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 "alloc_background.h"
   5 #include "btree_gc.h"
   6 #include "btree_update.h"
   7 #include "btree_update_interior.h"
   8 #include "btree_io.h"
   9 #include "buckets.h"
  10 #include "dirent.h"
  11 #include "ec.h"
  12 #include "error.h"
  13 #include "fs-common.h"
  14 #include "fsck.h"
  15 #include "journal_io.h"
  16 #include "journal_reclaim.h"
  17 #include "journal_seq_blacklist.h"
  18 #include "quota.h"
  19 #include "recovery.h"
  20 #include "replicas.h"
  21 #include "super-io.h"
  22
  23 #include <linux/sort.h>
  24 #include <linux/stat.h>
  25
  26 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  27
  28 /* iterate over keys read from the journal: */
  29
  30 struct journal_iter bch2_journal_iter_init(struct journal_keys *keys,
  31                                            enum btree_id id)
  32 {
  33         return (struct journal_iter) {
  34                 .keys           = keys,
  35                 .k              = keys->d,
  36                 .btree_id       = id,
  37         };
  38 }
  39
  40 struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
  41 {
  42         while (1) {
  43                 if (iter->k == iter->keys->d + iter->keys->nr)
  44                         return bkey_s_c_null;
  45
  46                 if (iter->k->btree_id == iter->btree_id)
  47                         return bkey_i_to_s_c(iter->k->k);
  48
  49                 iter->k++;
  50         }
  51
  52         return bkey_s_c_null;
  53 }
  54
  55 struct bkey_s_c bch2_journal_iter_next(struct journal_iter *iter)
  56 {
  57         if (iter->k == iter->keys->d + iter->keys->nr)
  58                 return bkey_s_c_null;
  59
  60         iter->k++;
  61         return bch2_journal_iter_peek(iter);
  62 }
  63
  64 /* sort and dedup all keys in the journal: */
  65
  66 static void journal_entries_free(struct list_head *list)
  67 {
  68
  69         while (!list_empty(list)) {
  70                 struct journal_replay *i =
  71                         list_first_entry(list, struct journal_replay, list);
  72                 list_del(&i->list);
  73                 kvpfree(i, offsetof(struct journal_replay, j) +
  74                         vstruct_bytes(&i->j));
  75         }
  76 }
  77
  78 static int journal_sort_key_cmp(const void *_l, const void *_r)
  79 {
  80         const struct journal_key *l = _l;
  81         const struct journal_key *r = _r;
  82
  83         return cmp_int(l->btree_id, r->btree_id) ?:
  84                 bkey_cmp(l->pos, r->pos) ?:
  85                 cmp_int(l->journal_seq, r->journal_seq) ?:
  86                 cmp_int(l->journal_offset, r->journal_offset);
  87 }
  88
  89 static int journal_sort_seq_cmp(const void *_l, const void *_r)
  90 {
  91         const struct journal_key *l = _l;
  92         const struct journal_key *r = _r;
  93
  94         return cmp_int(l->journal_seq, r->journal_seq) ?:
  95                 cmp_int(l->btree_id, r->btree_id) ?:
  96                 bkey_cmp(l->pos, r->pos);
  97 }
  98
  99 static void journal_keys_sift(struct journal_keys *keys, struct journal_key *i)
 100 {
 101         while (i + 1 < keys->d + keys->nr &&
 102                journal_sort_key_cmp(i, i + 1) > 0) {
 103                 swap(i[0], i[1]);
 104                 i++;
 105         }
 106 }
 107
 108 static void journal_keys_free(struct journal_keys *keys)
 109 {
 110         struct journal_key *i;
 111
 112         for_each_journal_key(*keys, i)
 113                 if (i->allocated)
 114                         kfree(i->k);
 115         kvfree(keys->d);
 116         keys->d = NULL;
 117         keys->nr = 0;
 118 }
 119
 120 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 121 {
 122         struct journal_replay *p;
 123         struct jset_entry *entry;
 124         struct bkey_i *k, *_n;
 125         struct journal_keys keys = { NULL }, keys_deduped = { NULL };
 126         struct journal_key *i;
 127         size_t nr_keys = 0;
 128
 129         list_for_each_entry(p, journal_entries, list)
 130                 for_each_jset_key(k, _n, entry, &p->j)
 131                         nr_keys++;
 132
 133         keys.journal_seq_base = keys_deduped.journal_seq_base =
 134                 le64_to_cpu(list_first_entry(journal_entries,
 135                                              struct journal_replay,
 136                                              list)->j.seq);
 137
 138         keys.d = kvmalloc(sizeof(keys.d[0]) * nr_keys, GFP_KERNEL);
 139         if (!keys.d)
 140                 goto err;
 141
 142         keys_deduped.d = kvmalloc(sizeof(keys.d[0]) * nr_keys * 2, GFP_KERNEL);
 143         if (!keys_deduped.d)
 144                 goto err;
 145
 146         list_for_each_entry(p, journal_entries, list)
 147                 for_each_jset_key(k, _n, entry, &p->j)
 148                         keys.d[keys.nr++] = (struct journal_key) {
 149                                 .btree_id       = entry->btree_id,
 150                                 .pos            = bkey_start_pos(&k->k),
 151                                 .k              = k,
 152                                 .journal_seq    = le64_to_cpu(p->j.seq) -
 153                                         keys.journal_seq_base,
 154                                 .journal_offset = k->_data - p->j._data,
 155                         };
 156
 157         sort(keys.d, nr_keys, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 158
 159         i = keys.d;
 160         while (i < keys.d + keys.nr) {
 161                 if (i + 1 < keys.d + keys.nr &&
 162                     i[0].btree_id == i[1].btree_id &&
 163                     !bkey_cmp(i[0].pos, i[1].pos)) {
 164                         if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) <= 0) {
 165                                 i++;
 166                         } else {
 167                                 bch2_cut_front(i[1].k->k.p, i[0].k);
 168                                 i[0].pos = i[1].k->k.p;
 169                                 journal_keys_sift(&keys, i);
 170                         }
 171                         continue;
 172                 }
 173
 174                 if (i + 1 < keys.d + keys.nr &&
 175                     i[0].btree_id == i[1].btree_id &&
 176                     bkey_cmp(i[0].k->k.p, bkey_start_pos(&i[1].k->k)) > 0) {
 177                         if ((cmp_int(i[0].journal_seq, i[1].journal_seq) ?:
 178                              cmp_int(i[0].journal_offset, i[1].journal_offset)) < 0) {
 179                                 if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) <= 0) {
 180                                         bch2_cut_back(bkey_start_pos(&i[1].k->k), i[0].k);
 181                                 } else {
 182                                         struct bkey_i *split =
 183                                                 kmalloc(bkey_bytes(i[0].k), GFP_KERNEL);
 184
 185                                         if (!split)
 186                                                 goto err;
 187
 188                                         bkey_copy(split, i[0].k);
 189                                         bch2_cut_back(bkey_start_pos(&i[1].k->k), split);
 190                                         keys_deduped.d[keys_deduped.nr++] = (struct journal_key) {
 191                                                 .btree_id       = i[0].btree_id,
 192                                                 .allocated      = true,
 193                                                 .pos            = bkey_start_pos(&split->k),
 194                                                 .k              = split,
 195                                                 .journal_seq    = i[0].journal_seq,
 196                                                 .journal_offset = i[0].journal_offset,
 197                                         };
 198
 199                                         bch2_cut_front(i[1].k->k.p, i[0].k);
 200                                         i[0].pos = i[1].k->k.p;
 201                                         journal_keys_sift(&keys, i);
 202                                         continue;
 203                                 }
 204                         } else {
 205                                 if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) >= 0) {
 206                                         i[1] = i[0];
 207                                         i++;
 208                                         continue;
 209                                 } else {
 210                                         bch2_cut_front(i[0].k->k.p, i[1].k);
 211                                         i[1].pos = i[0].k->k.p;
 212                                         journal_keys_sift(&keys, i + 1);
 213                                         continue;
 214                                 }
 215                         }
 216                 }
 217
 218                 keys_deduped.d[keys_deduped.nr++] = *i++;
 219         }
 220
 221         kvfree(keys.d);
 222         return keys_deduped;
 223 err:
 224         journal_keys_free(&keys_deduped);
 225         kvfree(keys.d);
 226         return (struct journal_keys) { NULL };
 227 }
 228
 229 /* journal replay: */
 230
 231 static void replay_now_at(struct journal *j, u64 seq)
 232 {
 233         BUG_ON(seq < j->replay_journal_seq);
 234         BUG_ON(seq > j->replay_journal_seq_end);
 235
 236         while (j->replay_journal_seq < seq)
 237                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 238 }
 239
 240 static int bch2_extent_replay_key(struct bch_fs *c, enum btree_id btree_id,
 241                                   struct bkey_i *k)
 242 {
 243         struct btree_trans trans;
 244         struct btree_iter *iter, *split_iter;
 245         /*
 246          * We might cause compressed extents to be split, so we need to pass in
 247          * a disk_reservation:
 248          */
 249         struct disk_reservation disk_res =
 250                 bch2_disk_reservation_init(c, 0);
 251         struct bkey_i *split;
 252         struct bpos atomic_end;
 253         /*
 254          * Some extents aren't equivalent - w.r.t. what the triggers do
 255          * - if they're split:
 256          */
 257         bool remark_if_split = bch2_bkey_sectors_compressed(bkey_i_to_s_c(k)) ||
 258                 k->k.type == KEY_TYPE_reflink_p;
 259         bool remark = false;
 260         int ret;
 261
 262         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 263 retry:
 264         bch2_trans_begin(&trans);
 265
 266         iter = bch2_trans_get_iter(&trans, btree_id,
 267                                    bkey_start_pos(&k->k),
 268                                    BTREE_ITER_INTENT);
 269
 270         do {
 271                 ret = bch2_btree_iter_traverse(iter);
 272                 if (ret)
 273                         goto err;
 274
 275                 atomic_end = bpos_min(k->k.p, iter->l[0].b->key.k.p);
 276
 277                 split_iter = bch2_trans_copy_iter(&trans, iter);
 278                 ret = PTR_ERR_OR_ZERO(split_iter);
 279                 if (ret)
 280                         goto err;
 281
 282                 split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
 283                 ret = PTR_ERR_OR_ZERO(split);
 284                 if (ret)
 285                         goto err;
 286
 287                 if (!remark &&
 288                     remark_if_split &&
 289                     bkey_cmp(atomic_end, k->k.p) < 0) {
 290                         ret = bch2_disk_reservation_add(c, &disk_res,
 291                                         k->k.size *
 292                                         bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(k)),
 293                                         BCH_DISK_RESERVATION_NOFAIL);
 294                         BUG_ON(ret);
 295
 296                         remark = true;
 297                 }
 298
 299                 bkey_copy(split, k);
 300                 bch2_cut_front(split_iter->pos, split);
 301                 bch2_cut_back(atomic_end, split);
 302
 303                 bch2_trans_update(&trans, split_iter, split);
 304                 bch2_btree_iter_set_pos(iter, split->k.p);
 305         } while (bkey_cmp(iter->pos, k->k.p) < 0);
 306
 307         if (remark) {
 308                 ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k),
 309                                           0, -((s64) k->k.size),
 310                                           BCH_BUCKET_MARK_OVERWRITE) ?:
 311                       bch2_trans_commit(&trans, &disk_res, NULL,
 312                                         BTREE_INSERT_NOFAIL|
 313                                         BTREE_INSERT_LAZY_RW|
 314                                         BTREE_INSERT_NOMARK_OVERWRITES);
 315         } else {
 316                 ret = bch2_trans_commit(&trans, &disk_res, NULL,
 317                                         BTREE_INSERT_NOFAIL|
 318                                         BTREE_INSERT_LAZY_RW|
 319                                         BTREE_INSERT_JOURNAL_REPLAY|
 320                                         BTREE_INSERT_NOMARK);
 321         }
 322
 323         if (ret)
 324                 goto err;
 325 err:
 326         if (ret == -EINTR)
 327                 goto retry;
 328
 329         bch2_disk_reservation_put(c, &disk_res);
 330
 331         return bch2_trans_exit(&trans) ?: ret;
 332 }
 333
 334 static int bch2_journal_replay(struct bch_fs *c,
 335                                struct journal_keys keys)
 336 {
 337         struct journal *j = &c->journal;
 338         struct journal_key *i;
 339         int ret;
 340
 341         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
 342
 343         for_each_journal_key(keys, i) {
 344                 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 345
 346                 if (i->btree_id == BTREE_ID_ALLOC)
 347                         ret = bch2_alloc_replay_key(c, i->k);
 348                 else if (btree_node_type_is_extents(i->btree_id))
 349                         ret = bch2_extent_replay_key(c, i->btree_id, i->k);
 350                 else
 351                         ret = bch2_btree_insert(c, i->btree_id, i->k,
 352                                                 NULL, NULL,
 353                                                 BTREE_INSERT_NOFAIL|
 354                                                 BTREE_INSERT_LAZY_RW|
 355                                                 BTREE_INSERT_JOURNAL_REPLAY|
 356                                                 BTREE_INSERT_NOMARK);
 357
 358                 if (ret) {
 359                         bch_err(c, "journal replay: error %d while replaying key",
 360                                 ret);
 361                         return ret;
 362                 }
 363
 364                 cond_resched();
 365         }
 366
 367         replay_now_at(j, j->replay_journal_seq_end);
 368         j->replay_journal_seq = 0;
 369
 370         bch2_journal_set_replay_done(j);
 371         bch2_journal_flush_all_pins(j);
 372         return bch2_journal_error(j);
 373 }
 374
 375 static bool journal_empty(struct list_head *journal)
 376 {
 377         return list_empty(journal) ||
 378                 journal_entry_empty(&list_last_entry(journal,
 379                                         struct journal_replay, list)->j);
 380 }
 381
 382 static int
 383 verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c,
 384                                                   struct list_head *journal)
 385 {
 386         struct journal_replay *i =
 387                 list_last_entry(journal, struct journal_replay, list);
 388         u64 start_seq   = le64_to_cpu(i->j.last_seq);
 389         u64 end_seq     = le64_to_cpu(i->j.seq);
 390         u64 seq         = start_seq;
 391         int ret = 0;
 392
 393         list_for_each_entry(i, journal, list) {
 394                 fsck_err_on(seq != le64_to_cpu(i->j.seq), c,
 395                         "journal entries %llu-%llu missing! (replaying %llu-%llu)",
 396                         seq, le64_to_cpu(i->j.seq) - 1,
 397                         start_seq, end_seq);
 398
 399                 seq = le64_to_cpu(i->j.seq);
 400
 401                 fsck_err_on(bch2_journal_seq_is_blacklisted(c, seq, false), c,
 402                             "found blacklisted journal entry %llu", seq);
 403
 404                 do {
 405                         seq++;
 406                 } while (bch2_journal_seq_is_blacklisted(c, seq, false));
 407         }
 408 fsck_err:
 409         return ret;
 410 }
 411
 412 /* journal replay early: */
 413
 414 static int journal_replay_entry_early(struct bch_fs *c,
 415                                       struct jset_entry *entry)
 416 {
 417         int ret = 0;
 418
 419         switch (entry->type) {
 420         case BCH_JSET_ENTRY_btree_root: {
 421                 struct btree_root *r;
 422
 423                 if (entry->btree_id >= BTREE_ID_NR) {
 424                         bch_err(c, "filesystem has unknown btree type %u",
 425                                 entry->btree_id);
 426                         return -EINVAL;
 427                 }
 428
 429                 r = &c->btree_roots[entry->btree_id];
 430
 431                 if (entry->u64s) {
 432                         r->level = entry->level;
 433                         bkey_copy(&r->key, &entry->start[0]);
 434                         r->error = 0;
 435                 } else {
 436                         r->error = -EIO;
 437                 }
 438                 r->alive = true;
 439                 break;
 440         }
 441         case BCH_JSET_ENTRY_usage: {
 442                 struct jset_entry_usage *u =
 443                         container_of(entry, struct jset_entry_usage, entry);
 444
 445                 switch (entry->btree_id) {
 446                 case FS_USAGE_RESERVED:
 447                         if (entry->level < BCH_REPLICAS_MAX)
 448                                 c->usage_base->persistent_reserved[entry->level] =
 449                                         le64_to_cpu(u->v);
 450                         break;
 451                 case FS_USAGE_INODES:
 452                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 453                         break;
 454                 case FS_USAGE_KEY_VERSION:
 455                         atomic64_set(&c->key_version,
 456                                      le64_to_cpu(u->v));
 457                         break;
 458                 }
 459
 460                 break;
 461         }
 462         case BCH_JSET_ENTRY_data_usage: {
 463                 struct jset_entry_data_usage *u =
 464                         container_of(entry, struct jset_entry_data_usage, entry);
 465                 ret = bch2_replicas_set_usage(c, &u->r,
 466                                               le64_to_cpu(u->v));
 467                 break;
 468         }
 469         case BCH_JSET_ENTRY_blacklist: {
 470                 struct jset_entry_blacklist *bl_entry =
 471                         container_of(entry, struct jset_entry_blacklist, entry);
 472
 473                 ret = bch2_journal_seq_blacklist_add(c,
 474                                 le64_to_cpu(bl_entry->seq),
 475                                 le64_to_cpu(bl_entry->seq) + 1);
 476                 break;
 477         }
 478         case BCH_JSET_ENTRY_blacklist_v2: {
 479                 struct jset_entry_blacklist_v2 *bl_entry =
 480                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 481
 482                 ret = bch2_journal_seq_blacklist_add(c,
 483                                 le64_to_cpu(bl_entry->start),
 484                                 le64_to_cpu(bl_entry->end) + 1);
 485                 break;
 486         }
 487         }
 488
 489         return ret;
 490 }
 491
 492 static int journal_replay_early(struct bch_fs *c,
 493                                 struct bch_sb_field_clean *clean,
 494                                 struct list_head *journal)
 495 {
 496         struct jset_entry *entry;
 497         int ret;
 498
 499         if (clean) {
 500                 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
 501                 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
 502
 503                 for (entry = clean->start;
 504                      entry != vstruct_end(&clean->field);
 505                      entry = vstruct_next(entry)) {
 506                         ret = journal_replay_entry_early(c, entry);
 507                         if (ret)
 508                                 return ret;
 509                 }
 510         } else {
 511                 struct journal_replay *i =
 512                         list_last_entry(journal, struct journal_replay, list);
 513
 514                 c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
 515                 c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
 516
 517                 list_for_each_entry(i, journal, list)
 518                         vstruct_for_each(&i->j, entry) {
 519                                 ret = journal_replay_entry_early(c, entry);
 520                                 if (ret)
 521                                         return ret;
 522                         }
 523         }
 524
 525         bch2_fs_usage_initialize(c);
 526
 527         return 0;
 528 }
 529
 530 /* sb clean section: */
 531
 532 static struct bkey_i *btree_root_find(struct bch_fs *c,
 533                                       struct bch_sb_field_clean *clean,
 534                                       struct jset *j,
 535                                       enum btree_id id, unsigned *level)
 536 {
 537         struct bkey_i *k;
 538         struct jset_entry *entry, *start, *end;
 539
 540         if (clean) {
 541                 start = clean->start;
 542                 end = vstruct_end(&clean->field);
 543         } else {
 544                 start = j->start;
 545                 end = vstruct_last(j);
 546         }
 547
 548         for (entry = start; entry < end; entry = vstruct_next(entry))
 549                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 550                     entry->btree_id == id)
 551                         goto found;
 552
 553         return NULL;
 554 found:
 555         if (!entry->u64s)
 556                 return ERR_PTR(-EINVAL);
 557
 558         k = entry->start;
 559         *level = entry->level;
 560         return k;
 561 }
 562
 563 static int verify_superblock_clean(struct bch_fs *c,
 564                                    struct bch_sb_field_clean **cleanp,
 565                                    struct jset *j)
 566 {
 567         unsigned i;
 568         struct bch_sb_field_clean *clean = *cleanp;
 569         int ret = 0;
 570
 571         if (!c->sb.clean || !j)
 572                 return 0;
 573
 574         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 575                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 576                         le64_to_cpu(clean->journal_seq),
 577                         le64_to_cpu(j->seq))) {
 578                 kfree(clean);
 579                 *cleanp = NULL;
 580                 return 0;
 581         }
 582
 583         mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
 584                         "superblock read clock doesn't match journal after clean shutdown");
 585         mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
 586                         "superblock read clock doesn't match journal after clean shutdown");
 587
 588         for (i = 0; i < BTREE_ID_NR; i++) {
 589                 struct bkey_i *k1, *k2;
 590                 unsigned l1 = 0, l2 = 0;
 591
 592                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 593                 k2 = btree_root_find(c, NULL, j, i, &l2);
 594
 595                 if (!k1 && !k2)
 596                         continue;
 597
 598                 mustfix_fsck_err_on(!k1 || !k2 ||
 599                                     IS_ERR(k1) ||
 600                                     IS_ERR(k2) ||
 601                                     k1->k.u64s != k2->k.u64s ||
 602                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 603                                     l1 != l2, c,
 604                         "superblock btree root doesn't match journal after clean shutdown");
 605         }
 606 fsck_err:
 607         return ret;
 608 }
 609
 610 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 611 {
 612         struct bch_sb_field_clean *clean, *sb_clean;
 613         int ret;
 614
 615         mutex_lock(&c->sb_lock);
 616         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 617
 618         if (fsck_err_on(!sb_clean, c,
 619                         "superblock marked clean but clean section not present")) {
 620                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 621                 c->sb.clean = false;
 622                 mutex_unlock(&c->sb_lock);
 623                 return NULL;
 624         }
 625
 626         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 627                         GFP_KERNEL);
 628         if (!clean) {
 629                 mutex_unlock(&c->sb_lock);
 630                 return ERR_PTR(-ENOMEM);
 631         }
 632
 633         if (le16_to_cpu(c->disk_sb.sb->version) <
 634             bcachefs_metadata_version_bkey_renumber)
 635                 bch2_sb_clean_renumber(clean, READ);
 636
 637         mutex_unlock(&c->sb_lock);
 638
 639         return clean;
 640 fsck_err:
 641         mutex_unlock(&c->sb_lock);
 642         return ERR_PTR(ret);
 643 }
 644
 645 static int read_btree_roots(struct bch_fs *c)
 646 {
 647         unsigned i;
 648         int ret = 0;
 649
 650         for (i = 0; i < BTREE_ID_NR; i++) {
 651                 struct btree_root *r = &c->btree_roots[i];
 652
 653                 if (!r->alive)
 654                         continue;
 655
 656                 if (i == BTREE_ID_ALLOC &&
 657                     c->opts.reconstruct_alloc) {
 658                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 659                         continue;
 660                 }
 661
 662
 663                 if (r->error) {
 664                         __fsck_err(c, i == BTREE_ID_ALLOC
 665                                    ? FSCK_CAN_IGNORE : 0,
 666                                    "invalid btree root %s",
 667                                    bch2_btree_ids[i]);
 668                         if (i == BTREE_ID_ALLOC)
 669                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 670                 }
 671
 672                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 673                 if (ret) {
 674                         __fsck_err(c, i == BTREE_ID_ALLOC
 675                                    ? FSCK_CAN_IGNORE : 0,
 676                                    "error reading btree root %s",
 677                                    bch2_btree_ids[i]);
 678                         if (i == BTREE_ID_ALLOC)
 679                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 680                 }
 681         }
 682
 683         for (i = 0; i < BTREE_ID_NR; i++)
 684                 if (!c->btree_roots[i].b)
 685                         bch2_btree_root_alloc(c, i);
 686 fsck_err:
 687         return ret;
 688 }
 689
 690 int bch2_fs_recovery(struct bch_fs *c)
 691 {
 692         const char *err = "cannot allocate memory";
 693         struct bch_sb_field_clean *clean = NULL;
 694         u64 journal_seq;
 695         LIST_HEAD(journal_entries);
 696         struct journal_keys journal_keys = { NULL };
 697         bool wrote = false, write_sb = false;
 698         int ret;
 699
 700         if (c->sb.clean)
 701                 clean = read_superblock_clean(c);
 702         ret = PTR_ERR_OR_ZERO(clean);
 703         if (ret)
 704                 goto err;
 705
 706         if (c->sb.clean)
 707                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 708                          le64_to_cpu(clean->journal_seq));
 709
 710         if (!c->replicas.entries) {
 711                 bch_info(c, "building replicas info");
 712                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 713         }
 714
 715         if (!c->sb.clean || c->opts.fsck) {
 716                 struct jset *j;
 717
 718                 ret = bch2_journal_read(c, &journal_entries);
 719                 if (ret)
 720                         goto err;
 721
 722                 if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&journal_entries), c,
 723                                 "filesystem marked clean but journal not empty")) {
 724                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 725                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 726                         c->sb.clean = false;
 727                 }
 728
 729                 if (!c->sb.clean && list_empty(&journal_entries)) {
 730                         bch_err(c, "no journal entries found");
 731                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
 732                         goto err;
 733                 }
 734
 735                 journal_keys = journal_keys_sort(&journal_entries);
 736                 if (!journal_keys.d) {
 737                         ret = -ENOMEM;
 738                         goto err;
 739                 }
 740
 741                 j = &list_last_entry(&journal_entries,
 742                                      struct journal_replay, list)->j;
 743
 744                 ret = verify_superblock_clean(c, &clean, j);
 745                 if (ret)
 746                         goto err;
 747
 748                 journal_seq = le64_to_cpu(j->seq) + 1;
 749         } else {
 750                 journal_seq = le64_to_cpu(clean->journal_seq) + 1;
 751         }
 752
 753         ret = journal_replay_early(c, clean, &journal_entries);
 754         if (ret)
 755                 goto err;
 756
 757         if (!c->sb.clean) {
 758                 ret = bch2_journal_seq_blacklist_add(c,
 759                                                      journal_seq,
 760                                                      journal_seq + 4);
 761                 if (ret) {
 762                         bch_err(c, "error creating new journal seq blacklist entry");
 763                         goto err;
 764                 }
 765
 766                 journal_seq += 4;
 767         }
 768
 769         ret = bch2_blacklist_table_initialize(c);
 770
 771         if (!list_empty(&journal_entries)) {
 772                 ret = verify_journal_entries_not_blacklisted_or_missing(c,
 773                                                         &journal_entries);
 774                 if (ret)
 775                         goto err;
 776         }
 777
 778         ret = bch2_fs_journal_start(&c->journal, journal_seq,
 779                                     &journal_entries);
 780         if (ret)
 781                 goto err;
 782
 783         ret = read_btree_roots(c);
 784         if (ret)
 785                 goto err;
 786
 787         bch_verbose(c, "starting alloc read");
 788         err = "error reading allocation information";
 789         ret = bch2_alloc_read(c, &journal_keys);
 790         if (ret)
 791                 goto err;
 792         bch_verbose(c, "alloc read done");
 793
 794         bch_verbose(c, "starting stripes_read");
 795         err = "error reading stripes";
 796         ret = bch2_stripes_read(c, &journal_keys);
 797         if (ret)
 798                 goto err;
 799         bch_verbose(c, "stripes_read done");
 800
 801         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 802
 803         if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
 804             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
 805                 /*
 806                  * interior btree node updates aren't consistent with the
 807                  * journal; after an unclean shutdown we have to walk all
 808                  * pointers to metadata:
 809                  */
 810                 bch_info(c, "starting metadata mark and sweep");
 811                 err = "error in mark and sweep";
 812                 ret = bch2_gc(c, NULL, true, true);
 813                 if (ret)
 814                         goto err;
 815                 bch_verbose(c, "mark and sweep done");
 816         }
 817
 818         if (c->opts.fsck ||
 819             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
 820             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
 821                 bch_info(c, "starting mark and sweep");
 822                 err = "error in mark and sweep";
 823                 ret = bch2_gc(c, &journal_keys, true, false);
 824                 if (ret)
 825                         goto err;
 826                 bch_verbose(c, "mark and sweep done");
 827         }
 828
 829         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 830         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
 831
 832         /*
 833          * Skip past versions that might have possibly been used (as nonces),
 834          * but hadn't had their pointers written:
 835          */
 836         if (c->sb.encryption_type && !c->sb.clean)
 837                 atomic64_add(1 << 16, &c->key_version);
 838
 839         if (c->opts.norecovery)
 840                 goto out;
 841
 842         bch_verbose(c, "starting journal replay");
 843         err = "journal replay failed";
 844         ret = bch2_journal_replay(c, journal_keys);
 845         if (ret)
 846                 goto err;
 847         bch_verbose(c, "journal replay done");
 848
 849         if (!c->opts.nochanges) {
 850                 /*
 851                  * note that even when filesystem was clean there might be work
 852                  * to do here, if we ran gc (because of fsck) which recalculated
 853                  * oldest_gen:
 854                  */
 855                 bch_verbose(c, "writing allocation info");
 856                 err = "error writing out alloc info";
 857                 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW, &wrote) ?:
 858                         bch2_alloc_write(c, BTREE_INSERT_LAZY_RW, &wrote);
 859                 if (ret) {
 860                         bch_err(c, "error writing alloc info");
 861                         goto err;
 862                 }
 863                 bch_verbose(c, "alloc write done");
 864
 865                 set_bit(BCH_FS_ALLOC_WRITTEN, &c->flags);
 866         }
 867
 868         if (!c->sb.clean) {
 869                 if (!(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
 870                         bch_info(c, "checking inode link counts");
 871                         err = "error in recovery";
 872                         ret = bch2_fsck_inode_nlink(c);
 873                         if (ret)
 874                                 goto err;
 875                         bch_verbose(c, "check inodes done");
 876
 877                 } else {
 878                         bch_verbose(c, "checking for deleted inodes");
 879                         err = "error in recovery";
 880                         ret = bch2_fsck_walk_inodes_only(c);
 881                         if (ret)
 882                                 goto err;
 883                         bch_verbose(c, "check inodes done");
 884                 }
 885         }
 886
 887         if (c->opts.fsck) {
 888                 bch_info(c, "starting fsck");
 889                 err = "error in fsck";
 890                 ret = bch2_fsck_full(c);
 891                 if (ret)
 892                         goto err;
 893                 bch_verbose(c, "fsck done");
 894         }
 895
 896         if (enabled_qtypes(c)) {
 897                 bch_verbose(c, "reading quotas");
 898                 ret = bch2_fs_quota_read(c);
 899                 if (ret)
 900                         goto err;
 901                 bch_verbose(c, "quotas done");
 902         }
 903
 904         mutex_lock(&c->sb_lock);
 905         if (c->opts.version_upgrade) {
 906                 if (c->sb.version < bcachefs_metadata_version_new_versioning)
 907                         c->disk_sb.sb->version_min =
 908                                 le16_to_cpu(bcachefs_metadata_version_min);
 909                 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
 910                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_siphash;
 911                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_extent_overwrite;
 912                 write_sb = true;
 913         }
 914
 915         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
 916                 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
 917                 write_sb = true;
 918         }
 919
 920         if (c->opts.fsck &&
 921             !test_bit(BCH_FS_ERROR, &c->flags)) {
 922                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
 923                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
 924                 write_sb = true;
 925         }
 926
 927         if (write_sb)
 928                 bch2_write_super(c);
 929         mutex_unlock(&c->sb_lock);
 930
 931         if (c->journal_seq_blacklist_table &&
 932             c->journal_seq_blacklist_table->nr > 128)
 933                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
 934 out:
 935         ret = 0;
 936 err:
 937 fsck_err:
 938         set_bit(BCH_FS_FSCK_DONE, &c->flags);
 939         bch2_flush_fsck_errs(c);
 940
 941         journal_keys_free(&journal_keys);
 942         journal_entries_free(&journal_entries);
 943         kfree(clean);
 944         if (ret)
 945                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
 946         else
 947                 bch_verbose(c, "ret %i", ret);
 948         return ret;
 949 }
 950
 951 int bch2_fs_initialize(struct bch_fs *c)
 952 {
 953         struct bch_inode_unpacked root_inode, lostfound_inode;
 954         struct bkey_inode_buf packed_inode;
 955         struct qstr lostfound = QSTR("lost+found");
 956         const char *err = "cannot allocate memory";
 957         struct bch_dev *ca;
 958         LIST_HEAD(journal);
 959         unsigned i;
 960         int ret;
 961
 962         bch_notice(c, "initializing new filesystem");
 963
 964         mutex_lock(&c->sb_lock);
 965         for_each_online_member(ca, c, i)
 966                 bch2_mark_dev_superblock(c, ca, 0);
 967         mutex_unlock(&c->sb_lock);
 968
 969         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 970         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
 971
 972         for (i = 0; i < BTREE_ID_NR; i++)
 973                 bch2_btree_root_alloc(c, i);
 974
 975         err = "unable to allocate journal buckets";
 976         for_each_online_member(ca, c, i) {
 977                 ret = bch2_dev_journal_alloc(ca);
 978                 if (ret) {
 979                         percpu_ref_put(&ca->io_ref);
 980                         goto err;
 981                 }
 982         }
 983
 984         /*
 985          * journal_res_get() will crash if called before this has
 986          * set up the journal.pin FIFO and journal.cur pointer:
 987          */
 988         bch2_fs_journal_start(&c->journal, 1, &journal);
 989         bch2_journal_set_replay_done(&c->journal);
 990
 991         bch2_inode_init(c, &root_inode, 0, 0,
 992                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
 993         root_inode.bi_inum = BCACHEFS_ROOT_INO;
 994         bch2_inode_pack(&packed_inode, &root_inode);
 995
 996         err = "error creating root directory";
 997         ret = bch2_btree_insert(c, BTREE_ID_INODES,
 998                                 &packed_inode.inode.k_i,
 999                                 NULL, NULL, BTREE_INSERT_LAZY_RW);
1000         if (ret)
1001                 goto err;
1002
1003         bch2_inode_init_early(c, &lostfound_inode);
1004
1005         err = "error creating lost+found";
1006         ret = bch2_trans_do(c, NULL, NULL, 0,
1007                 bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
1008                                   &root_inode, &lostfound_inode,
1009                                   &lostfound,
1010                                   0, 0, S_IFDIR|0700, 0,
1011                                   NULL, NULL));
1012         if (ret)
1013                 goto err;
1014
1015         if (enabled_qtypes(c)) {
1016                 ret = bch2_fs_quota_read(c);
1017                 if (ret)
1018                         goto err;
1019         }
1020
1021         err = "error writing first journal entry";
1022         ret = bch2_journal_meta(&c->journal);
1023         if (ret)
1024                 goto err;
1025
1026         mutex_lock(&c->sb_lock);
1027         c->disk_sb.sb->version = c->disk_sb.sb->version_min =
1028                 le16_to_cpu(bcachefs_metadata_version_current);
1029         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1030         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_siphash;
1031         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_extent_overwrite;
1032
1033         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1034         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1035
1036         bch2_write_super(c);
1037         mutex_unlock(&c->sb_lock);
1038
1039         return 0;
1040 err:
1041         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1042         return ret;
1043 }