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

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "bkey_buf.h"
   5 #include "btree_update.h"
   6 #include "dirent.h"
   7 #include "error.h"
   8 #include "fs-common.h"
   9 #include "fsck.h"
  10 #include "inode.h"
  11 #include "keylist.h"
  12 #include "super.h"
  13 #include "xattr.h"
  14
  15 #include <linux/bsearch.h>
  16 #include <linux/dcache.h> /* struct qstr */
  17
  18 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  19
  20 static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum)
  21 {
  22         struct btree_iter *iter;
  23         struct bkey_s_c k;
  24         u64 sectors = 0;
  25         int ret;
  26
  27         for_each_btree_key(trans, iter, BTREE_ID_extents,
  28                            POS(inum, 0), 0, k, ret) {
  29                 if (k.k->p.inode != inum)
  30                         break;
  31
  32                 if (bkey_extent_is_allocation(k.k))
  33                         sectors += k.k->size;
  34         }
  35
  36         bch2_trans_iter_free(trans, iter);
  37
  38         return ret ?: sectors;
  39 }
  40
  41 static int __lookup_inode(struct btree_trans *trans, u64 inode_nr,
  42                           struct bch_inode_unpacked *inode,
  43                           u32 *snapshot)
  44 {
  45         struct btree_iter *iter;
  46         struct bkey_s_c k;
  47         int ret;
  48
  49         iter = bch2_trans_get_iter(trans, BTREE_ID_inodes,
  50                         POS(0, inode_nr), 0);
  51         k = bch2_btree_iter_peek_slot(iter);
  52         ret = bkey_err(k);
  53         if (ret)
  54                 goto err;
  55
  56         if (snapshot)
  57                 *snapshot = iter->pos.snapshot;
  58         ret = k.k->type == KEY_TYPE_inode
  59                 ? bch2_inode_unpack(bkey_s_c_to_inode(k), inode)
  60                 : -ENOENT;
  61 err:
  62         bch2_trans_iter_free(trans, iter);
  63         return ret;
  64 }
  65
  66 static int lookup_inode(struct btree_trans *trans, u64 inode_nr,
  67                         struct bch_inode_unpacked *inode,
  68                         u32 *snapshot)
  69 {
  70         return lockrestart_do(trans, __lookup_inode(trans, inode_nr, inode, snapshot));
  71 }
  72
  73 static int __write_inode(struct btree_trans *trans,
  74                          struct bch_inode_unpacked *inode,
  75                          u32 snapshot)
  76 {
  77         struct btree_iter *inode_iter =
  78                 bch2_trans_get_iter(trans, BTREE_ID_inodes,
  79                                     SPOS(0, inode->bi_inum, snapshot),
  80                                     BTREE_ITER_INTENT);
  81         int ret = bch2_inode_write(trans, inode_iter, inode);
  82         bch2_trans_iter_put(trans, inode_iter);
  83         return ret;
  84 }
  85
  86 static int write_inode(struct btree_trans *trans,
  87                        struct bch_inode_unpacked *inode,
  88                        u32 snapshot)
  89 {
  90         int ret = __bch2_trans_do(trans, NULL, NULL,
  91                                   BTREE_INSERT_NOFAIL|
  92                                   BTREE_INSERT_LAZY_RW,
  93                                   __write_inode(trans, inode, snapshot));
  94         if (ret)
  95                 bch_err(trans->c, "error in fsck: error %i updating inode", ret);
  96         return ret;
  97 }
  98
  99 static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
 100 {
 101         struct bch_fs *c = trans->c;
 102         struct btree_iter *iter;
 103         struct bch_inode_unpacked dir_inode;
 104         struct bch_hash_info dir_hash_info;
 105         int ret;
 106
 107         ret = lookup_inode(trans, pos.inode, &dir_inode, NULL);
 108         if (ret)
 109                 return ret;
 110
 111         dir_hash_info = bch2_hash_info_init(c, &dir_inode);
 112
 113         iter = bch2_trans_get_iter(trans, BTREE_ID_dirents, pos, BTREE_ITER_INTENT);
 114
 115         ret = bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
 116                                   &dir_hash_info, iter);
 117         bch2_trans_iter_put(trans, iter);
 118         return ret;
 119 }
 120
 121 static int remove_dirent(struct btree_trans *trans, struct bpos pos)
 122 {
 123         int ret = __bch2_trans_do(trans, NULL, NULL,
 124                                   BTREE_INSERT_NOFAIL|
 125                                   BTREE_INSERT_LAZY_RW,
 126                                   __remove_dirent(trans, pos));
 127         if (ret)
 128                 bch_err(trans->c, "remove_dirent: err %i deleting dirent", ret);
 129         return ret;
 130 }
 131
 132 /* Get lost+found, create if it doesn't exist: */
 133 static int lookup_lostfound(struct btree_trans *trans,
 134                             struct bch_inode_unpacked *lostfound)
 135 {
 136         struct bch_fs *c = trans->c;
 137         struct bch_inode_unpacked root;
 138         struct bch_hash_info root_hash_info;
 139         struct qstr lostfound_str = QSTR("lost+found");
 140         u64 inum;
 141         u32 snapshot;
 142         int ret;
 143
 144         ret = lookup_inode(trans, BCACHEFS_ROOT_INO, &root, &snapshot);
 145         if (ret && ret != -ENOENT)
 146                 return ret;
 147
 148         root_hash_info = bch2_hash_info_init(c, &root);
 149         inum = bch2_dirent_lookup(c, BCACHEFS_ROOT_INO, &root_hash_info,
 150                                   &lostfound_str);
 151         if (!inum) {
 152                 bch_notice(c, "creating lost+found");
 153                 goto create_lostfound;
 154         }
 155
 156         ret = lookup_inode(trans, inum, lostfound, &snapshot);
 157         if (ret && ret != -ENOENT) {
 158                 /*
 159                  * The check_dirents pass has already run, dangling dirents
 160                  * shouldn't exist here:
 161                  */
 162                 bch_err(c, "error looking up lost+found: %i", ret);
 163                 return ret;
 164         }
 165
 166         if (ret == -ENOENT) {
 167 create_lostfound:
 168                 bch2_inode_init_early(c, lostfound);
 169
 170                 ret = __bch2_trans_do(trans, NULL, NULL,
 171                                       BTREE_INSERT_NOFAIL|
 172                                       BTREE_INSERT_LAZY_RW,
 173                         bch2_create_trans(trans,
 174                                           BCACHEFS_ROOT_INO, &root,
 175                                           lostfound,
 176                                           &lostfound_str,
 177                                           0, 0, S_IFDIR|0700, 0, NULL, NULL));
 178                 if (ret)
 179                         bch_err(c, "error creating lost+found: %i", ret);
 180         }
 181
 182         return 0;
 183 }
 184
 185 static int reattach_inode(struct btree_trans *trans,
 186                           struct bch_inode_unpacked *inode)
 187 {
 188         struct bch_hash_info dir_hash;
 189         struct bch_inode_unpacked lostfound;
 190         char name_buf[20];
 191         struct qstr name;
 192         u64 dir_offset = 0;
 193         int ret;
 194
 195         ret = lookup_lostfound(trans, &lostfound);
 196         if (ret)
 197                 return ret;
 198
 199         if (S_ISDIR(inode->bi_mode)) {
 200                 lostfound.bi_nlink++;
 201
 202                 ret = write_inode(trans, &lostfound, U32_MAX);
 203                 if (ret)
 204                         return ret;
 205         }
 206
 207         dir_hash = bch2_hash_info_init(trans->c, &lostfound);
 208
 209         snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
 210         name = (struct qstr) QSTR(name_buf);
 211
 212         ret = __bch2_trans_do(trans, NULL, NULL, BTREE_INSERT_LAZY_RW,
 213                 bch2_dirent_create(trans, lostfound.bi_inum, &dir_hash,
 214                                    mode_to_type(inode->bi_mode),
 215                                    &name, inode->bi_inum, &dir_offset,
 216                                    BCH_HASH_SET_MUST_CREATE));
 217         if (ret) {
 218                 bch_err(trans->c, "error %i reattaching inode %llu",
 219                         ret, inode->bi_inum);
 220                 return ret;
 221         }
 222
 223         inode->bi_dir           = lostfound.bi_inum;
 224         inode->bi_dir_offset    = dir_offset;
 225
 226         return write_inode(trans, inode, U32_MAX);
 227 }
 228
 229 static int remove_backpointer(struct btree_trans *trans,
 230                               struct bch_inode_unpacked *inode)
 231 {
 232         struct btree_iter *iter;
 233         struct bkey_s_c k;
 234         int ret;
 235
 236         iter = bch2_trans_get_iter(trans, BTREE_ID_dirents,
 237                                    POS(inode->bi_dir, inode->bi_dir_offset), 0);
 238         k = bch2_btree_iter_peek_slot(iter);
 239         ret = bkey_err(k);
 240         if (ret)
 241                 goto out;
 242         if (k.k->type != KEY_TYPE_dirent) {
 243                 ret = -ENOENT;
 244                 goto out;
 245         }
 246
 247         ret = remove_dirent(trans, k.k->p);
 248 out:
 249         bch2_trans_iter_put(trans, iter);
 250         return ret;
 251 }
 252
 253 struct inode_walker {
 254         bool                    first_this_inode;
 255         bool                    have_inode;
 256         u64                     cur_inum;
 257         u32                     snapshot;
 258         struct bch_inode_unpacked inode;
 259 };
 260
 261 static struct inode_walker inode_walker_init(void)
 262 {
 263         return (struct inode_walker) {
 264                 .cur_inum       = -1,
 265                 .have_inode     = false,
 266         };
 267 }
 268
 269 static int walk_inode(struct btree_trans *trans,
 270                       struct inode_walker *w, u64 inum)
 271 {
 272         if (inum != w->cur_inum) {
 273                 int ret = lookup_inode(trans, inum, &w->inode, &w->snapshot);
 274
 275                 if (ret && ret != -ENOENT)
 276                         return ret;
 277
 278                 w->have_inode   = !ret;
 279                 w->cur_inum     = inum;
 280                 w->first_this_inode = true;
 281         } else {
 282                 w->first_this_inode = false;
 283         }
 284
 285         return 0;
 286 }
 287
 288 static int hash_redo_key(struct btree_trans *trans,
 289                          const struct bch_hash_desc desc,
 290                          struct bch_hash_info *hash_info,
 291                          struct btree_iter *k_iter, struct bkey_s_c k)
 292 {
 293         struct bkey_i *delete;
 294         struct bkey_i *tmp;
 295
 296         delete = bch2_trans_kmalloc(trans, sizeof(*delete));
 297         if (IS_ERR(delete))
 298                 return PTR_ERR(delete);
 299
 300         tmp = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
 301         if (IS_ERR(tmp))
 302                 return PTR_ERR(tmp);
 303
 304         bkey_reassemble(tmp, k);
 305
 306         bkey_init(&delete->k);
 307         delete->k.p = k_iter->pos;
 308         bch2_trans_update(trans, k_iter, delete, 0);
 309
 310         return bch2_hash_set(trans, desc, hash_info, k_iter->pos.inode, tmp, 0);
 311 }
 312
 313 static int fsck_hash_delete_at(struct btree_trans *trans,
 314                                const struct bch_hash_desc desc,
 315                                struct bch_hash_info *info,
 316                                struct btree_iter *iter)
 317 {
 318         int ret;
 319 retry:
 320         ret   = bch2_hash_delete_at(trans, desc, info, iter) ?:
 321                 bch2_trans_commit(trans, NULL, NULL,
 322                                   BTREE_INSERT_NOFAIL|
 323                                   BTREE_INSERT_LAZY_RW);
 324         if (ret == -EINTR) {
 325                 ret = bch2_btree_iter_traverse(iter);
 326                 if (!ret)
 327                         goto retry;
 328         }
 329
 330         return ret;
 331 }
 332
 333 static int hash_check_key(struct btree_trans *trans,
 334                           const struct bch_hash_desc desc,
 335                           struct bch_hash_info *hash_info,
 336                           struct btree_iter *k_iter, struct bkey_s_c hash_k)
 337 {
 338         struct bch_fs *c = trans->c;
 339         struct btree_iter *iter = NULL;
 340         char buf[200];
 341         struct bkey_s_c k;
 342         u64 hash;
 343         int ret = 0;
 344
 345         if (hash_k.k->type != desc.key_type)
 346                 return 0;
 347
 348         hash = desc.hash_bkey(hash_info, hash_k);
 349
 350         if (likely(hash == hash_k.k->p.offset))
 351                 return 0;
 352
 353         if (hash_k.k->p.offset < hash)
 354                 goto bad_hash;
 355
 356         for_each_btree_key(trans, iter, desc.btree_id, POS(hash_k.k->p.inode, hash),
 357                            BTREE_ITER_SLOTS, k, ret) {
 358                 if (!bkey_cmp(k.k->p, hash_k.k->p))
 359                         break;
 360
 361                 if (fsck_err_on(k.k->type == desc.key_type &&
 362                                 !desc.cmp_bkey(k, hash_k), c,
 363                                 "duplicate hash table keys:\n%s",
 364                                 (bch2_bkey_val_to_text(&PBUF(buf), c,
 365                                                        hash_k), buf))) {
 366                         ret = fsck_hash_delete_at(trans, desc, hash_info, k_iter);
 367                         if (ret)
 368                                 return ret;
 369                         ret = 1;
 370                         break;
 371                 }
 372
 373                 if (bkey_deleted(k.k)) {
 374                         bch2_trans_iter_free(trans, iter);
 375                         goto bad_hash;
 376                 }
 377
 378         }
 379         bch2_trans_iter_free(trans, iter);
 380         return ret;
 381 bad_hash:
 382         if (fsck_err(c, "hash table key at wrong offset: btree %u inode %llu offset %llu, "
 383                      "hashed to %llu\n%s",
 384                      desc.btree_id, hash_k.k->p.inode, hash_k.k->p.offset, hash,
 385                      (bch2_bkey_val_to_text(&PBUF(buf), c, hash_k), buf)) == FSCK_ERR_IGNORE)
 386                 return 0;
 387
 388         ret = __bch2_trans_do(trans, NULL, NULL,
 389                               BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
 390                 hash_redo_key(trans, desc, hash_info, k_iter, hash_k));
 391         if (ret) {
 392                 bch_err(c, "hash_redo_key err %i", ret);
 393                 return ret;
 394         }
 395         return -EINTR;
 396 fsck_err:
 397         return ret;
 398 }
 399
 400 static int check_inode(struct btree_trans *trans,
 401                        struct btree_iter *iter,
 402                        struct bkey_s_c_inode inode)
 403 {
 404         struct bch_fs *c = trans->c;
 405         struct bch_inode_unpacked u;
 406         bool do_update = false;
 407         int ret = 0;
 408
 409         ret = bch2_inode_unpack(inode, &u);
 410
 411         if (bch2_fs_inconsistent_on(ret, c,
 412                          "error unpacking inode %llu in fsck",
 413                          inode.k->p.inode))
 414                 return ret;
 415
 416         if (u.bi_flags & BCH_INODE_UNLINKED &&
 417             (!c->sb.clean ||
 418              fsck_err(c, "filesystem marked clean, but inode %llu unlinked",
 419                       u.bi_inum))) {
 420                 bch_verbose(c, "deleting inode %llu", u.bi_inum);
 421
 422                 bch2_trans_unlock(trans);
 423                 bch2_fs_lazy_rw(c);
 424
 425                 ret = bch2_inode_rm(c, u.bi_inum, false);
 426                 if (ret)
 427                         bch_err(c, "error in fsck: error %i while deleting inode", ret);
 428                 return ret;
 429         }
 430
 431         if (u.bi_flags & BCH_INODE_I_SIZE_DIRTY &&
 432             (!c->sb.clean ||
 433              fsck_err(c, "filesystem marked clean, but inode %llu has i_size dirty",
 434                       u.bi_inum))) {
 435                 bch_verbose(c, "truncating inode %llu", u.bi_inum);
 436
 437                 bch2_trans_unlock(trans);
 438                 bch2_fs_lazy_rw(c);
 439
 440                 /*
 441                  * XXX: need to truncate partial blocks too here - or ideally
 442                  * just switch units to bytes and that issue goes away
 443                  */
 444                 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
 445                                 POS(u.bi_inum, round_up(u.bi_size, block_bytes(c))),
 446                                 POS(u.bi_inum, U64_MAX),
 447                                 NULL);
 448                 if (ret) {
 449                         bch_err(c, "error in fsck: error %i truncating inode", ret);
 450                         return ret;
 451                 }
 452
 453                 /*
 454                  * We truncated without our normal sector accounting hook, just
 455                  * make sure we recalculate it:
 456                  */
 457                 u.bi_flags |= BCH_INODE_I_SECTORS_DIRTY;
 458
 459                 u.bi_flags &= ~BCH_INODE_I_SIZE_DIRTY;
 460                 do_update = true;
 461         }
 462
 463         if (u.bi_flags & BCH_INODE_I_SECTORS_DIRTY &&
 464             (!c->sb.clean ||
 465              fsck_err(c, "filesystem marked clean, but inode %llu has i_sectors dirty",
 466                       u.bi_inum))) {
 467                 s64 sectors;
 468
 469                 bch_verbose(c, "recounting sectors for inode %llu",
 470                             u.bi_inum);
 471
 472                 sectors = bch2_count_inode_sectors(trans, u.bi_inum);
 473                 if (sectors < 0) {
 474                         bch_err(c, "error in fsck: error %i recounting inode sectors",
 475                                 (int) sectors);
 476                         return sectors;
 477                 }
 478
 479                 u.bi_sectors = sectors;
 480                 u.bi_flags &= ~BCH_INODE_I_SECTORS_DIRTY;
 481                 do_update = true;
 482         }
 483
 484         if (u.bi_flags & BCH_INODE_BACKPTR_UNTRUSTED) {
 485                 u.bi_dir = 0;
 486                 u.bi_dir_offset = 0;
 487                 u.bi_flags &= ~BCH_INODE_BACKPTR_UNTRUSTED;
 488                 do_update = true;
 489         }
 490
 491         if (do_update) {
 492                 ret = __bch2_trans_do(trans, NULL, NULL,
 493                                       BTREE_INSERT_NOFAIL|
 494                                       BTREE_INSERT_LAZY_RW,
 495                                 bch2_inode_write(trans, iter, &u));
 496                 if (ret)
 497                         bch_err(c, "error in fsck: error %i "
 498                                 "updating inode", ret);
 499         }
 500 fsck_err:
 501         return ret;
 502 }
 503
 504 noinline_for_stack
 505 static int check_inodes(struct bch_fs *c, bool full)
 506 {
 507         struct btree_trans trans;
 508         struct btree_iter *iter;
 509         struct bkey_s_c k;
 510         struct bkey_s_c_inode inode;
 511         int ret;
 512
 513         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 514
 515         for_each_btree_key(&trans, iter, BTREE_ID_inodes, POS_MIN, 0, k, ret) {
 516                 if (k.k->type != KEY_TYPE_inode)
 517                         continue;
 518
 519                 inode = bkey_s_c_to_inode(k);
 520
 521                 if (full ||
 522                     (inode.v->bi_flags & (BCH_INODE_I_SIZE_DIRTY|
 523                                           BCH_INODE_I_SECTORS_DIRTY|
 524                                           BCH_INODE_UNLINKED))) {
 525                         ret = check_inode(&trans, iter, inode);
 526                         if (ret)
 527                                 break;
 528                 }
 529         }
 530         bch2_trans_iter_put(&trans, iter);
 531
 532         BUG_ON(ret == -EINTR);
 533
 534         return bch2_trans_exit(&trans) ?: ret;
 535 }
 536
 537 static int fix_overlapping_extent(struct btree_trans *trans,
 538                                        struct bkey_s_c k, struct bpos cut_at)
 539 {
 540         struct btree_iter *iter;
 541         struct bkey_i *u;
 542         int ret;
 543
 544         u = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
 545         ret = PTR_ERR_OR_ZERO(u);
 546         if (ret)
 547                 return ret;
 548
 549         bkey_reassemble(u, k);
 550         bch2_cut_front(cut_at, u);
 551
 552
 553         /*
 554          * We don't want to go through the extent_handle_overwrites path:
 555          *
 556          * XXX: this is going to screw up disk accounting, extent triggers
 557          * assume things about extent overwrites - we should be running the
 558          * triggers manually here
 559          */
 560         iter = bch2_trans_get_iter(trans, BTREE_ID_extents, u->k.p,
 561                                    BTREE_ITER_INTENT|BTREE_ITER_NOT_EXTENTS);
 562
 563         BUG_ON(iter->flags & BTREE_ITER_IS_EXTENTS);
 564         bch2_trans_update(trans, iter, u, BTREE_TRIGGER_NORUN);
 565         bch2_trans_iter_put(trans, iter);
 566
 567         return bch2_trans_commit(trans, NULL, NULL,
 568                                  BTREE_INSERT_NOFAIL|
 569                                  BTREE_INSERT_LAZY_RW);
 570 }
 571
 572 static int inode_backpointer_exists(struct btree_trans *trans,
 573                                     struct bch_inode_unpacked *inode)
 574 {
 575         struct btree_iter *iter;
 576         struct bkey_s_c k;
 577         int ret;
 578
 579         iter = bch2_trans_get_iter(trans, BTREE_ID_dirents,
 580                                    POS(inode->bi_dir, inode->bi_dir_offset), 0);
 581         k = bch2_btree_iter_peek_slot(iter);
 582         ret = bkey_err(k);
 583         if (ret)
 584                 goto out;
 585         if (k.k->type != KEY_TYPE_dirent)
 586                 goto out;
 587
 588         ret = le64_to_cpu(bkey_s_c_to_dirent(k).v->d_inum) == inode->bi_inum;
 589 out:
 590         bch2_trans_iter_free(trans, iter);
 591         return ret;
 592 }
 593
 594 static bool inode_backpointer_matches(struct bkey_s_c_dirent d,
 595                                       struct bch_inode_unpacked *inode)
 596 {
 597         return d.k->p.inode == inode->bi_dir &&
 598                 d.k->p.offset == inode->bi_dir_offset;
 599 }
 600
 601 /*
 602  * Walk extents: verify that extents have a corresponding S_ISREG inode, and
 603  * that i_size an i_sectors are consistent
 604  */
 605 noinline_for_stack
 606 static int check_extents(struct bch_fs *c)
 607 {
 608         struct inode_walker w = inode_walker_init();
 609         struct btree_trans trans;
 610         struct btree_iter *iter;
 611         struct bkey_s_c k;
 612         struct bkey_buf prev;
 613         u64 i_sectors = 0;
 614         int ret = 0;
 615
 616         bch2_bkey_buf_init(&prev);
 617         prev.k->k = KEY(0, 0, 0);
 618         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 619
 620         bch_verbose(c, "checking extents");
 621
 622         iter = bch2_trans_get_iter(&trans, BTREE_ID_extents,
 623                                    POS(BCACHEFS_ROOT_INO, 0),
 624                                    BTREE_ITER_INTENT);
 625 retry:
 626         while ((k = bch2_btree_iter_peek(iter)).k &&
 627                !(ret = bkey_err(k))) {
 628                 if (w.have_inode &&
 629                     w.cur_inum != k.k->p.inode &&
 630                     !(w.inode.bi_flags & BCH_INODE_I_SECTORS_DIRTY) &&
 631                     fsck_err_on(w.inode.bi_sectors != i_sectors, c,
 632                                 "inode %llu has incorrect i_sectors: got %llu, should be %llu",
 633                                 w.inode.bi_inum,
 634                                 w.inode.bi_sectors, i_sectors)) {
 635                         w.inode.bi_sectors = i_sectors;
 636
 637                         ret = write_inode(&trans, &w.inode, w.snapshot);
 638                         if (ret)
 639                                 break;
 640                 }
 641
 642                 if (bkey_cmp(prev.k->k.p, bkey_start_pos(k.k)) > 0) {
 643                         char buf1[200];
 644                         char buf2[200];
 645
 646                         bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(prev.k));
 647                         bch2_bkey_val_to_text(&PBUF(buf2), c, k);
 648
 649                         if (fsck_err(c, "overlapping extents:\n%s\n%s", buf1, buf2))
 650                                 return fix_overlapping_extent(&trans, k, prev.k->k.p) ?: -EINTR;
 651                 }
 652
 653                 ret = walk_inode(&trans, &w, k.k->p.inode);
 654                 if (ret)
 655                         break;
 656
 657                 if (w.first_this_inode)
 658                         i_sectors = 0;
 659
 660                 if (fsck_err_on(!w.have_inode, c,
 661                                 "extent type %u for missing inode %llu",
 662                                 k.k->type, k.k->p.inode) ||
 663                     fsck_err_on(w.have_inode &&
 664                                 !S_ISREG(w.inode.bi_mode) && !S_ISLNK(w.inode.bi_mode), c,
 665                                 "extent type %u for non regular file, inode %llu mode %o",
 666                                 k.k->type, k.k->p.inode, w.inode.bi_mode)) {
 667                         bch2_fs_lazy_rw(c);
 668                         return bch2_btree_delete_range_trans(&trans, BTREE_ID_extents,
 669                                                        POS(k.k->p.inode, 0),
 670                                                        POS(k.k->p.inode, U64_MAX),
 671                                                        NULL) ?: -EINTR;
 672                 }
 673
 674                 if (fsck_err_on(w.have_inode &&
 675                                 !(w.inode.bi_flags & BCH_INODE_I_SIZE_DIRTY) &&
 676                                 k.k->type != KEY_TYPE_reservation &&
 677                                 k.k->p.offset > round_up(w.inode.bi_size, block_bytes(c)) >> 9, c,
 678                                 "extent type %u offset %llu past end of inode %llu, i_size %llu",
 679                                 k.k->type, k.k->p.offset, k.k->p.inode, w.inode.bi_size)) {
 680                         bch2_fs_lazy_rw(c);
 681                         return bch2_btree_delete_range_trans(&trans, BTREE_ID_extents,
 682                                         POS(k.k->p.inode, round_up(w.inode.bi_size, block_bytes(c))),
 683                                         POS(k.k->p.inode, U64_MAX),
 684                                         NULL) ?: -EINTR;
 685                 }
 686
 687                 if (bkey_extent_is_allocation(k.k))
 688                         i_sectors += k.k->size;
 689                 bch2_bkey_buf_reassemble(&prev, c, k);
 690
 691                 bch2_btree_iter_advance(iter);
 692         }
 693 fsck_err:
 694         if (ret == -EINTR)
 695                 goto retry;
 696         bch2_trans_iter_put(&trans, iter);
 697         bch2_bkey_buf_exit(&prev, c);
 698         return bch2_trans_exit(&trans) ?: ret;
 699 }
 700
 701 /*
 702  * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
 703  * validate d_type
 704  */
 705 noinline_for_stack
 706 static int check_dirents(struct bch_fs *c)
 707 {
 708         struct inode_walker w = inode_walker_init();
 709         struct bch_hash_info hash_info;
 710         struct btree_trans trans;
 711         struct btree_iter *iter;
 712         struct bkey_s_c k;
 713         char buf[200];
 714         unsigned nr_subdirs = 0;
 715         int ret = 0;
 716
 717         bch_verbose(c, "checking dirents");
 718
 719         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 720
 721         iter = bch2_trans_get_iter(&trans, BTREE_ID_dirents,
 722                                    POS(BCACHEFS_ROOT_INO, 0), 0);
 723 retry:
 724         while ((k = bch2_btree_iter_peek(iter)).k &&
 725                !(ret = bkey_err(k))) {
 726                 struct bkey_s_c_dirent d;
 727                 struct bch_inode_unpacked target;
 728                 u32 target_snapshot;
 729                 bool have_target;
 730                 bool backpointer_exists = true;
 731                 u64 d_inum;
 732
 733                 if (w.have_inode &&
 734                     w.cur_inum != k.k->p.inode &&
 735                     fsck_err_on(w.inode.bi_nlink != nr_subdirs, c,
 736                                 "directory %llu with wrong i_nlink: got %u, should be %u",
 737                                 w.inode.bi_inum, w.inode.bi_nlink, nr_subdirs)) {
 738                         w.inode.bi_nlink = nr_subdirs;
 739                         ret = write_inode(&trans, &w.inode, w.snapshot);
 740                         if (ret)
 741                                 break;
 742                 }
 743
 744                 ret = walk_inode(&trans, &w, k.k->p.inode);
 745                 if (ret)
 746                         break;
 747
 748                 if (w.first_this_inode)
 749                         nr_subdirs = 0;
 750
 751                 if (fsck_err_on(!w.have_inode, c,
 752                                 "dirent in nonexisting directory:\n%s",
 753                                 (bch2_bkey_val_to_text(&PBUF(buf), c,
 754                                                        k), buf)) ||
 755                     fsck_err_on(!S_ISDIR(w.inode.bi_mode), c,
 756                                 "dirent in non directory inode type %u:\n%s",
 757                                 mode_to_type(w.inode.bi_mode),
 758                                 (bch2_bkey_val_to_text(&PBUF(buf), c,
 759                                                        k), buf))) {
 760                         ret = lockrestart_do(&trans,
 761                                         bch2_btree_delete_at(&trans, iter, 0));
 762                         if (ret)
 763                                 goto err;
 764                         goto next;
 765                 }
 766
 767                 if (!w.have_inode)
 768                         goto next;
 769
 770                 if (w.first_this_inode)
 771                         hash_info = bch2_hash_info_init(c, &w.inode);
 772
 773                 ret = hash_check_key(&trans, bch2_dirent_hash_desc,
 774                                      &hash_info, iter, k);
 775                 if (ret > 0) {
 776                         ret = 0;
 777                         goto next;
 778                 }
 779                 if (ret)
 780                         goto fsck_err;
 781
 782                 if (k.k->type != KEY_TYPE_dirent)
 783                         goto next;
 784
 785                 d = bkey_s_c_to_dirent(k);
 786                 d_inum = le64_to_cpu(d.v->d_inum);
 787
 788                 ret = lookup_inode(&trans, d_inum, &target, &target_snapshot);
 789                 if (ret && ret != -ENOENT)
 790                         break;
 791
 792                 have_target = !ret;
 793                 ret = 0;
 794
 795                 if (fsck_err_on(!have_target, c,
 796                                 "dirent points to missing inode:\n%s",
 797                                 (bch2_bkey_val_to_text(&PBUF(buf), c,
 798                                                        k), buf))) {
 799                         ret = remove_dirent(&trans, d.k->p);
 800                         if (ret)
 801                                 goto err;
 802                         goto next;
 803                 }
 804
 805                 if (!have_target)
 806                         goto next;
 807
 808                 if (!target.bi_dir &&
 809                     !target.bi_dir_offset) {
 810                         target.bi_dir           = k.k->p.inode;
 811                         target.bi_dir_offset    = k.k->p.offset;
 812
 813                         ret = write_inode(&trans, &target, target_snapshot);
 814                         if (ret)
 815                                 goto err;
 816                 }
 817
 818                 if (!inode_backpointer_matches(d, &target)) {
 819                         ret = inode_backpointer_exists(&trans, &target);
 820                         if (ret < 0)
 821                                 goto err;
 822
 823                         backpointer_exists = ret;
 824                         ret = 0;
 825
 826                         if (fsck_err_on(S_ISDIR(target.bi_mode) &&
 827                                         backpointer_exists, c,
 828                                         "directory %llu with multiple links",
 829                                         target.bi_inum)) {
 830                                 ret = remove_dirent(&trans, d.k->p);
 831                                 if (ret)
 832                                         goto err;
 833                                 continue;
 834                         }
 835
 836                         if (fsck_err_on(backpointer_exists &&
 837                                         !target.bi_nlink, c,
 838                                         "inode %llu has multiple links but i_nlink 0",
 839                                         d_inum)) {
 840                                 target.bi_nlink++;
 841                                 target.bi_flags &= ~BCH_INODE_UNLINKED;
 842
 843                                 ret = write_inode(&trans, &target, target_snapshot);
 844                                 if (ret)
 845                                         goto err;
 846                         }
 847
 848                         if (fsck_err_on(!backpointer_exists, c,
 849                                         "inode %llu has wrong backpointer:\n"
 850                                         "got       %llu:%llu\n"
 851                                         "should be %llu:%llu",
 852                                         d_inum,
 853                                         target.bi_dir,
 854                                         target.bi_dir_offset,
 855                                         k.k->p.inode,
 856                                         k.k->p.offset)) {
 857                                 target.bi_dir           = k.k->p.inode;
 858                                 target.bi_dir_offset    = k.k->p.offset;
 859
 860                                 ret = write_inode(&trans, &target, target_snapshot);
 861                                 if (ret)
 862                                         goto err;
 863                         }
 864                 }
 865
 866                 if (fsck_err_on(d.v->d_type != mode_to_type(target.bi_mode), c,
 867                                 "incorrect d_type: should be %u:\n%s",
 868                                 mode_to_type(target.bi_mode),
 869                                 (bch2_bkey_val_to_text(&PBUF(buf), c,
 870                                                        k), buf))) {
 871                         struct bkey_i_dirent *n;
 872
 873                         n = kmalloc(bkey_bytes(d.k), GFP_KERNEL);
 874                         if (!n) {
 875                                 ret = -ENOMEM;
 876                                 goto err;
 877                         }
 878
 879                         bkey_reassemble(&n->k_i, d.s_c);
 880                         n->v.d_type = mode_to_type(target.bi_mode);
 881
 882                         ret = __bch2_trans_do(&trans, NULL, NULL,
 883                                               BTREE_INSERT_NOFAIL|
 884                                               BTREE_INSERT_LAZY_RW,
 885                                 (bch2_trans_update(&trans, iter, &n->k_i, 0), 0));
 886                         kfree(n);
 887                         if (ret)
 888                                 goto err;
 889
 890                 }
 891
 892                 nr_subdirs += d.v->d_type == DT_DIR;
 893 next:
 894                 bch2_btree_iter_advance(iter);
 895         }
 896 err:
 897 fsck_err:
 898         if (ret == -EINTR)
 899                 goto retry;
 900
 901         bch2_trans_iter_put(&trans, iter);
 902         return bch2_trans_exit(&trans) ?: ret;
 903 }
 904
 905 /*
 906  * Walk xattrs: verify that they all have a corresponding inode
 907  */
 908 noinline_for_stack
 909 static int check_xattrs(struct bch_fs *c)
 910 {
 911         struct inode_walker w = inode_walker_init();
 912         struct bch_hash_info hash_info;
 913         struct btree_trans trans;
 914         struct btree_iter *iter;
 915         struct bkey_s_c k;
 916         int ret = 0;
 917
 918         bch_verbose(c, "checking xattrs");
 919
 920         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 921
 922         iter = bch2_trans_get_iter(&trans, BTREE_ID_xattrs,
 923                                    POS(BCACHEFS_ROOT_INO, 0), 0);
 924 retry:
 925         while ((k = bch2_btree_iter_peek(iter)).k &&
 926                !(ret = bkey_err(k))) {
 927                 ret = walk_inode(&trans, &w, k.k->p.inode);
 928                 if (ret)
 929                         break;
 930
 931                 if (fsck_err_on(!w.have_inode, c,
 932                                 "xattr for missing inode %llu",
 933                                 k.k->p.inode)) {
 934                         ret = bch2_btree_delete_at(&trans, iter, 0);
 935                         if (ret)
 936                                 break;
 937                         continue;
 938                 }
 939
 940                 if (w.first_this_inode && w.have_inode)
 941                         hash_info = bch2_hash_info_init(c, &w.inode);
 942
 943                 ret = hash_check_key(&trans, bch2_xattr_hash_desc,
 944                                      &hash_info, iter, k);
 945                 if (ret)
 946                         break;
 947
 948                 bch2_btree_iter_advance(iter);
 949         }
 950 fsck_err:
 951         if (ret == -EINTR)
 952                 goto retry;
 953
 954         bch2_trans_iter_put(&trans, iter);
 955         return bch2_trans_exit(&trans) ?: ret;
 956 }
 957
 958 /* Get root directory, create if it doesn't exist: */
 959 static int check_root(struct bch_fs *c, struct bch_inode_unpacked *root_inode)
 960 {
 961         struct bkey_inode_buf packed;
 962         u32 snapshot;
 963         int ret;
 964
 965         bch_verbose(c, "checking root directory");
 966
 967         ret = bch2_trans_do(c, NULL, NULL, 0,
 968                 lookup_inode(&trans, BCACHEFS_ROOT_INO, root_inode, &snapshot));
 969         if (ret && ret != -ENOENT)
 970                 return ret;
 971
 972         if (fsck_err_on(ret, c, "root directory missing"))
 973                 goto create_root;
 974
 975         if (fsck_err_on(!S_ISDIR(root_inode->bi_mode), c,
 976                         "root inode not a directory"))
 977                 goto create_root;
 978
 979         return 0;
 980 fsck_err:
 981         return ret;
 982 create_root:
 983         bch2_inode_init(c, root_inode, 0, 0, S_IFDIR|0755,
 984                         0, NULL);
 985         root_inode->bi_inum = BCACHEFS_ROOT_INO;
 986
 987         bch2_inode_pack(c, &packed, root_inode);
 988
 989         return bch2_btree_insert(c, BTREE_ID_inodes, &packed.inode.k_i,
 990                                  NULL, NULL,
 991                                  BTREE_INSERT_NOFAIL|
 992                                  BTREE_INSERT_LAZY_RW);
 993 }
 994
 995 struct pathbuf {
 996         size_t          nr;
 997         size_t          size;
 998
 999         struct pathbuf_entry {
1000                 u64     inum;
1001         }               *entries;
1002 };
1003
1004 static int path_down(struct pathbuf *p, u64 inum)
1005 {
1006         if (p->nr == p->size) {
1007                 size_t new_size = max_t(size_t, 256UL, p->size * 2);
1008                 void *n = krealloc(p->entries,
1009                                    new_size * sizeof(p->entries[0]),
1010                                    GFP_KERNEL);
1011                 if (!n) {
1012                         return -ENOMEM;
1013                 }
1014
1015                 p->entries = n;
1016                 p->size = new_size;
1017         };
1018
1019         p->entries[p->nr++] = (struct pathbuf_entry) {
1020                 .inum = inum,
1021         };
1022         return 0;
1023 }
1024
1025 static int check_path(struct btree_trans *trans,
1026                       struct pathbuf *p,
1027                       struct bch_inode_unpacked *inode)
1028 {
1029         struct bch_fs *c = trans->c;
1030         u32 snapshot;
1031         size_t i;
1032         int ret = 0;
1033
1034         p->nr = 0;
1035
1036         while (inode->bi_inum != BCACHEFS_ROOT_INO) {
1037                 ret = lockrestart_do(trans,
1038                         inode_backpointer_exists(trans, inode));
1039                 if (ret < 0)
1040                         break;
1041
1042                 if (!ret) {
1043                         if (fsck_err(c,  "unreachable inode %llu, type %u nlink %u backptr %llu:%llu",
1044                                      inode->bi_inum,
1045                                      mode_to_type(inode->bi_mode),
1046                                      inode->bi_nlink,
1047                                      inode->bi_dir,
1048                                      inode->bi_dir_offset))
1049                                 ret = reattach_inode(trans, inode);
1050                         break;
1051                 }
1052                 ret = 0;
1053
1054                 if (!S_ISDIR(inode->bi_mode))
1055                         break;
1056
1057                 ret = path_down(p, inode->bi_inum);
1058                 if (ret) {
1059                         bch_err(c, "memory allocation failure");
1060                         return ret;
1061                 }
1062
1063                 for (i = 0; i < p->nr; i++) {
1064                         if (inode->bi_dir != p->entries[i].inum)
1065                                 continue;
1066
1067                         /* XXX print path */
1068                         if (!fsck_err(c, "directory structure loop"))
1069                                 return 0;
1070
1071                         ret = lockrestart_do(trans,
1072                                          remove_backpointer(trans, inode));
1073                         if (ret) {
1074                                 bch_err(c, "error removing dirent: %i", ret);
1075                                 break;
1076                         }
1077
1078                         ret = reattach_inode(trans, inode);
1079                         break;
1080                 }
1081
1082                 ret = lookup_inode(trans, inode->bi_dir, inode, &snapshot);
1083                 if (ret) {
1084                         /* Should have been caught in dirents pass */
1085                         bch_err(c, "error looking up parent directory: %i", ret);
1086                         break;
1087                 }
1088         }
1089 fsck_err:
1090         if (ret)
1091                 bch_err(c, "%s: err %i", __func__, ret);
1092         return ret;
1093 }
1094
1095 /*
1096  * Check for unreachable inodes, as well as loops in the directory structure:
1097  * After check_dirents(), if an inode backpointer doesn't exist that means it's
1098  * unreachable:
1099  */
1100 static int check_directory_structure(struct bch_fs *c)
1101 {
1102         struct btree_trans trans;
1103         struct btree_iter *iter;
1104         struct bkey_s_c k;
1105         struct bch_inode_unpacked u;
1106         struct pathbuf path = { 0, 0, NULL };
1107         int ret;
1108
1109         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1110
1111         for_each_btree_key(&trans, iter, BTREE_ID_inodes, POS_MIN, 0, k, ret) {
1112                 if (k.k->type != KEY_TYPE_inode)
1113                         continue;
1114
1115                 ret = bch2_inode_unpack(bkey_s_c_to_inode(k), &u);
1116                 if (ret) {
1117                         /* Should have been caught earlier in fsck: */
1118                         bch_err(c, "error unpacking inode %llu: %i", k.k->p.offset, ret);
1119                         break;
1120                 }
1121
1122                 ret = check_path(&trans, &path, &u);
1123                 if (ret)
1124                         break;
1125         }
1126         bch2_trans_iter_put(&trans, iter);
1127
1128         BUG_ON(ret == -EINTR);
1129
1130         kfree(path.entries);
1131
1132         return bch2_trans_exit(&trans) ?: ret;
1133 }
1134
1135 struct nlink_table {
1136         size_t          nr;
1137         size_t          size;
1138
1139         struct nlink {
1140                 u64     inum;
1141                 u32     snapshot;
1142                 u32     count;
1143         }               *d;
1144 };
1145
1146 static int add_nlink(struct nlink_table *t, u64 inum, u32 snapshot)
1147 {
1148         if (t->nr == t->size) {
1149                 size_t new_size = max_t(size_t, 128UL, t->size * 2);
1150                 void *d = kvmalloc(new_size * sizeof(t->d[0]), GFP_KERNEL);
1151                 if (!d) {
1152                         return -ENOMEM;
1153                 }
1154
1155                 memcpy(d, t->d, t->size * sizeof(t->d[0]));
1156                 kvfree(t->d);
1157
1158                 t->d = d;
1159                 t->size = new_size;
1160         }
1161
1162
1163         t->d[t->nr++] = (struct nlink) {
1164                 .inum           = inum,
1165                 .snapshot       = snapshot,
1166         };
1167
1168         return 0;
1169 }
1170
1171 static int nlink_cmp(const void *_l, const void *_r)
1172 {
1173         const struct nlink *l = _l;
1174         const struct nlink *r = _r;
1175
1176         return cmp_int(l->inum, r->inum) ?: cmp_int(l->snapshot, r->snapshot);
1177 }
1178
1179 static void inc_link(struct bch_fs *c, struct nlink_table *links,
1180                      u64 range_start, u64 range_end, u64 inum)
1181 {
1182         struct nlink *link, key = {
1183                 .inum = inum, .snapshot = U32_MAX,
1184         };
1185
1186         if (inum < range_start || inum >= range_end)
1187                 return;
1188
1189         link = __inline_bsearch(&key, links->d, links->nr,
1190                                 sizeof(links->d[0]), nlink_cmp);
1191         if (link)
1192                 link->count++;
1193 }
1194
1195 noinline_for_stack
1196 static int check_nlinks_find_hardlinks(struct bch_fs *c,
1197                                        struct nlink_table *t,
1198                                        u64 start, u64 *end)
1199 {
1200         struct btree_trans trans;
1201         struct btree_iter *iter;
1202         struct bkey_s_c k;
1203         struct bkey_s_c_inode inode;
1204         struct bch_inode_unpacked u;
1205         int ret = 0;
1206
1207         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1208
1209         for_each_btree_key(&trans, iter, BTREE_ID_inodes,
1210                            POS(0, start), 0, k, ret) {
1211                 if (k.k->type != KEY_TYPE_inode)
1212                         continue;
1213
1214                 inode = bkey_s_c_to_inode(k);
1215
1216                 /*
1217                  * Backpointer and directory structure checks are sufficient for
1218                  * directories, since they can't have hardlinks:
1219                  */
1220                 if (S_ISDIR(le16_to_cpu(inode.v->bi_mode)))
1221                         continue;
1222
1223                 /* Should never fail, checked by bch2_inode_invalid: */
1224                 BUG_ON(bch2_inode_unpack(inode, &u));
1225
1226                 if (!u.bi_nlink)
1227                         continue;
1228
1229                 ret = add_nlink(t, k.k->p.offset, k.k->p.snapshot);
1230                 if (ret) {
1231                         *end = k.k->p.offset;
1232                         ret = 0;
1233                         break;
1234                 }
1235
1236         }
1237         bch2_trans_iter_put(&trans, iter);
1238         bch2_trans_exit(&trans);
1239
1240         if (ret)
1241                 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
1242
1243         return ret;
1244 }
1245
1246 noinline_for_stack
1247 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
1248                                      u64 range_start, u64 range_end)
1249 {
1250         struct btree_trans trans;
1251         struct btree_iter *iter;
1252         struct bkey_s_c k;
1253         struct bkey_s_c_dirent d;
1254         int ret;
1255
1256         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1257
1258         for_each_btree_key(&trans, iter, BTREE_ID_dirents, POS_MIN, 0, k, ret) {
1259                 switch (k.k->type) {
1260                 case KEY_TYPE_dirent:
1261                         d = bkey_s_c_to_dirent(k);
1262
1263                         if (d.v->d_type != DT_DIR)
1264                                 inc_link(c, links, range_start, range_end,
1265                                          le64_to_cpu(d.v->d_inum));
1266                         break;
1267                 }
1268
1269                 bch2_trans_cond_resched(&trans);
1270         }
1271         bch2_trans_iter_put(&trans, iter);
1272
1273         ret = bch2_trans_exit(&trans) ?: ret;
1274         if (ret)
1275                 bch_err(c, "error in fsck: btree error %i while walking dirents", ret);
1276
1277         return ret;
1278 }
1279
1280 noinline_for_stack
1281 static int check_nlinks_update_hardlinks(struct bch_fs *c,
1282                                struct nlink_table *links,
1283                                u64 range_start, u64 range_end)
1284 {
1285         struct btree_trans trans;
1286         struct btree_iter *iter;
1287         struct bkey_s_c k;
1288         struct bkey_s_c_inode inode;
1289         struct bch_inode_unpacked u;
1290         struct nlink *link = links->d;
1291         int ret = 0;
1292
1293         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1294
1295         for_each_btree_key(&trans, iter, BTREE_ID_inodes,
1296                            POS(0, range_start), 0, k, ret) {
1297                 if (k.k->p.offset >= range_end)
1298                         break;
1299
1300                 if (k.k->type != KEY_TYPE_inode)
1301                         continue;
1302
1303                 inode = bkey_s_c_to_inode(k);
1304                 if (S_ISDIR(le16_to_cpu(inode.v->bi_mode)))
1305                         continue;
1306
1307                 BUG_ON(bch2_inode_unpack(inode, &u));
1308
1309                 if (!u.bi_nlink)
1310                         continue;
1311
1312                 while (link->inum < k.k->p.offset) {
1313                         link++;
1314                         BUG_ON(link >= links->d + links->nr);
1315                 }
1316
1317                 if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count, c,
1318                                 "inode %llu has wrong i_nlink (type %u i_nlink %u, should be %u)",
1319                                 u.bi_inum, mode_to_type(u.bi_mode),
1320                                 bch2_inode_nlink_get(&u), link->count)) {
1321                         bch2_inode_nlink_set(&u, link->count);
1322
1323                         ret = __bch2_trans_do(&trans, NULL, NULL,
1324                                               BTREE_INSERT_NOFAIL|
1325                                               BTREE_INSERT_LAZY_RW,
1326                                         bch2_inode_write(&trans, iter, &u));
1327                         if (ret)
1328                                 bch_err(c, "error in fsck: error %i updating inode", ret);
1329                 }
1330         }
1331 fsck_err:
1332         bch2_trans_iter_put(&trans, iter);
1333         bch2_trans_exit(&trans);
1334
1335         if (ret)
1336                 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
1337
1338         return ret;
1339 }
1340
1341 noinline_for_stack
1342 static int check_nlinks(struct bch_fs *c)
1343 {
1344         struct nlink_table links = { 0 };
1345         u64 this_iter_range_start, next_iter_range_start = 0;
1346         int ret = 0;
1347
1348         bch_verbose(c, "checking inode nlinks");
1349
1350         do {
1351                 this_iter_range_start = next_iter_range_start;
1352                 next_iter_range_start = U64_MAX;
1353
1354                 ret = check_nlinks_find_hardlinks(c, &links,
1355                                                   this_iter_range_start,
1356                                                   &next_iter_range_start);
1357
1358                 ret = check_nlinks_walk_dirents(c, &links,
1359                                           this_iter_range_start,
1360                                           next_iter_range_start);
1361                 if (ret)
1362                         break;
1363
1364                 ret = check_nlinks_update_hardlinks(c, &links,
1365                                          this_iter_range_start,
1366                                          next_iter_range_start);
1367                 if (ret)
1368                         break;
1369
1370                 links.nr = 0;
1371         } while (next_iter_range_start != U64_MAX);
1372
1373         kvfree(links.d);
1374
1375         return ret;
1376 }
1377
1378 /*
1379  * Checks for inconsistencies that shouldn't happen, unless we have a bug.
1380  * Doesn't fix them yet, mainly because they haven't yet been observed:
1381  */
1382 int bch2_fsck_full(struct bch_fs *c)
1383 {
1384         struct bch_inode_unpacked root_inode;
1385
1386         return  check_inodes(c, true) ?:
1387                 check_extents(c) ?:
1388                 check_dirents(c) ?:
1389                 check_xattrs(c) ?:
1390                 check_root(c, &root_inode) ?:
1391                 check_directory_structure(c) ?:
1392                 check_nlinks(c);
1393 }
1394
1395 int bch2_fsck_walk_inodes_only(struct bch_fs *c)
1396 {
1397         return check_inodes(c, false);
1398 }