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

   1
   2 #include "bcachefs.h"
   3 #include "alloc_background.h"
   4 #include "btree_gc.h"
   5 #include "btree_update.h"
   6 #include "btree_update_interior.h"
   7 #include "btree_io.h"
   8 #include "buckets.h"
   9 #include "dirent.h"
  10 #include "ec.h"
  11 #include "error.h"
  12 #include "fsck.h"
  13 #include "journal_io.h"
  14 #include "journal_reclaim.h"
  15 #include "journal_seq_blacklist.h"
  16 #include "quota.h"
  17 #include "recovery.h"
  18 #include "replicas.h"
  19 #include "super-io.h"
  20
  21 #include <linux/sort.h>
  22 #include <linux/stat.h>
  23
  24 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  25
  26 /* sort and dedup all keys in the journal: */
  27
  28 static void journal_entries_free(struct list_head *list)
  29 {
  30
  31         while (!list_empty(list)) {
  32                 struct journal_replay *i =
  33                         list_first_entry(list, struct journal_replay, list);
  34                 list_del(&i->list);
  35                 kvpfree(i, offsetof(struct journal_replay, j) +
  36                         vstruct_bytes(&i->j));
  37         }
  38 }
  39
  40 static int journal_sort_key_cmp(const void *_l, const void *_r)
  41 {
  42         const struct journal_key *l = _l;
  43         const struct journal_key *r = _r;
  44
  45         return cmp_int(l->btree_id, r->btree_id) ?:
  46                 bkey_cmp(l->pos, r->pos) ?:
  47                 cmp_int(l->journal_seq, r->journal_seq) ?:
  48                 cmp_int(l->journal_offset, r->journal_offset);
  49 }
  50
  51 static int journal_sort_seq_cmp(const void *_l, const void *_r)
  52 {
  53         const struct journal_key *l = _l;
  54         const struct journal_key *r = _r;
  55
  56         return cmp_int(l->journal_seq, r->journal_seq) ?:
  57                 cmp_int(l->btree_id, r->btree_id) ?:
  58                 bkey_cmp(l->pos, r->pos);
  59 }
  60
  61 static void journal_keys_sift(struct journal_keys *keys, struct journal_key *i)
  62 {
  63         while (i + 1 < keys->d + keys->nr &&
  64                journal_sort_key_cmp(i, i + 1) > 0) {
  65                 swap(i[0], i[1]);
  66                 i++;
  67         }
  68 }
  69
  70 static void journal_keys_free(struct journal_keys *keys)
  71 {
  72         struct journal_key *i;
  73
  74         for_each_journal_key(*keys, i)
  75                 if (i->allocated)
  76                         kfree(i->k);
  77         kvfree(keys->d);
  78         keys->d = NULL;
  79         keys->nr = 0;
  80 }
  81
  82 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
  83 {
  84         struct journal_replay *p;
  85         struct jset_entry *entry;
  86         struct bkey_i *k, *_n;
  87         struct journal_keys keys = { NULL }, keys_deduped = { NULL };
  88         struct journal_key *i;
  89         size_t nr_keys = 0;
  90
  91         list_for_each_entry(p, journal_entries, list)
  92                 for_each_jset_key(k, _n, entry, &p->j)
  93                         nr_keys++;
  94
  95         keys.journal_seq_base = keys_deduped.journal_seq_base =
  96                 le64_to_cpu(list_first_entry(journal_entries,
  97                                              struct journal_replay,
  98                                              list)->j.seq);
  99
 100         keys.d = kvmalloc(sizeof(keys.d[0]) * nr_keys, GFP_KERNEL);
 101         if (!keys.d)
 102                 goto err;
 103
 104         keys_deduped.d = kvmalloc(sizeof(keys.d[0]) * nr_keys * 2, GFP_KERNEL);
 105         if (!keys_deduped.d)
 106                 goto err;
 107
 108         list_for_each_entry(p, journal_entries, list)
 109                 for_each_jset_key(k, _n, entry, &p->j)
 110                         keys.d[keys.nr++] = (struct journal_key) {
 111                                 .btree_id       = entry->btree_id,
 112                                 .pos            = bkey_start_pos(&k->k),
 113                                 .k              = k,
 114                                 .journal_seq    = le64_to_cpu(p->j.seq) -
 115                                         keys.journal_seq_base,
 116                                 .journal_offset = k->_data - p->j._data,
 117                         };
 118
 119         sort(keys.d, nr_keys, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 120
 121         i = keys.d;
 122         while (i < keys.d + keys.nr) {
 123                 if (i + 1 < keys.d + keys.nr &&
 124                     i[0].btree_id == i[1].btree_id &&
 125                     !bkey_cmp(i[0].pos, i[1].pos)) {
 126                         if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) <= 0) {
 127                                 i++;
 128                         } else {
 129                                 bch2_cut_front(i[1].k->k.p, i[0].k);
 130                                 i[0].pos = i[1].k->k.p;
 131                                 journal_keys_sift(&keys, i);
 132                         }
 133                         continue;
 134                 }
 135
 136                 if (i + 1 < keys.d + keys.nr &&
 137                     i[0].btree_id == i[1].btree_id &&
 138                     bkey_cmp(i[0].k->k.p, bkey_start_pos(&i[1].k->k)) > 0) {
 139                         if ((cmp_int(i[0].journal_seq, i[1].journal_seq) ?:
 140                              cmp_int(i[0].journal_offset, i[1].journal_offset)) < 0) {
 141                                 if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) <= 0) {
 142                                         bch2_cut_back(bkey_start_pos(&i[1].k->k), &i[0].k->k);
 143                                 } else {
 144                                         struct bkey_i *split =
 145                                                 kmalloc(bkey_bytes(i[0].k), GFP_KERNEL);
 146
 147                                         if (!split)
 148                                                 goto err;
 149
 150                                         bkey_copy(split, i[0].k);
 151                                         bch2_cut_back(bkey_start_pos(&i[1].k->k), &split->k);
 152                                         keys_deduped.d[keys_deduped.nr++] = (struct journal_key) {
 153                                                 .btree_id       = i[0].btree_id,
 154                                                 .allocated      = true,
 155                                                 .pos            = bkey_start_pos(&split->k),
 156                                                 .k              = split,
 157                                                 .journal_seq    = i[0].journal_seq,
 158                                                 .journal_offset = i[0].journal_offset,
 159                                         };
 160
 161                                         bch2_cut_front(i[1].k->k.p, i[0].k);
 162                                         i[0].pos = i[1].k->k.p;
 163                                         journal_keys_sift(&keys, i);
 164                                         continue;
 165                                 }
 166                         } else {
 167                                 if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) >= 0) {
 168                                         i[1] = i[0];
 169                                         i++;
 170                                         continue;
 171                                 } else {
 172                                         bch2_cut_front(i[0].k->k.p, i[1].k);
 173                                         i[1].pos = i[0].k->k.p;
 174                                         journal_keys_sift(&keys, i + 1);
 175                                         continue;
 176                                 }
 177                         }
 178                 }
 179
 180                 keys_deduped.d[keys_deduped.nr++] = *i++;
 181         }
 182
 183         kvfree(keys.d);
 184         return keys_deduped;
 185 err:
 186         journal_keys_free(&keys_deduped);
 187         kvfree(keys.d);
 188         return (struct journal_keys) { NULL };
 189 }
 190
 191 /* journal replay: */
 192
 193 static void replay_now_at(struct journal *j, u64 seq)
 194 {
 195         BUG_ON(seq < j->replay_journal_seq);
 196         BUG_ON(seq > j->replay_journal_seq_end);
 197
 198         while (j->replay_journal_seq < seq)
 199                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 200 }
 201
 202 static int bch2_extent_replay_key(struct bch_fs *c, struct bkey_i *k)
 203 {
 204         struct btree_trans trans;
 205         struct btree_iter *iter, *split_iter;
 206         /*
 207          * We might cause compressed extents to be split, so we need to pass in
 208          * a disk_reservation:
 209          */
 210         struct disk_reservation disk_res =
 211                 bch2_disk_reservation_init(c, 0);
 212         struct bkey_i *split;
 213         bool split_compressed = false;
 214         int ret;
 215
 216         bch2_trans_init(&trans, c);
 217         bch2_trans_preload_iters(&trans);
 218 retry:
 219         bch2_trans_begin(&trans);
 220
 221         iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
 222                                    bkey_start_pos(&k->k),
 223                                    BTREE_ITER_INTENT);
 224
 225         do {
 226                 ret = bch2_btree_iter_traverse(iter);
 227                 if (ret)
 228                         goto err;
 229
 230                 split_iter = bch2_trans_copy_iter(&trans, iter);
 231                 ret = PTR_ERR_OR_ZERO(split_iter);
 232                 if (ret)
 233                         goto err;
 234
 235                 split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
 236                 ret = PTR_ERR_OR_ZERO(split);
 237                 if (ret)
 238                         goto err;
 239
 240                 if (!split_compressed &&
 241                     bch2_extent_is_compressed(bkey_i_to_s_c(k)) &&
 242                     !bch2_extent_is_atomic(k, split_iter)) {
 243                         ret = bch2_disk_reservation_add(c, &disk_res,
 244                                         k->k.size *
 245                                         bch2_bkey_nr_dirty_ptrs(bkey_i_to_s_c(k)),
 246                                         BCH_DISK_RESERVATION_NOFAIL);
 247                         BUG_ON(ret);
 248
 249                         split_compressed = true;
 250                 }
 251
 252                 bkey_copy(split, k);
 253                 bch2_cut_front(split_iter->pos, split);
 254                 bch2_extent_trim_atomic(split, split_iter);
 255
 256                 bch2_trans_update(&trans, BTREE_INSERT_ENTRY(split_iter, split));
 257                 bch2_btree_iter_set_pos(iter, split->k.p);
 258         } while (bkey_cmp(iter->pos, k->k.p) < 0);
 259
 260         if (split_compressed) {
 261                 memset(&trans.fs_usage_deltas.fs_usage, 0,
 262                        sizeof(trans.fs_usage_deltas.fs_usage));
 263                 trans.fs_usage_deltas.top = trans.fs_usage_deltas.d;
 264
 265                 ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k), false,
 266                                           -((s64) k->k.size),
 267                                           &trans.fs_usage_deltas) ?:
 268                       bch2_trans_commit(&trans, &disk_res, NULL,
 269                                         BTREE_INSERT_ATOMIC|
 270                                         BTREE_INSERT_NOFAIL|
 271                                         BTREE_INSERT_LAZY_RW|
 272                                         BTREE_INSERT_NOMARK_OVERWRITES|
 273                                         BTREE_INSERT_NO_CLEAR_REPLICAS);
 274         } else {
 275                 ret = bch2_trans_commit(&trans, &disk_res, NULL,
 276                                         BTREE_INSERT_ATOMIC|
 277                                         BTREE_INSERT_NOFAIL|
 278                                         BTREE_INSERT_LAZY_RW|
 279                                         BTREE_INSERT_JOURNAL_REPLAY|
 280                                         BTREE_INSERT_NOMARK);
 281         }
 282
 283         if (ret)
 284                 goto err;
 285 err:
 286         if (ret == -EINTR)
 287                 goto retry;
 288
 289         bch2_disk_reservation_put(c, &disk_res);
 290
 291         return bch2_trans_exit(&trans) ?: ret;
 292 }
 293
 294 static int bch2_journal_replay(struct bch_fs *c,
 295                                struct journal_keys keys)
 296 {
 297         struct journal *j = &c->journal;
 298         struct journal_key *i;
 299         int ret;
 300
 301         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
 302
 303         for_each_journal_key(keys, i) {
 304                 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 305
 306                 switch (i->btree_id) {
 307                 case BTREE_ID_ALLOC:
 308                         ret = bch2_alloc_replay_key(c, i->k);
 309                         break;
 310                 case BTREE_ID_EXTENTS:
 311                         ret = bch2_extent_replay_key(c, i->k);
 312                         break;
 313                 default:
 314                         ret = bch2_btree_insert(c, i->btree_id, i->k,
 315                                                 NULL, NULL,
 316                                                 BTREE_INSERT_NOFAIL|
 317                                                 BTREE_INSERT_LAZY_RW|
 318                                                 BTREE_INSERT_JOURNAL_REPLAY|
 319                                                 BTREE_INSERT_NOMARK);
 320                         break;
 321                 }
 322
 323                 if (ret) {
 324                         bch_err(c, "journal replay: error %d while replaying key",
 325                                 ret);
 326                         return ret;
 327                 }
 328
 329                 cond_resched();
 330         }
 331
 332         replay_now_at(j, j->replay_journal_seq_end);
 333         j->replay_journal_seq = 0;
 334
 335         bch2_journal_set_replay_done(j);
 336         bch2_journal_flush_all_pins(j);
 337         return bch2_journal_error(j);
 338 }
 339
 340 static bool journal_empty(struct list_head *journal)
 341 {
 342         return list_empty(journal) ||
 343                 journal_entry_empty(&list_last_entry(journal,
 344                                         struct journal_replay, list)->j);
 345 }
 346
 347 static int
 348 verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c,
 349                                                   struct list_head *journal)
 350 {
 351         struct journal_replay *i =
 352                 list_last_entry(journal, struct journal_replay, list);
 353         u64 start_seq   = le64_to_cpu(i->j.last_seq);
 354         u64 end_seq     = le64_to_cpu(i->j.seq);
 355         u64 seq         = start_seq;
 356         int ret = 0;
 357
 358         list_for_each_entry(i, journal, list) {
 359                 fsck_err_on(seq != le64_to_cpu(i->j.seq), c,
 360                         "journal entries %llu-%llu missing! (replaying %llu-%llu)",
 361                         seq, le64_to_cpu(i->j.seq) - 1,
 362                         start_seq, end_seq);
 363
 364                 seq = le64_to_cpu(i->j.seq);
 365
 366                 fsck_err_on(bch2_journal_seq_is_blacklisted(c, seq, false), c,
 367                             "found blacklisted journal entry %llu", seq);
 368
 369                 do {
 370                         seq++;
 371                 } while (bch2_journal_seq_is_blacklisted(c, seq, false));
 372         }
 373 fsck_err:
 374         return ret;
 375 }
 376
 377 /* journal replay early: */
 378
 379 static int journal_replay_entry_early(struct bch_fs *c,
 380                                       struct jset_entry *entry)
 381 {
 382         int ret = 0;
 383
 384         switch (entry->type) {
 385         case BCH_JSET_ENTRY_btree_root: {
 386                 struct btree_root *r = &c->btree_roots[entry->btree_id];
 387
 388                 if (entry->u64s) {
 389                         r->level = entry->level;
 390                         bkey_copy(&r->key, &entry->start[0]);
 391                         r->error = 0;
 392                 } else {
 393                         r->error = -EIO;
 394                 }
 395                 r->alive = true;
 396                 break;
 397         }
 398         case BCH_JSET_ENTRY_usage: {
 399                 struct jset_entry_usage *u =
 400                         container_of(entry, struct jset_entry_usage, entry);
 401
 402                 switch (entry->btree_id) {
 403                 case FS_USAGE_RESERVED:
 404                         if (entry->level < BCH_REPLICAS_MAX)
 405                                 c->usage_base->persistent_reserved[entry->level] =
 406                                         le64_to_cpu(u->v);
 407                         break;
 408                 case FS_USAGE_INODES:
 409                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 410                         break;
 411                 case FS_USAGE_KEY_VERSION:
 412                         atomic64_set(&c->key_version,
 413                                      le64_to_cpu(u->v));
 414                         break;
 415                 }
 416
 417                 break;
 418         }
 419         case BCH_JSET_ENTRY_data_usage: {
 420                 struct jset_entry_data_usage *u =
 421                         container_of(entry, struct jset_entry_data_usage, entry);
 422                 ret = bch2_replicas_set_usage(c, &u->r,
 423                                               le64_to_cpu(u->v));
 424                 break;
 425         }
 426         case BCH_JSET_ENTRY_blacklist: {
 427                 struct jset_entry_blacklist *bl_entry =
 428                         container_of(entry, struct jset_entry_blacklist, entry);
 429
 430                 ret = bch2_journal_seq_blacklist_add(c,
 431                                 le64_to_cpu(bl_entry->seq),
 432                                 le64_to_cpu(bl_entry->seq) + 1);
 433                 break;
 434         }
 435         case BCH_JSET_ENTRY_blacklist_v2: {
 436                 struct jset_entry_blacklist_v2 *bl_entry =
 437                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 438
 439                 ret = bch2_journal_seq_blacklist_add(c,
 440                                 le64_to_cpu(bl_entry->start),
 441                                 le64_to_cpu(bl_entry->end) + 1);
 442                 break;
 443         }
 444         }
 445
 446         return ret;
 447 }
 448
 449 static int journal_replay_early(struct bch_fs *c,
 450                                 struct bch_sb_field_clean *clean,
 451                                 struct list_head *journal)
 452 {
 453         struct jset_entry *entry;
 454         int ret;
 455
 456         if (clean) {
 457                 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
 458                 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
 459
 460                 for (entry = clean->start;
 461                      entry != vstruct_end(&clean->field);
 462                      entry = vstruct_next(entry)) {
 463                         ret = journal_replay_entry_early(c, entry);
 464                         if (ret)
 465                                 return ret;
 466                 }
 467         } else {
 468                 struct journal_replay *i =
 469                         list_last_entry(journal, struct journal_replay, list);
 470
 471                 c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
 472                 c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
 473
 474                 list_for_each_entry(i, journal, list)
 475                         vstruct_for_each(&i->j, entry) {
 476                                 ret = journal_replay_entry_early(c, entry);
 477                                 if (ret)
 478                                         return ret;
 479                         }
 480         }
 481
 482         bch2_fs_usage_initialize(c);
 483
 484         return 0;
 485 }
 486
 487 /* sb clean section: */
 488
 489 static struct bkey_i *btree_root_find(struct bch_fs *c,
 490                                       struct bch_sb_field_clean *clean,
 491                                       struct jset *j,
 492                                       enum btree_id id, unsigned *level)
 493 {
 494         struct bkey_i *k;
 495         struct jset_entry *entry, *start, *end;
 496
 497         if (clean) {
 498                 start = clean->start;
 499                 end = vstruct_end(&clean->field);
 500         } else {
 501                 start = j->start;
 502                 end = vstruct_last(j);
 503         }
 504
 505         for (entry = start; entry < end; entry = vstruct_next(entry))
 506                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 507                     entry->btree_id == id)
 508                         goto found;
 509
 510         return NULL;
 511 found:
 512         if (!entry->u64s)
 513                 return ERR_PTR(-EINVAL);
 514
 515         k = entry->start;
 516         *level = entry->level;
 517         return k;
 518 }
 519
 520 static int verify_superblock_clean(struct bch_fs *c,
 521                                    struct bch_sb_field_clean **cleanp,
 522                                    struct jset *j)
 523 {
 524         unsigned i;
 525         struct bch_sb_field_clean *clean = *cleanp;
 526         int ret = 0;
 527
 528         if (!c->sb.clean || !j)
 529                 return 0;
 530
 531         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 532                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 533                         le64_to_cpu(clean->journal_seq),
 534                         le64_to_cpu(j->seq))) {
 535                 kfree(clean);
 536                 *cleanp = NULL;
 537                 return 0;
 538         }
 539
 540         mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
 541                         "superblock read clock doesn't match journal after clean shutdown");
 542         mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
 543                         "superblock read clock doesn't match journal after clean shutdown");
 544
 545         for (i = 0; i < BTREE_ID_NR; i++) {
 546                 struct bkey_i *k1, *k2;
 547                 unsigned l1 = 0, l2 = 0;
 548
 549                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 550                 k2 = btree_root_find(c, NULL, j, i, &l2);
 551
 552                 if (!k1 && !k2)
 553                         continue;
 554
 555                 mustfix_fsck_err_on(!k1 || !k2 ||
 556                                     IS_ERR(k1) ||
 557                                     IS_ERR(k2) ||
 558                                     k1->k.u64s != k2->k.u64s ||
 559                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 560                                     l1 != l2, c,
 561                         "superblock btree root doesn't match journal after clean shutdown");
 562         }
 563 fsck_err:
 564         return ret;
 565 }
 566
 567 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 568 {
 569         struct bch_sb_field_clean *clean, *sb_clean;
 570         int ret;
 571
 572         mutex_lock(&c->sb_lock);
 573         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 574
 575         if (fsck_err_on(!sb_clean, c,
 576                         "superblock marked clean but clean section not present")) {
 577                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 578                 c->sb.clean = false;
 579                 mutex_unlock(&c->sb_lock);
 580                 return NULL;
 581         }
 582
 583         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 584                         GFP_KERNEL);
 585         if (!clean) {
 586                 mutex_unlock(&c->sb_lock);
 587                 return ERR_PTR(-ENOMEM);
 588         }
 589
 590         if (le16_to_cpu(c->disk_sb.sb->version) <
 591             bcachefs_metadata_version_bkey_renumber)
 592                 bch2_sb_clean_renumber(clean, READ);
 593
 594         mutex_unlock(&c->sb_lock);
 595
 596         return clean;
 597 fsck_err:
 598         mutex_unlock(&c->sb_lock);
 599         return ERR_PTR(ret);
 600 }
 601
 602 static int read_btree_roots(struct bch_fs *c)
 603 {
 604         unsigned i;
 605         int ret = 0;
 606
 607         for (i = 0; i < BTREE_ID_NR; i++) {
 608                 struct btree_root *r = &c->btree_roots[i];
 609
 610                 if (!r->alive)
 611                         continue;
 612
 613                 if (i == BTREE_ID_ALLOC &&
 614                     test_reconstruct_alloc(c)) {
 615                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 616                         continue;
 617                 }
 618
 619
 620                 if (r->error) {
 621                         __fsck_err(c, i == BTREE_ID_ALLOC
 622                                    ? FSCK_CAN_IGNORE : 0,
 623                                    "invalid btree root %s",
 624                                    bch2_btree_ids[i]);
 625                         if (i == BTREE_ID_ALLOC)
 626                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 627                 }
 628
 629                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 630                 if (ret) {
 631                         __fsck_err(c, i == BTREE_ID_ALLOC
 632                                    ? FSCK_CAN_IGNORE : 0,
 633                                    "error reading btree root %s",
 634                                    bch2_btree_ids[i]);
 635                         if (i == BTREE_ID_ALLOC)
 636                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 637                 }
 638         }
 639
 640         for (i = 0; i < BTREE_ID_NR; i++)
 641                 if (!c->btree_roots[i].b)
 642                         bch2_btree_root_alloc(c, i);
 643 fsck_err:
 644         return ret;
 645 }
 646
 647 int bch2_fs_recovery(struct bch_fs *c)
 648 {
 649         const char *err = "cannot allocate memory";
 650         struct bch_sb_field_clean *clean = NULL;
 651         u64 journal_seq;
 652         LIST_HEAD(journal_entries);
 653         struct journal_keys journal_keys = { NULL };
 654         bool wrote = false, write_sb = false;
 655         int ret;
 656
 657         if (c->sb.clean)
 658                 clean = read_superblock_clean(c);
 659         ret = PTR_ERR_OR_ZERO(clean);
 660         if (ret)
 661                 goto err;
 662
 663         if (c->sb.clean)
 664                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 665                          le64_to_cpu(clean->journal_seq));
 666
 667         if (!c->replicas.entries) {
 668                 bch_info(c, "building replicas info");
 669                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 670         }
 671
 672         if (!c->sb.clean || c->opts.fsck) {
 673                 struct jset *j;
 674
 675                 ret = bch2_journal_read(c, &journal_entries);
 676                 if (ret)
 677                         goto err;
 678
 679                 if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&journal_entries), c,
 680                                 "filesystem marked clean but journal not empty")) {
 681                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 682                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 683                         c->sb.clean = false;
 684                 }
 685
 686                 if (!c->sb.clean && list_empty(&journal_entries)) {
 687                         bch_err(c, "no journal entries found");
 688                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
 689                         goto err;
 690                 }
 691
 692                 journal_keys = journal_keys_sort(&journal_entries);
 693                 if (!journal_keys.d) {
 694                         ret = -ENOMEM;
 695                         goto err;
 696                 }
 697
 698                 j = &list_last_entry(&journal_entries,
 699                                      struct journal_replay, list)->j;
 700
 701                 ret = verify_superblock_clean(c, &clean, j);
 702                 if (ret)
 703                         goto err;
 704
 705                 journal_seq = le64_to_cpu(j->seq) + 1;
 706         } else {
 707                 journal_seq = le64_to_cpu(clean->journal_seq) + 1;
 708         }
 709
 710         ret = journal_replay_early(c, clean, &journal_entries);
 711         if (ret)
 712                 goto err;
 713
 714         if (!c->sb.clean) {
 715                 ret = bch2_journal_seq_blacklist_add(c,
 716                                                      journal_seq,
 717                                                      journal_seq + 4);
 718                 if (ret) {
 719                         bch_err(c, "error creating new journal seq blacklist entry");
 720                         goto err;
 721                 }
 722
 723                 journal_seq += 4;
 724         }
 725
 726         ret = bch2_blacklist_table_initialize(c);
 727
 728         ret = verify_journal_entries_not_blacklisted_or_missing(c,
 729                                                 &journal_entries);
 730         if (ret)
 731                 goto err;
 732
 733         ret = bch2_fs_journal_start(&c->journal, journal_seq,
 734                                     &journal_entries);
 735         if (ret)
 736                 goto err;
 737
 738         ret = read_btree_roots(c);
 739         if (ret)
 740                 goto err;
 741
 742         bch_verbose(c, "starting alloc read");
 743         err = "error reading allocation information";
 744         ret = bch2_alloc_read(c, &journal_keys);
 745         if (ret)
 746                 goto err;
 747         bch_verbose(c, "alloc read done");
 748
 749         bch_verbose(c, "starting stripes_read");
 750         err = "error reading stripes";
 751         ret = bch2_stripes_read(c, &journal_keys);
 752         if (ret)
 753                 goto err;
 754         bch_verbose(c, "stripes_read done");
 755
 756         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 757
 758         if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
 759             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
 760                 /*
 761                  * interior btree node updates aren't consistent with the
 762                  * journal; after an unclean shutdown we have to walk all
 763                  * pointers to metadata:
 764                  */
 765                 bch_info(c, "starting metadata mark and sweep");
 766                 err = "error in mark and sweep";
 767                 ret = bch2_gc(c, NULL, true, true);
 768                 if (ret)
 769                         goto err;
 770                 bch_verbose(c, "mark and sweep done");
 771         }
 772
 773         if (c->opts.fsck ||
 774             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
 775             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
 776                 bch_info(c, "starting mark and sweep");
 777                 err = "error in mark and sweep";
 778                 ret = bch2_gc(c, &journal_keys, true, false);
 779                 if (ret)
 780                         goto err;
 781                 bch_verbose(c, "mark and sweep done");
 782         }
 783
 784         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 785         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
 786
 787         /*
 788          * Skip past versions that might have possibly been used (as nonces),
 789          * but hadn't had their pointers written:
 790          */
 791         if (c->sb.encryption_type && !c->sb.clean)
 792                 atomic64_add(1 << 16, &c->key_version);
 793
 794         if (c->opts.norecovery)
 795                 goto out;
 796
 797         bch_verbose(c, "starting journal replay");
 798         err = "journal replay failed";
 799         ret = bch2_journal_replay(c, journal_keys);
 800         if (ret)
 801                 goto err;
 802         bch_verbose(c, "journal replay done");
 803
 804         if (!c->opts.nochanges) {
 805                 /*
 806                  * note that even when filesystem was clean there might be work
 807                  * to do here, if we ran gc (because of fsck) which recalculated
 808                  * oldest_gen:
 809                  */
 810                 bch_verbose(c, "writing allocation info");
 811                 err = "error writing out alloc info";
 812                 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW, &wrote) ?:
 813                         bch2_alloc_write(c, BTREE_INSERT_LAZY_RW, &wrote);
 814                 if (ret) {
 815                         bch_err(c, "error writing alloc info");
 816                         goto err;
 817                 }
 818                 bch_verbose(c, "alloc write done");
 819         }
 820
 821         if (!c->sb.clean) {
 822                 if (!(c->sb.features & (1 << BCH_FEATURE_ATOMIC_NLINK))) {
 823                         bch_info(c, "checking inode link counts");
 824                         err = "error in recovery";
 825                         ret = bch2_fsck_inode_nlink(c);
 826                         if (ret)
 827                                 goto err;
 828                         bch_verbose(c, "check inodes done");
 829
 830                 } else {
 831                         bch_verbose(c, "checking for deleted inodes");
 832                         err = "error in recovery";
 833                         ret = bch2_fsck_walk_inodes_only(c);
 834                         if (ret)
 835                                 goto err;
 836                         bch_verbose(c, "check inodes done");
 837                 }
 838         }
 839
 840         if (c->opts.fsck) {
 841                 bch_info(c, "starting fsck");
 842                 err = "error in fsck";
 843                 ret = bch2_fsck_full(c);
 844                 if (ret)
 845                         goto err;
 846                 bch_verbose(c, "fsck done");
 847         }
 848
 849         if (enabled_qtypes(c)) {
 850                 bch_verbose(c, "reading quotas");
 851                 ret = bch2_fs_quota_read(c);
 852                 if (ret)
 853                         goto err;
 854                 bch_verbose(c, "quotas done");
 855         }
 856
 857         mutex_lock(&c->sb_lock);
 858         if (c->opts.version_upgrade) {
 859                 if (c->sb.version < bcachefs_metadata_version_new_versioning)
 860                         c->disk_sb.sb->version_min =
 861                                 le16_to_cpu(bcachefs_metadata_version_min);
 862                 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
 863                 write_sb = true;
 864         }
 865
 866         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
 867                 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
 868                 write_sb = true;
 869         }
 870
 871         if (c->opts.fsck &&
 872             !test_bit(BCH_FS_ERROR, &c->flags)) {
 873                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_ATOMIC_NLINK;
 874                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
 875                 write_sb = true;
 876         }
 877
 878         if (write_sb)
 879                 bch2_write_super(c);
 880         mutex_unlock(&c->sb_lock);
 881
 882         if (c->journal_seq_blacklist_table &&
 883             c->journal_seq_blacklist_table->nr > 128)
 884                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
 885 out:
 886         ret = 0;
 887 err:
 888 fsck_err:
 889         bch2_flush_fsck_errs(c);
 890         journal_keys_free(&journal_keys);
 891         journal_entries_free(&journal_entries);
 892         kfree(clean);
 893         if (ret)
 894                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
 895         else
 896                 bch_verbose(c, "ret %i", ret);
 897         return ret;
 898 }
 899
 900 int bch2_fs_initialize(struct bch_fs *c)
 901 {
 902         struct bch_inode_unpacked root_inode, lostfound_inode;
 903         struct bkey_inode_buf packed_inode;
 904         struct bch_hash_info root_hash_info;
 905         struct qstr lostfound = QSTR("lost+found");
 906         const char *err = "cannot allocate memory";
 907         struct bch_dev *ca;
 908         LIST_HEAD(journal);
 909         unsigned i;
 910         int ret;
 911
 912         bch_notice(c, "initializing new filesystem");
 913
 914         mutex_lock(&c->sb_lock);
 915         for_each_online_member(ca, c, i)
 916                 bch2_mark_dev_superblock(c, ca, 0);
 917         mutex_unlock(&c->sb_lock);
 918
 919         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 920         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
 921
 922         for (i = 0; i < BTREE_ID_NR; i++)
 923                 bch2_btree_root_alloc(c, i);
 924
 925         err = "unable to allocate journal buckets";
 926         for_each_online_member(ca, c, i) {
 927                 ret = bch2_dev_journal_alloc(ca);
 928                 if (ret) {
 929                         percpu_ref_put(&ca->io_ref);
 930                         goto err;
 931                 }
 932         }
 933
 934         /*
 935          * journal_res_get() will crash if called before this has
 936          * set up the journal.pin FIFO and journal.cur pointer:
 937          */
 938         bch2_fs_journal_start(&c->journal, 1, &journal);
 939         bch2_journal_set_replay_done(&c->journal);
 940
 941         err = "error going read write";
 942         ret = __bch2_fs_read_write(c, true);
 943         if (ret)
 944                 goto err;
 945
 946         bch2_inode_init(c, &root_inode, 0, 0,
 947                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
 948         root_inode.bi_inum = BCACHEFS_ROOT_INO;
 949         root_inode.bi_nlink++; /* lost+found */
 950         bch2_inode_pack(&packed_inode, &root_inode);
 951
 952         err = "error creating root directory";
 953         ret = bch2_btree_insert(c, BTREE_ID_INODES,
 954                                 &packed_inode.inode.k_i,
 955                                 NULL, NULL, 0);
 956         if (ret)
 957                 goto err;
 958
 959         bch2_inode_init(c, &lostfound_inode, 0, 0,
 960                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0,
 961                         &root_inode);
 962         lostfound_inode.bi_inum = BCACHEFS_ROOT_INO + 1;
 963         bch2_inode_pack(&packed_inode, &lostfound_inode);
 964
 965         err = "error creating lost+found";
 966         ret = bch2_btree_insert(c, BTREE_ID_INODES,
 967                                 &packed_inode.inode.k_i,
 968                                 NULL, NULL, 0);
 969         if (ret)
 970                 goto err;
 971
 972         root_hash_info = bch2_hash_info_init(c, &root_inode);
 973
 974         ret = bch2_dirent_create(c, BCACHEFS_ROOT_INO, &root_hash_info, DT_DIR,
 975                                  &lostfound, lostfound_inode.bi_inum, NULL,
 976                                  BTREE_INSERT_NOFAIL);
 977         if (ret)
 978                 goto err;
 979
 980         if (enabled_qtypes(c)) {
 981                 ret = bch2_fs_quota_read(c);
 982                 if (ret)
 983                         goto err;
 984         }
 985
 986         err = "error writing first journal entry";
 987         ret = bch2_journal_meta(&c->journal);
 988         if (ret)
 989                 goto err;
 990
 991         mutex_lock(&c->sb_lock);
 992         c->disk_sb.sb->version = c->disk_sb.sb->version_min =
 993                 le16_to_cpu(bcachefs_metadata_version_current);
 994         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_ATOMIC_NLINK;
 995
 996         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
 997         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 998
 999         bch2_write_super(c);
1000         mutex_unlock(&c->sb_lock);
1001
1002         return 0;
1003 err:
1004         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1005         return ret;
1006 }