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 bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
  31 {
  32         while (iter->k) {
  33                 if (iter->k->btree_id == iter->btree_id)
  34                         return bkey_i_to_s_c(iter->k->k);
  35
  36                 iter->k++;
  37                 if (iter->k == iter->keys->d + iter->keys->nr)
  38                         iter->k = NULL;
  39         }
  40
  41         return bkey_s_c_null;
  42 }
  43
  44 struct bkey_s_c bch2_journal_iter_next(struct journal_iter *iter)
  45 {
  46         if (!iter->k)
  47                 return bkey_s_c_null;
  48
  49         iter->k++;
  50         if (iter->k == iter->keys->d + iter->keys->nr)
  51                 iter->k = NULL;
  52
  53         return bch2_journal_iter_peek(iter);
  54 }
  55
  56 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
  57 {
  58         switch (iter->last) {
  59         case none:
  60                 break;
  61         case btree:
  62                 bch2_btree_iter_next(iter->btree);
  63                 break;
  64         case journal:
  65                 bch2_journal_iter_next(&iter->journal);
  66                 break;
  67         }
  68
  69         iter->last = none;
  70 }
  71
  72 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
  73 {
  74         struct bkey_s_c ret;
  75
  76         while (1) {
  77                 struct bkey_s_c btree_k         = bch2_btree_iter_peek(iter->btree);
  78                 struct bkey_s_c journal_k       = bch2_journal_iter_peek(&iter->journal);
  79
  80                 if (btree_k.k && journal_k.k) {
  81                         int cmp = bkey_cmp(btree_k.k->p, journal_k.k->p);
  82
  83                         if (!cmp)
  84                                 bch2_btree_iter_next(iter->btree);
  85
  86                         iter->last = cmp < 0 ? btree : journal;
  87                 } else if (btree_k.k) {
  88                         iter->last = btree;
  89                 } else if (journal_k.k) {
  90                         iter->last = journal;
  91                 } else {
  92                         iter->last = none;
  93                         return bkey_s_c_null;
  94                 }
  95
  96                 ret = iter->last == journal ? journal_k : btree_k;
  97                 if (!bkey_deleted(ret.k))
  98                         break;
  99
 100                 bch2_btree_and_journal_iter_advance(iter);
 101         }
 102
 103         return ret;
 104 }
 105
 106 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 107 {
 108         bch2_btree_and_journal_iter_advance(iter);
 109
 110         return bch2_btree_and_journal_iter_peek(iter);
 111 }
 112
 113 struct journal_key *journal_key_search(struct journal_keys *journal_keys,
 114                                        enum btree_id id, struct bpos pos)
 115 {
 116         size_t l = 0, r = journal_keys->nr, m;
 117
 118         while (l < r) {
 119                 m = l + ((r - l) >> 1);
 120                 if ((cmp_int(id, journal_keys->d[m].btree_id) ?:
 121                      bkey_cmp(pos, journal_keys->d[m].k->k.p)) > 0)
 122                         l = m + 1;
 123                 else
 124                         r = m;
 125         }
 126
 127         BUG_ON(l < journal_keys->nr &&
 128                (cmp_int(id, journal_keys->d[l].btree_id) ?:
 129                 bkey_cmp(pos, journal_keys->d[l].k->k.p)) > 0);
 130
 131         BUG_ON(l &&
 132                (cmp_int(id, journal_keys->d[l - 1].btree_id) ?:
 133                 bkey_cmp(pos, journal_keys->d[l - 1].k->k.p)) <= 0);
 134
 135         return l < journal_keys->nr ? journal_keys->d + l : NULL;
 136 }
 137
 138 void bch2_btree_and_journal_iter_init(struct btree_and_journal_iter *iter,
 139                                       struct btree_trans *trans,
 140                                       struct journal_keys *journal_keys,
 141                                       enum btree_id id, struct bpos pos)
 142 {
 143         iter->journal.keys      = journal_keys;
 144         iter->journal.k         = journal_key_search(journal_keys, id, pos);
 145         iter->journal.btree_id  = id;
 146
 147         iter->btree = bch2_trans_get_iter(trans, id, pos, 0);
 148 }
 149
 150 /* sort and dedup all keys in the journal: */
 151
 152 static void journal_entries_free(struct list_head *list)
 153 {
 154
 155         while (!list_empty(list)) {
 156                 struct journal_replay *i =
 157                         list_first_entry(list, struct journal_replay, list);
 158                 list_del(&i->list);
 159                 kvpfree(i, offsetof(struct journal_replay, j) +
 160                         vstruct_bytes(&i->j));
 161         }
 162 }
 163
 164 /*
 165  * When keys compare equal, oldest compares first:
 166  */
 167 static int journal_sort_key_cmp(const void *_l, const void *_r)
 168 {
 169         const struct journal_key *l = _l;
 170         const struct journal_key *r = _r;
 171
 172         return cmp_int(l->btree_id, r->btree_id) ?:
 173                 bkey_cmp(l->k->k.p, r->k->k.p) ?:
 174                 cmp_int(l->journal_seq, r->journal_seq) ?:
 175                 cmp_int(l->journal_offset, r->journal_offset);
 176 }
 177
 178 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 179 {
 180         const struct journal_key *l = _l;
 181         const struct journal_key *r = _r;
 182
 183         return cmp_int(l->journal_seq, r->journal_seq) ?:
 184                 cmp_int(l->btree_id, r->btree_id) ?:
 185                 bkey_cmp(l->k->k.p, r->k->k.p);
 186 }
 187
 188 static void journal_keys_free(struct journal_keys *keys)
 189 {
 190         kvfree(keys->d);
 191         keys->d = NULL;
 192         keys->nr = 0;
 193 }
 194
 195 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 196 {
 197         struct journal_replay *p;
 198         struct jset_entry *entry;
 199         struct bkey_i *k, *_n;
 200         struct journal_keys keys = { NULL };
 201         struct journal_key *src, *dst;
 202         size_t nr_keys = 0;
 203
 204         list_for_each_entry(p, journal_entries, list)
 205                 for_each_jset_key(k, _n, entry, &p->j)
 206                         nr_keys++;
 207
 208         keys.journal_seq_base =
 209                 le64_to_cpu(list_first_entry(journal_entries,
 210                                              struct journal_replay,
 211                                              list)->j.seq);
 212
 213         keys.d = kvmalloc(sizeof(keys.d[0]) * nr_keys, GFP_KERNEL);
 214         if (!keys.d)
 215                 goto err;
 216
 217         list_for_each_entry(p, journal_entries, list)
 218                 for_each_jset_key(k, _n, entry, &p->j)
 219                         keys.d[keys.nr++] = (struct journal_key) {
 220                                 .btree_id       = entry->btree_id,
 221                                 .k              = k,
 222                                 .journal_seq    = le64_to_cpu(p->j.seq) -
 223                                         keys.journal_seq_base,
 224                                 .journal_offset = k->_data - p->j._data,
 225                         };
 226
 227         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 228
 229         src = dst = keys.d;
 230         while (src < keys.d + keys.nr) {
 231                 while (src + 1 < keys.d + keys.nr &&
 232                        src[0].btree_id == src[1].btree_id &&
 233                        !bkey_cmp(src[0].k->k.p, src[1].k->k.p))
 234                         src++;
 235
 236                 *dst++ = *src++;
 237         }
 238
 239         keys.nr = dst - keys.d;
 240 err:
 241         return keys;
 242 }
 243
 244 /* journal replay: */
 245
 246 static void replay_now_at(struct journal *j, u64 seq)
 247 {
 248         BUG_ON(seq < j->replay_journal_seq);
 249         BUG_ON(seq > j->replay_journal_seq_end);
 250
 251         while (j->replay_journal_seq < seq)
 252                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 253 }
 254
 255 static int bch2_extent_replay_key(struct bch_fs *c, enum btree_id btree_id,
 256                                   struct bkey_i *k)
 257 {
 258         struct btree_trans trans;
 259         struct btree_iter *iter, *split_iter;
 260         /*
 261          * We might cause compressed extents to be split, so we need to pass in
 262          * a disk_reservation:
 263          */
 264         struct disk_reservation disk_res =
 265                 bch2_disk_reservation_init(c, 0);
 266         struct bkey_i *split;
 267         struct bpos atomic_end;
 268         /*
 269          * Some extents aren't equivalent - w.r.t. what the triggers do
 270          * - if they're split:
 271          */
 272         bool remark_if_split = bch2_bkey_sectors_compressed(bkey_i_to_s_c(k)) ||
 273                 k->k.type == KEY_TYPE_reflink_p;
 274         bool remark = false;
 275         int ret;
 276
 277         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 278 retry:
 279         bch2_trans_begin(&trans);
 280
 281         iter = bch2_trans_get_iter(&trans, btree_id,
 282                                    bkey_start_pos(&k->k),
 283                                    BTREE_ITER_INTENT);
 284
 285         do {
 286                 ret = bch2_btree_iter_traverse(iter);
 287                 if (ret)
 288                         goto err;
 289
 290                 atomic_end = bpos_min(k->k.p, iter->l[0].b->key.k.p);
 291
 292                 split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
 293                 ret = PTR_ERR_OR_ZERO(split);
 294                 if (ret)
 295                         goto err;
 296
 297                 if (!remark &&
 298                     remark_if_split &&
 299                     bkey_cmp(atomic_end, k->k.p) < 0) {
 300                         ret = bch2_disk_reservation_add(c, &disk_res,
 301                                         k->k.size *
 302                                         bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(k)),
 303                                         BCH_DISK_RESERVATION_NOFAIL);
 304                         BUG_ON(ret);
 305
 306                         remark = true;
 307                 }
 308
 309                 bkey_copy(split, k);
 310                 bch2_cut_front(iter->pos, split);
 311                 bch2_cut_back(atomic_end, split);
 312
 313                 split_iter = bch2_trans_copy_iter(&trans, iter);
 314                 ret = PTR_ERR_OR_ZERO(split_iter);
 315                 if (ret)
 316                         goto err;
 317
 318                 /*
 319                  * It's important that we don't go through the
 320                  * extent_handle_overwrites() and extent_update_to_keys() path
 321                  * here: journal replay is supposed to treat extents like
 322                  * regular keys
 323                  */
 324                 __bch2_btree_iter_set_pos(split_iter, split->k.p, false);
 325                 bch2_trans_update(&trans, split_iter, split, !remark
 326                                   ? BTREE_TRIGGER_NORUN
 327                                   : BTREE_TRIGGER_NOOVERWRITES);
 328
 329                 bch2_btree_iter_set_pos(iter, split->k.p);
 330         } while (bkey_cmp(iter->pos, k->k.p) < 0);
 331
 332         if (remark) {
 333                 ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k),
 334                                           0, -((s64) k->k.size),
 335                                           BTREE_TRIGGER_OVERWRITE);
 336                 if (ret)
 337                         goto err;
 338         }
 339
 340         ret = bch2_trans_commit(&trans, &disk_res, NULL,
 341                                 BTREE_INSERT_NOFAIL|
 342                                 BTREE_INSERT_LAZY_RW|
 343                                 BTREE_INSERT_JOURNAL_REPLAY);
 344 err:
 345         if (ret == -EINTR)
 346                 goto retry;
 347
 348         bch2_disk_reservation_put(c, &disk_res);
 349
 350         return bch2_trans_exit(&trans) ?: ret;
 351 }
 352
 353 static int __bch2_journal_replay_key(struct btree_trans *trans,
 354                                      enum btree_id id, struct bkey_i *k)
 355 {
 356         struct btree_iter *iter;
 357         int ret;
 358
 359         iter = bch2_trans_get_iter(trans, id, k->k.p, BTREE_ITER_INTENT);
 360         if (IS_ERR(iter))
 361                 return PTR_ERR(iter);
 362
 363         /*
 364          * iter->flags & BTREE_ITER_IS_EXTENTS triggers the update path to run
 365          * extent_handle_overwrites() and extent_update_to_keys() - but we don't
 366          * want that here, journal replay is supposed to treat extents like
 367          * regular keys:
 368          */
 369         __bch2_btree_iter_set_pos(iter, k->k.p, false);
 370
 371         ret   = bch2_btree_iter_traverse(iter) ?:
 372                 bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 373         bch2_trans_iter_put(trans, iter);
 374         return ret;
 375 }
 376
 377 static int bch2_journal_replay_key(struct bch_fs *c, enum btree_id id,
 378                                    struct bkey_i *k)
 379 {
 380         return bch2_trans_do(c, NULL, NULL,
 381                              BTREE_INSERT_NOFAIL|
 382                              BTREE_INSERT_LAZY_RW|
 383                              BTREE_INSERT_JOURNAL_REPLAY,
 384                              __bch2_journal_replay_key(&trans, id, k));
 385 }
 386
 387 static int bch2_journal_replay(struct bch_fs *c,
 388                                struct journal_keys keys)
 389 {
 390         struct journal *j = &c->journal;
 391         struct journal_key *i;
 392         int ret;
 393
 394         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
 395
 396         for_each_journal_key(keys, i) {
 397                 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 398
 399                 if (i->btree_id == BTREE_ID_ALLOC)
 400                         ret = bch2_alloc_replay_key(c, i->k);
 401                 else if (i->k->k.size)
 402                         ret = bch2_extent_replay_key(c, i->btree_id, i->k);
 403                 else
 404                         ret = bch2_journal_replay_key(c, i->btree_id, i->k);
 405
 406                 if (ret) {
 407                         bch_err(c, "journal replay: error %d while replaying key",
 408                                 ret);
 409                         return ret;
 410                 }
 411
 412                 cond_resched();
 413         }
 414
 415         replay_now_at(j, j->replay_journal_seq_end);
 416         j->replay_journal_seq = 0;
 417
 418         bch2_journal_set_replay_done(j);
 419         bch2_journal_flush_all_pins(j);
 420         return bch2_journal_error(j);
 421 }
 422
 423 static bool journal_empty(struct list_head *journal)
 424 {
 425         return list_empty(journal) ||
 426                 journal_entry_empty(&list_last_entry(journal,
 427                                         struct journal_replay, list)->j);
 428 }
 429
 430 static int
 431 verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c,
 432                                                   struct list_head *journal)
 433 {
 434         struct journal_replay *i =
 435                 list_last_entry(journal, struct journal_replay, list);
 436         u64 start_seq   = le64_to_cpu(i->j.last_seq);
 437         u64 end_seq     = le64_to_cpu(i->j.seq);
 438         u64 seq         = start_seq;
 439         int ret = 0;
 440
 441         list_for_each_entry(i, journal, list) {
 442                 fsck_err_on(seq != le64_to_cpu(i->j.seq), c,
 443                         "journal entries %llu-%llu missing! (replaying %llu-%llu)",
 444                         seq, le64_to_cpu(i->j.seq) - 1,
 445                         start_seq, end_seq);
 446
 447                 seq = le64_to_cpu(i->j.seq);
 448
 449                 fsck_err_on(bch2_journal_seq_is_blacklisted(c, seq, false), c,
 450                             "found blacklisted journal entry %llu", seq);
 451
 452                 do {
 453                         seq++;
 454                 } while (bch2_journal_seq_is_blacklisted(c, seq, false));
 455         }
 456 fsck_err:
 457         return ret;
 458 }
 459
 460 /* journal replay early: */
 461
 462 static int journal_replay_entry_early(struct bch_fs *c,
 463                                       struct jset_entry *entry)
 464 {
 465         int ret = 0;
 466
 467         switch (entry->type) {
 468         case BCH_JSET_ENTRY_btree_root: {
 469                 struct btree_root *r;
 470
 471                 if (entry->btree_id >= BTREE_ID_NR) {
 472                         bch_err(c, "filesystem has unknown btree type %u",
 473                                 entry->btree_id);
 474                         return -EINVAL;
 475                 }
 476
 477                 r = &c->btree_roots[entry->btree_id];
 478
 479                 if (entry->u64s) {
 480                         r->level = entry->level;
 481                         bkey_copy(&r->key, &entry->start[0]);
 482                         r->error = 0;
 483                 } else {
 484                         r->error = -EIO;
 485                 }
 486                 r->alive = true;
 487                 break;
 488         }
 489         case BCH_JSET_ENTRY_usage: {
 490                 struct jset_entry_usage *u =
 491                         container_of(entry, struct jset_entry_usage, entry);
 492
 493                 switch (entry->btree_id) {
 494                 case FS_USAGE_RESERVED:
 495                         if (entry->level < BCH_REPLICAS_MAX)
 496                                 c->usage_base->persistent_reserved[entry->level] =
 497                                         le64_to_cpu(u->v);
 498                         break;
 499                 case FS_USAGE_INODES:
 500                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 501                         break;
 502                 case FS_USAGE_KEY_VERSION:
 503                         atomic64_set(&c->key_version,
 504                                      le64_to_cpu(u->v));
 505                         break;
 506                 }
 507
 508                 break;
 509         }
 510         case BCH_JSET_ENTRY_data_usage: {
 511                 struct jset_entry_data_usage *u =
 512                         container_of(entry, struct jset_entry_data_usage, entry);
 513                 ret = bch2_replicas_set_usage(c, &u->r,
 514                                               le64_to_cpu(u->v));
 515                 break;
 516         }
 517         case BCH_JSET_ENTRY_blacklist: {
 518                 struct jset_entry_blacklist *bl_entry =
 519                         container_of(entry, struct jset_entry_blacklist, entry);
 520
 521                 ret = bch2_journal_seq_blacklist_add(c,
 522                                 le64_to_cpu(bl_entry->seq),
 523                                 le64_to_cpu(bl_entry->seq) + 1);
 524                 break;
 525         }
 526         case BCH_JSET_ENTRY_blacklist_v2: {
 527                 struct jset_entry_blacklist_v2 *bl_entry =
 528                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 529
 530                 ret = bch2_journal_seq_blacklist_add(c,
 531                                 le64_to_cpu(bl_entry->start),
 532                                 le64_to_cpu(bl_entry->end) + 1);
 533                 break;
 534         }
 535         }
 536
 537         return ret;
 538 }
 539
 540 static int journal_replay_early(struct bch_fs *c,
 541                                 struct bch_sb_field_clean *clean,
 542                                 struct list_head *journal)
 543 {
 544         struct jset_entry *entry;
 545         int ret;
 546
 547         if (clean) {
 548                 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
 549                 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
 550
 551                 for (entry = clean->start;
 552                      entry != vstruct_end(&clean->field);
 553                      entry = vstruct_next(entry)) {
 554                         ret = journal_replay_entry_early(c, entry);
 555                         if (ret)
 556                                 return ret;
 557                 }
 558         } else {
 559                 struct journal_replay *i =
 560                         list_last_entry(journal, struct journal_replay, list);
 561
 562                 c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
 563                 c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
 564
 565                 list_for_each_entry(i, journal, list)
 566                         vstruct_for_each(&i->j, entry) {
 567                                 ret = journal_replay_entry_early(c, entry);
 568                                 if (ret)
 569                                         return ret;
 570                         }
 571         }
 572
 573         bch2_fs_usage_initialize(c);
 574
 575         return 0;
 576 }
 577
 578 /* sb clean section: */
 579
 580 static struct bkey_i *btree_root_find(struct bch_fs *c,
 581                                       struct bch_sb_field_clean *clean,
 582                                       struct jset *j,
 583                                       enum btree_id id, unsigned *level)
 584 {
 585         struct bkey_i *k;
 586         struct jset_entry *entry, *start, *end;
 587
 588         if (clean) {
 589                 start = clean->start;
 590                 end = vstruct_end(&clean->field);
 591         } else {
 592                 start = j->start;
 593                 end = vstruct_last(j);
 594         }
 595
 596         for (entry = start; entry < end; entry = vstruct_next(entry))
 597                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 598                     entry->btree_id == id)
 599                         goto found;
 600
 601         return NULL;
 602 found:
 603         if (!entry->u64s)
 604                 return ERR_PTR(-EINVAL);
 605
 606         k = entry->start;
 607         *level = entry->level;
 608         return k;
 609 }
 610
 611 static int verify_superblock_clean(struct bch_fs *c,
 612                                    struct bch_sb_field_clean **cleanp,
 613                                    struct jset *j)
 614 {
 615         unsigned i;
 616         struct bch_sb_field_clean *clean = *cleanp;
 617         int ret = 0;
 618
 619         if (!c->sb.clean || !j)
 620                 return 0;
 621
 622         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 623                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 624                         le64_to_cpu(clean->journal_seq),
 625                         le64_to_cpu(j->seq))) {
 626                 kfree(clean);
 627                 *cleanp = NULL;
 628                 return 0;
 629         }
 630
 631         mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
 632                         "superblock read clock doesn't match journal after clean shutdown");
 633         mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
 634                         "superblock read clock doesn't match journal after clean shutdown");
 635
 636         for (i = 0; i < BTREE_ID_NR; i++) {
 637                 struct bkey_i *k1, *k2;
 638                 unsigned l1 = 0, l2 = 0;
 639
 640                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 641                 k2 = btree_root_find(c, NULL, j, i, &l2);
 642
 643                 if (!k1 && !k2)
 644                         continue;
 645
 646                 mustfix_fsck_err_on(!k1 || !k2 ||
 647                                     IS_ERR(k1) ||
 648                                     IS_ERR(k2) ||
 649                                     k1->k.u64s != k2->k.u64s ||
 650                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 651                                     l1 != l2, c,
 652                         "superblock btree root doesn't match journal after clean shutdown");
 653         }
 654 fsck_err:
 655         return ret;
 656 }
 657
 658 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 659 {
 660         struct bch_sb_field_clean *clean, *sb_clean;
 661         int ret;
 662
 663         mutex_lock(&c->sb_lock);
 664         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 665
 666         if (fsck_err_on(!sb_clean, c,
 667                         "superblock marked clean but clean section not present")) {
 668                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 669                 c->sb.clean = false;
 670                 mutex_unlock(&c->sb_lock);
 671                 return NULL;
 672         }
 673
 674         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 675                         GFP_KERNEL);
 676         if (!clean) {
 677                 mutex_unlock(&c->sb_lock);
 678                 return ERR_PTR(-ENOMEM);
 679         }
 680
 681         if (le16_to_cpu(c->disk_sb.sb->version) <
 682             bcachefs_metadata_version_bkey_renumber)
 683                 bch2_sb_clean_renumber(clean, READ);
 684
 685         mutex_unlock(&c->sb_lock);
 686
 687         return clean;
 688 fsck_err:
 689         mutex_unlock(&c->sb_lock);
 690         return ERR_PTR(ret);
 691 }
 692
 693 static int read_btree_roots(struct bch_fs *c)
 694 {
 695         unsigned i;
 696         int ret = 0;
 697
 698         for (i = 0; i < BTREE_ID_NR; i++) {
 699                 struct btree_root *r = &c->btree_roots[i];
 700
 701                 if (!r->alive)
 702                         continue;
 703
 704                 if (i == BTREE_ID_ALLOC &&
 705                     c->opts.reconstruct_alloc) {
 706                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 707                         continue;
 708                 }
 709
 710
 711                 if (r->error) {
 712                         __fsck_err(c, i == BTREE_ID_ALLOC
 713                                    ? FSCK_CAN_IGNORE : 0,
 714                                    "invalid btree root %s",
 715                                    bch2_btree_ids[i]);
 716                         if (i == BTREE_ID_ALLOC)
 717                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 718                 }
 719
 720                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 721                 if (ret) {
 722                         __fsck_err(c, i == BTREE_ID_ALLOC
 723                                    ? FSCK_CAN_IGNORE : 0,
 724                                    "error reading btree root %s",
 725                                    bch2_btree_ids[i]);
 726                         if (i == BTREE_ID_ALLOC)
 727                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 728                 }
 729         }
 730
 731         for (i = 0; i < BTREE_ID_NR; i++)
 732                 if (!c->btree_roots[i].b)
 733                         bch2_btree_root_alloc(c, i);
 734 fsck_err:
 735         return ret;
 736 }
 737
 738 int bch2_fs_recovery(struct bch_fs *c)
 739 {
 740         const char *err = "cannot allocate memory";
 741         struct bch_sb_field_clean *clean = NULL;
 742         u64 journal_seq;
 743         LIST_HEAD(journal_entries);
 744         struct journal_keys journal_keys = { NULL };
 745         bool wrote = false, write_sb = false;
 746         int ret;
 747
 748         if (c->sb.clean)
 749                 clean = read_superblock_clean(c);
 750         ret = PTR_ERR_OR_ZERO(clean);
 751         if (ret)
 752                 goto err;
 753
 754         if (c->sb.clean)
 755                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 756                          le64_to_cpu(clean->journal_seq));
 757
 758         if (!c->replicas.entries) {
 759                 bch_info(c, "building replicas info");
 760                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 761         }
 762
 763         if (!c->sb.clean || c->opts.fsck) {
 764                 struct jset *j;
 765
 766                 ret = bch2_journal_read(c, &journal_entries);
 767                 if (ret)
 768                         goto err;
 769
 770                 if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&journal_entries), c,
 771                                 "filesystem marked clean but journal not empty")) {
 772                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 773                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 774                         c->sb.clean = false;
 775                 }
 776
 777                 if (!c->sb.clean && list_empty(&journal_entries)) {
 778                         bch_err(c, "no journal entries found");
 779                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
 780                         goto err;
 781                 }
 782
 783                 journal_keys = journal_keys_sort(&journal_entries);
 784                 if (!journal_keys.d) {
 785                         ret = -ENOMEM;
 786                         goto err;
 787                 }
 788
 789                 j = &list_last_entry(&journal_entries,
 790                                      struct journal_replay, list)->j;
 791
 792                 ret = verify_superblock_clean(c, &clean, j);
 793                 if (ret)
 794                         goto err;
 795
 796                 journal_seq = le64_to_cpu(j->seq) + 1;
 797         } else {
 798                 journal_seq = le64_to_cpu(clean->journal_seq) + 1;
 799         }
 800
 801         if (!c->sb.clean &&
 802             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
 803                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
 804                 ret = -EINVAL;
 805                 goto err;
 806         }
 807
 808         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_extents_above_btree_updates;
 809
 810         ret = journal_replay_early(c, clean, &journal_entries);
 811         if (ret)
 812                 goto err;
 813
 814         if (!c->sb.clean) {
 815                 ret = bch2_journal_seq_blacklist_add(c,
 816                                                      journal_seq,
 817                                                      journal_seq + 4);
 818                 if (ret) {
 819                         bch_err(c, "error creating new journal seq blacklist entry");
 820                         goto err;
 821                 }
 822
 823                 journal_seq += 4;
 824         }
 825
 826         ret = bch2_blacklist_table_initialize(c);
 827
 828         if (!list_empty(&journal_entries)) {
 829                 ret = verify_journal_entries_not_blacklisted_or_missing(c,
 830                                                         &journal_entries);
 831                 if (ret)
 832                         goto err;
 833         }
 834
 835         ret = bch2_fs_journal_start(&c->journal, journal_seq,
 836                                     &journal_entries);
 837         if (ret)
 838                 goto err;
 839
 840         ret = read_btree_roots(c);
 841         if (ret)
 842                 goto err;
 843
 844         bch_verbose(c, "starting alloc read");
 845         err = "error reading allocation information";
 846         ret = bch2_alloc_read(c, &journal_keys);
 847         if (ret)
 848                 goto err;
 849         bch_verbose(c, "alloc read done");
 850
 851         bch_verbose(c, "starting stripes_read");
 852         err = "error reading stripes";
 853         ret = bch2_stripes_read(c, &journal_keys);
 854         if (ret)
 855                 goto err;
 856         bch_verbose(c, "stripes_read done");
 857
 858         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 859
 860         if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
 861             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
 862                 /*
 863                  * interior btree node updates aren't consistent with the
 864                  * journal; after an unclean shutdown we have to walk all
 865                  * pointers to metadata:
 866                  */
 867                 bch_info(c, "starting metadata mark and sweep");
 868                 err = "error in mark and sweep";
 869                 ret = bch2_gc(c, NULL, true, true);
 870                 if (ret)
 871                         goto err;
 872                 bch_verbose(c, "mark and sweep done");
 873         }
 874
 875         if (c->opts.fsck ||
 876             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
 877             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
 878                 bch_info(c, "starting mark and sweep");
 879                 err = "error in mark and sweep";
 880                 ret = bch2_gc(c, &journal_keys, true, false);
 881                 if (ret)
 882                         goto err;
 883                 bch_verbose(c, "mark and sweep done");
 884         }
 885
 886         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 887         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
 888
 889         /*
 890          * Skip past versions that might have possibly been used (as nonces),
 891          * but hadn't had their pointers written:
 892          */
 893         if (c->sb.encryption_type && !c->sb.clean)
 894                 atomic64_add(1 << 16, &c->key_version);
 895
 896         if (c->opts.norecovery)
 897                 goto out;
 898
 899         bch_verbose(c, "starting journal replay");
 900         err = "journal replay failed";
 901         ret = bch2_journal_replay(c, journal_keys);
 902         if (ret)
 903                 goto err;
 904         bch_verbose(c, "journal replay done");
 905
 906         if (!c->opts.nochanges) {
 907                 /*
 908                  * note that even when filesystem was clean there might be work
 909                  * to do here, if we ran gc (because of fsck) which recalculated
 910                  * oldest_gen:
 911                  */
 912                 bch_verbose(c, "writing allocation info");
 913                 err = "error writing out alloc info";
 914                 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW, &wrote) ?:
 915                         bch2_alloc_write(c, BTREE_INSERT_LAZY_RW, &wrote);
 916                 if (ret) {
 917                         bch_err(c, "error writing alloc info");
 918                         goto err;
 919                 }
 920                 bch_verbose(c, "alloc write done");
 921
 922                 set_bit(BCH_FS_ALLOC_WRITTEN, &c->flags);
 923         }
 924
 925         if (!c->sb.clean) {
 926                 if (!(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
 927                         bch_info(c, "checking inode link counts");
 928                         err = "error in recovery";
 929                         ret = bch2_fsck_inode_nlink(c);
 930                         if (ret)
 931                                 goto err;
 932                         bch_verbose(c, "check inodes done");
 933
 934                 } else {
 935                         bch_verbose(c, "checking for deleted inodes");
 936                         err = "error in recovery";
 937                         ret = bch2_fsck_walk_inodes_only(c);
 938                         if (ret)
 939                                 goto err;
 940                         bch_verbose(c, "check inodes done");
 941                 }
 942         }
 943
 944         if (c->opts.fsck) {
 945                 bch_info(c, "starting fsck");
 946                 err = "error in fsck";
 947                 ret = bch2_fsck_full(c);
 948                 if (ret)
 949                         goto err;
 950                 bch_verbose(c, "fsck done");
 951         }
 952
 953         if (enabled_qtypes(c)) {
 954                 bch_verbose(c, "reading quotas");
 955                 ret = bch2_fs_quota_read(c);
 956                 if (ret)
 957                         goto err;
 958                 bch_verbose(c, "quotas done");
 959         }
 960
 961         mutex_lock(&c->sb_lock);
 962         if (c->opts.version_upgrade) {
 963                 if (c->sb.version < bcachefs_metadata_version_new_versioning)
 964                         c->disk_sb.sb->version_min =
 965                                 le16_to_cpu(bcachefs_metadata_version_min);
 966                 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
 967                 c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
 968                 write_sb = true;
 969         }
 970
 971         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
 972                 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
 973                 write_sb = true;
 974         }
 975
 976         if (c->opts.fsck &&
 977             !test_bit(BCH_FS_ERROR, &c->flags)) {
 978                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
 979                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
 980                 write_sb = true;
 981         }
 982
 983         if (write_sb)
 984                 bch2_write_super(c);
 985         mutex_unlock(&c->sb_lock);
 986
 987         if (c->journal_seq_blacklist_table &&
 988             c->journal_seq_blacklist_table->nr > 128)
 989                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
 990 out:
 991         ret = 0;
 992 err:
 993 fsck_err:
 994         set_bit(BCH_FS_FSCK_DONE, &c->flags);
 995         bch2_flush_fsck_errs(c);
 996
 997         journal_keys_free(&journal_keys);
 998         journal_entries_free(&journal_entries);
 999         kfree(clean);
1000         if (ret)
1001                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1002         else
1003                 bch_verbose(c, "ret %i", ret);
1004         return ret;
1005 }
1006
1007 int bch2_fs_initialize(struct bch_fs *c)
1008 {
1009         struct bch_inode_unpacked root_inode, lostfound_inode;
1010         struct bkey_inode_buf packed_inode;
1011         struct qstr lostfound = QSTR("lost+found");
1012         const char *err = "cannot allocate memory";
1013         struct bch_dev *ca;
1014         LIST_HEAD(journal);
1015         unsigned i;
1016         int ret;
1017
1018         bch_notice(c, "initializing new filesystem");
1019
1020         mutex_lock(&c->sb_lock);
1021         for_each_online_member(ca, c, i)
1022                 bch2_mark_dev_superblock(c, ca, 0);
1023         mutex_unlock(&c->sb_lock);
1024
1025         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1026         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1027
1028         for (i = 0; i < BTREE_ID_NR; i++)
1029                 bch2_btree_root_alloc(c, i);
1030
1031         err = "unable to allocate journal buckets";
1032         for_each_online_member(ca, c, i) {
1033                 ret = bch2_dev_journal_alloc(ca);
1034                 if (ret) {
1035                         percpu_ref_put(&ca->io_ref);
1036                         goto err;
1037                 }
1038         }
1039
1040         /*
1041          * journal_res_get() will crash if called before this has
1042          * set up the journal.pin FIFO and journal.cur pointer:
1043          */
1044         bch2_fs_journal_start(&c->journal, 1, &journal);
1045         bch2_journal_set_replay_done(&c->journal);
1046
1047         bch2_inode_init(c, &root_inode, 0, 0,
1048                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1049         root_inode.bi_inum = BCACHEFS_ROOT_INO;
1050         bch2_inode_pack(&packed_inode, &root_inode);
1051
1052         err = "error creating root directory";
1053         ret = bch2_btree_insert(c, BTREE_ID_INODES,
1054                                 &packed_inode.inode.k_i,
1055                                 NULL, NULL, BTREE_INSERT_LAZY_RW);
1056         if (ret)
1057                 goto err;
1058
1059         bch2_inode_init_early(c, &lostfound_inode);
1060
1061         err = "error creating lost+found";
1062         ret = bch2_trans_do(c, NULL, NULL, 0,
1063                 bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
1064                                   &root_inode, &lostfound_inode,
1065                                   &lostfound,
1066                                   0, 0, S_IFDIR|0700, 0,
1067                                   NULL, NULL));
1068         if (ret)
1069                 goto err;
1070
1071         if (enabled_qtypes(c)) {
1072                 ret = bch2_fs_quota_read(c);
1073                 if (ret)
1074                         goto err;
1075         }
1076
1077         err = "error writing first journal entry";
1078         ret = bch2_journal_meta(&c->journal);
1079         if (ret)
1080                 goto err;
1081
1082         mutex_lock(&c->sb_lock);
1083         c->disk_sb.sb->version = c->disk_sb.sb->version_min =
1084                 le16_to_cpu(bcachefs_metadata_version_current);
1085         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1086         c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1087
1088         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1089         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1090
1091         bch2_write_super(c);
1092         mutex_unlock(&c->sb_lock);
1093
1094         return 0;
1095 err:
1096         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1097         return ret;
1098 }