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

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "bkey_buf.h"
   5 #include "alloc_background.h"
   6 #include "btree_gc.h"
   7 #include "btree_update.h"
   8 #include "btree_update_interior.h"
   9 #include "btree_io.h"
  10 #include "buckets.h"
  11 #include "dirent.h"
  12 #include "ec.h"
  13 #include "error.h"
  14 #include "fs-common.h"
  15 #include "fsck.h"
  16 #include "journal_io.h"
  17 #include "journal_reclaim.h"
  18 #include "journal_seq_blacklist.h"
  19 #include "lru.h"
  20 #include "move.h"
  21 #include "quota.h"
  22 #include "recovery.h"
  23 #include "replicas.h"
  24 #include "subvolume.h"
  25 #include "super-io.h"
  26
  27 #include <linux/sort.h>
  28 #include <linux/stat.h>
  29
  30 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  31
  32 /* for -o reconstruct_alloc: */
  33 static void drop_alloc_keys(struct journal_keys *keys)
  34 {
  35         size_t src, dst;
  36
  37         for (src = 0, dst = 0; src < keys->nr; src++)
  38                 if (keys->d[src].btree_id != BTREE_ID_alloc)
  39                         keys->d[dst++] = keys->d[src];
  40
  41         keys->nr = dst;
  42 }
  43
  44 /*
  45  * Btree node pointers have a field to stack a pointer to the in memory btree
  46  * node; we need to zero out this field when reading in btree nodes, or when
  47  * reading in keys from the journal:
  48  */
  49 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
  50 {
  51         struct journal_key *i;
  52
  53         for (i = keys->d; i < keys->d + keys->nr; i++)
  54                 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
  55                         bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
  56 }
  57
  58 /* iterate over keys read from the journal: */
  59
  60 static int __journal_key_cmp(enum btree_id      l_btree_id,
  61                              unsigned           l_level,
  62                              struct bpos        l_pos,
  63                              const struct journal_key *r)
  64 {
  65         return (cmp_int(l_btree_id,     r->btree_id) ?:
  66                 cmp_int(l_level,        r->level) ?:
  67                 bpos_cmp(l_pos, r->k->k.p));
  68 }
  69
  70 static int journal_key_cmp(const struct journal_key *l, const struct journal_key *r)
  71 {
  72         return __journal_key_cmp(l->btree_id, l->level, l->k->k.p, r);
  73 }
  74
  75 size_t bch2_journal_key_search(struct journal_keys *journal_keys,
  76                                enum btree_id id, unsigned level,
  77                                struct bpos pos)
  78 {
  79         size_t l = 0, r = journal_keys->nr, m;
  80
  81         while (l < r) {
  82                 m = l + ((r - l) >> 1);
  83                 if (__journal_key_cmp(id, level, pos, &journal_keys->d[m]) > 0)
  84                         l = m + 1;
  85                 else
  86                         r = m;
  87         }
  88
  89         BUG_ON(l < journal_keys->nr &&
  90                __journal_key_cmp(id, level, pos, &journal_keys->d[l]) > 0);
  91
  92         BUG_ON(l &&
  93                __journal_key_cmp(id, level, pos, &journal_keys->d[l - 1]) <= 0);
  94
  95         return l;
  96 }
  97
  98 struct bkey_i *bch2_journal_keys_peek(struct bch_fs *c, enum btree_id btree_id,
  99                                       unsigned level, struct bpos pos)
 100 {
 101         struct journal_keys *keys = &c->journal_keys;
 102         struct journal_key *end = keys->d + keys->nr;
 103         struct journal_key *k = keys->d +
 104                 bch2_journal_key_search(keys, btree_id, level, pos);
 105
 106         while (k < end && k->overwritten)
 107                 k++;
 108
 109         if (k < end &&
 110             k->btree_id == btree_id &&
 111             k->level    == level)
 112                 return k->k;
 113         return NULL;
 114 }
 115
 116 static void journal_iter_fix(struct bch_fs *c, struct journal_iter *iter, unsigned idx)
 117 {
 118         struct bkey_i *n = iter->keys->d[idx].k;
 119         struct btree_and_journal_iter *biter =
 120                 container_of(iter, struct btree_and_journal_iter, journal);
 121
 122         if (iter->idx > idx ||
 123             (iter->idx == idx &&
 124              biter->last &&
 125              bpos_cmp(n->k.p, biter->unpacked.p) <= 0))
 126                 iter->idx++;
 127 }
 128
 129 int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
 130                                  unsigned level, struct bkey_i *k)
 131 {
 132         struct journal_key n = {
 133                 .btree_id       = id,
 134                 .level          = level,
 135                 .k              = k,
 136                 .allocated      = true,
 137                 /*
 138                  * Ensure these keys are done last by journal replay, to unblock
 139                  * journal reclaim:
 140                  */
 141                 .journal_seq    = U32_MAX,
 142         };
 143         struct journal_keys *keys = &c->journal_keys;
 144         struct journal_iter *iter;
 145         size_t idx = bch2_journal_key_search(keys, id, level, k->k.p);
 146
 147         BUG_ON(test_bit(BCH_FS_RW, &c->flags));
 148
 149         if (idx < keys->nr &&
 150             journal_key_cmp(&n, &keys->d[idx]) == 0) {
 151                 if (keys->d[idx].allocated)
 152                         kfree(keys->d[idx].k);
 153                 keys->d[idx] = n;
 154                 return 0;
 155         }
 156
 157         if (keys->nr == keys->size) {
 158                 struct journal_keys new_keys = {
 159                         .nr                     = keys->nr,
 160                         .size                   = keys->size * 2,
 161                         .journal_seq_base       = keys->journal_seq_base,
 162                 };
 163
 164                 new_keys.d = kvmalloc(sizeof(new_keys.d[0]) * new_keys.size, GFP_KERNEL);
 165                 if (!new_keys.d) {
 166                         bch_err(c, "%s: error allocating new key array (size %zu)",
 167                                 __func__, new_keys.size);
 168                         return -ENOMEM;
 169                 }
 170
 171                 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
 172                 kvfree(keys->d);
 173                 *keys = new_keys;
 174         }
 175
 176         array_insert_item(keys->d, keys->nr, idx, n);
 177
 178         list_for_each_entry(iter, &c->journal_iters, list)
 179                 journal_iter_fix(c, iter, idx);
 180
 181         return 0;
 182 }
 183
 184 /*
 185  * Can only be used from the recovery thread while we're still RO - can't be
 186  * used once we've got RW, as journal_keys is at that point used by multiple
 187  * threads:
 188  */
 189 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
 190                             unsigned level, struct bkey_i *k)
 191 {
 192         struct bkey_i *n;
 193         int ret;
 194
 195         n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
 196         if (!n)
 197                 return -ENOMEM;
 198
 199         bkey_copy(n, k);
 200         ret = bch2_journal_key_insert_take(c, id, level, n);
 201         if (ret)
 202                 kfree(n);
 203         return ret;
 204 }
 205
 206 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
 207                             unsigned level, struct bpos pos)
 208 {
 209         struct bkey_i whiteout;
 210
 211         bkey_init(&whiteout.k);
 212         whiteout.k.p = pos;
 213
 214         return bch2_journal_key_insert(c, id, level, &whiteout);
 215 }
 216
 217 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
 218                                   unsigned level, struct bpos pos)
 219 {
 220         struct journal_keys *keys = &c->journal_keys;
 221         size_t idx = bch2_journal_key_search(keys, btree, level, pos);
 222
 223         if (idx < keys->nr &&
 224             keys->d[idx].btree_id       == btree &&
 225             keys->d[idx].level          == level &&
 226             !bpos_cmp(keys->d[idx].k->k.p, pos))
 227                 keys->d[idx].overwritten = true;
 228 }
 229
 230 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
 231 {
 232         struct journal_key *k = iter->keys->d + iter->idx;
 233
 234         while (k < iter->keys->d + iter->keys->nr &&
 235                k->btree_id      == iter->btree_id &&
 236                k->level         == iter->level) {
 237                 if (!k->overwritten)
 238                         return k->k;
 239
 240                 iter->idx++;
 241                 k = iter->keys->d + iter->idx;
 242         }
 243
 244         return NULL;
 245 }
 246
 247 static void bch2_journal_iter_advance(struct journal_iter *iter)
 248 {
 249         if (iter->idx < iter->keys->nr)
 250                 iter->idx++;
 251 }
 252
 253 static void bch2_journal_iter_exit(struct journal_iter *iter)
 254 {
 255         list_del(&iter->list);
 256 }
 257
 258 static void bch2_journal_iter_init(struct bch_fs *c,
 259                                    struct journal_iter *iter,
 260                                    enum btree_id id, unsigned level,
 261                                    struct bpos pos)
 262 {
 263         iter->btree_id  = id;
 264         iter->level     = level;
 265         iter->keys      = &c->journal_keys;
 266         iter->idx       = bch2_journal_key_search(&c->journal_keys, id, level, pos);
 267 }
 268
 269 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 270 {
 271         return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 272                                                 iter->b, &iter->unpacked);
 273 }
 274
 275 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 276 {
 277         bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 278 }
 279
 280 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 281 {
 282         switch (iter->last) {
 283         case none:
 284                 break;
 285         case btree:
 286                 bch2_journal_iter_advance_btree(iter);
 287                 break;
 288         case journal:
 289                 bch2_journal_iter_advance(&iter->journal);
 290                 break;
 291         }
 292
 293         iter->last = none;
 294 }
 295
 296 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 297 {
 298         struct bkey_s_c ret;
 299
 300         while (1) {
 301                 struct bkey_s_c btree_k         =
 302                         bch2_journal_iter_peek_btree(iter);
 303                 struct bkey_s_c journal_k       =
 304                         bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
 305
 306                 if (btree_k.k && journal_k.k) {
 307                         int cmp = bpos_cmp(btree_k.k->p, journal_k.k->p);
 308
 309                         if (!cmp)
 310                                 bch2_journal_iter_advance_btree(iter);
 311
 312                         iter->last = cmp < 0 ? btree : journal;
 313                 } else if (btree_k.k) {
 314                         iter->last = btree;
 315                 } else if (journal_k.k) {
 316                         iter->last = journal;
 317                 } else {
 318                         iter->last = none;
 319                         return bkey_s_c_null;
 320                 }
 321
 322                 ret = iter->last == journal ? journal_k : btree_k;
 323
 324                 if (iter->b &&
 325                     bpos_cmp(ret.k->p, iter->b->data->max_key) > 0) {
 326                         iter->journal.idx = iter->journal.keys->nr;
 327                         iter->last = none;
 328                         return bkey_s_c_null;
 329                 }
 330
 331                 if (!bkey_deleted(ret.k))
 332                         break;
 333
 334                 bch2_btree_and_journal_iter_advance(iter);
 335         }
 336
 337         return ret;
 338 }
 339
 340 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 341 {
 342         bch2_btree_and_journal_iter_advance(iter);
 343
 344         return bch2_btree_and_journal_iter_peek(iter);
 345 }
 346
 347 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
 348 {
 349         bch2_journal_iter_exit(&iter->journal);
 350 }
 351
 352 void __bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 353                                                   struct bch_fs *c,
 354                                                   struct btree *b,
 355                                                   struct btree_node_iter node_iter,
 356                                                   struct bpos pos)
 357 {
 358         memset(iter, 0, sizeof(*iter));
 359
 360         iter->b = b;
 361         iter->node_iter = node_iter;
 362         bch2_journal_iter_init(c, &iter->journal, b->c.btree_id, b->c.level, pos);
 363         INIT_LIST_HEAD(&iter->journal.list);
 364 }
 365
 366 /*
 367  * this version is used by btree_gc before filesystem has gone RW and
 368  * multithreaded, so uses the journal_iters list:
 369  */
 370 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 371                                                 struct bch_fs *c,
 372                                                 struct btree *b)
 373 {
 374         struct btree_node_iter node_iter;
 375
 376         bch2_btree_node_iter_init_from_start(&node_iter, b);
 377         __bch2_btree_and_journal_iter_init_node_iter(iter, c, b, node_iter, b->data->min_key);
 378         list_add(&iter->journal.list, &c->journal_iters);
 379 }
 380
 381 /* sort and dedup all keys in the journal: */
 382
 383 void bch2_journal_entries_free(struct list_head *list)
 384 {
 385
 386         while (!list_empty(list)) {
 387                 struct journal_replay *i =
 388                         list_first_entry(list, struct journal_replay, list);
 389                 list_del(&i->list);
 390                 kvpfree(i, offsetof(struct journal_replay, j) +
 391                         vstruct_bytes(&i->j));
 392         }
 393 }
 394
 395 /*
 396  * When keys compare equal, oldest compares first:
 397  */
 398 static int journal_sort_key_cmp(const void *_l, const void *_r)
 399 {
 400         const struct journal_key *l = _l;
 401         const struct journal_key *r = _r;
 402
 403         return  journal_key_cmp(l, r) ?:
 404                 cmp_int(l->journal_seq, r->journal_seq) ?:
 405                 cmp_int(l->journal_offset, r->journal_offset);
 406 }
 407
 408 void bch2_journal_keys_free(struct journal_keys *keys)
 409 {
 410         struct journal_key *i;
 411
 412         for (i = keys->d; i < keys->d + keys->nr; i++)
 413                 if (i->allocated)
 414                         kfree(i->k);
 415
 416         kvfree(keys->d);
 417         keys->d = NULL;
 418         keys->nr = 0;
 419 }
 420
 421 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 422 {
 423         struct journal_replay *i;
 424         struct jset_entry *entry;
 425         struct bkey_i *k, *_n;
 426         struct journal_keys keys = { NULL };
 427         struct journal_key *src, *dst;
 428         size_t nr_keys = 0;
 429
 430         if (list_empty(journal_entries))
 431                 return keys;
 432
 433         list_for_each_entry(i, journal_entries, list) {
 434                 if (i->ignore)
 435                         continue;
 436
 437                 if (!keys.journal_seq_base)
 438                         keys.journal_seq_base = le64_to_cpu(i->j.seq);
 439
 440                 for_each_jset_key(k, _n, entry, &i->j)
 441                         nr_keys++;
 442         }
 443
 444         keys.size = roundup_pow_of_two(nr_keys);
 445
 446         keys.d = kvmalloc(sizeof(keys.d[0]) * keys.size, GFP_KERNEL);
 447         if (!keys.d)
 448                 goto err;
 449
 450         list_for_each_entry(i, journal_entries, list) {
 451                 if (i->ignore)
 452                         continue;
 453
 454                 BUG_ON(le64_to_cpu(i->j.seq) - keys.journal_seq_base > U32_MAX);
 455
 456                 for_each_jset_key(k, _n, entry, &i->j)
 457                         keys.d[keys.nr++] = (struct journal_key) {
 458                                 .btree_id       = entry->btree_id,
 459                                 .level          = entry->level,
 460                                 .k              = k,
 461                                 .journal_seq    = le64_to_cpu(i->j.seq) -
 462                                         keys.journal_seq_base,
 463                                 .journal_offset = k->_data - i->j._data,
 464                         };
 465         }
 466
 467         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 468
 469         src = dst = keys.d;
 470         while (src < keys.d + keys.nr) {
 471                 while (src + 1 < keys.d + keys.nr &&
 472                        src[0].btree_id  == src[1].btree_id &&
 473                        src[0].level     == src[1].level &&
 474                        !bpos_cmp(src[0].k->k.p, src[1].k->k.p))
 475                         src++;
 476
 477                 *dst++ = *src++;
 478         }
 479
 480         keys.nr = dst - keys.d;
 481 err:
 482         return keys;
 483 }
 484
 485 /* journal replay: */
 486
 487 static void replay_now_at(struct journal *j, u64 seq)
 488 {
 489         BUG_ON(seq < j->replay_journal_seq);
 490         BUG_ON(seq > j->replay_journal_seq_end);
 491
 492         while (j->replay_journal_seq < seq)
 493                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 494 }
 495
 496 static int bch2_journal_replay_key(struct btree_trans *trans,
 497                                    struct journal_key *k)
 498 {
 499         struct btree_iter iter;
 500         unsigned iter_flags =
 501                 BTREE_ITER_INTENT|
 502                 BTREE_ITER_NOT_EXTENTS;
 503         int ret;
 504
 505         if (!k->level && k->btree_id == BTREE_ID_alloc)
 506                 iter_flags |= BTREE_ITER_CACHED;
 507
 508         bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 509                                   BTREE_MAX_DEPTH, k->level,
 510                                   iter_flags);
 511         ret = bch2_btree_iter_traverse(&iter);
 512         if (ret)
 513                 goto out;
 514
 515         /* Must be checked with btree locked: */
 516         if (k->overwritten)
 517                 goto out;
 518
 519         ret = bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
 520 out:
 521         bch2_trans_iter_exit(trans, &iter);
 522         return ret;
 523 }
 524
 525 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 526 {
 527         const struct journal_key *l = *((const struct journal_key **)_l);
 528         const struct journal_key *r = *((const struct journal_key **)_r);
 529
 530         return cmp_int(l->journal_seq, r->journal_seq);
 531 }
 532
 533 static int bch2_journal_replay(struct bch_fs *c)
 534 {
 535         struct journal_keys *keys = &c->journal_keys;
 536         struct journal_key **keys_sorted, *k;
 537         struct journal *j = &c->journal;
 538         size_t i;
 539         int ret;
 540
 541         keys_sorted = kvmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL);
 542         if (!keys_sorted)
 543                 return -ENOMEM;
 544
 545         for (i = 0; i < keys->nr; i++)
 546                 keys_sorted[i] = &keys->d[i];
 547
 548         sort(keys_sorted, keys->nr,
 549              sizeof(keys_sorted[0]),
 550              journal_sort_seq_cmp, NULL);
 551
 552         if (keys->nr)
 553                 replay_now_at(j, keys->journal_seq_base);
 554
 555         for (i = 0; i < keys->nr; i++) {
 556                 k = keys_sorted[i];
 557
 558                 cond_resched();
 559
 560                 if (!k->allocated)
 561                         replay_now_at(j, keys->journal_seq_base + k->journal_seq);
 562
 563                 ret = bch2_trans_do(c, NULL, NULL,
 564                                     BTREE_INSERT_LAZY_RW|
 565                                     BTREE_INSERT_NOFAIL|
 566                                     (!k->allocated
 567                                      ? BTREE_INSERT_JOURNAL_REPLAY|JOURNAL_WATERMARK_reserved
 568                                      : 0),
 569                              bch2_journal_replay_key(&trans, k));
 570                 if (ret) {
 571                         bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
 572                                 ret, bch2_btree_ids[k->btree_id], k->level);
 573                         goto err;
 574                 }
 575         }
 576
 577         replay_now_at(j, j->replay_journal_seq_end);
 578         j->replay_journal_seq = 0;
 579
 580         bch2_journal_set_replay_done(j);
 581         bch2_journal_flush_all_pins(j);
 582         ret = bch2_journal_error(j);
 583
 584         if (keys->nr && !ret)
 585                 bch2_journal_log_msg(&c->journal, "journal replay finished");
 586 err:
 587         kvfree(keys_sorted);
 588         return ret;
 589 }
 590
 591 /* journal replay early: */
 592
 593 static int journal_replay_entry_early(struct bch_fs *c,
 594                                       struct jset_entry *entry)
 595 {
 596         int ret = 0;
 597
 598         switch (entry->type) {
 599         case BCH_JSET_ENTRY_btree_root: {
 600                 struct btree_root *r;
 601
 602                 if (entry->btree_id >= BTREE_ID_NR) {
 603                         bch_err(c, "filesystem has unknown btree type %u",
 604                                 entry->btree_id);
 605                         return -EINVAL;
 606                 }
 607
 608                 r = &c->btree_roots[entry->btree_id];
 609
 610                 if (entry->u64s) {
 611                         r->level = entry->level;
 612                         bkey_copy(&r->key, &entry->start[0]);
 613                         r->error = 0;
 614                 } else {
 615                         r->error = -EIO;
 616                 }
 617                 r->alive = true;
 618                 break;
 619         }
 620         case BCH_JSET_ENTRY_usage: {
 621                 struct jset_entry_usage *u =
 622                         container_of(entry, struct jset_entry_usage, entry);
 623
 624                 switch (entry->btree_id) {
 625                 case BCH_FS_USAGE_reserved:
 626                         if (entry->level < BCH_REPLICAS_MAX)
 627                                 c->usage_base->persistent_reserved[entry->level] =
 628                                         le64_to_cpu(u->v);
 629                         break;
 630                 case BCH_FS_USAGE_inodes:
 631                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 632                         break;
 633                 case BCH_FS_USAGE_key_version:
 634                         atomic64_set(&c->key_version,
 635                                      le64_to_cpu(u->v));
 636                         break;
 637                 }
 638
 639                 break;
 640         }
 641         case BCH_JSET_ENTRY_data_usage: {
 642                 struct jset_entry_data_usage *u =
 643                         container_of(entry, struct jset_entry_data_usage, entry);
 644
 645                 ret = bch2_replicas_set_usage(c, &u->r,
 646                                               le64_to_cpu(u->v));
 647                 break;
 648         }
 649         case BCH_JSET_ENTRY_dev_usage: {
 650                 struct jset_entry_dev_usage *u =
 651                         container_of(entry, struct jset_entry_dev_usage, entry);
 652                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 653                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 654
 655                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 656                 ca->usage_base->buckets_unavailable     = le64_to_cpu(u->buckets_unavailable);
 657
 658                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 659                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 660                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 661                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 662                 }
 663
 664                 break;
 665         }
 666         case BCH_JSET_ENTRY_blacklist: {
 667                 struct jset_entry_blacklist *bl_entry =
 668                         container_of(entry, struct jset_entry_blacklist, entry);
 669
 670                 ret = bch2_journal_seq_blacklist_add(c,
 671                                 le64_to_cpu(bl_entry->seq),
 672                                 le64_to_cpu(bl_entry->seq) + 1);
 673                 break;
 674         }
 675         case BCH_JSET_ENTRY_blacklist_v2: {
 676                 struct jset_entry_blacklist_v2 *bl_entry =
 677                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 678
 679                 ret = bch2_journal_seq_blacklist_add(c,
 680                                 le64_to_cpu(bl_entry->start),
 681                                 le64_to_cpu(bl_entry->end) + 1);
 682                 break;
 683         }
 684         case BCH_JSET_ENTRY_clock: {
 685                 struct jset_entry_clock *clock =
 686                         container_of(entry, struct jset_entry_clock, entry);
 687
 688                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 689         }
 690         }
 691
 692         return ret;
 693 }
 694
 695 static int journal_replay_early(struct bch_fs *c,
 696                                 struct bch_sb_field_clean *clean,
 697                                 struct list_head *journal)
 698 {
 699         struct journal_replay *i;
 700         struct jset_entry *entry;
 701         int ret;
 702
 703         if (clean) {
 704                 for (entry = clean->start;
 705                      entry != vstruct_end(&clean->field);
 706                      entry = vstruct_next(entry)) {
 707                         ret = journal_replay_entry_early(c, entry);
 708                         if (ret)
 709                                 return ret;
 710                 }
 711         } else {
 712                 list_for_each_entry(i, journal, list) {
 713                         if (i->ignore)
 714                                 continue;
 715
 716                         vstruct_for_each(&i->j, entry) {
 717                                 ret = journal_replay_entry_early(c, entry);
 718                                 if (ret)
 719                                         return ret;
 720                         }
 721                 }
 722         }
 723
 724         bch2_fs_usage_initialize(c);
 725
 726         return 0;
 727 }
 728
 729 /* sb clean section: */
 730
 731 static struct bkey_i *btree_root_find(struct bch_fs *c,
 732                                       struct bch_sb_field_clean *clean,
 733                                       struct jset *j,
 734                                       enum btree_id id, unsigned *level)
 735 {
 736         struct bkey_i *k;
 737         struct jset_entry *entry, *start, *end;
 738
 739         if (clean) {
 740                 start = clean->start;
 741                 end = vstruct_end(&clean->field);
 742         } else {
 743                 start = j->start;
 744                 end = vstruct_last(j);
 745         }
 746
 747         for (entry = start; entry < end; entry = vstruct_next(entry))
 748                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 749                     entry->btree_id == id)
 750                         goto found;
 751
 752         return NULL;
 753 found:
 754         if (!entry->u64s)
 755                 return ERR_PTR(-EINVAL);
 756
 757         k = entry->start;
 758         *level = entry->level;
 759         return k;
 760 }
 761
 762 static int verify_superblock_clean(struct bch_fs *c,
 763                                    struct bch_sb_field_clean **cleanp,
 764                                    struct jset *j)
 765 {
 766         unsigned i;
 767         struct bch_sb_field_clean *clean = *cleanp;
 768         struct printbuf buf1 = PRINTBUF;
 769         struct printbuf buf2 = PRINTBUF;
 770         int ret = 0;
 771
 772         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 773                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 774                         le64_to_cpu(clean->journal_seq),
 775                         le64_to_cpu(j->seq))) {
 776                 kfree(clean);
 777                 *cleanp = NULL;
 778                 return 0;
 779         }
 780
 781         for (i = 0; i < BTREE_ID_NR; i++) {
 782                 struct bkey_i *k1, *k2;
 783                 unsigned l1 = 0, l2 = 0;
 784
 785                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 786                 k2 = btree_root_find(c, NULL, j, i, &l2);
 787
 788                 if (!k1 && !k2)
 789                         continue;
 790
 791                 printbuf_reset(&buf1);
 792                 printbuf_reset(&buf2);
 793
 794                 if (k1)
 795                         bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(k1));
 796                 else
 797                         pr_buf(&buf1, "(none)");
 798
 799                 if (k2)
 800                         bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(k2));
 801                 else
 802                         pr_buf(&buf2, "(none)");
 803
 804                 mustfix_fsck_err_on(!k1 || !k2 ||
 805                                     IS_ERR(k1) ||
 806                                     IS_ERR(k2) ||
 807                                     k1->k.u64s != k2->k.u64s ||
 808                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 809                                     l1 != l2, c,
 810                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 811                         "sb:      l=%u %s\n"
 812                         "journal: l=%u %s\n", i,
 813                         l1, buf1.buf,
 814                         l2, buf2.buf);
 815         }
 816 fsck_err:
 817         printbuf_exit(&buf2);
 818         printbuf_exit(&buf1);
 819         return ret;
 820 }
 821
 822 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 823 {
 824         struct bch_sb_field_clean *clean, *sb_clean;
 825         int ret;
 826
 827         mutex_lock(&c->sb_lock);
 828         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 829
 830         if (fsck_err_on(!sb_clean, c,
 831                         "superblock marked clean but clean section not present")) {
 832                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 833                 c->sb.clean = false;
 834                 mutex_unlock(&c->sb_lock);
 835                 return NULL;
 836         }
 837
 838         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 839                         GFP_KERNEL);
 840         if (!clean) {
 841                 mutex_unlock(&c->sb_lock);
 842                 return ERR_PTR(-ENOMEM);
 843         }
 844
 845         ret = bch2_sb_clean_validate_late(c, clean, READ);
 846         if (ret) {
 847                 mutex_unlock(&c->sb_lock);
 848                 return ERR_PTR(ret);
 849         }
 850
 851         mutex_unlock(&c->sb_lock);
 852
 853         return clean;
 854 fsck_err:
 855         mutex_unlock(&c->sb_lock);
 856         return ERR_PTR(ret);
 857 }
 858
 859 static int read_btree_roots(struct bch_fs *c)
 860 {
 861         unsigned i;
 862         int ret = 0;
 863
 864         for (i = 0; i < BTREE_ID_NR; i++) {
 865                 struct btree_root *r = &c->btree_roots[i];
 866
 867                 if (!r->alive)
 868                         continue;
 869
 870                 if (i == BTREE_ID_alloc &&
 871                     c->opts.reconstruct_alloc) {
 872                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 873                         continue;
 874                 }
 875
 876                 if (r->error) {
 877                         __fsck_err(c, i == BTREE_ID_alloc
 878                                    ? FSCK_CAN_IGNORE : 0,
 879                                    "invalid btree root %s",
 880                                    bch2_btree_ids[i]);
 881                         if (i == BTREE_ID_alloc)
 882                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 883                 }
 884
 885                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 886                 if (ret) {
 887                         __fsck_err(c, i == BTREE_ID_alloc
 888                                    ? FSCK_CAN_IGNORE : 0,
 889                                    "error reading btree root %s",
 890                                    bch2_btree_ids[i]);
 891                         if (i == BTREE_ID_alloc)
 892                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 893                 }
 894         }
 895
 896         for (i = 0; i < BTREE_ID_NR; i++)
 897                 if (!c->btree_roots[i].b)
 898                         bch2_btree_root_alloc(c, i);
 899 fsck_err:
 900         return ret;
 901 }
 902
 903 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
 904 {
 905         struct bkey_i_snapshot  root_snapshot;
 906         struct bkey_i_subvolume root_volume;
 907         int ret;
 908
 909         bkey_snapshot_init(&root_snapshot.k_i);
 910         root_snapshot.k.p.offset = U32_MAX;
 911         root_snapshot.v.flags   = 0;
 912         root_snapshot.v.parent  = 0;
 913         root_snapshot.v.subvol  = BCACHEFS_ROOT_SUBVOL;
 914         root_snapshot.v.pad     = 0;
 915         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
 916
 917         ret = bch2_btree_insert(c, BTREE_ID_snapshots,
 918                                 &root_snapshot.k_i,
 919                                 NULL, NULL, 0);
 920         if (ret)
 921                 return ret;
 922
 923
 924         bkey_subvolume_init(&root_volume.k_i);
 925         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
 926         root_volume.v.flags     = 0;
 927         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
 928         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
 929
 930         ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
 931                                 &root_volume.k_i,
 932                                 NULL, NULL, 0);
 933         if (ret)
 934                 return ret;
 935
 936         return 0;
 937 }
 938
 939 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
 940 {
 941         struct btree_iter iter;
 942         struct bkey_s_c k;
 943         struct bch_inode_unpacked inode;
 944         int ret;
 945
 946         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
 947                              SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
 948         k = bch2_btree_iter_peek_slot(&iter);
 949         ret = bkey_err(k);
 950         if (ret)
 951                 goto err;
 952
 953         if (!bkey_is_inode(k.k)) {
 954                 bch_err(trans->c, "root inode not found");
 955                 ret = -ENOENT;
 956                 goto err;
 957         }
 958
 959         ret = bch2_inode_unpack(k, &inode);
 960         BUG_ON(ret);
 961
 962         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
 963
 964         ret = bch2_inode_write(trans, &iter, &inode);
 965 err:
 966         bch2_trans_iter_exit(trans, &iter);
 967         return ret;
 968 }
 969
 970 int bch2_fs_recovery(struct bch_fs *c)
 971 {
 972         const char *err = "cannot allocate memory";
 973         struct bch_sb_field_clean *clean = NULL;
 974         struct jset *last_journal_entry = NULL;
 975         u64 blacklist_seq, journal_seq;
 976         bool write_sb = false;
 977         int ret = 0;
 978
 979         if (c->sb.clean)
 980                 clean = read_superblock_clean(c);
 981         ret = PTR_ERR_OR_ZERO(clean);
 982         if (ret)
 983                 goto err;
 984
 985         if (c->sb.clean)
 986                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 987                          le64_to_cpu(clean->journal_seq));
 988         else
 989                 bch_info(c, "recovering from unclean shutdown");
 990
 991         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
 992                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
 993                 ret = -EINVAL;
 994                 goto err;
 995         }
 996
 997         if (!c->sb.clean &&
 998             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
 999                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1000                 ret = -EINVAL;
1001                 goto err;
1002         }
1003
1004         if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1005                 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
1006                 ret = -EINVAL;
1007                 goto err;
1008         }
1009
1010         if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
1011                 bch_info(c, "alloc_v2 feature bit not set, fsck required");
1012                 c->opts.fsck = true;
1013                 c->opts.fix_errors = FSCK_OPT_YES;
1014         }
1015
1016         if (!c->replicas.entries ||
1017             c->opts.rebuild_replicas) {
1018                 bch_info(c, "building replicas info");
1019                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1020         }
1021
1022         if (!c->opts.nochanges) {
1023                 if (c->sb.version < bcachefs_metadata_version_inode_backpointers) {
1024                         bch_info(c, "version prior to inode backpointers, upgrade and fsck required");
1025                         c->opts.version_upgrade = true;
1026                         c->opts.fsck            = true;
1027                         c->opts.fix_errors      = FSCK_OPT_YES;
1028                 } else if (c->sb.version < bcachefs_metadata_version_subvol_dirent) {
1029                         bch_info(c, "filesystem version is prior to subvol_dirent - upgrading");
1030                         c->opts.version_upgrade = true;
1031                         c->opts.fsck            = true;
1032                 } else if (c->sb.version < bcachefs_metadata_version_alloc_v4) {
1033                         bch_info(c, "filesystem version is prior to alloc_v4 - upgrading");
1034                         c->opts.version_upgrade = true;
1035                 }
1036         }
1037
1038         ret = bch2_blacklist_table_initialize(c);
1039         if (ret) {
1040                 bch_err(c, "error initializing blacklist table");
1041                 goto err;
1042         }
1043
1044         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1045                 struct journal_replay *i;
1046
1047                 bch_verbose(c, "starting journal read");
1048                 ret = bch2_journal_read(c, &c->journal_entries,
1049                                         &blacklist_seq, &journal_seq);
1050                 if (ret)
1051                         goto err;
1052
1053                 list_for_each_entry_reverse(i, &c->journal_entries, list)
1054                         if (!i->ignore) {
1055                                 last_journal_entry = &i->j;
1056                                 break;
1057                         }
1058
1059                 if (mustfix_fsck_err_on(c->sb.clean &&
1060                                         last_journal_entry &&
1061                                         !journal_entry_empty(last_journal_entry), c,
1062                                 "filesystem marked clean but journal not empty")) {
1063                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1064                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1065                         c->sb.clean = false;
1066                 }
1067
1068                 if (!last_journal_entry) {
1069                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
1070                         goto use_clean;
1071                 }
1072
1073                 c->journal_keys = journal_keys_sort(&c->journal_entries);
1074                 if (!c->journal_keys.d) {
1075                         ret = -ENOMEM;
1076                         goto err;
1077                 }
1078
1079                 if (c->sb.clean && last_journal_entry) {
1080                         ret = verify_superblock_clean(c, &clean,
1081                                                       last_journal_entry);
1082                         if (ret)
1083                                 goto err;
1084                 }
1085         } else {
1086 use_clean:
1087                 if (!clean) {
1088                         bch_err(c, "no superblock clean section found");
1089                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1090                         goto err;
1091
1092                 }
1093                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1094         }
1095
1096         if (c->opts.read_journal_only)
1097                 goto out;
1098
1099         if (c->opts.reconstruct_alloc) {
1100                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1101                 drop_alloc_keys(&c->journal_keys);
1102         }
1103
1104         zero_out_btree_mem_ptr(&c->journal_keys);
1105
1106         ret = journal_replay_early(c, clean, &c->journal_entries);
1107         if (ret)
1108                 goto err;
1109
1110         /*
1111          * After an unclean shutdown, skip then next few journal sequence
1112          * numbers as they may have been referenced by btree writes that
1113          * happened before their corresponding journal writes - those btree
1114          * writes need to be ignored, by skipping and blacklisting the next few
1115          * journal sequence numbers:
1116          */
1117         if (!c->sb.clean)
1118                 journal_seq += 8;
1119
1120         if (blacklist_seq != journal_seq) {
1121                 ret = bch2_journal_seq_blacklist_add(c,
1122                                         blacklist_seq, journal_seq);
1123                 if (ret) {
1124                         bch_err(c, "error creating new journal seq blacklist entry");
1125                         goto err;
1126                 }
1127         }
1128
1129         ret = bch2_fs_journal_start(&c->journal, journal_seq,
1130                                     &c->journal_entries);
1131         if (ret)
1132                 goto err;
1133
1134         ret = read_btree_roots(c);
1135         if (ret)
1136                 goto err;
1137
1138         bch_verbose(c, "starting alloc read");
1139         err = "error reading allocation information";
1140
1141         down_read(&c->gc_lock);
1142         ret = bch2_alloc_read(c);
1143         up_read(&c->gc_lock);
1144
1145         if (ret)
1146                 goto err;
1147         bch_verbose(c, "alloc read done");
1148
1149         bch_verbose(c, "starting stripes_read");
1150         err = "error reading stripes";
1151         ret = bch2_stripes_read(c);
1152         if (ret)
1153                 goto err;
1154         bch_verbose(c, "stripes_read done");
1155
1156         /*
1157          * If we're not running fsck, this ensures bch2_fsck_err() calls are
1158          * instead interpreted as bch2_inconsistent_err() calls:
1159          */
1160         if (!c->opts.fsck)
1161                 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1162
1163         if (c->opts.fsck ||
1164             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_info)) ||
1165             !(c->sb.compat & (1ULL << BCH_COMPAT_alloc_metadata)) ||
1166             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1167                 bool metadata_only = c->opts.norecovery;
1168
1169                 bch_info(c, "checking allocations");
1170                 err = "error checking allocations";
1171                 ret = bch2_gc(c, true, metadata_only);
1172                 if (ret)
1173                         goto err;
1174                 bch_verbose(c, "done checking allocations");
1175         }
1176
1177         if (c->opts.fsck) {
1178                 bch_info(c, "checking need_discard and freespace btrees");
1179                 err = "error checking need_discard and freespace btrees";
1180                 ret = bch2_check_alloc_info(c, true);
1181                 if (ret)
1182                         goto err;
1183
1184                 ret = bch2_check_lrus(c, true);
1185                 if (ret)
1186                         goto err;
1187                 bch_verbose(c, "done checking need_discard and freespace btrees");
1188         }
1189
1190         bch2_stripes_heap_start(c);
1191
1192         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1193         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1194         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1195
1196         /*
1197          * Skip past versions that might have possibly been used (as nonces),
1198          * but hadn't had their pointers written:
1199          */
1200         if (c->sb.encryption_type && !c->sb.clean)
1201                 atomic64_add(1 << 16, &c->key_version);
1202
1203         if (c->opts.norecovery)
1204                 goto out;
1205
1206         bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1207         err = "journal replay failed";
1208         ret = bch2_journal_replay(c);
1209         if (ret)
1210                 goto err;
1211         if (c->opts.verbose || !c->sb.clean)
1212                 bch_info(c, "journal replay done");
1213
1214         err = "error initializing freespace";
1215         ret = bch2_fs_freespace_init(c);
1216         if (ret)
1217                 goto err;
1218
1219         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1220                 bch2_fs_lazy_rw(c);
1221
1222                 err = "error creating root snapshot node";
1223                 ret = bch2_fs_initialize_subvolumes(c);
1224                 if (ret)
1225                         goto err;
1226         }
1227
1228         bch_verbose(c, "reading snapshots table");
1229         err = "error reading snapshots table";
1230         ret = bch2_fs_snapshots_start(c);
1231         if (ret)
1232                 goto err;
1233         bch_verbose(c, "reading snapshots done");
1234
1235         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1236                 /* set bi_subvol on root inode */
1237                 err = "error upgrade root inode for subvolumes";
1238                 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1239                                     bch2_fs_upgrade_for_subvolumes(&trans));
1240                 if (ret)
1241                         goto err;
1242         }
1243
1244         if (c->opts.fsck) {
1245                 bch_info(c, "starting fsck");
1246                 err = "error in fsck";
1247                 ret = bch2_fsck_full(c);
1248                 if (ret)
1249                         goto err;
1250                 bch_verbose(c, "fsck done");
1251         } else if (!c->sb.clean) {
1252                 bch_verbose(c, "checking for deleted inodes");
1253                 err = "error in recovery";
1254                 ret = bch2_fsck_walk_inodes_only(c);
1255                 if (ret)
1256                         goto err;
1257                 bch_verbose(c, "check inodes done");
1258         }
1259
1260         if (enabled_qtypes(c)) {
1261                 bch_verbose(c, "reading quotas");
1262                 ret = bch2_fs_quota_read(c);
1263                 if (ret)
1264                         goto err;
1265                 bch_verbose(c, "quotas done");
1266         }
1267
1268         mutex_lock(&c->sb_lock);
1269         if (c->opts.version_upgrade) {
1270                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1271                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1272                 write_sb = true;
1273         }
1274
1275         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1276                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1277                 write_sb = true;
1278         }
1279
1280         if (c->opts.fsck &&
1281             !test_bit(BCH_FS_ERROR, &c->flags) &&
1282             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1283                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1284                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1285                 write_sb = true;
1286         }
1287
1288         if (write_sb)
1289                 bch2_write_super(c);
1290         mutex_unlock(&c->sb_lock);
1291
1292         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1293             !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1294             le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) {
1295                 struct bch_move_stats stats;
1296
1297                 bch_move_stats_init(&stats, "recovery");
1298
1299                 bch_info(c, "scanning for old btree nodes");
1300                 ret = bch2_fs_read_write(c);
1301                 if (ret)
1302                         goto err;
1303
1304                 ret = bch2_scan_old_btree_nodes(c, &stats);
1305                 if (ret)
1306                         goto err;
1307                 bch_info(c, "scanning for old btree nodes done");
1308         }
1309
1310         if (c->journal_seq_blacklist_table &&
1311             c->journal_seq_blacklist_table->nr > 128)
1312                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1313
1314         ret = 0;
1315 out:
1316         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1317         bch2_flush_fsck_errs(c);
1318
1319         if (!c->opts.keep_journal) {
1320                 bch2_journal_keys_free(&c->journal_keys);
1321                 bch2_journal_entries_free(&c->journal_entries);
1322         }
1323         kfree(clean);
1324         if (ret)
1325                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1326         else
1327                 bch_verbose(c, "ret %i", ret);
1328         return ret;
1329 err:
1330 fsck_err:
1331         bch2_fs_emergency_read_only(c);
1332         goto out;
1333 }
1334
1335 int bch2_fs_initialize(struct bch_fs *c)
1336 {
1337         struct bch_inode_unpacked root_inode, lostfound_inode;
1338         struct bkey_inode_buf packed_inode;
1339         struct qstr lostfound = QSTR("lost+found");
1340         const char *err = "cannot allocate memory";
1341         struct bch_dev *ca;
1342         LIST_HEAD(journal);
1343         unsigned i;
1344         int ret;
1345
1346         bch_notice(c, "initializing new filesystem");
1347
1348         mutex_lock(&c->sb_lock);
1349         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1350         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1351
1352         if (c->opts.version_upgrade) {
1353                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1354                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1355                 bch2_write_super(c);
1356         }
1357         mutex_unlock(&c->sb_lock);
1358
1359         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1360         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1361         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1362
1363         for (i = 0; i < BTREE_ID_NR; i++)
1364                 bch2_btree_root_alloc(c, i);
1365
1366         err = "unable to allocate journal buckets";
1367         for_each_online_member(ca, c, i) {
1368                 ret = bch2_dev_journal_alloc(ca);
1369                 if (ret) {
1370                         percpu_ref_put(&ca->io_ref);
1371                         goto err;
1372                 }
1373         }
1374
1375         /*
1376          * journal_res_get() will crash if called before this has
1377          * set up the journal.pin FIFO and journal.cur pointer:
1378          */
1379         bch2_fs_journal_start(&c->journal, 1, &journal);
1380         bch2_journal_set_replay_done(&c->journal);
1381
1382         err = "error going read-write";
1383         ret = bch2_fs_read_write_early(c);
1384         if (ret)
1385                 goto err;
1386
1387         /*
1388          * Write out the superblock and journal buckets, now that we can do
1389          * btree updates
1390          */
1391         bch_verbose(c, "marking superblocks");
1392         err = "error marking superblock and journal";
1393         for_each_member_device(ca, c, i) {
1394                 ret = bch2_trans_mark_dev_sb(c, ca);
1395                 if (ret) {
1396                         percpu_ref_put(&ca->ref);
1397                         goto err;
1398                 }
1399
1400                 ca->new_fs_bucket_idx = 0;
1401         }
1402
1403         bch_verbose(c, "initializing freespace");
1404         err = "error initializing freespace";
1405         ret = bch2_fs_freespace_init(c);
1406         if (ret)
1407                 goto err;
1408
1409         err = "error creating root snapshot node";
1410         ret = bch2_fs_initialize_subvolumes(c);
1411         if (ret)
1412                 goto err;
1413
1414         bch_verbose(c, "reading snapshots table");
1415         err = "error reading snapshots table";
1416         ret = bch2_fs_snapshots_start(c);
1417         if (ret)
1418                 goto err;
1419         bch_verbose(c, "reading snapshots done");
1420
1421         bch2_inode_init(c, &root_inode, 0, 0,
1422                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1423         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1424         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1425         bch2_inode_pack(c, &packed_inode, &root_inode);
1426         packed_inode.inode.k.p.snapshot = U32_MAX;
1427
1428         err = "error creating root directory";
1429         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1430                                 &packed_inode.inode.k_i,
1431                                 NULL, NULL, 0);
1432         if (ret)
1433                 goto err;
1434
1435         bch2_inode_init_early(c, &lostfound_inode);
1436
1437         err = "error creating lost+found";
1438         ret = bch2_trans_do(c, NULL, NULL, 0,
1439                 bch2_create_trans(&trans,
1440                                   BCACHEFS_ROOT_SUBVOL_INUM,
1441                                   &root_inode, &lostfound_inode,
1442                                   &lostfound,
1443                                   0, 0, S_IFDIR|0700, 0,
1444                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1445         if (ret) {
1446                 bch_err(c, "error creating lost+found");
1447                 goto err;
1448         }
1449
1450         if (enabled_qtypes(c)) {
1451                 ret = bch2_fs_quota_read(c);
1452                 if (ret)
1453                         goto err;
1454         }
1455
1456         err = "error writing first journal entry";
1457         ret = bch2_journal_flush(&c->journal);
1458         if (ret)
1459                 goto err;
1460
1461         mutex_lock(&c->sb_lock);
1462         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1463         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1464
1465         bch2_write_super(c);
1466         mutex_unlock(&c->sb_lock);
1467
1468         return 0;
1469 err:
1470         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1471         return ret;
1472 }