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_take(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_insert(struct bch_fs *c, enum btree_id id,
 161                             unsigned level, struct bkey_i *k)
 162 {
 163         struct bkey_i *n;
 164         int ret;
 165
 166         n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
 167         if (!n)
 168                 return -ENOMEM;
 169
 170         bkey_copy(n, k);
 171         ret = bch2_journal_key_insert_take(c, id, level, n);
 172         if (ret)
 173                 kfree(n);
 174         return ret;
 175 }
 176
 177 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
 178                             unsigned level, struct bpos pos)
 179 {
 180         struct bkey_i whiteout;
 181
 182         bkey_init(&whiteout.k);
 183         whiteout.k.p = pos;
 184
 185         return bch2_journal_key_insert(c, id, level, &whiteout);
 186 }
 187
 188 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
 189                                   unsigned level, struct bpos pos)
 190 {
 191         struct journal_keys *keys = &c->journal_keys;
 192         size_t idx = journal_key_search(keys, btree, level, pos);
 193
 194         if (idx < keys->nr &&
 195             keys->d[idx].btree_id       == btree &&
 196             keys->d[idx].level          == level &&
 197             !bpos_cmp(keys->d[idx].k->k.p, pos))
 198                 keys->d[idx].overwritten = true;
 199 }
 200
 201 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
 202 {
 203         struct journal_key *k = iter->idx - iter->keys->nr
 204                 ? iter->keys->d + iter->idx : NULL;
 205
 206         if (k &&
 207             k->btree_id == iter->btree_id &&
 208             k->level    == iter->level)
 209                 return k->k;
 210
 211         iter->idx = iter->keys->nr;
 212         return NULL;
 213 }
 214
 215 static void bch2_journal_iter_advance(struct journal_iter *iter)
 216 {
 217         if (iter->idx < iter->keys->nr)
 218                 iter->idx++;
 219 }
 220
 221 static void bch2_journal_iter_exit(struct journal_iter *iter)
 222 {
 223         list_del(&iter->list);
 224 }
 225
 226 static void bch2_journal_iter_init(struct bch_fs *c,
 227                                    struct journal_iter *iter,
 228                                    enum btree_id id, unsigned level,
 229                                    struct bpos pos)
 230 {
 231         iter->btree_id  = id;
 232         iter->level     = level;
 233         iter->keys      = &c->journal_keys;
 234         iter->idx       = journal_key_search(&c->journal_keys, id, level, pos);
 235         list_add(&iter->list, &c->journal_iters);
 236 }
 237
 238 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 239 {
 240         return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 241                                                 iter->b, &iter->unpacked);
 242 }
 243
 244 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 245 {
 246         bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 247 }
 248
 249 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 250 {
 251         switch (iter->last) {
 252         case none:
 253                 break;
 254         case btree:
 255                 bch2_journal_iter_advance_btree(iter);
 256                 break;
 257         case journal:
 258                 bch2_journal_iter_advance(&iter->journal);
 259                 break;
 260         }
 261
 262         iter->last = none;
 263 }
 264
 265 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 266 {
 267         struct bkey_s_c ret;
 268
 269         while (1) {
 270                 struct bkey_s_c btree_k         =
 271                         bch2_journal_iter_peek_btree(iter);
 272                 struct bkey_s_c journal_k       =
 273                         bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
 274
 275                 if (btree_k.k && journal_k.k) {
 276                         int cmp = bpos_cmp(btree_k.k->p, journal_k.k->p);
 277
 278                         if (!cmp)
 279                                 bch2_journal_iter_advance_btree(iter);
 280
 281                         iter->last = cmp < 0 ? btree : journal;
 282                 } else if (btree_k.k) {
 283                         iter->last = btree;
 284                 } else if (journal_k.k) {
 285                         iter->last = journal;
 286                 } else {
 287                         iter->last = none;
 288                         return bkey_s_c_null;
 289                 }
 290
 291                 ret = iter->last == journal ? journal_k : btree_k;
 292
 293                 if (iter->b &&
 294                     bpos_cmp(ret.k->p, iter->b->data->max_key) > 0) {
 295                         iter->journal.idx = iter->journal.keys->nr;
 296                         iter->last = none;
 297                         return bkey_s_c_null;
 298                 }
 299
 300                 if (!bkey_deleted(ret.k))
 301                         break;
 302
 303                 bch2_btree_and_journal_iter_advance(iter);
 304         }
 305
 306         return ret;
 307 }
 308
 309 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 310 {
 311         bch2_btree_and_journal_iter_advance(iter);
 312
 313         return bch2_btree_and_journal_iter_peek(iter);
 314 }
 315
 316 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
 317 {
 318         bch2_journal_iter_exit(&iter->journal);
 319 }
 320
 321 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 322                                                 struct bch_fs *c,
 323                                                 struct btree *b)
 324 {
 325         memset(iter, 0, sizeof(*iter));
 326
 327         iter->b = b;
 328         bch2_btree_node_iter_init_from_start(&iter->node_iter, iter->b);
 329         bch2_journal_iter_init(c, &iter->journal,
 330                                b->c.btree_id, b->c.level, b->data->min_key);
 331 }
 332
 333 /* Walk btree, overlaying keys from the journal: */
 334
 335 static void btree_and_journal_iter_prefetch(struct bch_fs *c, struct btree *b,
 336                                            struct btree_and_journal_iter iter)
 337 {
 338         unsigned i = 0, nr = b->c.level > 1 ? 2 : 16;
 339         struct bkey_s_c k;
 340         struct bkey_buf tmp;
 341
 342         BUG_ON(!b->c.level);
 343
 344         bch2_bkey_buf_init(&tmp);
 345
 346         while (i < nr &&
 347                (k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 348                 bch2_bkey_buf_reassemble(&tmp, c, k);
 349
 350                 bch2_btree_node_prefetch(c, NULL, NULL, tmp.k,
 351                                         b->c.btree_id, b->c.level - 1);
 352
 353                 bch2_btree_and_journal_iter_advance(&iter);
 354                 i++;
 355         }
 356
 357         bch2_bkey_buf_exit(&tmp, c);
 358 }
 359
 360 static int bch2_btree_and_journal_walk_recurse(struct btree_trans *trans, struct btree *b,
 361                                 enum btree_id btree_id,
 362                                 btree_walk_key_fn key_fn)
 363 {
 364         struct bch_fs *c = trans->c;
 365         struct btree_and_journal_iter iter;
 366         struct bkey_s_c k;
 367         struct bkey_buf tmp;
 368         struct btree *child;
 369         int ret = 0;
 370
 371         bch2_bkey_buf_init(&tmp);
 372         bch2_btree_and_journal_iter_init_node_iter(&iter, c, b);
 373
 374         while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 375                 if (b->c.level) {
 376                         bch2_bkey_buf_reassemble(&tmp, c, k);
 377
 378                         child = bch2_btree_node_get_noiter(c, tmp.k,
 379                                                 b->c.btree_id, b->c.level - 1,
 380                                                 false);
 381
 382                         ret = PTR_ERR_OR_ZERO(child);
 383                         if (ret)
 384                                 break;
 385
 386                         btree_and_journal_iter_prefetch(c, b, iter);
 387
 388                         ret = bch2_btree_and_journal_walk_recurse(trans, child,
 389                                         btree_id, key_fn);
 390                         six_unlock_read(&child->c.lock);
 391                 } else {
 392                         ret = key_fn(trans, k);
 393                 }
 394
 395                 if (ret)
 396                         break;
 397
 398                 bch2_btree_and_journal_iter_advance(&iter);
 399         }
 400
 401         bch2_btree_and_journal_iter_exit(&iter);
 402         bch2_bkey_buf_exit(&tmp, c);
 403         return ret;
 404 }
 405
 406 int bch2_btree_and_journal_walk(struct btree_trans *trans, enum btree_id btree_id,
 407                                 btree_walk_key_fn key_fn)
 408 {
 409         struct bch_fs *c = trans->c;
 410         struct btree *b = c->btree_roots[btree_id].b;
 411         int ret = 0;
 412
 413         if (btree_node_fake(b))
 414                 return 0;
 415
 416         six_lock_read(&b->c.lock, NULL, NULL);
 417         ret = bch2_btree_and_journal_walk_recurse(trans, b, btree_id, key_fn);
 418         six_unlock_read(&b->c.lock);
 419
 420         return ret;
 421 }
 422
 423 /* sort and dedup all keys in the journal: */
 424
 425 void bch2_journal_entries_free(struct list_head *list)
 426 {
 427
 428         while (!list_empty(list)) {
 429                 struct journal_replay *i =
 430                         list_first_entry(list, struct journal_replay, list);
 431                 list_del(&i->list);
 432                 kvpfree(i, offsetof(struct journal_replay, j) +
 433                         vstruct_bytes(&i->j));
 434         }
 435 }
 436
 437 /*
 438  * When keys compare equal, oldest compares first:
 439  */
 440 static int journal_sort_key_cmp(const void *_l, const void *_r)
 441 {
 442         const struct journal_key *l = _l;
 443         const struct journal_key *r = _r;
 444
 445         return  cmp_int(l->btree_id,    r->btree_id) ?:
 446                 cmp_int(l->level,       r->level) ?:
 447                 bpos_cmp(l->k->k.p, r->k->k.p) ?:
 448                 cmp_int(l->journal_seq, r->journal_seq) ?:
 449                 cmp_int(l->journal_offset, r->journal_offset);
 450 }
 451
 452 void bch2_journal_keys_free(struct journal_keys *keys)
 453 {
 454         struct journal_key *i;
 455
 456         for (i = keys->d; i < keys->d + keys->nr; i++)
 457                 if (i->allocated)
 458                         kfree(i->k);
 459
 460         kvfree(keys->d);
 461         keys->d = NULL;
 462         keys->nr = 0;
 463 }
 464
 465 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 466 {
 467         struct journal_replay *i;
 468         struct jset_entry *entry;
 469         struct bkey_i *k, *_n;
 470         struct journal_keys keys = { NULL };
 471         struct journal_key *src, *dst;
 472         size_t nr_keys = 0;
 473
 474         if (list_empty(journal_entries))
 475                 return keys;
 476
 477         list_for_each_entry(i, journal_entries, list) {
 478                 if (i->ignore)
 479                         continue;
 480
 481                 if (!keys.journal_seq_base)
 482                         keys.journal_seq_base = le64_to_cpu(i->j.seq);
 483
 484                 for_each_jset_key(k, _n, entry, &i->j)
 485                         nr_keys++;
 486         }
 487
 488         keys.size = roundup_pow_of_two(nr_keys);
 489
 490         keys.d = kvmalloc(sizeof(keys.d[0]) * keys.size, GFP_KERNEL);
 491         if (!keys.d)
 492                 goto err;
 493
 494         list_for_each_entry(i, journal_entries, list) {
 495                 if (i->ignore)
 496                         continue;
 497
 498                 BUG_ON(le64_to_cpu(i->j.seq) - keys.journal_seq_base > U32_MAX);
 499
 500                 for_each_jset_key(k, _n, entry, &i->j)
 501                         keys.d[keys.nr++] = (struct journal_key) {
 502                                 .btree_id       = entry->btree_id,
 503                                 .level          = entry->level,
 504                                 .k              = k,
 505                                 .journal_seq    = le64_to_cpu(i->j.seq) -
 506                                         keys.journal_seq_base,
 507                                 .journal_offset = k->_data - i->j._data,
 508                         };
 509         }
 510
 511         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 512
 513         src = dst = keys.d;
 514         while (src < keys.d + keys.nr) {
 515                 while (src + 1 < keys.d + keys.nr &&
 516                        src[0].btree_id  == src[1].btree_id &&
 517                        src[0].level     == src[1].level &&
 518                        !bpos_cmp(src[0].k->k.p, src[1].k->k.p))
 519                         src++;
 520
 521                 *dst++ = *src++;
 522         }
 523
 524         keys.nr = dst - keys.d;
 525 err:
 526         return keys;
 527 }
 528
 529 /* journal replay: */
 530
 531 static void replay_now_at(struct journal *j, u64 seq)
 532 {
 533         BUG_ON(seq < j->replay_journal_seq);
 534         BUG_ON(seq > j->replay_journal_seq_end);
 535
 536         while (j->replay_journal_seq < seq)
 537                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 538 }
 539
 540 static int __bch2_journal_replay_key(struct btree_trans *trans,
 541                                      struct journal_key *k)
 542 {
 543         struct btree_iter iter;
 544         unsigned iter_flags =
 545                 BTREE_ITER_INTENT|
 546                 BTREE_ITER_NOT_EXTENTS;
 547         int ret;
 548
 549         /* Must be checked with btree locked: */
 550         if (k->overwritten)
 551                 return 0;
 552
 553         if (!k->level && k->btree_id == BTREE_ID_alloc)
 554                 iter_flags |= BTREE_ITER_CACHED|BTREE_ITER_CACHED_NOFILL;
 555
 556         bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 557                                   BTREE_MAX_DEPTH, k->level,
 558                                   iter_flags);
 559         ret   = bch2_btree_iter_traverse(&iter) ?:
 560                 bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
 561         bch2_trans_iter_exit(trans, &iter);
 562         return ret;
 563 }
 564
 565 static int bch2_journal_replay_key(struct bch_fs *c, struct journal_key *k)
 566 {
 567         unsigned commit_flags =
 568                 BTREE_INSERT_LAZY_RW|
 569                 BTREE_INSERT_NOFAIL|
 570                 BTREE_INSERT_JOURNAL_RESERVED;
 571
 572         if (!k->allocated)
 573                 commit_flags |= BTREE_INSERT_JOURNAL_REPLAY;
 574
 575         return bch2_trans_do(c, NULL, NULL, commit_flags,
 576                              __bch2_journal_replay_key(&trans, k));
 577 }
 578
 579 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 580 {
 581         const struct journal_key *l = *((const struct journal_key **)_l);
 582         const struct journal_key *r = *((const struct journal_key **)_r);
 583
 584         return  cmp_int(r->level,       l->level) ?:
 585                 cmp_int(l->journal_seq, r->journal_seq) ?:
 586                 cmp_int(l->btree_id,    r->btree_id) ?:
 587                 bpos_cmp(l->k->k.p,     r->k->k.p);
 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         struct bch_dev *ca;
 596         unsigned idx;
 597         size_t i;
 598         u64 seq;
 599         int ret;
 600
 601         keys_sorted = kmalloc_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         seq = j->replay_journal_seq;
 616
 617         /*
 618          * First replay updates to the alloc btree - these will only update the
 619          * btree key cache:
 620          */
 621         for (i = 0; i < keys->nr; i++) {
 622                 k = keys_sorted[i];
 623
 624                 cond_resched();
 625
 626                 if (!k->level && k->btree_id == BTREE_ID_alloc) {
 627                         j->replay_journal_seq = keys->journal_seq_base + k->journal_seq;
 628                         ret = bch2_journal_replay_key(c, k);
 629                         if (ret)
 630                                 goto err;
 631                 }
 632         }
 633
 634         /* Now we can start the allocator threads: */
 635         set_bit(BCH_FS_ALLOC_REPLAY_DONE, &c->flags);
 636         for_each_member_device(ca, c, idx)
 637                 bch2_wake_allocator(ca);
 638
 639         /*
 640          * Next replay updates to interior btree nodes:
 641          */
 642         for (i = 0; i < keys->nr; i++) {
 643                 k = keys_sorted[i];
 644
 645                 cond_resched();
 646
 647                 if (k->level) {
 648                         j->replay_journal_seq = keys->journal_seq_base + k->journal_seq;
 649                         ret = bch2_journal_replay_key(c, k);
 650                         if (ret)
 651                                 goto err;
 652                 }
 653         }
 654
 655         /*
 656          * Now that the btree is in a consistent state, we can start journal
 657          * reclaim (which will be flushing entries from the btree key cache back
 658          * to the btree:
 659          */
 660         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
 661         set_bit(JOURNAL_RECLAIM_STARTED, &j->flags);
 662         journal_reclaim_kick(j);
 663
 664         j->replay_journal_seq = seq;
 665
 666         /*
 667          * Now replay leaf node updates:
 668          */
 669         for (i = 0; i < keys->nr; i++) {
 670                 k = keys_sorted[i];
 671
 672                 cond_resched();
 673
 674                 if (k->level || k->btree_id == BTREE_ID_alloc)
 675                         continue;
 676
 677                 replay_now_at(j, keys->journal_seq_base + k->journal_seq);
 678
 679                 ret = bch2_journal_replay_key(c, k);
 680                 if (ret)
 681                         goto err;
 682         }
 683
 684         replay_now_at(j, j->replay_journal_seq_end);
 685         j->replay_journal_seq = 0;
 686
 687         bch2_journal_set_replay_done(j);
 688         bch2_journal_flush_all_pins(j);
 689         kfree(keys_sorted);
 690
 691         return bch2_journal_error(j);
 692 err:
 693         bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
 694                 ret, bch2_btree_ids[k->btree_id], k->level);
 695         kfree(keys_sorted);
 696
 697         return ret;
 698 }
 699
 700 /* journal replay early: */
 701
 702 static int journal_replay_entry_early(struct bch_fs *c,
 703                                       struct jset_entry *entry)
 704 {
 705         int ret = 0;
 706
 707         switch (entry->type) {
 708         case BCH_JSET_ENTRY_btree_root: {
 709                 struct btree_root *r;
 710
 711                 if (entry->btree_id >= BTREE_ID_NR) {
 712                         bch_err(c, "filesystem has unknown btree type %u",
 713                                 entry->btree_id);
 714                         return -EINVAL;
 715                 }
 716
 717                 r = &c->btree_roots[entry->btree_id];
 718
 719                 if (entry->u64s) {
 720                         r->level = entry->level;
 721                         bkey_copy(&r->key, &entry->start[0]);
 722                         r->error = 0;
 723                 } else {
 724                         r->error = -EIO;
 725                 }
 726                 r->alive = true;
 727                 break;
 728         }
 729         case BCH_JSET_ENTRY_usage: {
 730                 struct jset_entry_usage *u =
 731                         container_of(entry, struct jset_entry_usage, entry);
 732
 733                 switch (entry->btree_id) {
 734                 case BCH_FS_USAGE_reserved:
 735                         if (entry->level < BCH_REPLICAS_MAX)
 736                                 c->usage_base->persistent_reserved[entry->level] =
 737                                         le64_to_cpu(u->v);
 738                         break;
 739                 case BCH_FS_USAGE_inodes:
 740                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 741                         break;
 742                 case BCH_FS_USAGE_key_version:
 743                         atomic64_set(&c->key_version,
 744                                      le64_to_cpu(u->v));
 745                         break;
 746                 }
 747
 748                 break;
 749         }
 750         case BCH_JSET_ENTRY_data_usage: {
 751                 struct jset_entry_data_usage *u =
 752                         container_of(entry, struct jset_entry_data_usage, entry);
 753
 754                 ret = bch2_replicas_set_usage(c, &u->r,
 755                                               le64_to_cpu(u->v));
 756                 break;
 757         }
 758         case BCH_JSET_ENTRY_dev_usage: {
 759                 struct jset_entry_dev_usage *u =
 760                         container_of(entry, struct jset_entry_dev_usage, entry);
 761                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 762                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 763
 764                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 765                 ca->usage_base->buckets_unavailable     = le64_to_cpu(u->buckets_unavailable);
 766
 767                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 768                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 769                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 770                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 771                 }
 772
 773                 break;
 774         }
 775         case BCH_JSET_ENTRY_blacklist: {
 776                 struct jset_entry_blacklist *bl_entry =
 777                         container_of(entry, struct jset_entry_blacklist, entry);
 778
 779                 ret = bch2_journal_seq_blacklist_add(c,
 780                                 le64_to_cpu(bl_entry->seq),
 781                                 le64_to_cpu(bl_entry->seq) + 1);
 782                 break;
 783         }
 784         case BCH_JSET_ENTRY_blacklist_v2: {
 785                 struct jset_entry_blacklist_v2 *bl_entry =
 786                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 787
 788                 ret = bch2_journal_seq_blacklist_add(c,
 789                                 le64_to_cpu(bl_entry->start),
 790                                 le64_to_cpu(bl_entry->end) + 1);
 791                 break;
 792         }
 793         case BCH_JSET_ENTRY_clock: {
 794                 struct jset_entry_clock *clock =
 795                         container_of(entry, struct jset_entry_clock, entry);
 796
 797                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 798         }
 799         }
 800
 801         return ret;
 802 }
 803
 804 static int journal_replay_early(struct bch_fs *c,
 805                                 struct bch_sb_field_clean *clean,
 806                                 struct list_head *journal)
 807 {
 808         struct journal_replay *i;
 809         struct jset_entry *entry;
 810         int ret;
 811
 812         if (clean) {
 813                 for (entry = clean->start;
 814                      entry != vstruct_end(&clean->field);
 815                      entry = vstruct_next(entry)) {
 816                         ret = journal_replay_entry_early(c, entry);
 817                         if (ret)
 818                                 return ret;
 819                 }
 820         } else {
 821                 list_for_each_entry(i, journal, list) {
 822                         if (i->ignore)
 823                                 continue;
 824
 825                         vstruct_for_each(&i->j, entry) {
 826                                 ret = journal_replay_entry_early(c, entry);
 827                                 if (ret)
 828                                         return ret;
 829                         }
 830                 }
 831         }
 832
 833         bch2_fs_usage_initialize(c);
 834
 835         return 0;
 836 }
 837
 838 /* sb clean section: */
 839
 840 static struct bkey_i *btree_root_find(struct bch_fs *c,
 841                                       struct bch_sb_field_clean *clean,
 842                                       struct jset *j,
 843                                       enum btree_id id, unsigned *level)
 844 {
 845         struct bkey_i *k;
 846         struct jset_entry *entry, *start, *end;
 847
 848         if (clean) {
 849                 start = clean->start;
 850                 end = vstruct_end(&clean->field);
 851         } else {
 852                 start = j->start;
 853                 end = vstruct_last(j);
 854         }
 855
 856         for (entry = start; entry < end; entry = vstruct_next(entry))
 857                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 858                     entry->btree_id == id)
 859                         goto found;
 860
 861         return NULL;
 862 found:
 863         if (!entry->u64s)
 864                 return ERR_PTR(-EINVAL);
 865
 866         k = entry->start;
 867         *level = entry->level;
 868         return k;
 869 }
 870
 871 static int verify_superblock_clean(struct bch_fs *c,
 872                                    struct bch_sb_field_clean **cleanp,
 873                                    struct jset *j)
 874 {
 875         unsigned i;
 876         struct bch_sb_field_clean *clean = *cleanp;
 877         int ret = 0;
 878
 879         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 880                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 881                         le64_to_cpu(clean->journal_seq),
 882                         le64_to_cpu(j->seq))) {
 883                 kfree(clean);
 884                 *cleanp = NULL;
 885                 return 0;
 886         }
 887
 888         for (i = 0; i < BTREE_ID_NR; i++) {
 889                 char buf1[200], buf2[200];
 890                 struct bkey_i *k1, *k2;
 891                 unsigned l1 = 0, l2 = 0;
 892
 893                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 894                 k2 = btree_root_find(c, NULL, j, i, &l2);
 895
 896                 if (!k1 && !k2)
 897                         continue;
 898
 899                 mustfix_fsck_err_on(!k1 || !k2 ||
 900                                     IS_ERR(k1) ||
 901                                     IS_ERR(k2) ||
 902                                     k1->k.u64s != k2->k.u64s ||
 903                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 904                                     l1 != l2, c,
 905                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 906                         "sb:      l=%u %s\n"
 907                         "journal: l=%u %s\n", i,
 908                         l1, (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(k1)), buf1),
 909                         l2, (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(k2)), buf2));
 910         }
 911 fsck_err:
 912         return ret;
 913 }
 914
 915 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 916 {
 917         struct bch_sb_field_clean *clean, *sb_clean;
 918         int ret;
 919
 920         mutex_lock(&c->sb_lock);
 921         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 922
 923         if (fsck_err_on(!sb_clean, c,
 924                         "superblock marked clean but clean section not present")) {
 925                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 926                 c->sb.clean = false;
 927                 mutex_unlock(&c->sb_lock);
 928                 return NULL;
 929         }
 930
 931         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 932                         GFP_KERNEL);
 933         if (!clean) {
 934                 mutex_unlock(&c->sb_lock);
 935                 return ERR_PTR(-ENOMEM);
 936         }
 937
 938         ret = bch2_sb_clean_validate(c, clean, READ);
 939         if (ret) {
 940                 mutex_unlock(&c->sb_lock);
 941                 return ERR_PTR(ret);
 942         }
 943
 944         mutex_unlock(&c->sb_lock);
 945
 946         return clean;
 947 fsck_err:
 948         mutex_unlock(&c->sb_lock);
 949         return ERR_PTR(ret);
 950 }
 951
 952 static int read_btree_roots(struct bch_fs *c)
 953 {
 954         unsigned i;
 955         int ret = 0;
 956
 957         for (i = 0; i < BTREE_ID_NR; i++) {
 958                 struct btree_root *r = &c->btree_roots[i];
 959
 960                 if (!r->alive)
 961                         continue;
 962
 963                 if (i == BTREE_ID_alloc &&
 964                     c->opts.reconstruct_alloc) {
 965                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 966                         continue;
 967                 }
 968
 969                 if (r->error) {
 970                         __fsck_err(c, i == BTREE_ID_alloc
 971                                    ? FSCK_CAN_IGNORE : 0,
 972                                    "invalid btree root %s",
 973                                    bch2_btree_ids[i]);
 974                         if (i == BTREE_ID_alloc)
 975                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 976                 }
 977
 978                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 979                 if (ret) {
 980                         __fsck_err(c, i == BTREE_ID_alloc
 981                                    ? FSCK_CAN_IGNORE : 0,
 982                                    "error reading btree root %s",
 983                                    bch2_btree_ids[i]);
 984                         if (i == BTREE_ID_alloc)
 985                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 986                 }
 987         }
 988
 989         for (i = 0; i < BTREE_ID_NR; i++)
 990                 if (!c->btree_roots[i].b)
 991                         bch2_btree_root_alloc(c, i);
 992 fsck_err:
 993         return ret;
 994 }
 995
 996 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
 997 {
 998         struct bkey_i_snapshot  root_snapshot;
 999         struct bkey_i_subvolume root_volume;
1000         int ret;
1001
1002         bkey_snapshot_init(&root_snapshot.k_i);
1003         root_snapshot.k.p.offset = U32_MAX;
1004         root_snapshot.v.flags   = 0;
1005         root_snapshot.v.parent  = 0;
1006         root_snapshot.v.subvol  = BCACHEFS_ROOT_SUBVOL;
1007         root_snapshot.v.pad     = 0;
1008         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
1009
1010         ret = bch2_btree_insert(c, BTREE_ID_snapshots,
1011                                 &root_snapshot.k_i,
1012                                 NULL, NULL, 0);
1013         if (ret)
1014                 return ret;
1015
1016
1017         bkey_subvolume_init(&root_volume.k_i);
1018         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1019         root_volume.v.flags     = 0;
1020         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
1021         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
1022
1023         ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
1024                                 &root_volume.k_i,
1025                                 NULL, NULL, 0);
1026         if (ret)
1027                 return ret;
1028
1029         return 0;
1030 }
1031
1032 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
1033 {
1034         struct bch_fs *c = trans->c;
1035         struct btree_iter iter;
1036         struct bkey_s_c k;
1037         struct bch_inode_unpacked inode;
1038         int ret;
1039
1040         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
1041                              SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
1042         k = bch2_btree_iter_peek_slot(&iter);
1043         ret = bkey_err(k);
1044         if (ret)
1045                 goto err;
1046
1047         if (!bkey_is_inode(k.k)) {
1048                 bch_err(c, "root inode not found");
1049                 ret = -ENOENT;
1050                 goto err;
1051         }
1052
1053         ret = bch2_inode_unpack(k, &inode);
1054         BUG_ON(ret);
1055
1056         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1057
1058         ret = bch2_inode_write(trans, &iter, &inode);
1059 err:
1060         bch2_trans_iter_exit(trans, &iter);
1061         return ret;
1062 }
1063
1064 int bch2_fs_recovery(struct bch_fs *c)
1065 {
1066         const char *err = "cannot allocate memory";
1067         struct bch_sb_field_clean *clean = NULL;
1068         struct jset *last_journal_entry = NULL;
1069         u64 blacklist_seq, journal_seq;
1070         bool write_sb = false;
1071         int ret = 0;
1072
1073         if (c->sb.clean)
1074                 clean = read_superblock_clean(c);
1075         ret = PTR_ERR_OR_ZERO(clean);
1076         if (ret)
1077                 goto err;
1078
1079         if (c->sb.clean)
1080                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
1081                          le64_to_cpu(clean->journal_seq));
1082         else
1083                 bch_info(c, "recovering from unclean shutdown");
1084
1085         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
1086                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
1087                 ret = -EINVAL;
1088                 goto err;
1089         }
1090
1091         if (!c->sb.clean &&
1092             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1093                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1094                 ret = -EINVAL;
1095                 goto err;
1096         }
1097
1098         if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1099                 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
1100                 ret = -EINVAL;
1101                 goto err;
1102         }
1103
1104         if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
1105                 bch_info(c, "alloc_v2 feature bit not set, fsck required");
1106                 c->opts.fsck = true;
1107                 c->opts.fix_errors = FSCK_OPT_YES;
1108         }
1109
1110         if (!c->replicas.entries ||
1111             c->opts.rebuild_replicas) {
1112                 bch_info(c, "building replicas info");
1113                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1114         }
1115
1116         if (!c->opts.nochanges) {
1117                 if (c->sb.version < bcachefs_metadata_version_inode_backpointers) {
1118                         bch_info(c, "version prior to inode backpointers, upgrade and fsck required");
1119                         c->opts.version_upgrade = true;
1120                         c->opts.fsck            = true;
1121                         c->opts.fix_errors      = FSCK_OPT_YES;
1122                 } else if (c->sb.version < bcachefs_metadata_version_subvol_dirent) {
1123                         bch_info(c, "filesystem version is prior to subvol_dirent - upgrading");
1124                         c->opts.version_upgrade = true;
1125                         c->opts.fsck            = true;
1126                 } else if (c->sb.version < bcachefs_metadata_version_inode_v2) {
1127                         bch_info(c, "filesystem version is prior to inode_v2 - upgrading");
1128                         c->opts.version_upgrade = true;
1129                 }
1130         }
1131
1132         ret = bch2_blacklist_table_initialize(c);
1133         if (ret) {
1134                 bch_err(c, "error initializing blacklist table");
1135                 goto err;
1136         }
1137
1138         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1139                 struct journal_replay *i;
1140
1141                 ret = bch2_journal_read(c, &c->journal_entries,
1142                                         &blacklist_seq, &journal_seq);
1143                 if (ret)
1144                         goto err;
1145
1146                 list_for_each_entry_reverse(i, &c->journal_entries, list)
1147                         if (!i->ignore) {
1148                                 last_journal_entry = &i->j;
1149                                 break;
1150                         }
1151
1152                 if (mustfix_fsck_err_on(c->sb.clean &&
1153                                         last_journal_entry &&
1154                                         !journal_entry_empty(last_journal_entry), c,
1155                                 "filesystem marked clean but journal not empty")) {
1156                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1157                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1158                         c->sb.clean = false;
1159                 }
1160
1161                 if (!last_journal_entry) {
1162                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
1163                         goto use_clean;
1164                 }
1165
1166                 c->journal_keys = journal_keys_sort(&c->journal_entries);
1167                 if (!c->journal_keys.d) {
1168                         ret = -ENOMEM;
1169                         goto err;
1170                 }
1171
1172                 if (c->sb.clean && last_journal_entry) {
1173                         ret = verify_superblock_clean(c, &clean,
1174                                                       last_journal_entry);
1175                         if (ret)
1176                                 goto err;
1177                 }
1178         } else {
1179 use_clean:
1180                 if (!clean) {
1181                         bch_err(c, "no superblock clean section found");
1182                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1183                         goto err;
1184
1185                 }
1186                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1187         }
1188
1189         if (c->opts.reconstruct_alloc) {
1190                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1191                 drop_alloc_keys(&c->journal_keys);
1192         }
1193
1194         zero_out_btree_mem_ptr(&c->journal_keys);
1195
1196         ret = journal_replay_early(c, clean, &c->journal_entries);
1197         if (ret)
1198                 goto err;
1199
1200         if (blacklist_seq != journal_seq) {
1201                 ret = bch2_journal_seq_blacklist_add(c,
1202                                         blacklist_seq, journal_seq);
1203                 if (ret) {
1204                         bch_err(c, "error creating new journal seq blacklist entry");
1205                         goto err;
1206                 }
1207         }
1208
1209         ret = bch2_fs_journal_start(&c->journal, journal_seq,
1210                                     &c->journal_entries);
1211         if (ret)
1212                 goto err;
1213
1214         ret = read_btree_roots(c);
1215         if (ret)
1216                 goto err;
1217
1218         bch_verbose(c, "starting alloc read");
1219         err = "error reading allocation information";
1220         ret = bch2_alloc_read(c);
1221         if (ret)
1222                 goto err;
1223         bch_verbose(c, "alloc read done");
1224
1225         bch_verbose(c, "starting stripes_read");
1226         err = "error reading stripes";
1227         ret = bch2_stripes_read(c);
1228         if (ret)
1229                 goto err;
1230         bch_verbose(c, "stripes_read done");
1231
1232         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1233
1234         if (c->opts.fsck ||
1235             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)) ||
1236             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_metadata)) ||
1237             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1238                 bool metadata_only = c->opts.norecovery;
1239
1240                 bch_info(c, "starting mark and sweep");
1241                 err = "error in mark and sweep";
1242                 ret = bch2_gc(c, true, metadata_only);
1243                 if (ret)
1244                         goto err;
1245                 bch_verbose(c, "mark and sweep done");
1246         }
1247
1248         bch2_stripes_heap_start(c);
1249
1250         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1251         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1252
1253         /*
1254          * Skip past versions that might have possibly been used (as nonces),
1255          * but hadn't had their pointers written:
1256          */
1257         if (c->sb.encryption_type && !c->sb.clean)
1258                 atomic64_add(1 << 16, &c->key_version);
1259
1260         if (c->opts.norecovery)
1261                 goto out;
1262
1263         bch_verbose(c, "starting journal replay");
1264         err = "journal replay failed";
1265         ret = bch2_journal_replay(c);
1266         if (ret)
1267                 goto err;
1268         bch_verbose(c, "journal replay done");
1269
1270         if (test_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags) &&
1271             !c->opts.nochanges) {
1272                 /*
1273                  * note that even when filesystem was clean there might be work
1274                  * to do here, if we ran gc (because of fsck) which recalculated
1275                  * oldest_gen:
1276                  */
1277                 bch_verbose(c, "writing allocation info");
1278                 err = "error writing out alloc info";
1279                 ret = bch2_alloc_write_all(c, BTREE_INSERT_LAZY_RW);
1280                 if (ret) {
1281                         bch_err(c, "error writing alloc info");
1282                         goto err;
1283                 }
1284                 bch_verbose(c, "alloc write done");
1285         }
1286
1287         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1288                 bch2_fs_lazy_rw(c);
1289
1290                 err = "error creating root snapshot node";
1291                 ret = bch2_fs_initialize_subvolumes(c);
1292                 if (ret)
1293                         goto err;
1294         }
1295
1296         bch_verbose(c, "reading snapshots table");
1297         err = "error reading snapshots table";
1298         ret = bch2_fs_snapshots_start(c);
1299         if (ret)
1300                 goto err;
1301         bch_verbose(c, "reading snapshots done");
1302
1303         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1304                 /* set bi_subvol on root inode */
1305                 err = "error upgrade root inode for subvolumes";
1306                 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1307                                     bch2_fs_upgrade_for_subvolumes(&trans));
1308                 if (ret)
1309                         goto err;
1310         }
1311
1312         if (c->opts.fsck) {
1313                 bch_info(c, "starting fsck");
1314                 err = "error in fsck";
1315                 ret = bch2_fsck_full(c);
1316                 if (ret)
1317                         goto err;
1318                 bch_verbose(c, "fsck done");
1319         } else if (!c->sb.clean) {
1320                 bch_verbose(c, "checking for deleted inodes");
1321                 err = "error in recovery";
1322                 ret = bch2_fsck_walk_inodes_only(c);
1323                 if (ret)
1324                         goto err;
1325                 bch_verbose(c, "check inodes done");
1326         }
1327
1328         if (enabled_qtypes(c)) {
1329                 bch_verbose(c, "reading quotas");
1330                 ret = bch2_fs_quota_read(c);
1331                 if (ret)
1332                         goto err;
1333                 bch_verbose(c, "quotas done");
1334         }
1335
1336         mutex_lock(&c->sb_lock);
1337         /*
1338          * With journal replay done, we can clear the journal seq blacklist
1339          * table:
1340          */
1341         BUG_ON(!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags));
1342         if (le16_to_cpu(c->sb.version_min) >= bcachefs_metadata_version_btree_ptr_sectors_written)
1343                 bch2_sb_resize_journal_seq_blacklist(&c->disk_sb, 0);
1344
1345         if (c->opts.version_upgrade) {
1346                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1347                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1348                 write_sb = true;
1349         }
1350
1351         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1352                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1353                 write_sb = true;
1354         }
1355
1356         if (c->opts.fsck &&
1357             !test_bit(BCH_FS_ERROR, &c->flags) &&
1358             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1359                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1360                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1361                 write_sb = true;
1362         }
1363
1364         if (write_sb)
1365                 bch2_write_super(c);
1366         mutex_unlock(&c->sb_lock);
1367
1368         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1369             !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1370             le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) {
1371                 struct bch_move_stats stats;
1372
1373                 bch_move_stats_init(&stats, "recovery");
1374
1375                 bch_info(c, "scanning for old btree nodes");
1376                 ret = bch2_fs_read_write(c);
1377                 if (ret)
1378                         goto err;
1379
1380                 ret = bch2_scan_old_btree_nodes(c, &stats);
1381                 if (ret)
1382                         goto err;
1383                 bch_info(c, "scanning for old btree nodes done");
1384         }
1385
1386         ret = 0;
1387 out:
1388         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1389         bch2_flush_fsck_errs(c);
1390
1391         if (!c->opts.keep_journal) {
1392                 bch2_journal_keys_free(&c->journal_keys);
1393                 bch2_journal_entries_free(&c->journal_entries);
1394         }
1395         kfree(clean);
1396         if (ret)
1397                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1398         else
1399                 bch_verbose(c, "ret %i", ret);
1400         return ret;
1401 err:
1402 fsck_err:
1403         bch2_fs_emergency_read_only(c);
1404         goto out;
1405 }
1406
1407 int bch2_fs_initialize(struct bch_fs *c)
1408 {
1409         struct bch_inode_unpacked root_inode, lostfound_inode;
1410         struct bkey_inode_buf packed_inode;
1411         struct qstr lostfound = QSTR("lost+found");
1412         const char *err = "cannot allocate memory";
1413         struct bch_dev *ca;
1414         LIST_HEAD(journal);
1415         unsigned i;
1416         int ret;
1417
1418         bch_notice(c, "initializing new filesystem");
1419
1420         mutex_lock(&c->sb_lock);
1421         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1422         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1423
1424         if (c->opts.version_upgrade) {
1425                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1426                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1427                 bch2_write_super(c);
1428         }
1429         mutex_unlock(&c->sb_lock);
1430
1431         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1432         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1433
1434         for (i = 0; i < BTREE_ID_NR; i++)
1435                 bch2_btree_root_alloc(c, i);
1436
1437         set_bit(BCH_FS_ALLOC_REPLAY_DONE, &c->flags);
1438         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
1439         set_bit(JOURNAL_RECLAIM_STARTED, &c->journal.flags);
1440
1441         err = "unable to allocate journal buckets";
1442         for_each_online_member(ca, c, i) {
1443                 ret = bch2_dev_journal_alloc(ca);
1444                 if (ret) {
1445                         percpu_ref_put(&ca->io_ref);
1446                         goto err;
1447                 }
1448         }
1449
1450         /*
1451          * journal_res_get() will crash if called before this has
1452          * set up the journal.pin FIFO and journal.cur pointer:
1453          */
1454         bch2_fs_journal_start(&c->journal, 1, &journal);
1455         bch2_journal_set_replay_done(&c->journal);
1456
1457         err = "error going read-write";
1458         ret = bch2_fs_read_write_early(c);
1459         if (ret)
1460                 goto err;
1461
1462         /*
1463          * Write out the superblock and journal buckets, now that we can do
1464          * btree updates
1465          */
1466         err = "error marking superblock and journal";
1467         for_each_member_device(ca, c, i) {
1468                 ret = bch2_trans_mark_dev_sb(c, ca);
1469                 if (ret) {
1470                         percpu_ref_put(&ca->ref);
1471                         goto err;
1472                 }
1473
1474                 ca->new_fs_bucket_idx = 0;
1475         }
1476
1477         err = "error creating root snapshot node";
1478         ret = bch2_fs_initialize_subvolumes(c);
1479         if (ret)
1480                 goto err;
1481
1482         bch_verbose(c, "reading snapshots table");
1483         err = "error reading snapshots table";
1484         ret = bch2_fs_snapshots_start(c);
1485         if (ret)
1486                 goto err;
1487         bch_verbose(c, "reading snapshots done");
1488
1489         bch2_inode_init(c, &root_inode, 0, 0,
1490                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1491         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1492         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1493         bch2_inode_pack(c, &packed_inode, &root_inode);
1494         packed_inode.inode.k.p.snapshot = U32_MAX;
1495
1496         err = "error creating root directory";
1497         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1498                                 &packed_inode.inode.k_i,
1499                                 NULL, NULL, 0);
1500         if (ret)
1501                 goto err;
1502
1503         bch2_inode_init_early(c, &lostfound_inode);
1504
1505         err = "error creating lost+found";
1506         ret = bch2_trans_do(c, NULL, NULL, 0,
1507                 bch2_create_trans(&trans,
1508                                   BCACHEFS_ROOT_SUBVOL_INUM,
1509                                   &root_inode, &lostfound_inode,
1510                                   &lostfound,
1511                                   0, 0, S_IFDIR|0700, 0,
1512                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1513         if (ret) {
1514                 bch_err(c, "error creating lost+found");
1515                 goto err;
1516         }
1517
1518         if (enabled_qtypes(c)) {
1519                 ret = bch2_fs_quota_read(c);
1520                 if (ret)
1521                         goto err;
1522         }
1523
1524         err = "error writing first journal entry";
1525         ret = bch2_journal_flush(&c->journal);
1526         if (ret)
1527                 goto err;
1528
1529         mutex_lock(&c->sb_lock);
1530         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1531         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1532
1533         bch2_write_super(c);
1534         mutex_unlock(&c->sb_lock);
1535
1536         return 0;
1537 err:
1538         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1539         return ret;
1540 }