fs/bcachefs/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                                         BTREE_INSERT_NO_CLEAR_REPLICAS);
 316         } else {
 317                 ret = bch2_trans_commit(&trans, &disk_res, NULL,
 318                                         BTREE_INSERT_NOFAIL|
 319                                         BTREE_INSERT_LAZY_RW|
 320                                         BTREE_INSERT_JOURNAL_REPLAY|
 321                                         BTREE_INSERT_NOMARK);
 322         }
 323
 324         if (ret)
 325                 goto err;
 326 err:
 327         if (ret == -EINTR)
 328                 goto retry;
 329
 330         bch2_disk_reservation_put(c, &disk_res);
 331
 332         return bch2_trans_exit(&trans) ?: ret;
 333 }
 334
 335 static int bch2_journal_replay(struct bch_fs *c,
 336                                struct journal_keys keys)
 337 {
 338         struct journal *j = &c->journal;
 339         struct journal_key *i;
 340         int ret;
 341
 342         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
 343
 344         for_each_journal_key(keys, i) {
 345                 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 346
 347                 if (i->btree_id == BTREE_ID_ALLOC)
 348                         ret = bch2_alloc_replay_key(c, i->k);
 349                 else if (btree_node_type_is_extents(i->btree_id))
 350                         ret = bch2_extent_replay_key(c, i->btree_id, i->k);
 351                 else
 352                         ret = bch2_btree_insert(c, i->btree_id, i->k,
 353                                                 NULL, NULL,
 354                                                 BTREE_INSERT_NOFAIL|
 355                                                 BTREE_INSERT_LAZY_RW|
 356                                                 BTREE_INSERT_JOURNAL_REPLAY|
 357                                                 BTREE_INSERT_NOMARK);
 358
 359                 if (ret) {
 360                         bch_err(c, "journal replay: error %d while replaying key",
 361                                 ret);
 362                         return ret;
 363                 }
 364
 365                 cond_resched();
 366         }
 367
 368         replay_now_at(j, j->replay_journal_seq_end);
 369         j->replay_journal_seq = 0;
 370
 371         bch2_journal_set_replay_done(j);
 372         bch2_journal_flush_all_pins(j);
 373         return bch2_journal_error(j);
 374 }
 375
 376 static bool journal_empty(struct list_head *journal)
 377 {
 378         return list_empty(journal) ||
 379                 journal_entry_empty(&list_last_entry(journal,
 380                                         struct journal_replay, list)->j);
 381 }
 382
 383 static int
 384 verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c,
 385                                                   struct list_head *journal)
 386 {
 387         struct journal_replay *i =
 388                 list_last_entry(journal, struct journal_replay, list);
 389         u64 start_seq   = le64_to_cpu(i->j.last_seq);
 390         u64 end_seq     = le64_to_cpu(i->j.seq);
 391         u64 seq         = start_seq;
 392         int ret = 0;
 393
 394         list_for_each_entry(i, journal, list) {
 395                 fsck_err_on(seq != le64_to_cpu(i->j.seq), c,
 396                         "journal entries %llu-%llu missing! (replaying %llu-%llu)",
 397                         seq, le64_to_cpu(i->j.seq) - 1,
 398                         start_seq, end_seq);
 399
 400                 seq = le64_to_cpu(i->j.seq);
 401
 402                 fsck_err_on(bch2_journal_seq_is_blacklisted(c, seq, false), c,
 403                             "found blacklisted journal entry %llu", seq);
 404
 405                 do {
 406                         seq++;
 407                 } while (bch2_journal_seq_is_blacklisted(c, seq, false));
 408         }
 409 fsck_err:
 410         return ret;
 411 }
 412
 413 /* journal replay early: */
 414
 415 static int journal_replay_entry_early(struct bch_fs *c,
 416                                       struct jset_entry *entry)
 417 {
 418         int ret = 0;
 419
 420         switch (entry->type) {
 421         case BCH_JSET_ENTRY_btree_root: {
 422                 struct btree_root *r;
 423
 424                 if (entry->btree_id >= BTREE_ID_NR) {
 425                         bch_err(c, "filesystem has unknown btree type %u",
 426                                 entry->btree_id);
 427                         return -EINVAL;
 428                 }
 429
 430                 r = &c->btree_roots[entry->btree_id];
 431
 432                 if (entry->u64s) {
 433                         r->level = entry->level;
 434                         bkey_copy(&r->key, &entry->start[0]);
 435                         r->error = 0;
 436                 } else {
 437                         r->error = -EIO;
 438                 }
 439                 r->alive = true;
 440                 break;
 441         }
 442         case BCH_JSET_ENTRY_usage: {
 443                 struct jset_entry_usage *u =
 444                         container_of(entry, struct jset_entry_usage, entry);
 445
 446                 switch (entry->btree_id) {
 447                 case FS_USAGE_RESERVED:
 448                         if (entry->level < BCH_REPLICAS_MAX)
 449                                 c->usage_base->persistent_reserved[entry->level] =
 450                                         le64_to_cpu(u->v);
 451                         break;
 452                 case FS_USAGE_INODES:
 453                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 454                         break;
 455                 case FS_USAGE_KEY_VERSION:
 456                         atomic64_set(&c->key_version,
 457                                      le64_to_cpu(u->v));
 458                         break;
 459                 }
 460
 461                 break;
 462         }
 463         case BCH_JSET_ENTRY_data_usage: {
 464                 struct jset_entry_data_usage *u =
 465                         container_of(entry, struct jset_entry_data_usage, entry);
 466                 ret = bch2_replicas_set_usage(c, &u->r,
 467                                               le64_to_cpu(u->v));
 468                 break;
 469         }
 470         case BCH_JSET_ENTRY_blacklist: {
 471                 struct jset_entry_blacklist *bl_entry =
 472                         container_of(entry, struct jset_entry_blacklist, entry);
 473
 474                 ret = bch2_journal_seq_blacklist_add(c,
 475                                 le64_to_cpu(bl_entry->seq),
 476                                 le64_to_cpu(bl_entry->seq) + 1);
 477                 break;
 478         }
 479         case BCH_JSET_ENTRY_blacklist_v2: {
 480                 struct jset_entry_blacklist_v2 *bl_entry =
 481                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 482
 483                 ret = bch2_journal_seq_blacklist_add(c,
 484                                 le64_to_cpu(bl_entry->start),
 485                                 le64_to_cpu(bl_entry->end) + 1);
 486                 break;
 487         }
 488         }
 489
 490         return ret;
 491 }
 492
 493 static int journal_replay_early(struct bch_fs *c,
 494                                 struct bch_sb_field_clean *clean,
 495                                 struct list_head *journal)
 496 {
 497         struct jset_entry *entry;
 498         int ret;
 499
 500         if (clean) {
 501                 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
 502                 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
 503
 504                 for (entry = clean->start;
 505                      entry != vstruct_end(&clean->field);
 506                      entry = vstruct_next(entry)) {
 507                         ret = journal_replay_entry_early(c, entry);
 508                         if (ret)
 509                                 return ret;
 510                 }
 511         } else {
 512                 struct journal_replay *i =
 513                         list_last_entry(journal, struct journal_replay, list);
 514
 515                 c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
 516                 c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
 517
 518                 list_for_each_entry(i, journal, list)
 519                         vstruct_for_each(&i->j, entry) {
 520                                 ret = journal_replay_entry_early(c, entry);
 521                                 if (ret)
 522                                         return ret;
 523                         }
 524         }
 525
 526         bch2_fs_usage_initialize(c);
 527
 528         return 0;
 529 }
 530
 531 /* sb clean section: */
 532
 533 static struct bkey_i *btree_root_find(struct bch_fs *c,
 534                                       struct bch_sb_field_clean *clean,
 535                                       struct jset *j,
 536                                       enum btree_id id, unsigned *level)
 537 {
 538         struct bkey_i *k;
 539         struct jset_entry *entry, *start, *end;
 540
 541         if (clean) {
 542                 start = clean->start;
 543                 end = vstruct_end(&clean->field);
 544         } else {
 545                 start = j->start;
 546                 end = vstruct_last(j);
 547         }
 548
 549         for (entry = start; entry < end; entry = vstruct_next(entry))
 550                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 551                     entry->btree_id == id)
 552                         goto found;
 553
 554         return NULL;
 555 found:
 556         if (!entry->u64s)
 557                 return ERR_PTR(-EINVAL);
 558
 559         k = entry->start;
 560         *level = entry->level;
 561         return k;
 562 }
 563
 564 static int verify_superblock_clean(struct bch_fs *c,
 565                                    struct bch_sb_field_clean **cleanp,
 566                                    struct jset *j)
 567 {
 568         unsigned i;
 569         struct bch_sb_field_clean *clean = *cleanp;
 570         int ret = 0;
 571
 572         if (!c->sb.clean || !j)
 573                 return 0;
 574
 575         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 576                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 577                         le64_to_cpu(clean->journal_seq),
 578                         le64_to_cpu(j->seq))) {
 579                 kfree(clean);
 580                 *cleanp = NULL;
 581                 return 0;
 582         }
 583
 584         mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
 585                         "superblock read clock doesn't match journal after clean shutdown");
 586         mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
 587                         "superblock read clock doesn't match journal after clean shutdown");
 588
 589         for (i = 0; i < BTREE_ID_NR; i++) {
 590                 struct bkey_i *k1, *k2;
 591                 unsigned l1 = 0, l2 = 0;
 592
 593                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 594                 k2 = btree_root_find(c, NULL, j, i, &l2);
 595
 596                 if (!k1 && !k2)
 597                         continue;
 598
 599                 mustfix_fsck_err_on(!k1 || !k2 ||
 600                                     IS_ERR(k1) ||
 601                                     IS_ERR(k2) ||
 602                                     k1->k.u64s != k2->k.u64s ||
 603                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 604                                     l1 != l2, c,
 605                         "superblock btree root doesn't match journal after clean shutdown");
 606         }
 607 fsck_err:
 608         return ret;
 609 }
 610
 611 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 612 {
 613         struct bch_sb_field_clean *clean, *sb_clean;
 614         int ret;
 615
 616         mutex_lock(&c->sb_lock);
 617         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 618
 619         if (fsck_err_on(!sb_clean, c,
 620                         "superblock marked clean but clean section not present")) {
 621                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 622                 c->sb.clean = false;
 623                 mutex_unlock(&c->sb_lock);
 624                 return NULL;
 625         }
 626
 627         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 628                         GFP_KERNEL);
 629         if (!clean) {
 630                 mutex_unlock(&c->sb_lock);
 631                 return ERR_PTR(-ENOMEM);
 632         }
 633
 634         if (le16_to_cpu(c->disk_sb.sb->version) <
 635             bcachefs_metadata_version_bkey_renumber)
 636                 bch2_sb_clean_renumber(clean, READ);
 637
 638         mutex_unlock(&c->sb_lock);
 639
 640         return clean;
 641 fsck_err:
 642         mutex_unlock(&c->sb_lock);
 643         return ERR_PTR(ret);
 644 }
 645
 646 static int read_btree_roots(struct bch_fs *c)
 647 {
 648         unsigned i;
 649         int ret = 0;
 650
 651         for (i = 0; i < BTREE_ID_NR; i++) {
 652                 struct btree_root *r = &c->btree_roots[i];
 653
 654                 if (!r->alive)
 655                         continue;
 656
 657                 if (i == BTREE_ID_ALLOC &&
 658                     c->opts.reconstruct_alloc) {
 659                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 660                         continue;
 661                 }
 662
 663
 664                 if (r->error) {
 665                         __fsck_err(c, i == BTREE_ID_ALLOC
 666                                    ? FSCK_CAN_IGNORE : 0,
 667                                    "invalid btree root %s",
 668                                    bch2_btree_ids[i]);
 669                         if (i == BTREE_ID_ALLOC)
 670                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 671                 }
 672
 673                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 674                 if (ret) {
 675                         __fsck_err(c, i == BTREE_ID_ALLOC
 676                                    ? FSCK_CAN_IGNORE : 0,
 677                                    "error reading btree root %s",
 678                                    bch2_btree_ids[i]);
 679                         if (i == BTREE_ID_ALLOC)
 680                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 681                 }
 682         }
 683
 684         for (i = 0; i < BTREE_ID_NR; i++)
 685                 if (!c->btree_roots[i].b)
 686                         bch2_btree_root_alloc(c, i);
 687 fsck_err:
 688         return ret;
 689 }
 690
 691 int bch2_fs_recovery(struct bch_fs *c)
 692 {
 693         const char *err = "cannot allocate memory";
 694         struct bch_sb_field_clean *clean = NULL;
 695         u64 journal_seq;
 696         LIST_HEAD(journal_entries);
 697         struct journal_keys journal_keys = { NULL };
 698         bool wrote = false, write_sb = false;
 699         int ret;
 700
 701         if (c->sb.clean)
 702                 clean = read_superblock_clean(c);
 703         ret = PTR_ERR_OR_ZERO(clean);
 704         if (ret)
 705                 goto err;
 706
 707         if (c->sb.clean)
 708                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 709                          le64_to_cpu(clean->journal_seq));
 710
 711         if (!c->replicas.entries) {
 712                 bch_info(c, "building replicas info");
 713                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 714         }
 715
 716         if (!c->sb.clean || c->opts.fsck) {
 717                 struct jset *j;
 718
 719                 ret = bch2_journal_read(c, &journal_entries);
 720                 if (ret)
 721                         goto err;
 722
 723                 if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&journal_entries), c,
 724                                 "filesystem marked clean but journal not empty")) {
 725                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 726                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 727                         c->sb.clean = false;
 728                 }
 729
 730                 if (!c->sb.clean && list_empty(&journal_entries)) {
 731                         bch_err(c, "no journal entries found");
 732                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
 733                         goto err;
 734                 }
 735
 736                 journal_keys = journal_keys_sort(&journal_entries);
 737                 if (!journal_keys.d) {
 738                         ret = -ENOMEM;
 739                         goto err;
 740                 }
 741
 742                 j = &list_last_entry(&journal_entries,
 743                                      struct journal_replay, list)->j;
 744
 745                 ret = verify_superblock_clean(c, &clean, j);
 746                 if (ret)
 747                         goto err;
 748
 749                 journal_seq = le64_to_cpu(j->seq) + 1;
 750         } else {
 751                 journal_seq = le64_to_cpu(clean->journal_seq) + 1;
 752         }
 753
 754         ret = journal_replay_early(c, clean, &journal_entries);
 755         if (ret)
 756                 goto err;
 757
 758         if (!c->sb.clean) {
 759                 ret = bch2_journal_seq_blacklist_add(c,
 760                                                      journal_seq,
 761                                                      journal_seq + 4);
 762                 if (ret) {
 763                         bch_err(c, "error creating new journal seq blacklist entry");
 764                         goto err;
 765                 }
 766
 767                 journal_seq += 4;
 768         }
 769
 770         ret = bch2_blacklist_table_initialize(c);
 771
 772         if (!list_empty(&journal_entries)) {
 773                 ret = verify_journal_entries_not_blacklisted_or_missing(c,
 774                                                         &journal_entries);
 775                 if (ret)
 776                         goto err;
 777         }
 778
 779         ret = bch2_fs_journal_start(&c->journal, journal_seq,
 780                                     &journal_entries);
 781         if (ret)
 782                 goto err;
 783
 784         ret = read_btree_roots(c);
 785         if (ret)
 786                 goto err;
 787
 788         bch_verbose(c, "starting alloc read");
 789         err = "error reading allocation information";
 790         ret = bch2_alloc_read(c, &journal_keys);
 791         if (ret)
 792                 goto err;
 793         bch_verbose(c, "alloc read done");
 794
 795         bch_verbose(c, "starting stripes_read");
 796         err = "error reading stripes";
 797         ret = bch2_stripes_read(c, &journal_keys);
 798         if (ret)
 799                 goto err;
 800         bch_verbose(c, "stripes_read done");
 801
 802         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 803
 804         if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
 805             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
 806                 /*
 807                  * interior btree node updates aren't consistent with the
 808                  * journal; after an unclean shutdown we have to walk all
 809                  * pointers to metadata:
 810                  */
 811                 bch_info(c, "starting metadata mark and sweep");
 812                 err = "error in mark and sweep";
 813                 ret = bch2_gc(c, NULL, true, true);
 814                 if (ret)
 815                         goto err;
 816                 bch_verbose(c, "mark and sweep done");
 817         }
 818
 819         if (c->opts.fsck ||
 820             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
 821             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
 822                 bch_info(c, "starting mark and sweep");
 823                 err = "error in mark and sweep";
 824                 ret = bch2_gc(c, &journal_keys, true, false);
 825                 if (ret)
 826                         goto err;
 827                 bch_verbose(c, "mark and sweep done");
 828         }
 829
 830         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 831         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
 832
 833         /*
 834          * Skip past versions that might have possibly been used (as nonces),
 835          * but hadn't had their pointers written:
 836          */
 837         if (c->sb.encryption_type && !c->sb.clean)
 838                 atomic64_add(1 << 16, &c->key_version);
 839
 840         if (c->opts.norecovery)
 841                 goto out;
 842
 843         bch_verbose(c, "starting journal replay");
 844         err = "journal replay failed";
 845         ret = bch2_journal_replay(c, journal_keys);
 846         if (ret)
 847                 goto err;
 848         bch_verbose(c, "journal replay done");
 849
 850         if (!c->opts.nochanges) {
 851                 /*
 852                  * note that even when filesystem was clean there might be work
 853                  * to do here, if we ran gc (because of fsck) which recalculated
 854                  * oldest_gen:
 855                  */
 856                 bch_verbose(c, "writing allocation info");
 857                 err = "error writing out alloc info";
 858                 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW, &wrote) ?:
 859                         bch2_alloc_write(c, BTREE_INSERT_LAZY_RW, &wrote);
 860                 if (ret) {
 861                         bch_err(c, "error writing alloc info");
 862                         goto err;
 863                 }
 864                 bch_verbose(c, "alloc write done");
 865
 866                 set_bit(BCH_FS_ALLOC_WRITTEN, &c->flags);
 867         }
 868
 869         if (!c->sb.clean) {
 870                 if (!(c->sb.features & (1 << BCH_FEATURE_ATOMIC_NLINK))) {
 871                         bch_info(c, "checking inode link counts");
 872                         err = "error in recovery";
 873                         ret = bch2_fsck_inode_nlink(c);
 874                         if (ret)
 875                                 goto err;
 876                         bch_verbose(c, "check inodes done");
 877
 878                 } else {
 879                         bch_verbose(c, "checking for deleted inodes");
 880                         err = "error in recovery";
 881                         ret = bch2_fsck_walk_inodes_only(c);
 882                         if (ret)
 883                                 goto err;
 884                         bch_verbose(c, "check inodes done");
 885                 }
 886         }
 887
 888         if (c->opts.fsck) {
 889                 bch_info(c, "starting fsck");
 890                 err = "error in fsck";
 891                 ret = bch2_fsck_full(c);
 892                 if (ret)
 893                         goto err;
 894                 bch_verbose(c, "fsck done");
 895         }
 896
 897         if (enabled_qtypes(c)) {
 898                 bch_verbose(c, "reading quotas");
 899                 ret = bch2_fs_quota_read(c);
 900                 if (ret)
 901                         goto err;
 902                 bch_verbose(c, "quotas done");
 903         }
 904
 905         mutex_lock(&c->sb_lock);
 906         if (c->opts.version_upgrade) {
 907                 if (c->sb.version < bcachefs_metadata_version_new_versioning)
 908                         c->disk_sb.sb->version_min =
 909                                 le16_to_cpu(bcachefs_metadata_version_min);
 910                 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
 911                 write_sb = true;
 912         }
 913
 914         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
 915                 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
 916                 write_sb = true;
 917         }
 918
 919         if (c->opts.fsck &&
 920             !test_bit(BCH_FS_ERROR, &c->flags)) {
 921                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_ATOMIC_NLINK;
 922                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
 923                 write_sb = true;
 924         }
 925
 926         if (write_sb)
 927                 bch2_write_super(c);
 928         mutex_unlock(&c->sb_lock);
 929
 930         if (c->journal_seq_blacklist_table &&
 931             c->journal_seq_blacklist_table->nr > 128)
 932                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
 933 out:
 934         ret = 0;
 935 err:
 936 fsck_err:
 937         set_bit(BCH_FS_FSCK_DONE, &c->flags);
 938         bch2_flush_fsck_errs(c);
 939
 940         journal_keys_free(&journal_keys);
 941         journal_entries_free(&journal_entries);
 942         kfree(clean);
 943         if (ret)
 944                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
 945         else
 946                 bch_verbose(c, "ret %i", ret);
 947         return ret;
 948 }
 949
 950 int bch2_fs_initialize(struct bch_fs *c)
 951 {
 952         struct bch_inode_unpacked root_inode, lostfound_inode;
 953         struct bkey_inode_buf packed_inode;
 954         struct qstr lostfound = QSTR("lost+found");
 955         const char *err = "cannot allocate memory";
 956         struct bch_dev *ca;
 957         LIST_HEAD(journal);
 958         unsigned i;
 959         int ret;
 960
 961         bch_notice(c, "initializing new filesystem");
 962
 963         mutex_lock(&c->sb_lock);
 964         for_each_online_member(ca, c, i)
 965                 bch2_mark_dev_superblock(c, ca, 0);
 966         mutex_unlock(&c->sb_lock);
 967
 968         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 969         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
 970
 971         for (i = 0; i < BTREE_ID_NR; i++)
 972                 bch2_btree_root_alloc(c, i);
 973
 974         err = "unable to allocate journal buckets";
 975         for_each_online_member(ca, c, i) {
 976                 ret = bch2_dev_journal_alloc(ca);
 977                 if (ret) {
 978                         percpu_ref_put(&ca->io_ref);
 979                         goto err;
 980                 }
 981         }
 982
 983         /*
 984          * journal_res_get() will crash if called before this has
 985          * set up the journal.pin FIFO and journal.cur pointer:
 986          */
 987         bch2_fs_journal_start(&c->journal, 1, &journal);
 988         bch2_journal_set_replay_done(&c->journal);
 989
 990         err = "error going read write";
 991         ret = __bch2_fs_read_write(c, true);
 992         if (ret)
 993                 goto err;
 994
 995         bch2_inode_init(c, &root_inode, 0, 0,
 996                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
 997         root_inode.bi_inum = BCACHEFS_ROOT_INO;
 998         bch2_inode_pack(&packed_inode, &root_inode);
 999
1000         err = "error creating root directory";
1001         ret = bch2_btree_insert(c, BTREE_ID_INODES,
1002                                 &packed_inode.inode.k_i,
1003                                 NULL, NULL, 0);
1004         if (ret)
1005                 goto err;
1006
1007         bch2_inode_init_early(c, &lostfound_inode);
1008
1009         err = "error creating lost+found";
1010         ret = bch2_trans_do(c, NULL, NULL, 0,
1011                 bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
1012                                   &root_inode, &lostfound_inode,
1013                                   &lostfound,
1014                                   0, 0, S_IFDIR|0700, 0,
1015                                   NULL, NULL));
1016         if (ret)
1017                 goto err;
1018
1019         if (enabled_qtypes(c)) {
1020                 ret = bch2_fs_quota_read(c);
1021                 if (ret)
1022                         goto err;
1023         }
1024
1025         err = "error writing first journal entry";
1026         ret = bch2_journal_meta(&c->journal);
1027         if (ret)
1028                 goto err;
1029
1030         mutex_lock(&c->sb_lock);
1031         c->disk_sb.sb->version = c->disk_sb.sb->version_min =
1032                 le16_to_cpu(bcachefs_metadata_version_current);
1033         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_ATOMIC_NLINK;
1034
1035         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1036         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1037
1038         bch2_write_super(c);
1039         mutex_unlock(&c->sb_lock);
1040
1041         return 0;
1042 err:
1043         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1044         return ret;
1045 }