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->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->k);
 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_extent_is_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                 split_iter = bch2_trans_copy_iter(&trans, iter);
 276                 ret = PTR_ERR_OR_ZERO(split_iter);
 277                 if (ret)
 278                         goto err;
 279
 280                 split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
 281                 ret = PTR_ERR_OR_ZERO(split);
 282                 if (ret)
 283                         goto err;
 284
 285                 ret = bch2_extent_atomic_end(split_iter, k, &atomic_end);
 286                 if (ret)
 287                         goto err;
 288
 289                 if (!remark &&
 290                     remark_if_split &&
 291                     bkey_cmp(atomic_end, k->k.p) < 0) {
 292                         ret = bch2_disk_reservation_add(c, &disk_res,
 293                                         k->k.size *
 294                                         bch2_bkey_nr_dirty_ptrs(bkey_i_to_s_c(k)),
 295                                         BCH_DISK_RESERVATION_NOFAIL);
 296                         BUG_ON(ret);
 297
 298                         remark = true;
 299                 }
 300
 301                 bkey_copy(split, k);
 302                 bch2_cut_front(split_iter->pos, split);
 303                 bch2_cut_back(atomic_end, &split->k);
 304
 305                 bch2_trans_update(&trans, split_iter, split);
 306                 bch2_btree_iter_set_pos(iter, split->k.p);
 307         } while (bkey_cmp(iter->pos, k->k.p) < 0);
 308
 309         if (remark) {
 310                 ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k),
 311                                           0, -((s64) k->k.size),
 312                                           BCH_BUCKET_MARK_OVERWRITE) ?:
 313                       bch2_trans_commit(&trans, &disk_res, NULL,
 314                                         BTREE_INSERT_ATOMIC|
 315                                         BTREE_INSERT_NOFAIL|
 316                                         BTREE_INSERT_LAZY_RW|
 317                                         BTREE_INSERT_NOMARK_OVERWRITES|
 318                                         BTREE_INSERT_NO_CLEAR_REPLICAS);
 319         } else {
 320                 ret = bch2_trans_commit(&trans, &disk_res, NULL,
 321                                         BTREE_INSERT_ATOMIC|
 322                                         BTREE_INSERT_NOFAIL|
 323                                         BTREE_INSERT_LAZY_RW|
 324                                         BTREE_INSERT_JOURNAL_REPLAY|
 325                                         BTREE_INSERT_NOMARK);
 326         }
 327
 328         if (ret)
 329                 goto err;
 330 err:
 331         if (ret == -EINTR)
 332                 goto retry;
 333
 334         bch2_disk_reservation_put(c, &disk_res);
 335
 336         return bch2_trans_exit(&trans) ?: ret;
 337 }
 338
 339 static int bch2_journal_replay(struct bch_fs *c,
 340                                struct journal_keys keys)
 341 {
 342         struct journal *j = &c->journal;
 343         struct journal_key *i;
 344         int ret;
 345
 346         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
 347
 348         for_each_journal_key(keys, i) {
 349                 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 350
 351                 if (i->btree_id == BTREE_ID_ALLOC)
 352                         ret = bch2_alloc_replay_key(c, i->k);
 353                 else if (btree_node_type_is_extents(i->btree_id))
 354                         ret = bch2_extent_replay_key(c, i->btree_id, i->k);
 355                 else
 356                         ret = bch2_btree_insert(c, i->btree_id, i->k,
 357                                                 NULL, NULL,
 358                                                 BTREE_INSERT_NOFAIL|
 359                                                 BTREE_INSERT_LAZY_RW|
 360                                                 BTREE_INSERT_JOURNAL_REPLAY|
 361                                                 BTREE_INSERT_NOMARK);
 362
 363                 if (ret) {
 364                         bch_err(c, "journal replay: error %d while replaying key",
 365                                 ret);
 366                         return ret;
 367                 }
 368
 369                 cond_resched();
 370         }
 371
 372         replay_now_at(j, j->replay_journal_seq_end);
 373         j->replay_journal_seq = 0;
 374
 375         bch2_journal_set_replay_done(j);
 376         bch2_journal_flush_all_pins(j);
 377         return bch2_journal_error(j);
 378 }
 379
 380 static bool journal_empty(struct list_head *journal)
 381 {
 382         return list_empty(journal) ||
 383                 journal_entry_empty(&list_last_entry(journal,
 384                                         struct journal_replay, list)->j);
 385 }
 386
 387 static int
 388 verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c,
 389                                                   struct list_head *journal)
 390 {
 391         struct journal_replay *i =
 392                 list_last_entry(journal, struct journal_replay, list);
 393         u64 start_seq   = le64_to_cpu(i->j.last_seq);
 394         u64 end_seq     = le64_to_cpu(i->j.seq);
 395         u64 seq         = start_seq;
 396         int ret = 0;
 397
 398         list_for_each_entry(i, journal, list) {
 399                 fsck_err_on(seq != le64_to_cpu(i->j.seq), c,
 400                         "journal entries %llu-%llu missing! (replaying %llu-%llu)",
 401                         seq, le64_to_cpu(i->j.seq) - 1,
 402                         start_seq, end_seq);
 403
 404                 seq = le64_to_cpu(i->j.seq);
 405
 406                 fsck_err_on(bch2_journal_seq_is_blacklisted(c, seq, false), c,
 407                             "found blacklisted journal entry %llu", seq);
 408
 409                 do {
 410                         seq++;
 411                 } while (bch2_journal_seq_is_blacklisted(c, seq, false));
 412         }
 413 fsck_err:
 414         return ret;
 415 }
 416
 417 /* journal replay early: */
 418
 419 static int journal_replay_entry_early(struct bch_fs *c,
 420                                       struct jset_entry *entry)
 421 {
 422         int ret = 0;
 423
 424         switch (entry->type) {
 425         case BCH_JSET_ENTRY_btree_root: {
 426                 struct btree_root *r;
 427
 428                 if (entry->btree_id >= BTREE_ID_NR) {
 429                         bch_err(c, "filesystem has unknown btree type %u",
 430                                 entry->btree_id);
 431                         return -EINVAL;
 432                 }
 433
 434                 r = &c->btree_roots[entry->btree_id];
 435
 436                 if (entry->u64s) {
 437                         r->level = entry->level;
 438                         bkey_copy(&r->key, &entry->start[0]);
 439                         r->error = 0;
 440                 } else {
 441                         r->error = -EIO;
 442                 }
 443                 r->alive = true;
 444                 break;
 445         }
 446         case BCH_JSET_ENTRY_usage: {
 447                 struct jset_entry_usage *u =
 448                         container_of(entry, struct jset_entry_usage, entry);
 449
 450                 switch (entry->btree_id) {
 451                 case FS_USAGE_RESERVED:
 452                         if (entry->level < BCH_REPLICAS_MAX)
 453                                 c->usage_base->persistent_reserved[entry->level] =
 454                                         le64_to_cpu(u->v);
 455                         break;
 456                 case FS_USAGE_INODES:
 457                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 458                         break;
 459                 case FS_USAGE_KEY_VERSION:
 460                         atomic64_set(&c->key_version,
 461                                      le64_to_cpu(u->v));
 462                         break;
 463                 }
 464
 465                 break;
 466         }
 467         case BCH_JSET_ENTRY_data_usage: {
 468                 struct jset_entry_data_usage *u =
 469                         container_of(entry, struct jset_entry_data_usage, entry);
 470                 ret = bch2_replicas_set_usage(c, &u->r,
 471                                               le64_to_cpu(u->v));
 472                 break;
 473         }
 474         case BCH_JSET_ENTRY_blacklist: {
 475                 struct jset_entry_blacklist *bl_entry =
 476                         container_of(entry, struct jset_entry_blacklist, entry);
 477
 478                 ret = bch2_journal_seq_blacklist_add(c,
 479                                 le64_to_cpu(bl_entry->seq),
 480                                 le64_to_cpu(bl_entry->seq) + 1);
 481                 break;
 482         }
 483         case BCH_JSET_ENTRY_blacklist_v2: {
 484                 struct jset_entry_blacklist_v2 *bl_entry =
 485                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 486
 487                 ret = bch2_journal_seq_blacklist_add(c,
 488                                 le64_to_cpu(bl_entry->start),
 489                                 le64_to_cpu(bl_entry->end) + 1);
 490                 break;
 491         }
 492         }
 493
 494         return ret;
 495 }
 496
 497 static int journal_replay_early(struct bch_fs *c,
 498                                 struct bch_sb_field_clean *clean,
 499                                 struct list_head *journal)
 500 {
 501         struct jset_entry *entry;
 502         int ret;
 503
 504         if (clean) {
 505                 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
 506                 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
 507
 508                 for (entry = clean->start;
 509                      entry != vstruct_end(&clean->field);
 510                      entry = vstruct_next(entry)) {
 511                         ret = journal_replay_entry_early(c, entry);
 512                         if (ret)
 513                                 return ret;
 514                 }
 515         } else {
 516                 struct journal_replay *i =
 517                         list_last_entry(journal, struct journal_replay, list);
 518
 519                 c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
 520                 c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
 521
 522                 list_for_each_entry(i, journal, list)
 523                         vstruct_for_each(&i->j, entry) {
 524                                 ret = journal_replay_entry_early(c, entry);
 525                                 if (ret)
 526                                         return ret;
 527                         }
 528         }
 529
 530         bch2_fs_usage_initialize(c);
 531
 532         return 0;
 533 }
 534
 535 /* sb clean section: */
 536
 537 static struct bkey_i *btree_root_find(struct bch_fs *c,
 538                                       struct bch_sb_field_clean *clean,
 539                                       struct jset *j,
 540                                       enum btree_id id, unsigned *level)
 541 {
 542         struct bkey_i *k;
 543         struct jset_entry *entry, *start, *end;
 544
 545         if (clean) {
 546                 start = clean->start;
 547                 end = vstruct_end(&clean->field);
 548         } else {
 549                 start = j->start;
 550                 end = vstruct_last(j);
 551         }
 552
 553         for (entry = start; entry < end; entry = vstruct_next(entry))
 554                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 555                     entry->btree_id == id)
 556                         goto found;
 557
 558         return NULL;
 559 found:
 560         if (!entry->u64s)
 561                 return ERR_PTR(-EINVAL);
 562
 563         k = entry->start;
 564         *level = entry->level;
 565         return k;
 566 }
 567
 568 static int verify_superblock_clean(struct bch_fs *c,
 569                                    struct bch_sb_field_clean **cleanp,
 570                                    struct jset *j)
 571 {
 572         unsigned i;
 573         struct bch_sb_field_clean *clean = *cleanp;
 574         int ret = 0;
 575
 576         if (!c->sb.clean || !j)
 577                 return 0;
 578
 579         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 580                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 581                         le64_to_cpu(clean->journal_seq),
 582                         le64_to_cpu(j->seq))) {
 583                 kfree(clean);
 584                 *cleanp = NULL;
 585                 return 0;
 586         }
 587
 588         mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
 589                         "superblock read clock doesn't match journal after clean shutdown");
 590         mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
 591                         "superblock read clock doesn't match journal after clean shutdown");
 592
 593         for (i = 0; i < BTREE_ID_NR; i++) {
 594                 struct bkey_i *k1, *k2;
 595                 unsigned l1 = 0, l2 = 0;
 596
 597                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 598                 k2 = btree_root_find(c, NULL, j, i, &l2);
 599
 600                 if (!k1 && !k2)
 601                         continue;
 602
 603                 mustfix_fsck_err_on(!k1 || !k2 ||
 604                                     IS_ERR(k1) ||
 605                                     IS_ERR(k2) ||
 606                                     k1->k.u64s != k2->k.u64s ||
 607                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 608                                     l1 != l2, c,
 609                         "superblock btree root doesn't match journal after clean shutdown");
 610         }
 611 fsck_err:
 612         return ret;
 613 }
 614
 615 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 616 {
 617         struct bch_sb_field_clean *clean, *sb_clean;
 618         int ret;
 619
 620         mutex_lock(&c->sb_lock);
 621         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 622
 623         if (fsck_err_on(!sb_clean, c,
 624                         "superblock marked clean but clean section not present")) {
 625                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 626                 c->sb.clean = false;
 627                 mutex_unlock(&c->sb_lock);
 628                 return NULL;
 629         }
 630
 631         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 632                         GFP_KERNEL);
 633         if (!clean) {
 634                 mutex_unlock(&c->sb_lock);
 635                 return ERR_PTR(-ENOMEM);
 636         }
 637
 638         if (le16_to_cpu(c->disk_sb.sb->version) <
 639             bcachefs_metadata_version_bkey_renumber)
 640                 bch2_sb_clean_renumber(clean, READ);
 641
 642         mutex_unlock(&c->sb_lock);
 643
 644         return clean;
 645 fsck_err:
 646         mutex_unlock(&c->sb_lock);
 647         return ERR_PTR(ret);
 648 }
 649
 650 static int read_btree_roots(struct bch_fs *c)
 651 {
 652         unsigned i;
 653         int ret = 0;
 654
 655         for (i = 0; i < BTREE_ID_NR; i++) {
 656                 struct btree_root *r = &c->btree_roots[i];
 657
 658                 if (!r->alive)
 659                         continue;
 660
 661                 if (i == BTREE_ID_ALLOC &&
 662                     c->opts.reconstruct_alloc) {
 663                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 664                         continue;
 665                 }
 666
 667
 668                 if (r->error) {
 669                         __fsck_err(c, i == BTREE_ID_ALLOC
 670                                    ? FSCK_CAN_IGNORE : 0,
 671                                    "invalid btree root %s",
 672                                    bch2_btree_ids[i]);
 673                         if (i == BTREE_ID_ALLOC)
 674                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 675                 }
 676
 677                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 678                 if (ret) {
 679                         __fsck_err(c, i == BTREE_ID_ALLOC
 680                                    ? FSCK_CAN_IGNORE : 0,
 681                                    "error reading btree root %s",
 682                                    bch2_btree_ids[i]);
 683                         if (i == BTREE_ID_ALLOC)
 684                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 685                 }
 686         }
 687
 688         for (i = 0; i < BTREE_ID_NR; i++)
 689                 if (!c->btree_roots[i].b)
 690                         bch2_btree_root_alloc(c, i);
 691 fsck_err:
 692         return ret;
 693 }
 694
 695 int bch2_fs_recovery(struct bch_fs *c)
 696 {
 697         const char *err = "cannot allocate memory";
 698         struct bch_sb_field_clean *clean = NULL;
 699         u64 journal_seq;
 700         LIST_HEAD(journal_entries);
 701         struct journal_keys journal_keys = { NULL };
 702         bool wrote = false, write_sb = false;
 703         int ret;
 704
 705         if (c->sb.clean)
 706                 clean = read_superblock_clean(c);
 707         ret = PTR_ERR_OR_ZERO(clean);
 708         if (ret)
 709                 goto err;
 710
 711         if (c->sb.clean)
 712                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 713                          le64_to_cpu(clean->journal_seq));
 714
 715         if (!c->replicas.entries) {
 716                 bch_info(c, "building replicas info");
 717                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 718         }
 719
 720         if (!c->sb.clean || c->opts.fsck) {
 721                 struct jset *j;
 722
 723                 ret = bch2_journal_read(c, &journal_entries);
 724                 if (ret)
 725                         goto err;
 726
 727                 if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&journal_entries), c,
 728                                 "filesystem marked clean but journal not empty")) {
 729                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 730                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 731                         c->sb.clean = false;
 732                 }
 733
 734                 if (!c->sb.clean && list_empty(&journal_entries)) {
 735                         bch_err(c, "no journal entries found");
 736                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
 737                         goto err;
 738                 }
 739
 740                 journal_keys = journal_keys_sort(&journal_entries);
 741                 if (!journal_keys.d) {
 742                         ret = -ENOMEM;
 743                         goto err;
 744                 }
 745
 746                 j = &list_last_entry(&journal_entries,
 747                                      struct journal_replay, list)->j;
 748
 749                 ret = verify_superblock_clean(c, &clean, j);
 750                 if (ret)
 751                         goto err;
 752
 753                 journal_seq = le64_to_cpu(j->seq) + 1;
 754         } else {
 755                 journal_seq = le64_to_cpu(clean->journal_seq) + 1;
 756         }
 757
 758         ret = journal_replay_early(c, clean, &journal_entries);
 759         if (ret)
 760                 goto err;
 761
 762         if (!c->sb.clean) {
 763                 ret = bch2_journal_seq_blacklist_add(c,
 764                                                      journal_seq,
 765                                                      journal_seq + 4);
 766                 if (ret) {
 767                         bch_err(c, "error creating new journal seq blacklist entry");
 768                         goto err;
 769                 }
 770
 771                 journal_seq += 4;
 772         }
 773
 774         ret = bch2_blacklist_table_initialize(c);
 775
 776         if (!list_empty(&journal_entries)) {
 777                 ret = verify_journal_entries_not_blacklisted_or_missing(c,
 778                                                         &journal_entries);
 779                 if (ret)
 780                         goto err;
 781         }
 782
 783         ret = bch2_fs_journal_start(&c->journal, journal_seq,
 784                                     &journal_entries);
 785         if (ret)
 786                 goto err;
 787
 788         ret = read_btree_roots(c);
 789         if (ret)
 790                 goto err;
 791
 792         bch_verbose(c, "starting alloc read");
 793         err = "error reading allocation information";
 794         ret = bch2_alloc_read(c, &journal_keys);
 795         if (ret)
 796                 goto err;
 797         bch_verbose(c, "alloc read done");
 798
 799         bch_verbose(c, "starting stripes_read");
 800         err = "error reading stripes";
 801         ret = bch2_stripes_read(c, &journal_keys);
 802         if (ret)
 803                 goto err;
 804         bch_verbose(c, "stripes_read done");
 805
 806         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 807
 808         if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
 809             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
 810                 /*
 811                  * interior btree node updates aren't consistent with the
 812                  * journal; after an unclean shutdown we have to walk all
 813                  * pointers to metadata:
 814                  */
 815                 bch_info(c, "starting metadata mark and sweep");
 816                 err = "error in mark and sweep";
 817                 ret = bch2_gc(c, NULL, true, true);
 818                 if (ret)
 819                         goto err;
 820                 bch_verbose(c, "mark and sweep done");
 821         }
 822
 823         if (c->opts.fsck ||
 824             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
 825             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
 826                 bch_info(c, "starting mark and sweep");
 827                 err = "error in mark and sweep";
 828                 ret = bch2_gc(c, &journal_keys, true, false);
 829                 if (ret)
 830                         goto err;
 831                 bch_verbose(c, "mark and sweep done");
 832         }
 833
 834         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 835         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
 836
 837         /*
 838          * Skip past versions that might have possibly been used (as nonces),
 839          * but hadn't had their pointers written:
 840          */
 841         if (c->sb.encryption_type && !c->sb.clean)
 842                 atomic64_add(1 << 16, &c->key_version);
 843
 844         if (c->opts.norecovery)
 845                 goto out;
 846
 847         bch_verbose(c, "starting journal replay");
 848         err = "journal replay failed";
 849         ret = bch2_journal_replay(c, journal_keys);
 850         if (ret)
 851                 goto err;
 852         bch_verbose(c, "journal replay done");
 853
 854         if (!c->opts.nochanges) {
 855                 /*
 856                  * note that even when filesystem was clean there might be work
 857                  * to do here, if we ran gc (because of fsck) which recalculated
 858                  * oldest_gen:
 859                  */
 860                 bch_verbose(c, "writing allocation info");
 861                 err = "error writing out alloc info";
 862                 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW, &wrote) ?:
 863                         bch2_alloc_write(c, BTREE_INSERT_LAZY_RW, &wrote);
 864                 if (ret) {
 865                         bch_err(c, "error writing alloc info");
 866                         goto err;
 867                 }
 868                 bch_verbose(c, "alloc write done");
 869         }
 870
 871         if (!c->sb.clean) {
 872                 if (!(c->sb.features & (1 << BCH_FEATURE_ATOMIC_NLINK))) {
 873                         bch_info(c, "checking inode link counts");
 874                         err = "error in recovery";
 875                         ret = bch2_fsck_inode_nlink(c);
 876                         if (ret)
 877                                 goto err;
 878                         bch_verbose(c, "check inodes done");
 879
 880                 } else {
 881                         bch_verbose(c, "checking for deleted inodes");
 882                         err = "error in recovery";
 883                         ret = bch2_fsck_walk_inodes_only(c);
 884                         if (ret)
 885                                 goto err;
 886                         bch_verbose(c, "check inodes done");
 887                 }
 888         }
 889
 890         if (c->opts.fsck) {
 891                 bch_info(c, "starting fsck");
 892                 err = "error in fsck";
 893                 ret = bch2_fsck_full(c);
 894                 if (ret)
 895                         goto err;
 896                 bch_verbose(c, "fsck done");
 897         }
 898
 899         if (enabled_qtypes(c)) {
 900                 bch_verbose(c, "reading quotas");
 901                 ret = bch2_fs_quota_read(c);
 902                 if (ret)
 903                         goto err;
 904                 bch_verbose(c, "quotas done");
 905         }
 906
 907         mutex_lock(&c->sb_lock);
 908         if (c->opts.version_upgrade) {
 909                 if (c->sb.version < bcachefs_metadata_version_new_versioning)
 910                         c->disk_sb.sb->version_min =
 911                                 le16_to_cpu(bcachefs_metadata_version_min);
 912                 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
 913                 write_sb = true;
 914         }
 915
 916         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
 917                 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
 918                 write_sb = true;
 919         }
 920
 921         if (c->opts.fsck &&
 922             !test_bit(BCH_FS_ERROR, &c->flags)) {
 923                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_ATOMIC_NLINK;
 924                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
 925                 write_sb = true;
 926         }
 927
 928         if (write_sb)
 929                 bch2_write_super(c);
 930         mutex_unlock(&c->sb_lock);
 931
 932         if (c->journal_seq_blacklist_table &&
 933             c->journal_seq_blacklist_table->nr > 128)
 934                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
 935 out:
 936         ret = 0;
 937 err:
 938 fsck_err:
 939         set_bit(BCH_FS_FSCK_DONE, &c->flags);
 940         bch2_flush_fsck_errs(c);
 941
 942         journal_keys_free(&journal_keys);
 943         journal_entries_free(&journal_entries);
 944         kfree(clean);
 945         if (ret)
 946                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
 947         else
 948                 bch_verbose(c, "ret %i", ret);
 949         return ret;
 950 }
 951
 952 int bch2_fs_initialize(struct bch_fs *c)
 953 {
 954         struct bch_inode_unpacked root_inode, lostfound_inode;
 955         struct bkey_inode_buf packed_inode;
 956         struct qstr lostfound = QSTR("lost+found");
 957         const char *err = "cannot allocate memory";
 958         struct bch_dev *ca;
 959         LIST_HEAD(journal);
 960         unsigned i;
 961         int ret;
 962
 963         bch_notice(c, "initializing new filesystem");
 964
 965         mutex_lock(&c->sb_lock);
 966         for_each_online_member(ca, c, i)
 967                 bch2_mark_dev_superblock(c, ca, 0);
 968         mutex_unlock(&c->sb_lock);
 969
 970         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 971         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
 972
 973         for (i = 0; i < BTREE_ID_NR; i++)
 974                 bch2_btree_root_alloc(c, i);
 975
 976         err = "unable to allocate journal buckets";
 977         for_each_online_member(ca, c, i) {
 978                 ret = bch2_dev_journal_alloc(ca);
 979                 if (ret) {
 980                         percpu_ref_put(&ca->io_ref);
 981                         goto err;
 982                 }
 983         }
 984
 985         /*
 986          * journal_res_get() will crash if called before this has
 987          * set up the journal.pin FIFO and journal.cur pointer:
 988          */
 989         bch2_fs_journal_start(&c->journal, 1, &journal);
 990         bch2_journal_set_replay_done(&c->journal);
 991
 992         err = "error going read write";
 993         ret = __bch2_fs_read_write(c, true);
 994         if (ret)
 995                 goto err;
 996
 997         bch2_inode_init(c, &root_inode, 0, 0,
 998                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
 999         root_inode.bi_inum = BCACHEFS_ROOT_INO;
1000         bch2_inode_pack(&packed_inode, &root_inode);
1001
1002         err = "error creating root directory";
1003         ret = bch2_btree_insert(c, BTREE_ID_INODES,
1004                                 &packed_inode.inode.k_i,
1005                                 NULL, NULL, 0);
1006         if (ret)
1007                 goto err;
1008
1009         bch2_inode_init_early(c, &lostfound_inode);
1010
1011         err = "error creating lost+found";
1012         ret = bch2_trans_do(c, NULL, BTREE_INSERT_ATOMIC,
1013                 bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
1014                                   &root_inode, &lostfound_inode,
1015                                   &lostfound,
1016                                   0, 0, 0755, 0,
1017                                   NULL, NULL));
1018         if (ret)
1019                 goto err;
1020
1021         if (enabled_qtypes(c)) {
1022                 ret = bch2_fs_quota_read(c);
1023                 if (ret)
1024                         goto err;
1025         }
1026
1027         err = "error writing first journal entry";
1028         ret = bch2_journal_meta(&c->journal);
1029         if (ret)
1030                 goto err;
1031
1032         mutex_lock(&c->sb_lock);
1033         c->disk_sb.sb->version = c->disk_sb.sb->version_min =
1034                 le16_to_cpu(bcachefs_metadata_version_current);
1035         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_ATOMIC_NLINK;
1036
1037         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1038         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1039
1040         bch2_write_super(c);
1041         mutex_unlock(&c->sb_lock);
1042
1043         return 0;
1044 err:
1045         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1046         return ret;
1047 }