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 "subvolume.h"
  13 #include "super.h"
  14 #include "xattr.h"
  15
  16 #include <linux/bsearch.h>
  17 #include <linux/dcache.h> /* struct qstr */
  18
  19 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  20
  21 static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum,
  22                                     u32 snapshot)
  23 {
  24         struct btree_iter iter;
  25         struct bkey_s_c k;
  26         u64 sectors = 0;
  27         int ret;
  28
  29         for_each_btree_key(trans, iter, BTREE_ID_extents,
  30                            SPOS(inum, 0, snapshot), 0, k, ret) {
  31                 if (k.k->p.inode != inum)
  32                         break;
  33
  34                 if (bkey_extent_is_allocation(k.k))
  35                         sectors += k.k->size;
  36         }
  37
  38         bch2_trans_iter_exit(trans, &iter);
  39
  40         return ret ?: sectors;
  41 }
  42
  43 static s64 bch2_count_subdirs(struct btree_trans *trans, u64 inum,
  44                                     u32 snapshot)
  45 {
  46         struct btree_iter iter;
  47         struct bkey_s_c k;
  48         struct bkey_s_c_dirent d;
  49         u64 subdirs = 0;
  50         int ret;
  51
  52         for_each_btree_key(trans, iter, BTREE_ID_dirents,
  53                            SPOS(inum, 0, snapshot), 0, k, ret) {
  54                 if (k.k->p.inode != inum)
  55                         break;
  56
  57                 if (k.k->type != KEY_TYPE_dirent)
  58                         continue;
  59
  60                 d = bkey_s_c_to_dirent(k);
  61                 if (d.v->d_type == DT_DIR)
  62                         subdirs++;
  63         }
  64
  65         bch2_trans_iter_exit(trans, &iter);
  66
  67         return ret ?: subdirs;
  68 }
  69
  70 static int __snapshot_lookup_subvol(struct btree_trans *trans, u32 snapshot,
  71                                     u32 *subvol)
  72 {
  73         struct btree_iter iter;
  74         struct bkey_s_c k;
  75         int ret;
  76
  77         bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots,
  78                              POS(0, snapshot), 0);
  79         k = bch2_btree_iter_peek_slot(&iter);
  80         ret = bkey_err(k);
  81         if (ret)
  82                 goto err;
  83
  84         if (k.k->type != KEY_TYPE_snapshot) {
  85                 bch_err(trans->c, "snapshot %u not fonud", snapshot);
  86                 ret = -ENOENT;
  87                 goto err;
  88         }
  89
  90         *subvol = le32_to_cpu(bkey_s_c_to_snapshot(k).v->subvol);
  91 err:
  92         bch2_trans_iter_exit(trans, &iter);
  93         return ret;
  94
  95 }
  96
  97 static int snapshot_lookup_subvol(struct btree_trans *trans, u32 snapshot,
  98                                   u32 *subvol)
  99 {
 100         return lockrestart_do(trans, __snapshot_lookup_subvol(trans, snapshot, subvol));
 101 }
 102
 103 static int __subvol_lookup(struct btree_trans *trans, u32 subvol,
 104                            u32 *snapshot, u64 *inum)
 105 {
 106         struct bch_subvolume s;
 107         int ret;
 108
 109         ret = bch2_subvolume_get(trans, subvol, false, 0, &s);
 110
 111         *snapshot = le32_to_cpu(s.snapshot);
 112         *inum = le64_to_cpu(s.inode);
 113         return ret;
 114 }
 115
 116 static int subvol_lookup(struct btree_trans *trans, u32 subvol,
 117                          u32 *snapshot, u64 *inum)
 118 {
 119         return lockrestart_do(trans, __subvol_lookup(trans, subvol, snapshot, inum));
 120 }
 121
 122 static int __lookup_inode(struct btree_trans *trans, u64 inode_nr,
 123                           struct bch_inode_unpacked *inode,
 124                           u32 *snapshot)
 125 {
 126         struct btree_iter iter;
 127         struct bkey_s_c k;
 128         int ret;
 129
 130         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
 131                              SPOS(0, inode_nr, *snapshot), 0);
 132         k = bch2_btree_iter_peek_slot(&iter);
 133         ret = bkey_err(k);
 134         if (ret)
 135                 goto err;
 136
 137         *snapshot = iter.pos.snapshot;
 138         ret = k.k->type == KEY_TYPE_inode
 139                 ? bch2_inode_unpack(bkey_s_c_to_inode(k), inode)
 140                 : -ENOENT;
 141 err:
 142         bch2_trans_iter_exit(trans, &iter);
 143         return ret;
 144 }
 145
 146 static int lookup_inode(struct btree_trans *trans, u64 inode_nr,
 147                         struct bch_inode_unpacked *inode,
 148                         u32 *snapshot)
 149 {
 150         return lockrestart_do(trans, __lookup_inode(trans, inode_nr, inode, snapshot));
 151 }
 152
 153 static int __lookup_dirent(struct btree_trans *trans,
 154                            struct bch_hash_info hash_info,
 155                            subvol_inum dir, struct qstr *name,
 156                            u64 *target, unsigned *type)
 157 {
 158         struct btree_iter iter;
 159         struct bkey_s_c_dirent d;
 160         int ret;
 161
 162         ret = bch2_hash_lookup(trans, &iter, bch2_dirent_hash_desc,
 163                                &hash_info, dir, name, 0);
 164         if (ret)
 165                 return ret;
 166
 167         d = bkey_s_c_to_dirent(bch2_btree_iter_peek_slot(&iter));
 168         *target = le64_to_cpu(d.v->d_inum);
 169         *type = d.v->d_type;
 170         bch2_trans_iter_exit(trans, &iter);
 171         return 0;
 172 }
 173
 174 static int lookup_dirent(struct btree_trans *trans,
 175                          struct bch_hash_info hash_info,
 176                          subvol_inum dir, struct qstr *name,
 177                          u64 *target, unsigned *type)
 178 {
 179         return lockrestart_do(trans,
 180                 __lookup_dirent(trans, hash_info, dir, name, target, type));
 181 }
 182
 183 static int __write_inode(struct btree_trans *trans,
 184                          struct bch_inode_unpacked *inode,
 185                          u32 snapshot)
 186 {
 187         struct btree_iter iter;
 188         int ret;
 189
 190         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
 191                             SPOS(0, inode->bi_inum, snapshot),
 192                             BTREE_ITER_INTENT);
 193
 194         ret   = bch2_btree_iter_traverse(&iter) ?:
 195                 bch2_inode_write(trans, &iter, inode);
 196         bch2_trans_iter_exit(trans, &iter);
 197         return ret;
 198 }
 199
 200 static int write_inode(struct btree_trans *trans,
 201                        struct bch_inode_unpacked *inode,
 202                        u32 snapshot)
 203 {
 204         int ret = __bch2_trans_do(trans, NULL, NULL,
 205                                   BTREE_INSERT_NOFAIL|
 206                                   BTREE_INSERT_LAZY_RW,
 207                                   __write_inode(trans, inode, snapshot));
 208         if (ret)
 209                 bch_err(trans->c, "error in fsck: error %i updating inode", ret);
 210         return ret;
 211 }
 212
 213 static int fsck_inode_rm(struct btree_trans *trans, u64 inum, u32 snapshot)
 214 {
 215         struct btree_iter iter = { NULL };
 216         struct bkey_i_inode_generation delete;
 217         struct bch_inode_unpacked inode_u;
 218         struct bkey_s_c k;
 219         int ret;
 220
 221         ret   = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
 222                                               SPOS(inum, 0, snapshot),
 223                                               SPOS(inum, U64_MAX, snapshot),
 224                                               0, NULL) ?:
 225                 bch2_btree_delete_range_trans(trans, BTREE_ID_dirents,
 226                                               SPOS(inum, 0, snapshot),
 227                                               SPOS(inum, U64_MAX, snapshot),
 228                                               0, NULL) ?:
 229                 bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs,
 230                                               SPOS(inum, 0, snapshot),
 231                                               SPOS(inum, U64_MAX, snapshot),
 232                                               0, NULL);
 233         if (ret)
 234                 goto err;
 235 retry:
 236         bch2_trans_begin(trans);
 237
 238         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
 239                              SPOS(0, inum, snapshot), BTREE_ITER_INTENT);
 240         k = bch2_btree_iter_peek_slot(&iter);
 241
 242         ret = bkey_err(k);
 243         if (ret)
 244                 goto err;
 245
 246         if (k.k->type != KEY_TYPE_inode) {
 247                 bch2_fs_inconsistent(trans->c,
 248                                      "inode %llu:%u not found when deleting",
 249                                      inum, snapshot);
 250                 ret = -EIO;
 251                 goto err;
 252         }
 253
 254         bch2_inode_unpack(bkey_s_c_to_inode(k), &inode_u);
 255
 256         /* Subvolume root? */
 257         if (inode_u.bi_subvol) {
 258                 ret = bch2_subvolume_delete(trans, inode_u.bi_subvol, -1);
 259                 if (ret)
 260                         goto err;
 261         }
 262
 263         bkey_inode_generation_init(&delete.k_i);
 264         delete.k.p = iter.pos;
 265         delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
 266
 267         ret   = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
 268                 bch2_trans_commit(trans, NULL, NULL,
 269                                 BTREE_INSERT_NOFAIL);
 270 err:
 271         bch2_trans_iter_exit(trans, &iter);
 272         if (ret == -EINTR)
 273                 goto retry;
 274
 275         return ret;
 276 }
 277
 278 static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
 279 {
 280         struct bch_fs *c = trans->c;
 281         struct btree_iter iter;
 282         struct bch_inode_unpacked dir_inode;
 283         struct bch_hash_info dir_hash_info;
 284         int ret;
 285
 286         ret = lookup_inode(trans, pos.inode, &dir_inode, NULL);
 287         if (ret)
 288                 return ret;
 289
 290         dir_hash_info = bch2_hash_info_init(c, &dir_inode);
 291
 292         bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_INTENT);
 293
 294         ret = bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
 295                                   &dir_hash_info, &iter, 0);
 296         bch2_trans_iter_exit(trans, &iter);
 297         return ret;
 298 }
 299
 300 static int remove_dirent(struct btree_trans *trans, struct bpos pos)
 301 {
 302         int ret = __bch2_trans_do(trans, NULL, NULL,
 303                                   BTREE_INSERT_NOFAIL|
 304                                   BTREE_INSERT_LAZY_RW,
 305                                   __remove_dirent(trans, pos));
 306         if (ret)
 307                 bch_err(trans->c, "remove_dirent: err %i deleting dirent", ret);
 308         return ret;
 309 }
 310
 311 /* Get lost+found, create if it doesn't exist: */
 312 static int lookup_lostfound(struct btree_trans *trans, u32 subvol,
 313                             struct bch_inode_unpacked *lostfound)
 314 {
 315         struct bch_fs *c = trans->c;
 316         struct bch_inode_unpacked root;
 317         struct bch_hash_info root_hash_info;
 318         struct qstr lostfound_str = QSTR("lost+found");
 319         subvol_inum root_inum = { .subvol = subvol };
 320         u64 inum = 0;
 321         unsigned d_type = 0;
 322         u32 snapshot;
 323         int ret;
 324
 325         ret = subvol_lookup(trans, subvol, &snapshot, &root_inum.inum);
 326         if (ret)
 327                 return ret;
 328
 329         ret = lookup_inode(trans, root_inum.inum, &root, &snapshot);
 330         if (ret) {
 331                 bch_err(c, "error fetching subvol root: %i", ret);
 332                 return ret;
 333         }
 334
 335         root_hash_info = bch2_hash_info_init(c, &root);
 336
 337         ret = lookup_dirent(trans, root_hash_info, root_inum,
 338                             &lostfound_str, &inum, &d_type);
 339         if (ret == -ENOENT) {
 340                 bch_notice(c, "creating lost+found");
 341                 goto create_lostfound;
 342         }
 343
 344         if (ret) {
 345                 bch_err(c, "error looking up lost+found: %i", ret);
 346                 return ret;
 347         }
 348
 349         if (d_type != DT_DIR) {
 350                 bch_err(c, "error looking up lost+found: not a directory");
 351                 return ret;
 352
 353         }
 354
 355         ret = lookup_inode(trans, inum, lostfound, &snapshot);
 356         if (ret && ret != -ENOENT) {
 357                 /*
 358                  * The check_dirents pass has already run, dangling dirents
 359                  * shouldn't exist here:
 360                  */
 361                 bch_err(c, "error looking up lost+found: %i", ret);
 362                 return ret;
 363         }
 364
 365         if (ret == -ENOENT) {
 366 create_lostfound:
 367                 bch2_inode_init_early(c, lostfound);
 368
 369                 ret = __bch2_trans_do(trans, NULL, NULL,
 370                                       BTREE_INSERT_NOFAIL|
 371                                       BTREE_INSERT_LAZY_RW,
 372                         bch2_create_trans(trans, root_inum, &root,
 373                                           lostfound, &lostfound_str,
 374                                           0, 0, S_IFDIR|0700, 0, NULL, NULL,
 375                                           (subvol_inum) { }, 0));
 376                 if (ret)
 377                         bch_err(c, "error creating lost+found: %i", ret);
 378         }
 379
 380         return 0;
 381 }
 382
 383 static int reattach_inode(struct btree_trans *trans,
 384                           struct bch_inode_unpacked *inode,
 385                           u32 inode_snapshot)
 386 {
 387         struct bch_hash_info dir_hash;
 388         struct bch_inode_unpacked lostfound;
 389         char name_buf[20];
 390         struct qstr name;
 391         u64 dir_offset = 0;
 392         u32 subvol;
 393         int ret;
 394
 395         ret = snapshot_lookup_subvol(trans, inode_snapshot, &subvol);
 396         if (ret)
 397                 return ret;
 398
 399         ret = lookup_lostfound(trans, subvol, &lostfound);
 400         if (ret)
 401                 return ret;
 402
 403         if (S_ISDIR(inode->bi_mode)) {
 404                 lostfound.bi_nlink++;
 405
 406                 ret = write_inode(trans, &lostfound, U32_MAX);
 407                 if (ret)
 408                         return ret;
 409         }
 410
 411         dir_hash = bch2_hash_info_init(trans->c, &lostfound);
 412
 413         snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
 414         name = (struct qstr) QSTR(name_buf);
 415
 416         ret = __bch2_trans_do(trans, NULL, NULL, BTREE_INSERT_LAZY_RW,
 417                         bch2_dirent_create(trans,
 418                                            (subvol_inum) {
 419                                                 .subvol = subvol,
 420                                                 .inum = lostfound.bi_inum,
 421                                            },
 422                                            &dir_hash,
 423                                            mode_to_type(inode->bi_mode),
 424                                            &name, inode->bi_inum, &dir_offset,
 425                                            BCH_HASH_SET_MUST_CREATE));
 426         if (ret) {
 427                 bch_err(trans->c, "error %i reattaching inode %llu",
 428                         ret, inode->bi_inum);
 429                 return ret;
 430         }
 431
 432         inode->bi_dir           = lostfound.bi_inum;
 433         inode->bi_dir_offset    = dir_offset;
 434
 435         return write_inode(trans, inode, inode_snapshot);
 436 }
 437
 438 static int remove_backpointer(struct btree_trans *trans,
 439                               struct bch_inode_unpacked *inode)
 440 {
 441         struct btree_iter iter;
 442         struct bkey_s_c k;
 443         int ret;
 444
 445         bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents,
 446                              POS(inode->bi_dir, inode->bi_dir_offset), 0);
 447         k = bch2_btree_iter_peek_slot(&iter);
 448         ret = bkey_err(k);
 449         if (ret)
 450                 goto out;
 451         if (k.k->type != KEY_TYPE_dirent) {
 452                 ret = -ENOENT;
 453                 goto out;
 454         }
 455
 456         ret = remove_dirent(trans, k.k->p);
 457 out:
 458         bch2_trans_iter_exit(trans, &iter);
 459         return ret;
 460 }
 461
 462 static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s, struct bpos pos)
 463 {
 464         pos.snapshot = snapshot_t(c, pos.snapshot)->equiv;
 465
 466         if (bkey_cmp(s->pos, pos))
 467                 s->nr = 0;
 468         s->pos = pos;
 469
 470         /* Might get called multiple times due to lock restarts */
 471         if (s->nr && s->d[s->nr - 1] == pos.snapshot)
 472                 return 0;
 473
 474         return snapshots_seen_add(c, s, pos.snapshot);
 475 }
 476
 477 /**
 478  * key_visible_in_snapshot - returns true if @id is a descendent of @ancestor,
 479  * and @ancestor hasn't been overwritten in @seen
 480  *
 481  * That is, returns whether key in @ancestor snapshot is visible in @id snapshot
 482  */
 483 static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *seen,
 484                                     u32 id, u32 ancestor)
 485 {
 486         ssize_t i;
 487
 488         BUG_ON(id > ancestor);
 489
 490         id              = snapshot_t(c, id)->equiv;
 491         ancestor        = snapshot_t(c, ancestor)->equiv;
 492
 493         /* @ancestor should be the snapshot most recently added to @seen */
 494         BUG_ON(!seen->nr || seen->d[seen->nr - 1] != ancestor);
 495         BUG_ON(seen->pos.snapshot != ancestor);
 496
 497         if (id == ancestor)
 498                 return true;
 499
 500         if (!bch2_snapshot_is_ancestor(c, id, ancestor))
 501                 return false;
 502
 503         for (i = seen->nr - 2;
 504              i >= 0 && seen->d[i] >= id;
 505              --i)
 506                 if (bch2_snapshot_is_ancestor(c, id, seen->d[i]) &&
 507                     bch2_snapshot_is_ancestor(c, seen->d[i], ancestor))
 508                         return false;
 509
 510         return true;
 511 }
 512
 513 /**
 514  * ref_visible - given a key with snapshot id @src that points to a key with
 515  * snapshot id @dst, test whether there is some snapshot in which @dst is
 516  * visible.
 517  *
 518  * This assumes we're visiting @src keys in natural key order.
 519  *
 520  * @s   - list of snapshot IDs already seen at @src
 521  * @src - snapshot ID of src key
 522  * @dst - snapshot ID of dst key
 523  */
 524 static int ref_visible(struct bch_fs *c, struct snapshots_seen *s,
 525                        u32 src, u32 dst)
 526 {
 527         return dst <= src
 528                 ? key_visible_in_snapshot(c, s, dst, src)
 529                 : bch2_snapshot_is_ancestor(c, src, dst);
 530 }
 531
 532 #define for_each_visible_inode(_c, _s, _w, _snapshot, _i)       \
 533         for (_i = (_w)->d; _i < (_w)->d + (_w)->nr && (_i)->snapshot <= (_snapshot); _i++)\
 534                 if (key_visible_in_snapshot(_c, _s, _i->snapshot, _snapshot))
 535
 536 struct inode_walker {
 537         bool                            first_this_inode;
 538         u64                             cur_inum;
 539
 540         size_t                          nr;
 541         size_t                          size;
 542         struct inode_walker_entry {
 543                 struct bch_inode_unpacked inode;
 544                 u32                     snapshot;
 545                 u64                     count;
 546         } *d;
 547 };
 548
 549 static void inode_walker_exit(struct inode_walker *w)
 550 {
 551         kfree(w->d);
 552         w->d = NULL;
 553 }
 554
 555 static struct inode_walker inode_walker_init(void)
 556 {
 557         return (struct inode_walker) { 0, };
 558 }
 559
 560 static int inode_walker_realloc(struct inode_walker *w)
 561 {
 562         if (w->nr == w->size) {
 563                 size_t new_size = max_t(size_t, 8UL, w->size * 2);
 564                 void *d = krealloc(w->d, new_size * sizeof(w->d[0]),
 565                                    GFP_KERNEL);
 566                 if (!d)
 567                         return -ENOMEM;
 568
 569                 w->d = d;
 570                 w->size = new_size;
 571         }
 572
 573         return 0;
 574 }
 575
 576 static int add_inode(struct bch_fs *c, struct inode_walker *w,
 577                      struct bkey_s_c_inode inode)
 578 {
 579         struct bch_inode_unpacked u;
 580         int ret;
 581
 582         ret = inode_walker_realloc(w);
 583         if (ret)
 584                 return ret;
 585
 586         BUG_ON(bch2_inode_unpack(inode, &u));
 587
 588         w->d[w->nr++] = (struct inode_walker_entry) {
 589                 .inode          = u,
 590                 .snapshot       = snapshot_t(c, inode.k->p.snapshot)->equiv,
 591         };
 592
 593         return 0;
 594 }
 595
 596 static int __walk_inode(struct btree_trans *trans,
 597                         struct inode_walker *w, struct bpos pos)
 598 {
 599         struct bch_fs *c = trans->c;
 600         struct btree_iter iter;
 601         struct bkey_s_c k;
 602         unsigned i, ancestor_pos;
 603         int ret;
 604
 605         pos.snapshot = snapshot_t(c, pos.snapshot)->equiv;
 606
 607         if (pos.inode == w->cur_inum) {
 608                 w->first_this_inode = false;
 609                 goto lookup_snapshot;
 610         }
 611
 612         w->nr = 0;
 613
 614         for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, pos.inode),
 615                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
 616                 if (k.k->p.offset != pos.inode)
 617                         break;
 618
 619                 if (k.k->type == KEY_TYPE_inode)
 620                         add_inode(c, w, bkey_s_c_to_inode(k));
 621         }
 622         bch2_trans_iter_exit(trans, &iter);
 623
 624         if (ret)
 625                 return ret;
 626
 627         w->cur_inum             = pos.inode;
 628         w->first_this_inode     = true;
 629 lookup_snapshot:
 630         for (i = 0; i < w->nr; i++)
 631                 if (bch2_snapshot_is_ancestor(c, pos.snapshot, w->d[i].snapshot))
 632                         goto found;
 633         return INT_MAX;
 634 found:
 635         BUG_ON(pos.snapshot > w->d[i].snapshot);
 636
 637         if (pos.snapshot != w->d[i].snapshot) {
 638                 ancestor_pos = i;
 639
 640                 while (i && w->d[i - 1].snapshot > pos.snapshot)
 641                         --i;
 642
 643                 ret = inode_walker_realloc(w);
 644                 if (ret)
 645                         return ret;
 646
 647                 array_insert_item(w->d, w->nr, i, w->d[ancestor_pos]);
 648                 w->d[i].snapshot = pos.snapshot;
 649                 w->d[i].count   = 0;
 650         }
 651
 652         return i;
 653 }
 654
 655 static int walk_inode(struct btree_trans *trans,
 656                       struct inode_walker *w, struct bpos pos)
 657 {
 658         return lockrestart_do(trans, __walk_inode(trans, w, pos));
 659 }
 660
 661 static int __get_visible_inodes(struct btree_trans *trans,
 662                                 struct inode_walker *w,
 663                                 struct snapshots_seen *s,
 664                                 u64 inum)
 665 {
 666         struct bch_fs *c = trans->c;
 667         struct btree_iter iter;
 668         struct bkey_s_c k;
 669         int ret;
 670
 671         w->nr = 0;
 672
 673         for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, inum),
 674                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
 675                 if (k.k->p.offset != inum)
 676                         break;
 677
 678                 if (k.k->type != KEY_TYPE_inode)
 679                         continue;
 680
 681                 if (ref_visible(c, s, s->pos.snapshot, k.k->p.snapshot)) {
 682                         add_inode(c, w, bkey_s_c_to_inode(k));
 683                         if (k.k->p.snapshot >= s->pos.snapshot)
 684                                 break;
 685                 }
 686         }
 687         bch2_trans_iter_exit(trans, &iter);
 688
 689         return ret;
 690 }
 691
 692 static int check_key_has_snapshot(struct btree_trans *trans,
 693                                   struct btree_iter *iter,
 694                                   struct bkey_s_c k)
 695 {
 696         struct bch_fs *c = trans->c;
 697         char buf[200];
 698         int ret = 0;
 699
 700         if (fsck_err_on(!snapshot_t(c, k.k->p.snapshot)->equiv, c,
 701                         "key in missing snapshot: %s",
 702                         (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf))) {
 703                 ret = __bch2_trans_do(trans, NULL, NULL, BTREE_INSERT_LAZY_RW,
 704                         bch2_btree_delete_at(trans, iter,
 705                                              BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE));
 706                 return ret ?: -EINTR;
 707         }
 708 fsck_err:
 709         return ret;
 710 }
 711
 712 static int hash_redo_key(struct btree_trans *trans,
 713                          const struct bch_hash_desc desc,
 714                          struct bch_hash_info *hash_info,
 715                          struct btree_iter *k_iter, struct bkey_s_c k)
 716 {
 717         bch_err(trans->c, "hash_redo_key() not implemented yet");
 718         return -EINVAL;
 719 #if 0
 720         struct bkey_i *delete;
 721         struct bkey_i *tmp;
 722
 723         delete = bch2_trans_kmalloc(trans, sizeof(*delete));
 724         if (IS_ERR(delete))
 725                 return PTR_ERR(delete);
 726
 727         tmp = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
 728         if (IS_ERR(tmp))
 729                 return PTR_ERR(tmp);
 730
 731         bkey_reassemble(tmp, k);
 732
 733         bkey_init(&delete->k);
 734         delete->k.p = k_iter->pos;
 735         return  bch2_btree_iter_traverse(k_iter) ?:
 736                 bch2_trans_update(trans, k_iter, delete, 0) ?:
 737                 bch2_hash_set(trans, desc, hash_info, k_iter->pos.inode, tmp, 0);
 738 #endif
 739 }
 740
 741 static int fsck_hash_delete_at(struct btree_trans *trans,
 742                                const struct bch_hash_desc desc,
 743                                struct bch_hash_info *info,
 744                                struct btree_iter *iter)
 745 {
 746         int ret;
 747 retry:
 748         ret   = bch2_hash_delete_at(trans, desc, info, iter, 0) ?:
 749                 bch2_trans_commit(trans, NULL, NULL,
 750                                   BTREE_INSERT_NOFAIL|
 751                                   BTREE_INSERT_LAZY_RW);
 752         if (ret == -EINTR) {
 753                 ret = bch2_btree_iter_traverse(iter);
 754                 if (!ret)
 755                         goto retry;
 756         }
 757
 758         return ret;
 759 }
 760
 761 static int hash_check_key(struct btree_trans *trans,
 762                           const struct bch_hash_desc desc,
 763                           struct bch_hash_info *hash_info,
 764                           struct btree_iter *k_iter, struct bkey_s_c hash_k)
 765 {
 766         struct bch_fs *c = trans->c;
 767         struct btree_iter iter = { NULL };
 768         char buf[200];
 769         struct bkey_s_c k;
 770         u64 hash;
 771         int ret = 0;
 772
 773         if (hash_k.k->type != desc.key_type)
 774                 return 0;
 775
 776         hash = desc.hash_bkey(hash_info, hash_k);
 777
 778         if (likely(hash == hash_k.k->p.offset))
 779                 return 0;
 780
 781         if (hash_k.k->p.offset < hash)
 782                 goto bad_hash;
 783
 784         for_each_btree_key(trans, iter, desc.btree_id, POS(hash_k.k->p.inode, hash),
 785                            BTREE_ITER_SLOTS, k, ret) {
 786                 if (!bkey_cmp(k.k->p, hash_k.k->p))
 787                         break;
 788
 789                 if (fsck_err_on(k.k->type == desc.key_type &&
 790                                 !desc.cmp_bkey(k, hash_k), c,
 791                                 "duplicate hash table keys:\n%s",
 792                                 (bch2_bkey_val_to_text(&PBUF(buf), c,
 793                                                        hash_k), buf))) {
 794                         ret = fsck_hash_delete_at(trans, desc, hash_info, k_iter);
 795                         if (ret)
 796                                 return ret;
 797                         ret = 1;
 798                         break;
 799                 }
 800
 801                 if (bkey_deleted(k.k)) {
 802                         bch2_trans_iter_exit(trans, &iter);
 803                         goto bad_hash;
 804                 }
 805
 806         }
 807         bch2_trans_iter_exit(trans, &iter);
 808         return ret;
 809 bad_hash:
 810         if (fsck_err(c, "hash table key at wrong offset: btree %u inode %llu offset %llu, "
 811                      "hashed to %llu\n%s",
 812                      desc.btree_id, hash_k.k->p.inode, hash_k.k->p.offset, hash,
 813                      (bch2_bkey_val_to_text(&PBUF(buf), c, hash_k), buf)) == FSCK_ERR_IGNORE)
 814                 return 0;
 815
 816         ret = __bch2_trans_do(trans, NULL, NULL,
 817                               BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
 818                 hash_redo_key(trans, desc, hash_info, k_iter, hash_k));
 819         if (ret) {
 820                 bch_err(c, "hash_redo_key err %i", ret);
 821                 return ret;
 822         }
 823         return -EINTR;
 824 fsck_err:
 825         return ret;
 826 }
 827
 828 static int check_inode(struct btree_trans *trans,
 829                        struct btree_iter *iter,
 830                        struct bch_inode_unpacked *prev,
 831                        struct bch_inode_unpacked u)
 832 {
 833         struct bch_fs *c = trans->c;
 834         bool do_update = false;
 835         int ret = 0;
 836
 837         if (fsck_err_on(prev &&
 838                         (prev->bi_hash_seed             != u.bi_hash_seed ||
 839                          mode_to_type(prev->bi_mode) != mode_to_type(u.bi_mode)), c,
 840                         "inodes in different snapshots don't match")) {
 841                 bch_err(c, "repair not implemented yet");
 842                 return -EINVAL;
 843         }
 844
 845         if (u.bi_flags & BCH_INODE_UNLINKED &&
 846             (!c->sb.clean ||
 847              fsck_err(c, "filesystem marked clean, but inode %llu unlinked",
 848                       u.bi_inum))) {
 849                 bch2_trans_unlock(trans);
 850                 bch2_fs_lazy_rw(c);
 851
 852                 ret = fsck_inode_rm(trans, u.bi_inum, iter->pos.snapshot);
 853                 if (ret)
 854                         bch_err(c, "error in fsck: error %i while deleting inode", ret);
 855                 return ret;
 856         }
 857
 858         if (u.bi_flags & BCH_INODE_I_SIZE_DIRTY &&
 859             (!c->sb.clean ||
 860              fsck_err(c, "filesystem marked clean, but inode %llu has i_size dirty",
 861                       u.bi_inum))) {
 862                 bch_verbose(c, "truncating inode %llu", u.bi_inum);
 863
 864                 bch2_trans_unlock(trans);
 865                 bch2_fs_lazy_rw(c);
 866
 867                 /*
 868                  * XXX: need to truncate partial blocks too here - or ideally
 869                  * just switch units to bytes and that issue goes away
 870                  */
 871                 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
 872                                 SPOS(u.bi_inum, round_up(u.bi_size, block_bytes(c)) >> 9,
 873                                      iter->pos.snapshot),
 874                                 POS(u.bi_inum, U64_MAX),
 875                                 0, NULL);
 876                 if (ret) {
 877                         bch_err(c, "error in fsck: error %i truncating inode", ret);
 878                         return ret;
 879                 }
 880
 881                 /*
 882                  * We truncated without our normal sector accounting hook, just
 883                  * make sure we recalculate it:
 884                  */
 885                 u.bi_flags |= BCH_INODE_I_SECTORS_DIRTY;
 886
 887                 u.bi_flags &= ~BCH_INODE_I_SIZE_DIRTY;
 888                 do_update = true;
 889         }
 890
 891         if (u.bi_flags & BCH_INODE_I_SECTORS_DIRTY &&
 892             (!c->sb.clean ||
 893              fsck_err(c, "filesystem marked clean, but inode %llu has i_sectors dirty",
 894                       u.bi_inum))) {
 895                 s64 sectors;
 896
 897                 bch_verbose(c, "recounting sectors for inode %llu",
 898                             u.bi_inum);
 899
 900                 sectors = bch2_count_inode_sectors(trans, u.bi_inum, iter->pos.snapshot);
 901                 if (sectors < 0) {
 902                         bch_err(c, "error in fsck: error %i recounting inode sectors",
 903                                 (int) sectors);
 904                         return sectors;
 905                 }
 906
 907                 u.bi_sectors = sectors;
 908                 u.bi_flags &= ~BCH_INODE_I_SECTORS_DIRTY;
 909                 do_update = true;
 910         }
 911
 912         if (u.bi_flags & BCH_INODE_BACKPTR_UNTRUSTED) {
 913                 u.bi_dir = 0;
 914                 u.bi_dir_offset = 0;
 915                 u.bi_flags &= ~BCH_INODE_BACKPTR_UNTRUSTED;
 916                 do_update = true;
 917         }
 918
 919         if (do_update) {
 920                 ret = write_inode(trans, &u, iter->pos.snapshot);
 921                 if (ret)
 922                         bch_err(c, "error in fsck: error %i "
 923                                 "updating inode", ret);
 924         }
 925 fsck_err:
 926         return ret;
 927 }
 928
 929 noinline_for_stack
 930 static int check_inodes(struct bch_fs *c, bool full)
 931 {
 932         struct btree_trans trans;
 933         struct btree_iter iter;
 934         struct bkey_s_c k;
 935         struct bkey_s_c_inode inode;
 936         struct bch_inode_unpacked prev, u;
 937         int ret;
 938
 939         memset(&prev, 0, sizeof(prev));
 940
 941         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 942
 943         for_each_btree_key(&trans, iter, BTREE_ID_inodes, POS_MIN,
 944                            BTREE_ITER_INTENT|
 945                            BTREE_ITER_PREFETCH|
 946                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
 947                 ret = check_key_has_snapshot(&trans, &iter, k);
 948                 if (ret)
 949                         break;
 950
 951                 /*
 952                  * if snapshot id isn't a leaf node, skip it - deletion in
 953                  * particular is not atomic, so on the internal snapshot nodes
 954                  * we can see inodes marked for deletion after a clean shutdown
 955                  */
 956                 if (bch2_snapshot_internal_node(c, k.k->p.snapshot))
 957                         continue;
 958
 959                 if (k.k->type != KEY_TYPE_inode)
 960                         continue;
 961
 962                 inode = bkey_s_c_to_inode(k);
 963
 964                 if (!full &&
 965                     !(inode.v->bi_flags & (BCH_INODE_I_SIZE_DIRTY|
 966                                            BCH_INODE_I_SECTORS_DIRTY|
 967                                            BCH_INODE_UNLINKED)))
 968                         continue;
 969
 970                 BUG_ON(bch2_inode_unpack(inode, &u));
 971
 972                 ret = check_inode(&trans, &iter,
 973                                   full && prev.bi_inum == u.bi_inum
 974                                   ? &prev : NULL, u);
 975                 if (ret)
 976                         break;
 977
 978                 prev = u;
 979         }
 980         bch2_trans_iter_exit(&trans, &iter);
 981
 982         BUG_ON(ret == -EINTR);
 983
 984         return bch2_trans_exit(&trans) ?: ret;
 985 }
 986
 987 noinline_for_stack
 988 static int check_subvols(struct bch_fs *c)
 989 {
 990         struct btree_trans trans;
 991         struct btree_iter iter;
 992         struct bkey_s_c k;
 993         int ret;
 994
 995         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 996
 997         for_each_btree_key(&trans, iter, BTREE_ID_subvolumes, POS_MIN,
 998                            0, k, ret) {
 999         }
1000         bch2_trans_iter_exit(&trans, &iter);
1001
1002         bch2_trans_exit(&trans);
1003         return ret;
1004 }
1005
1006 /*
1007  * Checking for overlapping extents needs to be reimplemented
1008  */
1009 #if 0
1010 static int fix_overlapping_extent(struct btree_trans *trans,
1011                                        struct bkey_s_c k, struct bpos cut_at)
1012 {
1013         struct btree_iter iter;
1014         struct bkey_i *u;
1015         int ret;
1016
1017         u = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
1018         ret = PTR_ERR_OR_ZERO(u);
1019         if (ret)
1020                 return ret;
1021
1022         bkey_reassemble(u, k);
1023         bch2_cut_front(cut_at, u);
1024
1025
1026         /*
1027          * We don't want to go through the extent_handle_overwrites path:
1028          *
1029          * XXX: this is going to screw up disk accounting, extent triggers
1030          * assume things about extent overwrites - we should be running the
1031          * triggers manually here
1032          */
1033         bch2_trans_iter_init(trans, &iter, BTREE_ID_extents, u->k.p,
1034                              BTREE_ITER_INTENT|BTREE_ITER_NOT_EXTENTS);
1035
1036         BUG_ON(iter.flags & BTREE_ITER_IS_EXTENTS);
1037         ret   = bch2_btree_iter_traverse(&iter) ?:
1038                 bch2_trans_update(trans, &iter, u, BTREE_TRIGGER_NORUN) ?:
1039                 bch2_trans_commit(trans, NULL, NULL,
1040                                   BTREE_INSERT_NOFAIL|
1041                                   BTREE_INSERT_LAZY_RW);
1042         bch2_trans_iter_exit(trans, &iter);
1043         return ret;
1044 }
1045 #endif
1046
1047 static int inode_backpointer_exists(struct btree_trans *trans,
1048                                     struct bch_inode_unpacked *inode,
1049                                     u32 snapshot)
1050 {
1051         struct btree_iter iter;
1052         struct bkey_s_c k;
1053         int ret;
1054
1055         bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents,
1056                         SPOS(inode->bi_dir, inode->bi_dir_offset, snapshot), 0);
1057         k = bch2_btree_iter_peek_slot(&iter);
1058         ret = bkey_err(k);
1059         if (ret)
1060                 goto out;
1061         if (k.k->type != KEY_TYPE_dirent)
1062                 goto out;
1063
1064         ret = le64_to_cpu(bkey_s_c_to_dirent(k).v->d_inum) == inode->bi_inum;
1065 out:
1066         bch2_trans_iter_exit(trans, &iter);
1067         return ret;
1068 }
1069
1070 static bool inode_backpointer_matches(struct bkey_s_c_dirent d,
1071                                       struct bch_inode_unpacked *inode)
1072 {
1073         return d.k->p.inode == inode->bi_dir &&
1074                 d.k->p.offset == inode->bi_dir_offset;
1075 }
1076
1077 static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
1078 {
1079         struct bch_fs *c = trans->c;
1080         struct inode_walker_entry *i;
1081         int ret = 0, ret2 = 0;
1082         s64 count2;
1083
1084         for (i = w->d; i < w->d + w->nr; i++) {
1085                 if (i->inode.bi_sectors == i->count)
1086                         continue;
1087
1088                 count2 = lockrestart_do(trans,
1089                         bch2_count_inode_sectors(trans, w->cur_inum, i->snapshot));
1090
1091                 if (i->count != count2) {
1092                         bch_err(c, "fsck counted i_sectors wrong: got %llu should be %llu",
1093                                 i->count, count2);
1094                         i->count = count2;
1095                         if (i->inode.bi_sectors == i->count)
1096                                 continue;
1097                 }
1098
1099                 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SECTORS_DIRTY), c,
1100                             "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
1101                             w->cur_inum, i->snapshot,
1102                             i->inode.bi_sectors, i->count) == FSCK_ERR_IGNORE)
1103                         continue;
1104
1105                 i->inode.bi_sectors = i->count;
1106                 ret = write_inode(trans, &i->inode, i->snapshot);
1107                 if (ret)
1108                         break;
1109                 ret2 = -EINTR;
1110         }
1111 fsck_err:
1112         return ret ?: ret2;
1113 }
1114
1115 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
1116                         struct inode_walker *inode,
1117                         struct snapshots_seen *s)
1118 {
1119         struct bch_fs *c = trans->c;
1120         struct bkey_s_c k;
1121         struct inode_walker_entry *i;
1122         char buf[200];
1123         int ret = 0;
1124
1125         k = bch2_btree_iter_peek(iter);
1126         if (!k.k)
1127                 return 0;
1128
1129         ret = bkey_err(k);
1130         if (ret)
1131                 return ret;
1132
1133         ret = check_key_has_snapshot(trans, iter, k);
1134         if (ret)
1135                 return ret;
1136
1137         ret = snapshots_seen_update(c, s, k.k->p);
1138         if (ret)
1139                 return ret;
1140
1141         if (k.k->type == KEY_TYPE_whiteout)
1142                 return 0;
1143
1144         if (inode->cur_inum != k.k->p.inode) {
1145                 ret = check_i_sectors(trans, inode);
1146                 if (ret)
1147                         return ret;
1148         }
1149 #if 0
1150         if (bkey_cmp(prev.k->k.p, bkey_start_pos(k.k)) > 0) {
1151                 char buf1[200];
1152                 char buf2[200];
1153
1154                 bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(prev.k));
1155                 bch2_bkey_val_to_text(&PBUF(buf2), c, k);
1156
1157                 if (fsck_err(c, "overlapping extents:\n%s\n%s", buf1, buf2))
1158                         return fix_overlapping_extent(trans, k, prev.k->k.p) ?: -EINTR;
1159         }
1160 #endif
1161         ret = __walk_inode(trans, inode, k.k->p);
1162         if (ret < 0)
1163                 return ret;
1164
1165         if (fsck_err_on(ret == INT_MAX, c,
1166                         "extent in missing inode:\n  %s",
1167                         (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)))
1168                 return __bch2_trans_do(trans, NULL, NULL, BTREE_INSERT_LAZY_RW,
1169                         bch2_btree_delete_at(trans, iter,
1170                                              BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE));
1171
1172         if (ret == INT_MAX)
1173                 return 0;
1174
1175         i = inode->d + ret;
1176         ret = 0;
1177
1178         if (fsck_err_on(!S_ISREG(i->inode.bi_mode) &&
1179                         !S_ISLNK(i->inode.bi_mode), c,
1180                         "extent in non regular inode mode %o:\n  %s",
1181                         i->inode.bi_mode,
1182                         (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)))
1183                 return __bch2_trans_do(trans, NULL, NULL, BTREE_INSERT_LAZY_RW,
1184                          bch2_btree_delete_at(trans, iter,
1185                                               BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE));
1186
1187         if (!bch2_snapshot_internal_node(c, k.k->p.snapshot)) {
1188                 for_each_visible_inode(c, s, inode, k.k->p.snapshot, i) {
1189                         if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SIZE_DIRTY) &&
1190                                         k.k->type != KEY_TYPE_reservation &&
1191                                         k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9, c,
1192                                         "extent type %u offset %llu past end of inode %llu, i_size %llu",
1193                                         k.k->type, k.k->p.offset, k.k->p.inode, i->inode.bi_size)) {
1194                                 bch2_fs_lazy_rw(c);
1195                                 return bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
1196                                                 SPOS(k.k->p.inode, round_up(i->inode.bi_size, block_bytes(c)) >> 9,
1197                                                      k.k->p.snapshot),
1198                                                 POS(k.k->p.inode, U64_MAX),
1199                                                 0, NULL) ?: -EINTR;
1200                         }
1201                 }
1202         }
1203
1204         if (bkey_extent_is_allocation(k.k))
1205                 for_each_visible_inode(c, s, inode, k.k->p.snapshot, i)
1206                         i->count += k.k->size;
1207 #if 0
1208         bch2_bkey_buf_reassemble(&prev, c, k);
1209 #endif
1210
1211 fsck_err:
1212         return ret;
1213 }
1214
1215 /*
1216  * Walk extents: verify that extents have a corresponding S_ISREG inode, and
1217  * that i_size an i_sectors are consistent
1218  */
1219 noinline_for_stack
1220 static int check_extents(struct bch_fs *c)
1221 {
1222         struct inode_walker w = inode_walker_init();
1223         struct snapshots_seen s;
1224         struct btree_trans trans;
1225         struct btree_iter iter;
1226         int ret = 0;
1227
1228 #if 0
1229         struct bkey_buf prev;
1230         bch2_bkey_buf_init(&prev);
1231         prev.k->k = KEY(0, 0, 0);
1232 #endif
1233         snapshots_seen_init(&s);
1234         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1235
1236         bch_verbose(c, "checking extents");
1237
1238         bch2_trans_iter_init(&trans, &iter, BTREE_ID_extents,
1239                              POS(BCACHEFS_ROOT_INO, 0),
1240                              BTREE_ITER_INTENT|
1241                              BTREE_ITER_PREFETCH|
1242                              BTREE_ITER_ALL_SNAPSHOTS);
1243
1244         do {
1245                 ret = lockrestart_do(&trans,
1246                         check_extent(&trans, &iter, &w, &s));
1247                 if (ret)
1248                         break;
1249         } while (bch2_btree_iter_advance(&iter));
1250         bch2_trans_iter_exit(&trans, &iter);
1251 #if 0
1252         bch2_bkey_buf_exit(&prev, c);
1253 #endif
1254         inode_walker_exit(&w);
1255         bch2_trans_exit(&trans);
1256         snapshots_seen_exit(&s);
1257
1258         return ret;
1259 }
1260
1261 static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
1262 {
1263         struct bch_fs *c = trans->c;
1264         struct inode_walker_entry *i;
1265         int ret = 0, ret2 = 0;
1266         s64 count2;
1267
1268         for (i = w->d; i < w->d + w->nr; i++) {
1269                 if (i->inode.bi_nlink == i->count)
1270                         continue;
1271
1272                 count2 = lockrestart_do(trans,
1273                                 bch2_count_subdirs(trans, w->cur_inum, i->snapshot));
1274
1275                 if (i->count != count2) {
1276                         bch_err(c, "fsck counted subdirectories wrong: got %llu should be %llu",
1277                                 i->count, count2);
1278                         i->count = count2;
1279                         if (i->inode.bi_nlink == i->count)
1280                                 continue;
1281                 }
1282
1283                 if (fsck_err_on(i->inode.bi_nlink != i->count, c,
1284                                 "directory %llu:%u with wrong i_nlink: got %u, should be %llu",
1285                                 w->cur_inum, i->snapshot, i->inode.bi_nlink, i->count)) {
1286                         i->inode.bi_nlink = i->count;
1287                         ret = write_inode(trans, &i->inode, i->snapshot);
1288                         if (ret)
1289                                 break;
1290                         ret2 = -EINTR;
1291                 }
1292         }
1293 fsck_err:
1294         return ret ?: ret2;
1295 }
1296
1297 static int check_dirent_target(struct btree_trans *trans,
1298                                struct btree_iter *iter,
1299                                struct bkey_s_c_dirent d,
1300                                struct bch_inode_unpacked *target,
1301                                u32 target_snapshot)
1302 {
1303         struct bch_fs *c = trans->c;
1304         bool backpointer_exists = true;
1305         char buf[200];
1306         int ret = 0;
1307
1308         if (!target->bi_dir &&
1309             !target->bi_dir_offset) {
1310                 target->bi_dir          = d.k->p.inode;
1311                 target->bi_dir_offset   = d.k->p.offset;
1312
1313                 ret = write_inode(trans, target, target_snapshot);
1314                 if (ret)
1315                         goto err;
1316         }
1317
1318         if (!inode_backpointer_matches(d, target)) {
1319                 ret = inode_backpointer_exists(trans, target, d.k->p.snapshot);
1320                 if (ret < 0)
1321                         goto err;
1322
1323                 backpointer_exists = ret;
1324                 ret = 0;
1325
1326                 if (fsck_err_on(S_ISDIR(target->bi_mode) &&
1327                                 backpointer_exists, c,
1328                                 "directory %llu with multiple links",
1329                                 target->bi_inum)) {
1330                         ret = remove_dirent(trans, d.k->p);
1331                         if (ret)
1332                                 goto err;
1333                         return 0;
1334                 }
1335
1336                 if (fsck_err_on(backpointer_exists &&
1337                                 !target->bi_nlink, c,
1338                                 "inode %llu has multiple links but i_nlink 0",
1339                                 target->bi_inum)) {
1340                         target->bi_nlink++;
1341                         target->bi_flags &= ~BCH_INODE_UNLINKED;
1342
1343                         ret = write_inode(trans, target, target_snapshot);
1344                         if (ret)
1345                                 goto err;
1346                 }
1347
1348                 if (fsck_err_on(!backpointer_exists, c,
1349                                 "inode %llu has wrong backpointer:\n"
1350                                 "got       %llu:%llu\n"
1351                                 "should be %llu:%llu",
1352                                 target->bi_inum,
1353                                 target->bi_dir,
1354                                 target->bi_dir_offset,
1355                                 d.k->p.inode,
1356                                 d.k->p.offset)) {
1357                         target->bi_dir          = d.k->p.inode;
1358                         target->bi_dir_offset   = d.k->p.offset;
1359
1360                         ret = write_inode(trans, target, target_snapshot);
1361                         if (ret)
1362                                 goto err;
1363                 }
1364         }
1365
1366         if (fsck_err_on(vfs_d_type(d.v->d_type) != mode_to_type(target->bi_mode), c,
1367                         "incorrect d_type: should be %u:\n%s",
1368                         mode_to_type(target->bi_mode),
1369                         (bch2_bkey_val_to_text(&PBUF(buf), c, d.s_c), buf))) {
1370                 struct bkey_i_dirent *n;
1371
1372                 n = kmalloc(bkey_bytes(d.k), GFP_KERNEL);
1373                 if (!n) {
1374                         ret = -ENOMEM;
1375                         goto err;
1376                 }
1377
1378                 bkey_reassemble(&n->k_i, d.s_c);
1379                 n->v.d_type = mode_to_type(target->bi_mode);
1380
1381                 ret = __bch2_trans_do(trans, NULL, NULL,
1382                                       BTREE_INSERT_NOFAIL|
1383                                       BTREE_INSERT_LAZY_RW,
1384                         bch2_trans_update(trans, iter, &n->k_i, 0));
1385                 kfree(n);
1386                 if (ret)
1387                         goto err;
1388         }
1389 err:
1390 fsck_err:
1391         return ret;
1392 }
1393
1394 static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
1395                         struct bch_hash_info *hash_info,
1396                         struct inode_walker *dir,
1397                         struct inode_walker *target,
1398                         struct snapshots_seen *s)
1399 {
1400         struct bch_fs *c = trans->c;
1401         struct bkey_s_c k;
1402         struct bkey_s_c_dirent d;
1403         struct inode_walker_entry *i;
1404         u32 target_snapshot;
1405         u32 target_subvol;
1406         u64 target_inum;
1407         char buf[200];
1408         int ret;
1409
1410         k = bch2_btree_iter_peek(iter);
1411         if (!k.k)
1412                 return 0;
1413
1414         ret = bkey_err(k);
1415         if (ret)
1416                 return ret;
1417
1418         ret = check_key_has_snapshot(trans, iter, k);
1419         if (ret)
1420                 return ret;
1421
1422         ret = snapshots_seen_update(c, s, k.k->p);
1423         if (ret)
1424                 return ret;
1425
1426         if (k.k->type == KEY_TYPE_whiteout)
1427                 return 0;
1428
1429         if (dir->cur_inum != k.k->p.inode) {
1430                 ret = check_subdir_count(trans, dir);
1431                 if (ret)
1432                         return ret;
1433         }
1434
1435         ret = __walk_inode(trans, dir, k.k->p);
1436         if (ret < 0)
1437                 return ret;
1438
1439         if (fsck_err_on(ret == INT_MAX, c,
1440                         "dirent in nonexisting directory:\n%s",
1441                         (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)))
1442                 return __bch2_trans_do(trans, NULL, NULL, BTREE_INSERT_LAZY_RW,
1443                                 bch2_btree_delete_at(trans, iter,
1444                                                      BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE));
1445
1446         if (ret == INT_MAX)
1447                 return 0;
1448
1449         i = dir->d + ret;
1450         ret = 0;
1451
1452         if (fsck_err_on(!S_ISDIR(i->inode.bi_mode), c,
1453                         "dirent in non directory inode type %u:\n%s",
1454                         mode_to_type(i->inode.bi_mode),
1455                         (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)))
1456                 return __bch2_trans_do(trans, NULL, NULL, 0,
1457                                 bch2_btree_delete_at(trans, iter, 0));
1458
1459         if (dir->first_this_inode)
1460                 *hash_info = bch2_hash_info_init(c, &dir->d[0].inode);
1461
1462         ret = hash_check_key(trans, bch2_dirent_hash_desc,
1463                              hash_info, iter, k);
1464         if (ret < 0)
1465                 return ret;
1466         if (ret) /* dirent has been deleted */
1467                 return 0;
1468
1469         if (k.k->type != KEY_TYPE_dirent)
1470                 return 0;
1471
1472         d = bkey_s_c_to_dirent(k);
1473
1474         ret = __bch2_dirent_read_target(trans, d,
1475                                         &target_subvol,
1476                                         &target_snapshot,
1477                                         &target_inum,
1478                                         true);
1479         if (ret && ret != -ENOENT)
1480                 return ret;
1481
1482         if (fsck_err_on(ret, c,
1483                         "dirent points to missing subvolume %llu",
1484                         le64_to_cpu(d.v->d_inum)))
1485                 return remove_dirent(trans, d.k->p);
1486
1487         if (target_subvol) {
1488                 struct bch_inode_unpacked subvol_root;
1489
1490                 ret = __lookup_inode(trans, target_inum,
1491                                    &subvol_root, &target_snapshot);
1492                 if (ret && ret != -ENOENT)
1493                         return ret;
1494
1495                 if (fsck_err_on(ret, c,
1496                                 "subvolume %u points to missing subvolume root %llu",
1497                                 target_subvol,
1498                                 target_inum)) {
1499                         bch_err(c, "repair not implemented yet");
1500                         return -EINVAL;
1501                 }
1502
1503                 if (fsck_err_on(subvol_root.bi_subvol != target_subvol, c,
1504                                 "subvol root %llu has wrong bi_subvol field: got %u, should be %u",
1505                                 target_inum,
1506                                 subvol_root.bi_subvol, target_subvol)) {
1507                         subvol_root.bi_subvol = target_subvol;
1508                         ret = write_inode(trans, &subvol_root, target_snapshot);
1509                         if (ret)
1510                                 return ret;
1511                 }
1512
1513                 ret = check_dirent_target(trans, iter, d, &subvol_root,
1514                                           target_snapshot);
1515                 if (ret)
1516                         return ret;
1517         } else {
1518                 ret = __get_visible_inodes(trans, target, s, target_inum);
1519                 if (ret)
1520                         return ret;
1521
1522                 if (fsck_err_on(!target->nr, c,
1523                                 "dirent points to missing inode:\n%s",
1524                                 (bch2_bkey_val_to_text(&PBUF(buf), c,
1525                                                        k), buf))) {
1526                         ret = remove_dirent(trans, d.k->p);
1527                         if (ret)
1528                                 return ret;
1529                 }
1530
1531                 for (i = target->d; i < target->d + target->nr; i++) {
1532                         ret = check_dirent_target(trans, iter, d,
1533                                                   &i->inode, i->snapshot);
1534                         if (ret)
1535                                 return ret;
1536                 }
1537         }
1538
1539         if (d.v->d_type == DT_DIR)
1540                 for_each_visible_inode(c, s, dir, d.k->p.snapshot, i)
1541                         i->count++;
1542
1543 fsck_err:
1544         return ret;
1545 }
1546
1547 /*
1548  * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
1549  * validate d_type
1550  */
1551 noinline_for_stack
1552 static int check_dirents(struct bch_fs *c)
1553 {
1554         struct inode_walker dir = inode_walker_init();
1555         struct inode_walker target = inode_walker_init();
1556         struct snapshots_seen s;
1557         struct bch_hash_info hash_info;
1558         struct btree_trans trans;
1559         struct btree_iter iter;
1560         int ret = 0;
1561
1562         bch_verbose(c, "checking dirents");
1563
1564         snapshots_seen_init(&s);
1565         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1566
1567         bch2_trans_iter_init(&trans, &iter, BTREE_ID_dirents,
1568                              POS(BCACHEFS_ROOT_INO, 0),
1569                              BTREE_ITER_INTENT|
1570                              BTREE_ITER_PREFETCH|
1571                              BTREE_ITER_ALL_SNAPSHOTS);
1572
1573         do {
1574                 ret = lockrestart_do(&trans,
1575                         check_dirent(&trans, &iter, &hash_info,
1576                                      &dir, &target, &s));
1577                 if (ret)
1578                         break;
1579         } while (bch2_btree_iter_advance(&iter));
1580         bch2_trans_iter_exit(&trans, &iter);
1581
1582         bch2_trans_exit(&trans);
1583         snapshots_seen_exit(&s);
1584         inode_walker_exit(&dir);
1585         inode_walker_exit(&target);
1586         return ret;
1587 }
1588
1589 /*
1590  * Walk xattrs: verify that they all have a corresponding inode
1591  */
1592 noinline_for_stack
1593 static int check_xattrs(struct bch_fs *c)
1594 {
1595         struct inode_walker w = inode_walker_init();
1596         struct bch_hash_info hash_info;
1597         struct btree_trans trans;
1598         struct btree_iter iter;
1599         struct bkey_s_c k;
1600         int ret = 0;
1601
1602         bch_verbose(c, "checking xattrs");
1603
1604         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1605
1606         bch2_trans_iter_init(&trans, &iter, BTREE_ID_xattrs,
1607                              POS(BCACHEFS_ROOT_INO, 0),
1608                              BTREE_ITER_INTENT|
1609                              BTREE_ITER_PREFETCH|
1610                              BTREE_ITER_ALL_SNAPSHOTS);
1611 retry:
1612         bch2_trans_begin(&trans);
1613
1614         while ((k = bch2_btree_iter_peek(&iter)).k &&
1615                !(ret = bkey_err(k))) {
1616                 ret = check_key_has_snapshot(&trans, &iter, k);
1617                 if (ret)
1618                         break;
1619
1620                 ret = walk_inode(&trans, &w, k.k->p);
1621                 if (ret < 0)
1622                         break;
1623
1624                 if (fsck_err_on(ret == INT_MAX, c,
1625                                 "xattr for missing inode %llu",
1626                                 k.k->p.inode)) {
1627                         ret = bch2_btree_delete_at(&trans, &iter, 0);
1628                         if (ret)
1629                                 break;
1630                         continue;
1631                 }
1632
1633                 if (ret == INT_MAX)
1634                         goto next;
1635                 ret = 0;
1636
1637                 if (w.first_this_inode)
1638                         hash_info = bch2_hash_info_init(c, &w.d[0].inode);
1639
1640                 ret = hash_check_key(&trans, bch2_xattr_hash_desc,
1641                                      &hash_info, &iter, k);
1642                 if (ret)
1643                         break;
1644 next:
1645                 bch2_btree_iter_advance(&iter);
1646         }
1647 fsck_err:
1648         if (ret == -EINTR)
1649                 goto retry;
1650
1651         bch2_trans_iter_exit(&trans, &iter);
1652         return bch2_trans_exit(&trans) ?: ret;
1653 }
1654
1655 /* Get root directory, create if it doesn't exist: */
1656 static int check_root(struct bch_fs *c)
1657 {
1658         struct btree_trans trans;
1659         struct bch_inode_unpacked root_inode;
1660         u32 snapshot;
1661         u64 inum;
1662         int ret;
1663
1664         bch2_trans_init(&trans, c, 0, 0);
1665
1666         bch_verbose(c, "checking root directory");
1667
1668         ret = subvol_lookup(&trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
1669         if (ret && ret != -ENOENT)
1670                 return ret;
1671
1672         if (mustfix_fsck_err_on(ret, c, "root subvol missing")) {
1673                 struct bkey_i_subvolume root_subvol;
1674
1675                 snapshot        = U32_MAX;
1676                 inum            = BCACHEFS_ROOT_INO;
1677
1678                 bkey_subvolume_init(&root_subvol.k_i);
1679                 root_subvol.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1680                 root_subvol.v.flags     = 0;
1681                 root_subvol.v.snapshot  = cpu_to_le32(snapshot);
1682                 root_subvol.v.inode     = cpu_to_le64(inum);
1683                 ret = __bch2_trans_do(&trans, NULL, NULL,
1684                                       BTREE_INSERT_NOFAIL|
1685                                       BTREE_INSERT_LAZY_RW,
1686                         __bch2_btree_insert(&trans, BTREE_ID_subvolumes, &root_subvol.k_i));
1687                 if (ret) {
1688                         bch_err(c, "error writing root subvol: %i", ret);
1689                         goto err;
1690                 }
1691
1692         }
1693
1694         ret = lookup_inode(&trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
1695         if (ret && ret != -ENOENT)
1696                 return ret;
1697
1698         if (mustfix_fsck_err_on(ret, c, "root directory missing") ||
1699             mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode), c,
1700                                 "root inode not a directory")) {
1701                 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755,
1702                                 0, NULL);
1703                 root_inode.bi_inum = inum;
1704
1705                 ret = write_inode(&trans, &root_inode, snapshot);
1706                 if (ret)
1707                         bch_err(c, "error writing root inode: %i", ret);
1708         }
1709 err:
1710 fsck_err:
1711         bch2_trans_exit(&trans);
1712         return ret;
1713 }
1714
1715 struct pathbuf {
1716         size_t          nr;
1717         size_t          size;
1718
1719         struct pathbuf_entry {
1720                 u64     inum;
1721         }               *entries;
1722 };
1723
1724 static int path_down(struct pathbuf *p, u64 inum)
1725 {
1726         if (p->nr == p->size) {
1727                 size_t new_size = max_t(size_t, 256UL, p->size * 2);
1728                 void *n = krealloc(p->entries,
1729                                    new_size * sizeof(p->entries[0]),
1730                                    GFP_KERNEL);
1731                 if (!n) {
1732                         return -ENOMEM;
1733                 }
1734
1735                 p->entries = n;
1736                 p->size = new_size;
1737         };
1738
1739         p->entries[p->nr++] = (struct pathbuf_entry) {
1740                 .inum = inum,
1741         };
1742         return 0;
1743 }
1744
1745 static int check_path(struct btree_trans *trans,
1746                       struct pathbuf *p,
1747                       struct bch_inode_unpacked *inode,
1748                       u32 snapshot)
1749 {
1750         struct bch_fs *c = trans->c;
1751         size_t i;
1752         int ret = 0;
1753
1754         snapshot = snapshot_t(c, snapshot)->equiv;
1755         p->nr = 0;
1756
1757         while (inode->bi_inum != BCACHEFS_ROOT_INO) {
1758                 ret = lockrestart_do(trans,
1759                         inode_backpointer_exists(trans, inode, snapshot));
1760                 if (ret < 0)
1761                         break;
1762
1763                 if (!ret) {
1764                         if (fsck_err(c,  "unreachable inode %llu:%u, type %u nlink %u backptr %llu:%llu",
1765                                      inode->bi_inum, snapshot,
1766                                      mode_to_type(inode->bi_mode),
1767                                      inode->bi_nlink,
1768                                      inode->bi_dir,
1769                                      inode->bi_dir_offset))
1770                                 ret = reattach_inode(trans, inode, snapshot);
1771                         break;
1772                 }
1773                 ret = 0;
1774
1775                 if (!S_ISDIR(inode->bi_mode))
1776                         break;
1777
1778                 ret = path_down(p, inode->bi_inum);
1779                 if (ret) {
1780                         bch_err(c, "memory allocation failure");
1781                         return ret;
1782                 }
1783
1784                 for (i = 0; i < p->nr; i++) {
1785                         if (inode->bi_dir != p->entries[i].inum)
1786                                 continue;
1787
1788                         /* XXX print path */
1789                         if (!fsck_err(c, "directory structure loop"))
1790                                 return 0;
1791
1792                         ret = lockrestart_do(trans,
1793                                         remove_backpointer(trans, inode));
1794                         if (ret) {
1795                                 bch_err(c, "error removing dirent: %i", ret);
1796                                 break;
1797                         }
1798
1799                         ret = reattach_inode(trans, inode, snapshot);
1800                         break;
1801                 }
1802
1803                 ret = lookup_inode(trans, inode->bi_dir, inode, &snapshot);
1804                 if (ret) {
1805                         /* Should have been caught in dirents pass */
1806                         bch_err(c, "error looking up parent directory: %i", ret);
1807                         break;
1808                 }
1809         }
1810 fsck_err:
1811         if (ret)
1812                 bch_err(c, "%s: err %i", __func__, ret);
1813         return ret;
1814 }
1815
1816 /*
1817  * Check for unreachable inodes, as well as loops in the directory structure:
1818  * After check_dirents(), if an inode backpointer doesn't exist that means it's
1819  * unreachable:
1820  */
1821 static int check_directory_structure(struct bch_fs *c)
1822 {
1823         struct btree_trans trans;
1824         struct btree_iter iter;
1825         struct bkey_s_c k;
1826         struct bch_inode_unpacked u;
1827         struct pathbuf path = { 0, 0, NULL };
1828         int ret;
1829
1830         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1831
1832         for_each_btree_key(&trans, iter, BTREE_ID_inodes, POS_MIN,
1833                            BTREE_ITER_INTENT|
1834                            BTREE_ITER_PREFETCH|
1835                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
1836                 if (k.k->type != KEY_TYPE_inode)
1837                         continue;
1838
1839                 ret = bch2_inode_unpack(bkey_s_c_to_inode(k), &u);
1840                 if (ret) {
1841                         /* Should have been caught earlier in fsck: */
1842                         bch_err(c, "error unpacking inode %llu: %i", k.k->p.offset, ret);
1843                         break;
1844                 }
1845
1846                 if (u.bi_flags & BCH_INODE_UNLINKED)
1847                         continue;
1848
1849                 ret = check_path(&trans, &path, &u, iter.pos.snapshot);
1850                 if (ret)
1851                         break;
1852         }
1853         bch2_trans_iter_exit(&trans, &iter);
1854
1855         BUG_ON(ret == -EINTR);
1856
1857         kfree(path.entries);
1858
1859         return bch2_trans_exit(&trans) ?: ret;
1860 }
1861
1862 struct nlink_table {
1863         size_t          nr;
1864         size_t          size;
1865
1866         struct nlink {
1867                 u64     inum;
1868                 u32     snapshot;
1869                 u32     count;
1870         }               *d;
1871 };
1872
1873 static int add_nlink(struct nlink_table *t, u64 inum, u32 snapshot)
1874 {
1875         if (t->nr == t->size) {
1876                 size_t new_size = max_t(size_t, 128UL, t->size * 2);
1877                 void *d = kvmalloc(new_size * sizeof(t->d[0]), GFP_KERNEL);
1878                 if (!d) {
1879                         return -ENOMEM;
1880                 }
1881
1882                 if (t->d)
1883                         memcpy(d, t->d, t->size * sizeof(t->d[0]));
1884                 kvfree(t->d);
1885
1886                 t->d = d;
1887                 t->size = new_size;
1888         }
1889
1890
1891         t->d[t->nr++] = (struct nlink) {
1892                 .inum           = inum,
1893                 .snapshot       = snapshot,
1894         };
1895
1896         return 0;
1897 }
1898
1899 static int nlink_cmp(const void *_l, const void *_r)
1900 {
1901         const struct nlink *l = _l;
1902         const struct nlink *r = _r;
1903
1904         return cmp_int(l->inum, r->inum) ?: cmp_int(l->snapshot, r->snapshot);
1905 }
1906
1907 static void inc_link(struct bch_fs *c, struct snapshots_seen *s,
1908                      struct nlink_table *links,
1909                      u64 range_start, u64 range_end, u64 inum, u32 snapshot)
1910 {
1911         struct nlink *link, key = {
1912                 .inum = inum, .snapshot = U32_MAX,
1913         };
1914
1915         if (inum < range_start || inum >= range_end)
1916                 return;
1917
1918         link = __inline_bsearch(&key, links->d, links->nr,
1919                                 sizeof(links->d[0]), nlink_cmp);
1920         if (!link)
1921                 return;
1922
1923         while (link > links->d && link[0].inum == link[-1].inum)
1924                 --link;
1925
1926         for (; link < links->d + links->nr && link->inum == inum; link++)
1927                 if (ref_visible(c, s, snapshot, link->snapshot)) {
1928                         link->count++;
1929                         if (link->snapshot >= snapshot)
1930                                 break;
1931                 }
1932 }
1933
1934 noinline_for_stack
1935 static int check_nlinks_find_hardlinks(struct bch_fs *c,
1936                                        struct nlink_table *t,
1937                                        u64 start, u64 *end)
1938 {
1939         struct btree_trans trans;
1940         struct btree_iter iter;
1941         struct bkey_s_c k;
1942         struct bkey_s_c_inode inode;
1943         struct bch_inode_unpacked u;
1944         int ret = 0;
1945
1946         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1947
1948         for_each_btree_key(&trans, iter, BTREE_ID_inodes,
1949                            POS(0, start),
1950                            BTREE_ITER_INTENT|
1951                            BTREE_ITER_PREFETCH|
1952                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
1953                 if (k.k->type != KEY_TYPE_inode)
1954                         continue;
1955
1956                 inode = bkey_s_c_to_inode(k);
1957
1958                 /*
1959                  * Backpointer and directory structure checks are sufficient for
1960                  * directories, since they can't have hardlinks:
1961                  */
1962                 if (S_ISDIR(le16_to_cpu(inode.v->bi_mode)))
1963                         continue;
1964
1965                 /* Should never fail, checked by bch2_inode_invalid: */
1966                 BUG_ON(bch2_inode_unpack(inode, &u));
1967
1968                 if (!u.bi_nlink)
1969                         continue;
1970
1971                 ret = add_nlink(t, k.k->p.offset, k.k->p.snapshot);
1972                 if (ret) {
1973                         *end = k.k->p.offset;
1974                         ret = 0;
1975                         break;
1976                 }
1977
1978         }
1979         bch2_trans_iter_exit(&trans, &iter);
1980         bch2_trans_exit(&trans);
1981
1982         if (ret)
1983                 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
1984
1985         return ret;
1986 }
1987
1988 noinline_for_stack
1989 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
1990                                      u64 range_start, u64 range_end)
1991 {
1992         struct btree_trans trans;
1993         struct snapshots_seen s;
1994         struct btree_iter iter;
1995         struct bkey_s_c k;
1996         struct bkey_s_c_dirent d;
1997         int ret;
1998
1999         snapshots_seen_init(&s);
2000
2001         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2002
2003         for_each_btree_key(&trans, iter, BTREE_ID_dirents, POS_MIN,
2004                            BTREE_ITER_INTENT|
2005                            BTREE_ITER_PREFETCH|
2006                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2007                 ret = snapshots_seen_update(c, &s, k.k->p);
2008                 if (ret)
2009                         break;
2010
2011                 switch (k.k->type) {
2012                 case KEY_TYPE_dirent:
2013                         d = bkey_s_c_to_dirent(k);
2014
2015                         if (d.v->d_type != DT_DIR &&
2016                             d.v->d_type != DT_SUBVOL)
2017                                 inc_link(c, &s, links, range_start, range_end,
2018                                          le64_to_cpu(d.v->d_inum),
2019                                          d.k->p.snapshot);
2020                         break;
2021                 }
2022
2023                 bch2_trans_cond_resched(&trans);
2024         }
2025         bch2_trans_iter_exit(&trans, &iter);
2026
2027         if (ret)
2028                 bch_err(c, "error in fsck: btree error %i while walking dirents", ret);
2029
2030         bch2_trans_exit(&trans);
2031         snapshots_seen_exit(&s);
2032         return ret;
2033 }
2034
2035 noinline_for_stack
2036 static int check_nlinks_update_hardlinks(struct bch_fs *c,
2037                                struct nlink_table *links,
2038                                u64 range_start, u64 range_end)
2039 {
2040         struct btree_trans trans;
2041         struct btree_iter iter;
2042         struct bkey_s_c k;
2043         struct bkey_s_c_inode inode;
2044         struct bch_inode_unpacked u;
2045         struct nlink *link = links->d;
2046         int ret = 0;
2047
2048         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2049
2050         for_each_btree_key(&trans, iter, BTREE_ID_inodes,
2051                            POS(0, range_start),
2052                            BTREE_ITER_INTENT|
2053                            BTREE_ITER_PREFETCH|
2054                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2055                 if (k.k->p.offset >= range_end)
2056                         break;
2057
2058                 if (k.k->type != KEY_TYPE_inode)
2059                         continue;
2060
2061                 inode = bkey_s_c_to_inode(k);
2062                 if (S_ISDIR(le16_to_cpu(inode.v->bi_mode)))
2063                         continue;
2064
2065                 BUG_ON(bch2_inode_unpack(inode, &u));
2066
2067                 if (!u.bi_nlink)
2068                         continue;
2069
2070                 while ((cmp_int(link->inum, k.k->p.offset) ?:
2071                         cmp_int(link->snapshot, k.k->p.snapshot)) < 0) {
2072                         link++;
2073                         BUG_ON(link >= links->d + links->nr);
2074                 }
2075
2076                 if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count, c,
2077                                 "inode %llu has wrong i_nlink (type %u i_nlink %u, should be %u)",
2078                                 u.bi_inum, mode_to_type(u.bi_mode),
2079                                 bch2_inode_nlink_get(&u), link->count)) {
2080                         bch2_inode_nlink_set(&u, link->count);
2081
2082                         ret = write_inode(&trans, &u, k.k->p.snapshot);
2083                         if (ret)
2084                                 bch_err(c, "error in fsck: error %i updating inode", ret);
2085                 }
2086         }
2087 fsck_err:
2088         bch2_trans_iter_exit(&trans, &iter);
2089         bch2_trans_exit(&trans);
2090
2091         if (ret)
2092                 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
2093
2094         return ret;
2095 }
2096
2097 noinline_for_stack
2098 static int check_nlinks(struct bch_fs *c)
2099 {
2100         struct nlink_table links = { 0 };
2101         u64 this_iter_range_start, next_iter_range_start = 0;
2102         int ret = 0;
2103
2104         bch_verbose(c, "checking inode nlinks");
2105
2106         do {
2107                 this_iter_range_start = next_iter_range_start;
2108                 next_iter_range_start = U64_MAX;
2109
2110                 ret = check_nlinks_find_hardlinks(c, &links,
2111                                                   this_iter_range_start,
2112                                                   &next_iter_range_start);
2113
2114                 ret = check_nlinks_walk_dirents(c, &links,
2115                                           this_iter_range_start,
2116                                           next_iter_range_start);
2117                 if (ret)
2118                         break;
2119
2120                 ret = check_nlinks_update_hardlinks(c, &links,
2121                                          this_iter_range_start,
2122                                          next_iter_range_start);
2123                 if (ret)
2124                         break;
2125
2126                 links.nr = 0;
2127         } while (next_iter_range_start != U64_MAX);
2128
2129         kvfree(links.d);
2130
2131         return ret;
2132 }
2133
2134 /*
2135  * Checks for inconsistencies that shouldn't happen, unless we have a bug.
2136  * Doesn't fix them yet, mainly because they haven't yet been observed:
2137  */
2138 int bch2_fsck_full(struct bch_fs *c)
2139 {
2140         return  bch2_fs_snapshots_check(c) ?:
2141                 check_inodes(c, true) ?:
2142                 check_subvols(c) ?:
2143                 check_extents(c) ?:
2144                 check_dirents(c) ?:
2145                 check_xattrs(c) ?:
2146                 check_root(c) ?:
2147                 check_directory_structure(c) ?:
2148                 check_nlinks(c);
2149 }
2150
2151 int bch2_fsck_walk_inodes_only(struct bch_fs *c)
2152 {
2153         return check_inodes(c, false);
2154 }