);
 908         if (ret) {
 909                 mutex_unlock(&c->sb_lock);
 910                 return ERR_PTR(ret);
 911         }
 912
 913         mutex_unlock(&c->sb_lock);
 914
 915         return clean;
 916 fsck_err:
 917         mutex_unlock(&c->sb_lock);
 918         return ERR_PTR(ret);
 919 }
 920
 921 static int read_btree_roots(struct bch_fs *c)
 922 {
 923         unsigned i;
 924         int ret = 0;
 925
 926         for (i = 0; i < BTREE_ID_NR; i++) {
 927                 struct btree_root *r = &c->btree_roots[i];
 928
 929                 if (!r->alive)
 930                         continue;
 931
 932                 if (i == BTREE_ID_alloc &&
 933                     c->opts.reconstruct_alloc) {
 934                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 935                         continue;
 936                 }
 937
 938                 if (r->error) {
 939                         __fsck_err(c, i == BTREE_ID_alloc
 940                                    ? FSCK_CAN_IGNORE : 0,
 941                                    "invalid btree root %s",
 942                                    bch2_btree_ids[i]);
 943                         if (i == BTREE_ID_alloc)
 944                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 945                 }
 946
 947                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 948                 if (ret) {
 949                         __fsck_err(c, i == BTREE_ID_alloc
 950                                    ? FSCK_CAN_IGNORE : 0,
 951                                    "error reading btree root %s",
 952                                    bch2_btree_ids[i]);
 953                         if (i == BTREE_ID_alloc)
 954                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 955                 }
 956         }
 957
 958         for (i = 0; i < BTREE_ID_NR; i++)
 959                 if (!c->btree_roots[i].b)
 960                         bch2_btree_root_alloc(c, i);
 961 fsck_err:
 962         return ret;
 963 }
 964
 965 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
 966 {
 967         struct bkey_i_snapshot  root_snapshot;
 968         struct bkey_i_subvolume root_volume;
 969         int ret;
 970
 971         bkey_snapshot_init(&root_snapshot.k_i);
 972         root_snapshot.k.p.offset = U32_MAX;
 973         root_snapshot.v.flags   = 0;
 974         root_snapshot.v.parent  = 0;
 975         root_snapshot.v.subvol  = BCACHEFS_ROOT_SUBVOL;
 976         root_snapshot.v.pad     = 0;
 977         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
 978
 979         ret = bch2_btree_insert(c, BTREE_ID_snapshots,
 980                                 &root_snapshot.k_i,
 981                                 NULL, NULL, 0);
 982         if (ret)
 983                 return ret;
 984
 985
 986         bkey_subvolume_init(&root_volume.k_i);
 987         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
 988         root_volume.v.flags     = 0;
 989         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
 990         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
 991
 992         ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
 993                                 &root_volume.k_i,
 994                                 NULL, NULL, 0);
 995         if (ret)
 996                 return ret;
 997
 998         return 0;
 999 }
1000
1001 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
1002 {
1003         struct bch_fs *c = trans->c;
1004         struct btree_iter iter;
1005         struct bkey_s_c k;
1006         struct bch_inode_unpacked inode;
1007         int ret;
1008
1009         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
1010                              SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
1011         k = bch2_btree_iter_peek_slot(&iter);
1012         ret = bkey_err(k);
1013         if (ret)
1014                 goto err;
1015
1016         if (k.k->type != KEY_TYPE_inode) {
1017                 bch_err(c, "root inode not found");
1018                 ret = -ENOENT;
1019                 goto err;
1020         }
1021
1022         ret = bch2_inode_unpack(bkey_s_c_to_inode(k), &inode);
1023         BUG_ON(ret);
1024
1025         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1026
1027         ret = bch2_inode_write(trans, &iter, &inode);
1028 err:
1029         bch2_trans_iter_exit(trans, &iter);
1030         return ret;
1031 }
1032
1033 int bch2_fs_recovery(struct bch_fs *c)
1034 {
1035         const char *err = "cannot allocate memory";
1036         struct bch_sb_field_clean *clean = NULL;
1037         struct jset *last_journal_entry = NULL;
1038         u64 blacklist_seq, journal_seq;
1039         bool write_sb = false;
1040         int ret = 0;
1041
1042         if (c->sb.clean)
1043                 clean = read_superblock_clean(c);
1044         ret = PTR_ERR_OR_ZERO(clean);
1045         if (ret)
1046                 goto err;
1047
1048         if (c->sb.clean)
1049                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
1050                          le64_to_cpu(clean->journal_seq));
1051
1052         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
1053                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
1054                 ret = -EINVAL;
1055                 goto err;
1056         }
1057
1058         if (!c->sb.clean &&
1059             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1060                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1061                 ret = -EINVAL;
1062                 goto err;
1063         }
1064
1065         if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1066                 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
1067                 ret = -EINVAL;
1068                 goto err;
1069
1070         }
1071
1072         if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
1073                 bch_info(c, "alloc_v2 feature bit not set, fsck required");
1074                 c->opts.fsck = true;
1075                 c->opts.fix_errors = FSCK_OPT_YES;
1076         }
1077
1078         if (!c->replicas.entries ||
1079             c->opts.rebuild_replicas) {
1080                 bch_info(c, "building replicas info");
1081                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1082         }
1083
1084         if (c->sb.version < bcachefs_metadata_version_inode_backpointers) {
1085                 bch_info(c, "version prior to inode backpointers, upgrade and fsck required");
1086                 c->opts.version_upgrade = true;
1087                 c->opts.fsck            = true;
1088                 c->opts.fix_errors      = FSCK_OPT_YES;
1089         } else if (c->sb.version < bcachefs_metadata_version_subvol_dirent) {
1090                 bch_info(c, "filesystem version is prior to subvol_dirent - upgrading");
1091                 c->opts.version_upgrade = true;
1092                 c->opts.fsck            = true;
1093         }
1094
1095         ret = bch2_blacklist_table_initialize(c);
1096         if (ret) {
1097                 bch_err(c, "error initializing blacklist table");
1098                 goto err;
1099         }
1100
1101         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1102                 struct journal_replay *i;
1103
1104                 ret = bch2_journal_read(c, &c->journal_entries,
1105                                         &blacklist_seq, &journal_seq);
1106                 if (ret)
1107                         goto err;
1108
1109                 list_for_each_entry_reverse(i, &c->journal_entries, list)
1110                         if (!i->ignore) {
1111                                 last_journal_entry = &i->j;
1112                                 break;
1113                         }
1114
1115                 if (mustfix_fsck_err_on(c->sb.clean &&
1116                                         last_journal_entry &&
1117                                         !journal_entry_empty(last_journal_entry), c,
1118                                 "filesystem marked clean but journal not empty")) {
1119                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1120                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1121                         c->sb.clean = false;
1122                 }
1123
1124                 if (!last_journal_entry) {
1125                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
1126                         goto use_clean;
1127                 }
1128
1129                 c->journal_keys = journal_keys_sort(&c->journal_entries);
1130                 if (!c->journal_keys.d) {
1131                         ret = -ENOMEM;
1132                         goto err;
1133                 }
1134
1135                 if (c->sb.clean && last_journal_entry) {
1136                         ret = verify_superblock_clean(c, &clean,
1137                                                       last_journal_entry);
1138                         if (ret)
1139                                 goto err;
1140                 }
1141         } else {
1142 use_clean:
1143                 if (!clean) {
1144                         bch_err(c, "no superblock clean section found");
1145                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1146                         goto err;
1147
1148                 }
1149                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1150         }
1151
1152         if (c->opts.reconstruct_alloc) {
1153                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1154                 drop_alloc_keys(&c->journal_keys);
1155         }
1156
1157         zero_out_btree_mem_ptr(&c->journal_keys);
1158
1159         ret = journal_replay_early(c, clean, &c->journal_entries);
1160         if (ret)
1161                 goto err;
1162
1163         /*
1164          * After an unclean shutdown, skip then next few journal sequence
1165          * numbers as they may have been referenced by btree writes that
1166          * happened before their corresponding journal writes - those btree
1167          * writes need to be ignored, by skipping and blacklisting the next few
1168          * journal sequence numbers:
1169          */
1170         if (!c->sb.clean)
1171                 journal_seq += 8;
1172
1173         if (blacklist_seq != journal_seq) {
1174                 ret = bch2_journal_seq_blacklist_add(c,
1175                                         blacklist_seq, journal_seq);
1176                 if (ret) {
1177                         bch_err(c, "error creating new journal seq blacklist entry");
1178                         goto err;
1179                 }
1180         }
1181
1182         ret = bch2_fs_journal_start(&c->journal, journal_seq,
1183                                     &c->journal_entries);
1184         if (ret)
1185                 goto err;
1186
1187         ret = read_btree_roots(c);
1188         if (ret)
1189                 goto err;
1190
1191         bch_verbose(c, "starting alloc read");
1192         err = "error reading allocation information";
1193         ret = bch2_alloc_read(c);
1194         if (ret)
1195                 goto err;
1196         bch_verbose(c, "alloc read done");
1197
1198         bch_verbose(c, "starting stripes_read");
1199         err = "error reading stripes";
1200         ret = bch2_stripes_read(c);
1201         if (ret)
1202                 goto err;
1203         bch_verbose(c, "stripes_read done");
1204
1205         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1206
1207         if (c->opts.fsck ||
1208             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)) ||
1209             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_metadata)) ||
1210             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1211                 bool metadata_only = c->opts.norecovery;
1212
1213                 bch_info(c, "starting mark and sweep");
1214                 err = "error in mark and sweep";
1215                 ret = bch2_gc(c, true, metadata_only);
1216                 if (ret)
1217                         goto err;
1218                 bch_verbose(c, "mark and sweep done");
1219         }
1220
1221         bch2_stripes_heap_start(c);
1222
1223         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1224         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1225
1226         /*
1227          * Skip past versions that might have possibly been used (as nonces),
1228          * but hadn't had their pointers written:
1229          */
1230         if (c->sb.encryption_type && !c->sb.clean)
1231                 atomic64_add(1 << 16, &c->key_version);
1232
1233         if (c->opts.norecovery)
1234                 goto out;
1235
1236         bch_verbose(c, "starting journal replay");
1237         err = "journal replay failed";
1238         ret = bch2_journal_replay(c, c->journal_keys);
1239         if (ret)
1240                 goto err;
1241         bch_verbose(c, "journal replay done");
1242
1243         if (test_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags) &&
1244             !c->opts.nochanges) {
1245                 /*
1246                  * note that even when filesystem was clean there might be work
1247                  * to do here, if we ran gc (because of fsck) which recalculated
1248                  * oldest_gen:
1249                  */
1250                 bch_verbose(c, "writing allocation info");
1251                 err = "error writing out alloc info";
1252                 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW) ?:
1253                         bch2_alloc_write(c, BTREE_INSERT_LAZY_RW);
1254                 if (ret) {
1255                         bch_err(c, "error writing alloc info");
1256                         goto err;
1257                 }
1258                 bch_verbose(c, "alloc write done");
1259         }
1260
1261         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1262                 bch2_fs_lazy_rw(c);
1263
1264                 err = "error creating root snapshot node";
1265                 ret = bch2_fs_initialize_subvolumes(c);
1266                 if (ret)
1267                         goto err;
1268         }
1269
1270         bch_verbose(c, "reading snapshots table");
1271         err = "error reading snapshots table";
1272         ret = bch2_fs_snapshots_start(c);
1273         if (ret)
1274                 goto err;
1275         bch_verbose(c, "reading snapshots done");
1276
1277         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1278                 /* set bi_subvol on root inode */
1279                 err = "error upgrade root inode for subvolumes";
1280                 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1281                                     bch2_fs_upgrade_for_subvolumes(&trans));
1282                 if (ret)
1283                         goto err;
1284         }
1285
1286         if (c->opts.fsck) {
1287                 bch_info(c, "starting fsck");
1288                 err = "error in fsck";
1289                 ret = bch2_fsck_full(c);
1290                 if (ret)
1291                         goto err;
1292                 bch_verbose(c, "fsck done");
1293         } else if (!c->sb.clean) {
1294                 bch_verbose(c, "checking for deleted inodes");
1295                 err = "error in recovery";
1296                 ret = bch2_fsck_walk_inodes_only(c);
1297                 if (ret)
1298                         goto err;
1299                 bch_verbose(c, "check inodes done");
1300         }
1301
1302         if (enabled_qtypes(c)) {
1303                 bch_verbose(c, "reading quotas");
1304                 ret = bch2_fs_quota_read(c);
1305                 if (ret)
1306                         goto err;
1307                 bch_verbose(c, "quotas done");
1308         }
1309
1310         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1311             !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1312                 struct bch_move_stats stats;
1313
1314                 bch_move_stats_init(&stats, "recovery");
1315
1316                 bch_info(c, "scanning for old btree nodes");
1317                 ret = bch2_fs_read_write(c);
1318                 if (ret)
1319                         goto err;
1320
1321                 ret = bch2_scan_old_btree_nodes(c, &stats);
1322                 if (ret)
1323                         goto err;
1324                 bch_info(c, "scanning for old btree nodes done");
1325         }
1326
1327         mutex_lock(&c->sb_lock);
1328         if (c->opts.version_upgrade) {
1329                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1330                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1331                 write_sb = true;
1332         }
1333
1334         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1335                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1336                 write_sb = true;
1337         }
1338
1339         if (c->opts.fsck &&
1340             !test_bit(BCH_FS_ERROR, &c->flags) &&
1341             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1342                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1343                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1344                 write_sb = true;
1345         }
1346
1347         if (write_sb)
1348                 bch2_write_super(c);
1349         mutex_unlock(&c->sb_lock);
1350
1351         if (c->journal_seq_blacklist_table &&
1352             c->journal_seq_blacklist_table->nr > 128)
1353                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1354
1355         ret = 0;
1356 out:
1357         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1358         bch2_flush_fsck_errs(c);
1359
1360         if (!c->opts.keep_journal) {
1361                 bch2_journal_keys_free(&c->journal_keys);
1362                 bch2_journal_entries_free(&c->journal_entries);
1363         }
1364         kfree(clean);
1365         if (ret)
1366                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1367         else
1368                 bch_verbose(c, "ret %i", ret);
1369         return ret;
1370 err:
1371 fsck_err:
1372         bch2_fs_emergency_read_only(c);
1373         goto out;
1374 }
1375
1376 int bch2_fs_initialize(struct bch_fs *c)
1377 {
1378         struct bch_inode_unpacked root_inode, lostfound_inode;
1379         struct bkey_inode_buf packed_inode;
1380         struct qstr lostfound = QSTR("lost+found");
1381         const char *err = "cannot allocate memory";
1382         struct bch_dev *ca;
1383         LIST_HEAD(journal);
1384         unsigned i;
1385         int ret;
1386
1387         bch_notice(c, "initializing new filesystem");
1388
1389         mutex_lock(&c->sb_lock);
1390         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1391         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1392
1393         if (c->opts.version_upgrade) {
1394                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1395                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1396                 bch2_write_super(c);
1397         }
1398
1399         for_each_online_member(ca, c, i)
1400                 bch2_mark_dev_superblock(c, ca, 0);
1401         mutex_unlock(&c->sb_lock);
1402
1403         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1404         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1405
1406         for (i = 0; i < BTREE_ID_NR; i++)
1407                 bch2_btree_root_alloc(c, i);
1408
1409         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
1410         set_bit(JOURNAL_RECLAIM_STARTED, &c->journal.flags);
1411
1412         err = "unable to allocate journal buckets";
1413         for_each_online_member(ca, c, i) {
1414                 ret = bch2_dev_journal_alloc(ca);
1415                 if (ret) {
1416                         percpu_ref_put(&ca->io_ref);
1417                         goto err;
1418                 }
1419         }
1420
1421         /*
1422          * journal_res_get() will crash if called before this has
1423          * set up the journal.pin FIFO and journal.cur pointer:
1424          */
1425         bch2_fs_journal_start(&c->journal, 1, &journal);
1426         bch2_journal_set_replay_done(&c->journal);
1427
1428         err = "error going read-write";
1429         ret = bch2_fs_read_write_early(c);
1430         if (ret)
1431                 goto err;
1432
1433         /*
1434          * Write out the superblock and journal buckets, now that we can do
1435          * btree updates
1436          */
1437         err = "error marking superblock and journal";
1438         for_each_member_device(ca, c, i) {
1439                 ret = bch2_trans_mark_dev_sb(c, ca);
1440                 if (ret) {
1441                         percpu_ref_put(&ca->ref);
1442                         goto err;
1443                 }
1444         }
1445
1446         err = "error creating root snapshot node";
1447         ret = bch2_fs_initialize_subvolumes(c);
1448         if (ret)
1449                 goto err;
1450
1451         bch_verbose(c, "reading snapshots table");
1452         err = "error reading snapshots table";
1453         ret = bch2_fs_snapshots_start(c);
1454         if (ret)
1455                 goto err;
1456         bch_verbose(c, "reading snapshots done");
1457
1458         bch2_inode_init(c, &root_inode, 0, 0,
1459                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1460         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1461         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1462         bch2_inode_pack(c, &packed_inode, &root_inode);
1463         packed_inode.inode.k.p.snapshot = U32_MAX;
1464
1465         err = "error creating root directory";
1466         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1467                                 &packed_inode.inode.k_i,
1468                                 NULL, NULL, 0);
1469         if (ret)
1470                 goto err;
1471
1472         bch2_inode_init_early(c, &lostfound_inode);
1473
1474         err = "error creating lost+found";
1475         ret = bch2_trans_do(c, NULL, NULL, 0,
1476                 bch2_create_trans(&trans,
1477                                   BCACHEFS_ROOT_SUBVOL_INUM,
1478                                   &root_inode, &lostfound_inode,
1479                                   &lostfound,
1480                                   0, 0, S_IFDIR|0700, 0,
1481                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1482         if (ret) {
1483                 bch_err(c, "error creating lost+found");
1484                 goto err;
1485         }
1486
1487         if (enabled_qtypes(c)) {
1488                 ret = bch2_fs_quota_read(c);
1489                 if (ret)
1490                         goto err;
1491         }
1492
1493         err = "error writing first journal entry";
1494         ret = bch2_journal_meta(&c->journal);
1495         if (ret)
1496                 goto err;
1497
1498         mutex_lock(&c->sb_lock);
1499         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1500         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1501
1502         bch2_write_super(c);
1503         mutex_unlock(&c->sb_lock);
1504
1505         return 0;
1506 err:
1507         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1508         return ret;
1509 }