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

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "bkey_buf.h"
   5 #include "alloc_background.h"
   6 #include "btree_gc.h"
   7 #include "btree_update.h"
   8 #include "btree_update_interior.h"
   9 #include "btree_io.h"
  10 #include "buckets.h"
  11 #include "dirent.h"
  12 #include "ec.h"
  13 #include "error.h"
  14 #include "fs-common.h"
  15 #include "fsck.h"
  16 #include "journal_io.h"
  17 #include "journal_reclaim.h"
  18 #include "journal_seq_blacklist.h"
  19 #include "lru.h"
  20 #include "move.h"
  21 #include "quota.h"
  22 #include "recovery.h"
  23 #include "replicas.h"
  24 #include "subvolume.h"
  25 #include "super-io.h"
  26
  27 #include <linux/sort.h>
  28 #include <linux/stat.h>
  29
  30 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  31
  32 /* for -o reconstruct_alloc: */
  33 static void drop_alloc_keys(struct journal_keys *keys)
  34 {
  35         size_t src, dst;
  36
  37         for (src = 0, dst = 0; src < keys->nr; src++)
  38                 if (keys->d[src].btree_id != BTREE_ID_alloc)
  39                         keys->d[dst++] = keys->d[src];
  40
  41         keys->nr = dst;
  42 }
  43
  44 /*
  45  * Btree node pointers have a field to stack a pointer to the in memory btree
  46  * node; we need to zero out this field when reading in btree nodes, or when
  47  * reading in keys from the journal:
  48  */
  49 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
  50 {
  51         struct journal_key *i;
  52
  53         for (i = keys->d; i < keys->d + keys->nr; i++)
  54                 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
  55                         bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
  56 }
  57
  58 /* iterate over keys read from the journal: */
  59
  60 static int __journal_key_cmp(enum btree_id      l_btree_id,
  61                              unsigned           l_level,
  62                              struct bpos        l_pos,
  63                              const struct journal_key *r)
  64 {
  65         return (cmp_int(l_btree_id,     r->btree_id) ?:
  66                 cmp_int(l_level,        r->level) ?:
  67                 bpos_cmp(l_pos, r->k->k.p));
  68 }
  69
  70 static int journal_key_cmp(const struct journal_key *l, const struct journal_key *r)
  71 {
  72         return __journal_key_cmp(l->btree_id, l->level, l->k->k.p, r);
  73 }
  74
  75 static inline size_t idx_to_pos(struct journal_keys *keys, size_t idx)
  76 {
  77         size_t gap_size = keys->size - keys->nr;
  78
  79         if (idx >= keys->gap)
  80                 idx += gap_size;
  81         return idx;
  82 }
  83
  84 static inline struct journal_key *idx_to_key(struct journal_keys *keys, size_t idx)
  85 {
  86         return keys->d + idx_to_pos(keys, idx);
  87 }
  88
  89 static size_t bch2_journal_key_search(struct journal_keys *keys,
  90                                       enum btree_id id, unsigned level,
  91                                       struct bpos pos)
  92 {
  93         size_t l = 0, r = keys->nr, m;
  94
  95         while (l < r) {
  96                 m = l + ((r - l) >> 1);
  97                 if (__journal_key_cmp(id, level, pos, idx_to_key(keys, m)) > 0)
  98                         l = m + 1;
  99                 else
 100                         r = m;
 101         }
 102
 103         BUG_ON(l < keys->nr &&
 104                __journal_key_cmp(id, level, pos, idx_to_key(keys, l)) > 0);
 105
 106         BUG_ON(l &&
 107                __journal_key_cmp(id, level, pos, idx_to_key(keys, l - 1)) <= 0);
 108
 109         return idx_to_pos(keys, l);
 110 }
 111
 112 struct bkey_i *bch2_journal_keys_peek_upto(struct bch_fs *c, enum btree_id btree_id,
 113                                            unsigned level, struct bpos pos,
 114                                            struct bpos end_pos, size_t *idx)
 115 {
 116         struct journal_keys *keys = &c->journal_keys;
 117         unsigned iters = 0;
 118 search:
 119         if (!*idx)
 120                 *idx = bch2_journal_key_search(keys, btree_id, level, pos);
 121
 122         while (*idx < keys->size &&
 123                keys->d[*idx].btree_id == btree_id &&
 124                keys->d[*idx].level == level &&
 125                bpos_cmp(keys->d[*idx].k->k.p, end_pos) <= 0) {
 126                 if (bpos_cmp(keys->d[*idx].k->k.p, pos) >= 0 &&
 127                     !keys->d[*idx].overwritten)
 128                         return keys->d[*idx].k;
 129
 130                 (*idx)++;
 131                 if (*idx == keys->gap)
 132                         *idx += keys->size - keys->nr;
 133
 134                 iters++;
 135                 if (iters == 10) {
 136                         *idx = 0;
 137                         goto search;
 138                 }
 139         }
 140
 141         return NULL;
 142 }
 143
 144 struct bkey_i *bch2_journal_keys_peek_slot(struct bch_fs *c, enum btree_id btree_id,
 145                                            unsigned level, struct bpos pos)
 146 {
 147         size_t idx = 0;
 148
 149         return bch2_journal_keys_peek_upto(c, btree_id, level, pos, pos, &idx);
 150 }
 151
 152 static void journal_iters_fix(struct bch_fs *c)
 153 {
 154         struct journal_keys *keys = &c->journal_keys;
 155         /* The key we just inserted is immediately before the gap: */
 156         size_t gap_end = keys->gap + (keys->size - keys->nr);
 157         struct btree_and_journal_iter *iter;
 158
 159         /*
 160          * If an iterator points one after the key we just inserted, decrement
 161          * the iterator so it points at the key we just inserted - if the
 162          * decrement was unnecessary, bch2_btree_and_journal_iter_peek() will
 163          * handle that:
 164          */
 165         list_for_each_entry(iter, &c->journal_iters, journal.list)
 166                 if (iter->journal.idx == gap_end)
 167                         iter->journal.idx = keys->gap - 1;
 168 }
 169
 170 static void journal_iters_move_gap(struct bch_fs *c, size_t old_gap, size_t new_gap)
 171 {
 172         struct journal_keys *keys = &c->journal_keys;
 173         struct journal_iter *iter;
 174         size_t gap_size = keys->size - keys->nr;
 175
 176         list_for_each_entry(iter, &c->journal_iters, list) {
 177                 if (iter->idx > old_gap)
 178                         iter->idx -= gap_size;
 179                 if (iter->idx >= new_gap)
 180                         iter->idx += gap_size;
 181         }
 182 }
 183
 184 int bch2_journal_key_insert_take(struct bch_fs *c, enum btree_id id,
 185                                  unsigned level, struct bkey_i *k)
 186 {
 187         struct journal_key n = {
 188                 .btree_id       = id,
 189                 .level          = level,
 190                 .k              = k,
 191                 .allocated      = true,
 192                 /*
 193                  * Ensure these keys are done last by journal replay, to unblock
 194                  * journal reclaim:
 195                  */
 196                 .journal_seq    = U32_MAX,
 197         };
 198         struct journal_keys *keys = &c->journal_keys;
 199         size_t idx = bch2_journal_key_search(keys, id, level, k->k.p);
 200
 201         BUG_ON(test_bit(BCH_FS_RW, &c->flags));
 202
 203         if (idx < keys->size &&
 204             journal_key_cmp(&n, &keys->d[idx]) == 0) {
 205                 if (keys->d[idx].allocated)
 206                         kfree(keys->d[idx].k);
 207                 keys->d[idx] = n;
 208                 return 0;
 209         }
 210
 211         if (idx > keys->gap)
 212                 idx -= keys->size - keys->nr;
 213
 214         if (keys->nr == keys->size) {
 215                 struct journal_keys new_keys = {
 216                         .nr                     = keys->nr,
 217                         .size                   = max_t(size_t, keys->size, 8) * 2,
 218                         .journal_seq_base       = keys->journal_seq_base,
 219                 };
 220
 221                 new_keys.d = kvmalloc(sizeof(new_keys.d[0]) * new_keys.size, GFP_KERNEL);
 222                 if (!new_keys.d) {
 223                         bch_err(c, "%s: error allocating new key array (size %zu)",
 224                                 __func__, new_keys.size);
 225                         return -ENOMEM;
 226                 }
 227
 228                 /* Since @keys was full, there was no gap: */
 229                 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
 230                 kvfree(keys->d);
 231                 *keys = new_keys;
 232
 233                 /* And now the gap is at the end: */
 234                 keys->gap = keys->nr;
 235         }
 236
 237         journal_iters_move_gap(c, keys->gap, idx);
 238
 239         move_gap(keys->d, keys->nr, keys->size, keys->gap, idx);
 240         keys->gap = idx;
 241
 242         keys->nr++;
 243         keys->d[keys->gap++] = n;
 244
 245         journal_iters_fix(c);
 246
 247         return 0;
 248 }
 249
 250 /*
 251  * Can only be used from the recovery thread while we're still RO - can't be
 252  * used once we've got RW, as journal_keys is at that point used by multiple
 253  * threads:
 254  */
 255 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
 256                             unsigned level, struct bkey_i *k)
 257 {
 258         struct bkey_i *n;
 259         int ret;
 260
 261         n = kmalloc(bkey_bytes(&k->k), GFP_KERNEL);
 262         if (!n)
 263                 return -ENOMEM;
 264
 265         bkey_copy(n, k);
 266         ret = bch2_journal_key_insert_take(c, id, level, n);
 267         if (ret)
 268                 kfree(n);
 269         return ret;
 270 }
 271
 272 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
 273                             unsigned level, struct bpos pos)
 274 {
 275         struct bkey_i whiteout;
 276
 277         bkey_init(&whiteout.k);
 278         whiteout.k.p = pos;
 279
 280         return bch2_journal_key_insert(c, id, level, &whiteout);
 281 }
 282
 283 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
 284                                   unsigned level, struct bpos pos)
 285 {
 286         struct journal_keys *keys = &c->journal_keys;
 287         size_t idx = bch2_journal_key_search(keys, btree, level, pos);
 288
 289         if (idx < keys->size &&
 290             keys->d[idx].btree_id       == btree &&
 291             keys->d[idx].level          == level &&
 292             !bpos_cmp(keys->d[idx].k->k.p, pos))
 293                 keys->d[idx].overwritten = true;
 294 }
 295
 296 static void bch2_journal_iter_advance(struct journal_iter *iter)
 297 {
 298         if (iter->idx < iter->keys->size) {
 299                 iter->idx++;
 300                 if (iter->idx == iter->keys->gap)
 301                         iter->idx += iter->keys->size - iter->keys->nr;
 302         }
 303 }
 304
 305 struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
 306 {
 307         struct journal_key *k = iter->keys->d + iter->idx;
 308
 309         while (k < iter->keys->d + iter->keys->size &&
 310                k->btree_id      == iter->btree_id &&
 311                k->level         == iter->level) {
 312                 if (!k->overwritten)
 313                         return bkey_i_to_s_c(k->k);
 314
 315                 bch2_journal_iter_advance(iter);
 316                 k = iter->keys->d + iter->idx;
 317         }
 318
 319         return bkey_s_c_null;
 320 }
 321
 322 static void bch2_journal_iter_exit(struct journal_iter *iter)
 323 {
 324         list_del(&iter->list);
 325 }
 326
 327 static void bch2_journal_iter_init(struct bch_fs *c,
 328                                    struct journal_iter *iter,
 329                                    enum btree_id id, unsigned level,
 330                                    struct bpos pos)
 331 {
 332         iter->btree_id  = id;
 333         iter->level     = level;
 334         iter->keys      = &c->journal_keys;
 335         iter->idx       = bch2_journal_key_search(&c->journal_keys, id, level, pos);
 336 }
 337
 338 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 339 {
 340         return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 341                                                 iter->b, &iter->unpacked);
 342 }
 343
 344 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 345 {
 346         bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 347 }
 348
 349 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 350 {
 351         if (!bpos_cmp(iter->pos, SPOS_MAX))
 352                 iter->at_end = true;
 353         else
 354                 iter->pos = bpos_successor(iter->pos);
 355 }
 356
 357 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 358 {
 359         struct bkey_s_c btree_k, journal_k, ret;
 360 again:
 361         if (iter->at_end)
 362                 return bkey_s_c_null;
 363
 364         while ((btree_k = bch2_journal_iter_peek_btree(iter)).k &&
 365                bpos_cmp(btree_k.k->p, iter->pos) < 0)
 366                 bch2_journal_iter_advance_btree(iter);
 367
 368         while ((journal_k = bch2_journal_iter_peek(&iter->journal)).k &&
 369                bpos_cmp(journal_k.k->p, iter->pos) < 0)
 370                 bch2_journal_iter_advance(&iter->journal);
 371
 372         ret = journal_k.k &&
 373                 (!btree_k.k || bpos_cmp(journal_k.k->p, btree_k.k->p) <= 0)
 374                 ? journal_k
 375                 : btree_k;
 376
 377         if (ret.k && iter->b && bpos_cmp(ret.k->p, iter->b->data->max_key) > 0)
 378                 ret = bkey_s_c_null;
 379
 380         if (ret.k) {
 381                 iter->pos = ret.k->p;
 382                 if (bkey_deleted(ret.k)) {
 383                         bch2_btree_and_journal_iter_advance(iter);
 384                         goto again;
 385                 }
 386         } else {
 387                 iter->pos = SPOS_MAX;
 388                 iter->at_end = true;
 389         }
 390
 391         return ret;
 392 }
 393
 394 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
 395 {
 396         bch2_journal_iter_exit(&iter->journal);
 397 }
 398
 399 void __bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 400                                                   struct bch_fs *c,
 401                                                   struct btree *b,
 402                                                   struct btree_node_iter node_iter,
 403                                                   struct bpos pos)
 404 {
 405         memset(iter, 0, sizeof(*iter));
 406
 407         iter->b = b;
 408         iter->node_iter = node_iter;
 409         bch2_journal_iter_init(c, &iter->journal, b->c.btree_id, b->c.level, pos);
 410         INIT_LIST_HEAD(&iter->journal.list);
 411         iter->pos = b->data->min_key;
 412         iter->at_end = false;
 413 }
 414
 415 /*
 416  * this version is used by btree_gc before filesystem has gone RW and
 417  * multithreaded, so uses the journal_iters list:
 418  */
 419 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 420                                                 struct bch_fs *c,
 421                                                 struct btree *b)
 422 {
 423         struct btree_node_iter node_iter;
 424
 425         bch2_btree_node_iter_init_from_start(&node_iter, b);
 426         __bch2_btree_and_journal_iter_init_node_iter(iter, c, b, node_iter, b->data->min_key);
 427         list_add(&iter->journal.list, &c->journal_iters);
 428 }
 429
 430 /* sort and dedup all keys in the journal: */
 431
 432 void bch2_journal_entries_free(struct bch_fs *c)
 433 {
 434         struct journal_replay **i;
 435         struct genradix_iter iter;
 436
 437         genradix_for_each(&c->journal_entries, iter, i)
 438                 if (*i)
 439                         kvpfree(*i, offsetof(struct journal_replay, j) +
 440                                 vstruct_bytes(&(*i)->j));
 441         genradix_free(&c->journal_entries);
 442 }
 443
 444 /*
 445  * When keys compare equal, oldest compares first:
 446  */
 447 static int journal_sort_key_cmp(const void *_l, const void *_r)
 448 {
 449         const struct journal_key *l = _l;
 450         const struct journal_key *r = _r;
 451
 452         return  journal_key_cmp(l, r) ?:
 453                 cmp_int(l->journal_seq, r->journal_seq) ?:
 454                 cmp_int(l->journal_offset, r->journal_offset);
 455 }
 456
 457 void bch2_journal_keys_free(struct journal_keys *keys)
 458 {
 459         struct journal_key *i;
 460
 461         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 462         keys->gap = keys->nr;
 463
 464         for (i = keys->d; i < keys->d + keys->nr; i++)
 465                 if (i->allocated)
 466                         kfree(i->k);
 467
 468         kvfree(keys->d);
 469         keys->d = NULL;
 470         keys->nr = keys->gap = keys->size = 0;
 471 }
 472
 473 static int journal_keys_sort(struct bch_fs *c)
 474 {
 475         struct genradix_iter iter;
 476         struct journal_replay *i, **_i;
 477         struct jset_entry *entry;
 478         struct bkey_i *k, *_n;
 479         struct journal_keys *keys = &c->journal_keys;
 480         struct journal_key *src, *dst;
 481         size_t nr_keys = 0;
 482
 483         genradix_for_each(&c->journal_entries, iter, _i) {
 484                 i = *_i;
 485
 486                 if (!i || i->ignore)
 487                         continue;
 488
 489                 if (!keys->journal_seq_base)
 490                         keys->journal_seq_base = le64_to_cpu(i->j.seq);
 491
 492                 for_each_jset_key(k, _n, entry, &i->j)
 493                         nr_keys++;
 494         }
 495
 496         if (!nr_keys)
 497                 return 0;
 498
 499         keys->size = roundup_pow_of_two(nr_keys);
 500
 501         keys->d = kvmalloc(sizeof(keys->d[0]) * keys->size, GFP_KERNEL);
 502         if (!keys->d)
 503                 return -ENOMEM;
 504
 505         genradix_for_each(&c->journal_entries, iter, _i) {
 506                 i = *_i;
 507
 508                 if (!i || i->ignore)
 509                         continue;
 510
 511                 BUG_ON(le64_to_cpu(i->j.seq) - keys->journal_seq_base > U32_MAX);
 512
 513                 for_each_jset_key(k, _n, entry, &i->j)
 514                         keys->d[keys->nr++] = (struct journal_key) {
 515                                 .btree_id       = entry->btree_id,
 516                                 .level          = entry->level,
 517                                 .k              = k,
 518                                 .journal_seq    = le64_to_cpu(i->j.seq) -
 519                                         keys->journal_seq_base,
 520                                 .journal_offset = k->_data - i->j._data,
 521                         };
 522         }
 523
 524         sort(keys->d, keys->nr, sizeof(keys->d[0]), journal_sort_key_cmp, NULL);
 525
 526         src = dst = keys->d;
 527         while (src < keys->d + keys->nr) {
 528                 while (src + 1 < keys->d + keys->nr &&
 529                        src[0].btree_id  == src[1].btree_id &&
 530                        src[0].level     == src[1].level &&
 531                        !bpos_cmp(src[0].k->k.p, src[1].k->k.p))
 532                         src++;
 533
 534                 *dst++ = *src++;
 535         }
 536
 537         keys->nr = dst - keys->d;
 538         keys->gap = keys->nr;
 539         return 0;
 540 }
 541
 542 /* journal replay: */
 543
 544 static void replay_now_at(struct journal *j, u64 seq)
 545 {
 546         BUG_ON(seq < j->replay_journal_seq);
 547
 548         seq = min(seq, j->replay_journal_seq_end);
 549
 550         while (j->replay_journal_seq < seq)
 551                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 552 }
 553
 554 static int bch2_journal_replay_key(struct btree_trans *trans,
 555                                    struct journal_key *k)
 556 {
 557         struct btree_iter iter;
 558         unsigned iter_flags =
 559                 BTREE_ITER_INTENT|
 560                 BTREE_ITER_NOT_EXTENTS;
 561         int ret;
 562
 563         if (!k->level && k->btree_id == BTREE_ID_alloc)
 564                 iter_flags |= BTREE_ITER_CACHED;
 565
 566         bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 567                                   BTREE_MAX_DEPTH, k->level,
 568                                   iter_flags);
 569         ret = bch2_btree_iter_traverse(&iter);
 570         if (ret)
 571                 goto out;
 572
 573         /* Must be checked with btree locked: */
 574         if (k->overwritten)
 575                 goto out;
 576
 577         ret = bch2_trans_update(trans, &iter, k->k, BTREE_TRIGGER_NORUN);
 578 out:
 579         bch2_trans_iter_exit(trans, &iter);
 580         return ret;
 581 }
 582
 583 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 584 {
 585         const struct journal_key *l = *((const struct journal_key **)_l);
 586         const struct journal_key *r = *((const struct journal_key **)_r);
 587
 588         return cmp_int(l->journal_seq, r->journal_seq);
 589 }
 590
 591 static int bch2_journal_replay(struct bch_fs *c)
 592 {
 593         struct journal_keys *keys = &c->journal_keys;
 594         struct journal_key **keys_sorted, *k;
 595         struct journal *j = &c->journal;
 596         size_t i;
 597         int ret;
 598
 599         move_gap(keys->d, keys->nr, keys->size, keys->gap, keys->nr);
 600         keys->gap = keys->nr;
 601
 602         keys_sorted = kvmalloc_array(sizeof(*keys_sorted), keys->nr, GFP_KERNEL);
 603         if (!keys_sorted)
 604                 return -ENOMEM;
 605
 606         for (i = 0; i < keys->nr; i++)
 607                 keys_sorted[i] = &keys->d[i];
 608
 609         sort(keys_sorted, keys->nr,
 610              sizeof(keys_sorted[0]),
 611              journal_sort_seq_cmp, NULL);
 612
 613         if (keys->nr)
 614                 replay_now_at(j, keys->journal_seq_base);
 615
 616         for (i = 0; i < keys->nr; i++) {
 617                 k = keys_sorted[i];
 618
 619                 cond_resched();
 620
 621                 replay_now_at(j, keys->journal_seq_base + k->journal_seq);
 622
 623                 ret = bch2_trans_do(c, NULL, NULL,
 624                                     BTREE_INSERT_LAZY_RW|
 625                                     BTREE_INSERT_NOFAIL|
 626                                     (!k->allocated
 627                                      ? BTREE_INSERT_JOURNAL_REPLAY|JOURNAL_WATERMARK_reserved
 628                                      : 0),
 629                              bch2_journal_replay_key(&trans, k));
 630                 if (ret) {
 631                         bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
 632                                 ret, bch2_btree_ids[k->btree_id], k->level);
 633                         goto err;
 634                 }
 635         }
 636
 637         replay_now_at(j, j->replay_journal_seq_end);
 638         j->replay_journal_seq = 0;
 639
 640         bch2_journal_set_replay_done(j);
 641         bch2_journal_flush_all_pins(j);
 642         ret = bch2_journal_error(j);
 643
 644         if (keys->nr && !ret)
 645                 bch2_journal_log_msg(&c->journal, "journal replay finished");
 646 err:
 647         kvfree(keys_sorted);
 648         return ret;
 649 }
 650
 651 /* journal replay early: */
 652
 653 static int journal_replay_entry_early(struct bch_fs *c,
 654                                       struct jset_entry *entry)
 655 {
 656         int ret = 0;
 657
 658         switch (entry->type) {
 659         case BCH_JSET_ENTRY_btree_root: {
 660                 struct btree_root *r;
 661
 662                 if (entry->btree_id >= BTREE_ID_NR) {
 663                         bch_err(c, "filesystem has unknown btree type %u",
 664                                 entry->btree_id);
 665                         return -EINVAL;
 666                 }
 667
 668                 r = &c->btree_roots[entry->btree_id];
 669
 670                 if (entry->u64s) {
 671                         r->level = entry->level;
 672                         bkey_copy(&r->key, &entry->start[0]);
 673                         r->error = 0;
 674                 } else {
 675                         r->error = -EIO;
 676                 }
 677                 r->alive = true;
 678                 break;
 679         }
 680         case BCH_JSET_ENTRY_usage: {
 681                 struct jset_entry_usage *u =
 682                         container_of(entry, struct jset_entry_usage, entry);
 683
 684                 switch (entry->btree_id) {
 685                 case BCH_FS_USAGE_reserved:
 686                         if (entry->level < BCH_REPLICAS_MAX)
 687                                 c->usage_base->persistent_reserved[entry->level] =
 688                                         le64_to_cpu(u->v);
 689                         break;
 690                 case BCH_FS_USAGE_inodes:
 691                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 692                         break;
 693                 case BCH_FS_USAGE_key_version:
 694                         atomic64_set(&c->key_version,
 695                                      le64_to_cpu(u->v));
 696                         break;
 697                 }
 698
 699                 break;
 700         }
 701         case BCH_JSET_ENTRY_data_usage: {
 702                 struct jset_entry_data_usage *u =
 703                         container_of(entry, struct jset_entry_data_usage, entry);
 704
 705                 ret = bch2_replicas_set_usage(c, &u->r,
 706                                               le64_to_cpu(u->v));
 707                 break;
 708         }
 709         case BCH_JSET_ENTRY_dev_usage: {
 710                 struct jset_entry_dev_usage *u =
 711                         container_of(entry, struct jset_entry_dev_usage, entry);
 712                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 713                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 714
 715                 ca->usage_base->buckets_ec              = le64_to_cpu(u->buckets_ec);
 716
 717                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 718                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 719                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 720                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 721                 }
 722
 723                 break;
 724         }
 725         case BCH_JSET_ENTRY_blacklist: {
 726                 struct jset_entry_blacklist *bl_entry =
 727                         container_of(entry, struct jset_entry_blacklist, entry);
 728
 729                 ret = bch2_journal_seq_blacklist_add(c,
 730                                 le64_to_cpu(bl_entry->seq),
 731                                 le64_to_cpu(bl_entry->seq) + 1);
 732                 break;
 733         }
 734         case BCH_JSET_ENTRY_blacklist_v2: {
 735                 struct jset_entry_blacklist_v2 *bl_entry =
 736                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 737
 738                 ret = bch2_journal_seq_blacklist_add(c,
 739                                 le64_to_cpu(bl_entry->start),
 740                                 le64_to_cpu(bl_entry->end) + 1);
 741                 break;
 742         }
 743         case BCH_JSET_ENTRY_clock: {
 744                 struct jset_entry_clock *clock =
 745                         container_of(entry, struct jset_entry_clock, entry);
 746
 747                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 748         }
 749         }
 750
 751         return ret;
 752 }
 753
 754 static int journal_replay_early(struct bch_fs *c,
 755                                 struct bch_sb_field_clean *clean)
 756 {
 757         struct jset_entry *entry;
 758         int ret;
 759
 760         if (clean) {
 761                 for (entry = clean->start;
 762                      entry != vstruct_end(&clean->field);
 763                      entry = vstruct_next(entry)) {
 764                         ret = journal_replay_entry_early(c, entry);
 765                         if (ret)
 766                                 return ret;
 767                 }
 768         } else {
 769                 struct genradix_iter iter;
 770                 struct journal_replay *i, **_i;
 771
 772                 genradix_for_each(&c->journal_entries, iter, _i) {
 773                         i = *_i;
 774
 775                         if (!i || i->ignore)
 776                                 continue;
 777
 778                         vstruct_for_each(&i->j, entry) {
 779                                 ret = journal_replay_entry_early(c, entry);
 780                                 if (ret)
 781                                         return ret;
 782                         }
 783                 }
 784         }
 785
 786         bch2_fs_usage_initialize(c);
 787
 788         return 0;
 789 }
 790
 791 /* sb clean section: */
 792
 793 static struct bkey_i *btree_root_find(struct bch_fs *c,
 794                                       struct bch_sb_field_clean *clean,
 795                                       struct jset *j,
 796                                       enum btree_id id, unsigned *level)
 797 {
 798         struct bkey_i *k;
 799         struct jset_entry *entry, *start, *end;
 800
 801         if (clean) {
 802                 start = clean->start;
 803                 end = vstruct_end(&clean->field);
 804         } else {
 805                 start = j->start;
 806                 end = vstruct_last(j);
 807         }
 808
 809         for (entry = start; entry < end; entry = vstruct_next(entry))
 810                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 811                     entry->btree_id == id)
 812                         goto found;
 813
 814         return NULL;
 815 found:
 816         if (!entry->u64s)
 817                 return ERR_PTR(-EINVAL);
 818
 819         k = entry->start;
 820         *level = entry->level;
 821         return k;
 822 }
 823
 824 static int verify_superblock_clean(struct bch_fs *c,
 825                                    struct bch_sb_field_clean **cleanp,
 826                                    struct jset *j)
 827 {
 828         unsigned i;
 829         struct bch_sb_field_clean *clean = *cleanp;
 830         struct printbuf buf1 = PRINTBUF;
 831         struct printbuf buf2 = PRINTBUF;
 832         int ret = 0;
 833
 834         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 835                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 836                         le64_to_cpu(clean->journal_seq),
 837                         le64_to_cpu(j->seq))) {
 838                 kfree(clean);
 839                 *cleanp = NULL;
 840                 return 0;
 841         }
 842
 843         for (i = 0; i < BTREE_ID_NR; i++) {
 844                 struct bkey_i *k1, *k2;
 845                 unsigned l1 = 0, l2 = 0;
 846
 847                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 848                 k2 = btree_root_find(c, NULL, j, i, &l2);
 849
 850                 if (!k1 && !k2)
 851                         continue;
 852
 853                 printbuf_reset(&buf1);
 854                 printbuf_reset(&buf2);
 855
 856                 if (k1)
 857                         bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(k1));
 858                 else
 859                         pr_buf(&buf1, "(none)");
 860
 861                 if (k2)
 862                         bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(k2));
 863                 else
 864                         pr_buf(&buf2, "(none)");
 865
 866                 mustfix_fsck_err_on(!k1 || !k2 ||
 867                                     IS_ERR(k1) ||
 868                                     IS_ERR(k2) ||
 869                                     k1->k.u64s != k2->k.u64s ||
 870                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 871                                     l1 != l2, c,
 872                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 873                         "sb:      l=%u %s\n"
 874                         "journal: l=%u %s\n", i,
 875                         l1, buf1.buf,
 876                         l2, buf2.buf);
 877         }
 878 fsck_err:
 879         printbuf_exit(&buf2);
 880         printbuf_exit(&buf1);
 881         return ret;
 882 }
 883
 884 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 885 {
 886         struct bch_sb_field_clean *clean, *sb_clean;
 887         int ret;
 888
 889         mutex_lock(&c->sb_lock);
 890         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 891
 892         if (fsck_err_on(!sb_clean, c,
 893                         "superblock marked clean but clean section not present")) {
 894                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 895                 c->sb.clean = false;
 896                 mutex_unlock(&c->sb_lock);
 897                 return NULL;
 898         }
 899
 900         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 901                         GFP_KERNEL);
 902         if (!clean) {
 903                 mutex_unlock(&c->sb_lock);
 904                 return ERR_PTR(-ENOMEM);
 905         }
 906
 907         ret = bch2_sb_clean_validate_late(c, clean, READ);
 908         if (ret) {
 909                 mutex_unlock(&c->sb_lock);
 910                 return ERR_PTR(ret);
 911         }
 912
 913         mutex_unlock(&c->sb_lock);
 914
 915         return clean;
 916 fsck_err:
 917         mutex_unlock(&c->sb_lock);
 918         return ERR_PTR(ret);
 919 }
 920
 921 static int read_btree_roots(struct bch_fs *c)
 922 {
 923         unsigned i;
 924         int ret = 0;
 925
 926         for (i = 0; i < BTREE_ID_NR; i++) {
 927                 struct btree_root *r = &c->btree_roots[i];
 928
 929                 if (!r->alive)
 930                         continue;
 931
 932                 if (i == BTREE_ID_alloc &&
 933                     c->opts.reconstruct_alloc) {
 934                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 935                         continue;
 936                 }
 937
 938                 if (r->error) {
 939                         __fsck_err(c, i == BTREE_ID_alloc
 940                                    ? FSCK_CAN_IGNORE : 0,
 941                                    "invalid btree root %s",
 942                                    bch2_btree_ids[i]);
 943                         if (i == BTREE_ID_alloc)
 944                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 945                 }
 946
 947                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 948                 if (ret) {
 949                         __fsck_err(c, i == BTREE_ID_alloc
 950                                    ? FSCK_CAN_IGNORE : 0,
 951                                    "error reading btree root %s",
 952                                    bch2_btree_ids[i]);
 953                         if (i == BTREE_ID_alloc)
 954                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 955                 }
 956         }
 957
 958         for (i = 0; i < BTREE_ID_NR; i++)
 959                 if (!c->btree_roots[i].b)
 960                         bch2_btree_root_alloc(c, i);
 961 fsck_err:
 962         return ret;
 963 }
 964
 965 static int bch2_fs_initialize_subvolumes(struct bch_fs *c)
 966 {
 967         struct bkey_i_snapshot  root_snapshot;
 968         struct bkey_i_subvolume root_volume;
 969         int ret;
 970
 971         bkey_snapshot_init(&root_snapshot.k_i);
 972         root_snapshot.k.p.offset = U32_MAX;
 973         root_snapshot.v.flags   = 0;
 974         root_snapshot.v.parent  = 0;
 975         root_snapshot.v.subvol  = BCACHEFS_ROOT_SUBVOL;
 976         root_snapshot.v.pad     = 0;
 977         SET_BCH_SNAPSHOT_SUBVOL(&root_snapshot.v, true);
 978
 979         ret = bch2_btree_insert(c, BTREE_ID_snapshots,
 980                                 &root_snapshot.k_i,
 981                                 NULL, NULL, 0);
 982         if (ret)
 983                 return ret;
 984
 985         bkey_subvolume_init(&root_volume.k_i);
 986         root_volume.k.p.offset = BCACHEFS_ROOT_SUBVOL;
 987         root_volume.v.flags     = 0;
 988         root_volume.v.snapshot  = cpu_to_le32(U32_MAX);
 989         root_volume.v.inode     = cpu_to_le64(BCACHEFS_ROOT_INO);
 990
 991         ret = bch2_btree_insert(c, BTREE_ID_subvolumes,
 992                                 &root_volume.k_i,
 993                                 NULL, NULL, 0);
 994         if (ret)
 995                 return ret;
 996
 997         return 0;
 998 }
 999
1000 static int bch2_fs_upgrade_for_subvolumes(struct btree_trans *trans)
1001 {
1002         struct btree_iter iter;
1003         struct bkey_s_c k;
1004         struct bch_inode_unpacked inode;
1005         int ret;
1006
1007         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
1008                              SPOS(0, BCACHEFS_ROOT_INO, U32_MAX), 0);
1009         k = bch2_btree_iter_peek_slot(&iter);
1010         ret = bkey_err(k);
1011         if (ret)
1012                 goto err;
1013
1014         if (!bkey_is_inode(k.k)) {
1015                 bch_err(trans->c, "root inode not found");
1016                 ret = -ENOENT;
1017                 goto err;
1018         }
1019
1020         ret = bch2_inode_unpack(k, &inode);
1021         BUG_ON(ret);
1022
1023         inode.bi_subvol = BCACHEFS_ROOT_SUBVOL;
1024
1025         ret = bch2_inode_write(trans, &iter, &inode);
1026 err:
1027         bch2_trans_iter_exit(trans, &iter);
1028         return ret;
1029 }
1030
1031 int bch2_fs_recovery(struct bch_fs *c)
1032 {
1033         const char *err = "cannot allocate memory";
1034         struct bch_sb_field_clean *clean = NULL;
1035         struct jset *last_journal_entry = NULL;
1036         u64 blacklist_seq, journal_seq;
1037         bool write_sb = false;
1038         int ret = 0;
1039
1040         if (c->sb.clean)
1041                 clean = read_superblock_clean(c);
1042         ret = PTR_ERR_OR_ZERO(clean);
1043         if (ret)
1044                 goto err;
1045
1046         if (c->sb.clean)
1047                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
1048                          le64_to_cpu(clean->journal_seq));
1049         else
1050                 bch_info(c, "recovering from unclean shutdown");
1051
1052         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
1053                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
1054                 ret = -EINVAL;
1055                 goto err;
1056         }
1057
1058         if (!c->sb.clean &&
1059             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1060                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1061                 ret = -EINVAL;
1062                 goto err;
1063         }
1064
1065         if (!(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) {
1066                 bch_err(c, "filesystem may have incompatible bkey formats; run fsck from the compat branch to fix");
1067                 ret = -EINVAL;
1068                 goto err;
1069         }
1070
1071         if (!(c->sb.features & (1ULL << BCH_FEATURE_alloc_v2))) {
1072                 bch_info(c, "alloc_v2 feature bit not set, fsck required");
1073                 c->opts.fsck = true;
1074                 c->opts.fix_errors = FSCK_OPT_YES;
1075         }
1076
1077         if (!c->opts.nochanges) {
1078                 if (c->sb.version < bcachefs_metadata_version_new_data_types) {
1079                         bch_info(c, "version prior to new_data_types, upgrade and fsck required");
1080                         c->opts.version_upgrade = true;
1081                         c->opts.fsck            = true;
1082                         c->opts.fix_errors      = FSCK_OPT_YES;
1083                 }
1084         }
1085
1086         if (c->opts.fsck && c->opts.norecovery) {
1087                 bch_err(c, "cannot select both norecovery and fsck");
1088                 ret = -EINVAL;
1089                 goto err;
1090         }
1091
1092         ret = bch2_blacklist_table_initialize(c);
1093         if (ret) {
1094                 bch_err(c, "error initializing blacklist table");
1095                 goto err;
1096         }
1097
1098         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1099                 struct genradix_iter iter;
1100                 struct journal_replay **i;
1101
1102                 bch_verbose(c, "starting journal read");
1103                 ret = bch2_journal_read(c, &blacklist_seq, &journal_seq);
1104                 if (ret)
1105                         goto err;
1106
1107                 genradix_for_each_reverse(&c->journal_entries, iter, i)
1108                         if (*i && !(*i)->ignore) {
1109                                 last_journal_entry = &(*i)->j;
1110                                 break;
1111                         }
1112
1113                 if (mustfix_fsck_err_on(c->sb.clean &&
1114                                         last_journal_entry &&
1115                                         !journal_entry_empty(last_journal_entry), c,
1116                                 "filesystem marked clean but journal not empty")) {
1117                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1118                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1119                         c->sb.clean = false;
1120                 }
1121
1122                 if (!last_journal_entry) {
1123                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
1124                         goto use_clean;
1125                 }
1126
1127                 ret = journal_keys_sort(c);
1128                 if (ret)
1129                         goto err;
1130
1131                 if (c->sb.clean && last_journal_entry) {
1132                         ret = verify_superblock_clean(c, &clean,
1133                                                       last_journal_entry);
1134                         if (ret)
1135                                 goto err;
1136                 }
1137         } else {
1138 use_clean:
1139                 if (!clean) {
1140                         bch_err(c, "no superblock clean section found");
1141                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1142                         goto err;
1143
1144                 }
1145                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1146         }
1147
1148         if (c->opts.reconstruct_alloc) {
1149                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
1150                 drop_alloc_keys(&c->journal_keys);
1151         }
1152
1153         zero_out_btree_mem_ptr(&c->journal_keys);
1154
1155         ret = journal_replay_early(c, clean);
1156         if (ret)
1157                 goto err;
1158
1159         /*
1160          * After an unclean shutdown, skip then next few journal sequence
1161          * numbers as they may have been referenced by btree writes that
1162          * happened before their corresponding journal writes - those btree
1163          * writes need to be ignored, by skipping and blacklisting the next few
1164          * journal sequence numbers:
1165          */
1166         if (!c->sb.clean)
1167                 journal_seq += 8;
1168
1169         if (blacklist_seq != journal_seq) {
1170                 ret = bch2_journal_seq_blacklist_add(c,
1171                                         blacklist_seq, journal_seq);
1172                 if (ret) {
1173                         bch_err(c, "error creating new journal seq blacklist entry");
1174                         goto err;
1175                 }
1176         }
1177
1178         /*
1179          * note: cmd_list_journal needs the blacklist table fully up to date so
1180          * it can asterisk ignored journal entries:
1181          */
1182         if (c->opts.read_journal_only)
1183                 goto out;
1184
1185         ret = bch2_fs_journal_start(&c->journal, journal_seq);
1186         if (ret)
1187                 goto err;
1188
1189         /*
1190          * Skip past versions that might have possibly been used (as nonces),
1191          * but hadn't had their pointers written:
1192          */
1193         if (c->sb.encryption_type && !c->sb.clean)
1194                 atomic64_add(1 << 16, &c->key_version);
1195
1196         ret = read_btree_roots(c);
1197         if (ret)
1198                 goto err;
1199
1200         bch_verbose(c, "starting alloc read");
1201         err = "error reading allocation information";
1202
1203         down_read(&c->gc_lock);
1204         ret = bch2_alloc_read(c);
1205         up_read(&c->gc_lock);
1206
1207         if (ret)
1208                 goto err;
1209         bch_verbose(c, "alloc read done");
1210
1211         bch_verbose(c, "starting stripes_read");
1212         err = "error reading stripes";
1213         ret = bch2_stripes_read(c);
1214         if (ret)
1215                 goto err;
1216         bch_verbose(c, "stripes_read done");
1217
1218         bch2_stripes_heap_start(c);
1219
1220         if (c->opts.fsck) {
1221                 bool metadata_only = c->opts.norecovery;
1222
1223                 bch_info(c, "checking allocations");
1224                 err = "error checking allocations";
1225                 ret = bch2_gc(c, true, metadata_only);
1226                 if (ret)
1227                         goto err;
1228                 bch_verbose(c, "done checking allocations");
1229
1230                 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1231
1232                 bch_info(c, "checking need_discard and freespace btrees");
1233                 err = "error checking need_discard and freespace btrees";
1234                 ret = bch2_check_alloc_info(c);
1235                 if (ret)
1236                         goto err;
1237                 bch_verbose(c, "done checking need_discard and freespace btrees");
1238
1239                 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1240
1241                 bch_info(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1242                 err = "journal replay failed";
1243                 ret = bch2_journal_replay(c);
1244                 if (ret)
1245                         goto err;
1246                 if (c->opts.verbose || !c->sb.clean)
1247                         bch_info(c, "journal replay done");
1248
1249                 bch_info(c, "checking lrus");
1250                 err = "error checking lrus";
1251                 ret = bch2_check_lrus(c);
1252                 if (ret)
1253                         goto err;
1254                 bch_verbose(c, "done checking lrus");
1255                 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1256
1257                 bch_info(c, "checking alloc to lru refs");
1258                 err = "error checking alloc to lru refs";
1259                 ret = bch2_check_alloc_to_lru_refs(c);
1260                 if (ret)
1261                         goto err;
1262                 bch_verbose(c, "done checking alloc to lru refs");
1263                 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1264         } else {
1265                 set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1266                 set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1267                 set_bit(BCH_FS_CHECK_LRUS_DONE, &c->flags);
1268                 set_bit(BCH_FS_CHECK_ALLOC_TO_LRU_REFS_DONE, &c->flags);
1269                 set_bit(BCH_FS_FSCK_DONE, &c->flags);
1270
1271                 if (c->opts.norecovery)
1272                         goto out;
1273
1274                 bch_verbose(c, "starting journal replay, %zu keys", c->journal_keys.nr);
1275                 err = "journal replay failed";
1276                 ret = bch2_journal_replay(c);
1277                 if (ret)
1278                         goto err;
1279                 if (c->opts.verbose || !c->sb.clean)
1280                         bch_info(c, "journal replay done");
1281         }
1282
1283         err = "error initializing freespace";
1284         ret = bch2_fs_freespace_init(c);
1285         if (ret)
1286                 goto err;
1287
1288         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1289                 bch2_fs_lazy_rw(c);
1290
1291                 err = "error creating root snapshot node";
1292                 ret = bch2_fs_initialize_subvolumes(c);
1293                 if (ret)
1294                         goto err;
1295         }
1296
1297         bch_verbose(c, "reading snapshots table");
1298         err = "error reading snapshots table";
1299         ret = bch2_fs_snapshots_start(c);
1300         if (ret)
1301                 goto err;
1302         bch_verbose(c, "reading snapshots done");
1303
1304         if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
1305                 /* set bi_subvol on root inode */
1306                 err = "error upgrade root inode for subvolumes";
1307                 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_LAZY_RW,
1308                                     bch2_fs_upgrade_for_subvolumes(&trans));
1309                 if (ret)
1310                         goto err;
1311         }
1312
1313         if (c->opts.fsck) {
1314                 bch_info(c, "starting fsck");
1315                 err = "error in fsck";
1316                 ret = bch2_fsck_full(c);
1317                 if (ret)
1318                         goto err;
1319                 bch_verbose(c, "fsck done");
1320         } else if (!c->sb.clean) {
1321                 bch_verbose(c, "checking for deleted inodes");
1322                 err = "error in recovery";
1323                 ret = bch2_fsck_walk_inodes_only(c);
1324                 if (ret)
1325                         goto err;
1326                 bch_verbose(c, "check inodes done");
1327         }
1328
1329         if (enabled_qtypes(c)) {
1330                 bch_verbose(c, "reading quotas");
1331                 ret = bch2_fs_quota_read(c);
1332                 if (ret)
1333                         goto err;
1334                 bch_verbose(c, "quotas done");
1335         }
1336
1337         mutex_lock(&c->sb_lock);
1338         if (c->opts.version_upgrade) {
1339                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1340                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1341                 write_sb = true;
1342         }
1343
1344         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1345                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1346                 write_sb = true;
1347         }
1348
1349         if (c->opts.fsck &&
1350             !test_bit(BCH_FS_ERROR, &c->flags) &&
1351             !test_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags)) {
1352                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1353                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
1354                 write_sb = true;
1355         }
1356
1357         if (write_sb)
1358                 bch2_write_super(c);
1359         mutex_unlock(&c->sb_lock);
1360
1361         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
1362             !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) ||
1363             le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) {
1364                 struct bch_move_stats stats;
1365
1366                 bch_move_stats_init(&stats, "recovery");
1367
1368                 bch_info(c, "scanning for old btree nodes");
1369                 ret = bch2_fs_read_write(c);
1370                 if (ret)
1371                         goto err;
1372
1373                 ret = bch2_scan_old_btree_nodes(c, &stats);
1374                 if (ret)
1375                         goto err;
1376                 bch_info(c, "scanning for old btree nodes done");
1377         }
1378
1379         if (c->journal_seq_blacklist_table &&
1380             c->journal_seq_blacklist_table->nr > 128)
1381                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1382
1383         ret = 0;
1384 out:
1385         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1386         bch2_flush_fsck_errs(c);
1387
1388         if (!c->opts.keep_journal) {
1389                 bch2_journal_keys_free(&c->journal_keys);
1390                 bch2_journal_entries_free(c);
1391         }
1392         kfree(clean);
1393         if (ret)
1394                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1395         else
1396                 bch_verbose(c, "ret %i", ret);
1397         return ret;
1398 err:
1399 fsck_err:
1400         bch2_fs_emergency_read_only(c);
1401         goto out;
1402 }
1403
1404 int bch2_fs_initialize(struct bch_fs *c)
1405 {
1406         struct bch_inode_unpacked root_inode, lostfound_inode;
1407         struct bkey_inode_buf packed_inode;
1408         struct qstr lostfound = QSTR("lost+found");
1409         const char *err = "cannot allocate memory";
1410         struct bch_dev *ca;
1411         unsigned i;
1412         int ret;
1413
1414         bch_notice(c, "initializing new filesystem");
1415
1416         mutex_lock(&c->sb_lock);
1417         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1418         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1419
1420         if (c->opts.version_upgrade) {
1421                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1422                 c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1423                 bch2_write_super(c);
1424         }
1425         mutex_unlock(&c->sb_lock);
1426
1427         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1428         set_bit(BCH_FS_MAY_GO_RW, &c->flags);
1429         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1430
1431         for (i = 0; i < BTREE_ID_NR; i++)
1432                 bch2_btree_root_alloc(c, i);
1433
1434         for_each_online_member(ca, c, i)
1435                 bch2_dev_usage_init(ca);
1436
1437         err = "unable to allocate journal buckets";
1438         for_each_online_member(ca, c, i) {
1439                 ret = bch2_dev_journal_alloc(ca);
1440                 if (ret) {
1441                         percpu_ref_put(&ca->io_ref);
1442                         goto err;
1443                 }
1444         }
1445
1446         /*
1447          * journal_res_get() will crash if called before this has
1448          * set up the journal.pin FIFO and journal.cur pointer:
1449          */
1450         bch2_fs_journal_start(&c->journal, 1);
1451         bch2_journal_set_replay_done(&c->journal);
1452
1453         err = "error going read-write";
1454         ret = bch2_fs_read_write_early(c);
1455         if (ret)
1456                 goto err;
1457
1458         /*
1459          * Write out the superblock and journal buckets, now that we can do
1460          * btree updates
1461          */
1462         bch_verbose(c, "marking superblocks");
1463         err = "error marking superblock and journal";
1464         for_each_member_device(ca, c, i) {
1465                 ret = bch2_trans_mark_dev_sb(c, ca);
1466                 if (ret) {
1467                         percpu_ref_put(&ca->ref);
1468                         goto err;
1469                 }
1470
1471                 ca->new_fs_bucket_idx = 0;
1472         }
1473
1474         bch_verbose(c, "initializing freespace");
1475         err = "error initializing freespace";
1476         ret = bch2_fs_freespace_init(c);
1477         if (ret)
1478                 goto err;
1479
1480         err = "error creating root snapshot node";
1481         ret = bch2_fs_initialize_subvolumes(c);
1482         if (ret)
1483                 goto err;
1484
1485         bch_verbose(c, "reading snapshots table");
1486         err = "error reading snapshots table";
1487         ret = bch2_fs_snapshots_start(c);
1488         if (ret)
1489                 goto err;
1490         bch_verbose(c, "reading snapshots done");
1491
1492         bch2_inode_init(c, &root_inode, 0, 0,
1493                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1494         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
1495         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1496         bch2_inode_pack(c, &packed_inode, &root_inode);
1497         packed_inode.inode.k.p.snapshot = U32_MAX;
1498
1499         err = "error creating root directory";
1500         ret = bch2_btree_insert(c, BTREE_ID_inodes,
1501                                 &packed_inode.inode.k_i,
1502                                 NULL, NULL, 0);
1503         if (ret)
1504                 goto err;
1505
1506         bch2_inode_init_early(c, &lostfound_inode);
1507
1508         err = "error creating lost+found";
1509         ret = bch2_trans_do(c, NULL, NULL, 0,
1510                 bch2_create_trans(&trans,
1511                                   BCACHEFS_ROOT_SUBVOL_INUM,
1512                                   &root_inode, &lostfound_inode,
1513                                   &lostfound,
1514                                   0, 0, S_IFDIR|0700, 0,
1515                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1516         if (ret) {
1517                 bch_err(c, "error creating lost+found");
1518                 goto err;
1519         }
1520
1521         if (enabled_qtypes(c)) {
1522                 ret = bch2_fs_quota_read(c);
1523                 if (ret)
1524                         goto err;
1525         }
1526
1527         err = "error writing first journal entry";
1528         ret = bch2_journal_flush(&c->journal);
1529         if (ret)
1530                 goto err;
1531
1532         mutex_lock(&c->sb_lock);
1533         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1534         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1535
1536         bch2_write_super(c);
1537         mutex_unlock(&c->sb_lock);
1538
1539         return 0;
1540 err:
1541         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1542         return ret;
1543 }