fs/bcachefs/fsck.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "bkey_buf.h"
   5 #include "btree_cache.h"
   6 #include "btree_update.h"
   7 #include "buckets.h"
   8 #include "darray.h"
   9 #include "dirent.h"
  10 #include "error.h"
  11 #include "fs-common.h"
  12 #include "fsck.h"
  13 #include "inode.h"
  14 #include "keylist.h"
  15 #include "recovery.h"
  16 #include "snapshot.h"
  17 #include "super.h"
  18 #include "xattr.h"
  19
  20 #include <linux/bsearch.h>
  21 #include <linux/dcache.h> /* struct qstr */
  22
  23 /*
  24  * XXX: this is handling transaction restarts without returning
  25  * -BCH_ERR_transaction_restart_nested, this is not how we do things anymore:
  26  */
  27 static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum,
  28                                     u32 snapshot)
  29 {
  30         u64 sectors = 0;
  31
  32         int ret = for_each_btree_key_upto(trans, iter, BTREE_ID_extents,
  33                                 SPOS(inum, 0, snapshot),
  34                                 POS(inum, U64_MAX),
  35                                 0, k, ({
  36                 if (bkey_extent_is_allocation(k.k))
  37                         sectors += k.k->size;
  38                 0;
  39         }));
  40
  41         return ret ?: sectors;
  42 }
  43
  44 static s64 bch2_count_subdirs(struct btree_trans *trans, u64 inum,
  45                                     u32 snapshot)
  46 {
  47         u64 subdirs = 0;
  48
  49         int ret = for_each_btree_key_upto(trans, iter, BTREE_ID_dirents,
  50                                     SPOS(inum, 0, snapshot),
  51                                     POS(inum, U64_MAX),
  52                                     0, k, ({
  53                 if (k.k->type == KEY_TYPE_dirent &&
  54                     bkey_s_c_to_dirent(k).v->d_type == DT_DIR)
  55                         subdirs++;
  56                 0;
  57         }));
  58
  59         return ret ?: subdirs;
  60 }
  61
  62 static int subvol_lookup(struct btree_trans *trans, u32 subvol,
  63                          u32 *snapshot, u64 *inum)
  64 {
  65         struct bch_subvolume s;
  66         int ret;
  67
  68         ret = bch2_subvolume_get(trans, subvol, false, 0, &s);
  69
  70         *snapshot = le32_to_cpu(s.snapshot);
  71         *inum = le64_to_cpu(s.inode);
  72         return ret;
  73 }
  74
  75 static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr,
  76                               struct bch_inode_unpacked *inode)
  77 {
  78         struct btree_iter iter;
  79         struct bkey_s_c k;
  80         int ret;
  81
  82         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
  83                              POS(0, inode_nr),
  84                              BTREE_ITER_ALL_SNAPSHOTS);
  85         k = bch2_btree_iter_peek(&iter);
  86         ret = bkey_err(k);
  87         if (ret)
  88                 goto err;
  89
  90         if (!k.k || !bkey_eq(k.k->p, POS(0, inode_nr))) {
  91                 ret = -BCH_ERR_ENOENT_inode;
  92                 goto err;
  93         }
  94
  95         ret = bch2_inode_unpack(k, inode);
  96 err:
  97         bch_err_msg(trans->c, ret, "fetching inode %llu", inode_nr);
  98         bch2_trans_iter_exit(trans, &iter);
  99         return ret;
 100 }
 101
 102 static int lookup_inode(struct btree_trans *trans, u64 inode_nr,
 103                         struct bch_inode_unpacked *inode,
 104                         u32 *snapshot)
 105 {
 106         struct btree_iter iter;
 107         struct bkey_s_c k;
 108         int ret;
 109
 110         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
 111                                SPOS(0, inode_nr, *snapshot), 0);
 112         ret = bkey_err(k);
 113         if (ret)
 114                 goto err;
 115
 116         ret = bkey_is_inode(k.k)
 117                 ? bch2_inode_unpack(k, inode)
 118                 : -BCH_ERR_ENOENT_inode;
 119         if (!ret)
 120                 *snapshot = iter.pos.snapshot;
 121 err:
 122         bch2_trans_iter_exit(trans, &iter);
 123         return ret;
 124 }
 125
 126 static int lookup_dirent_in_snapshot(struct btree_trans *trans,
 127                            struct bch_hash_info hash_info,
 128                            subvol_inum dir, struct qstr *name,
 129                            u64 *target, unsigned *type, u32 snapshot)
 130 {
 131         struct btree_iter iter;
 132         struct bkey_s_c_dirent d;
 133         int ret = bch2_hash_lookup_in_snapshot(trans, &iter, bch2_dirent_hash_desc,
 134                                &hash_info, dir, name, 0, snapshot);
 135         if (ret)
 136                 return ret;
 137
 138         d = bkey_s_c_to_dirent(bch2_btree_iter_peek_slot(&iter));
 139         *target = le64_to_cpu(d.v->d_inum);
 140         *type = d.v->d_type;
 141         bch2_trans_iter_exit(trans, &iter);
 142         return 0;
 143 }
 144
 145 static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
 146 {
 147         struct bch_fs *c = trans->c;
 148         struct btree_iter iter;
 149         struct bch_inode_unpacked dir_inode;
 150         struct bch_hash_info dir_hash_info;
 151         int ret;
 152
 153         ret = lookup_first_inode(trans, pos.inode, &dir_inode);
 154         if (ret)
 155                 goto err;
 156
 157         dir_hash_info = bch2_hash_info_init(c, &dir_inode);
 158
 159         bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_INTENT);
 160
 161         ret = bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
 162                                   &dir_hash_info, &iter,
 163                                   BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 164         bch2_trans_iter_exit(trans, &iter);
 165 err:
 166         bch_err_fn(c, ret);
 167         return ret;
 168 }
 169
 170 /* Get lost+found, create if it doesn't exist: */
 171 static int lookup_lostfound(struct btree_trans *trans, u32 snapshot,
 172                             struct bch_inode_unpacked *lostfound)
 173 {
 174         struct bch_fs *c = trans->c;
 175         struct qstr lostfound_str = QSTR("lost+found");
 176         u64 inum = 0;
 177         unsigned d_type = 0;
 178         int ret;
 179
 180         struct bch_snapshot_tree st;
 181         ret = bch2_snapshot_tree_lookup(trans,
 182                         bch2_snapshot_tree(c, snapshot), &st);
 183         if (ret)
 184                 return ret;
 185
 186         subvol_inum root_inum = { .subvol = le32_to_cpu(st.master_subvol) };
 187         u32 subvol_snapshot;
 188
 189         ret = subvol_lookup(trans, le32_to_cpu(st.master_subvol),
 190                             &subvol_snapshot, &root_inum.inum);
 191         bch_err_msg(c, ret, "looking up root subvol");
 192         if (ret)
 193                 return ret;
 194
 195         struct bch_inode_unpacked root_inode;
 196         struct bch_hash_info root_hash_info;
 197         u32 root_inode_snapshot = snapshot;
 198         ret = lookup_inode(trans, root_inum.inum, &root_inode, &root_inode_snapshot);
 199         bch_err_msg(c, ret, "looking up root inode");
 200         if (ret)
 201                 return ret;
 202
 203         root_hash_info = bch2_hash_info_init(c, &root_inode);
 204
 205         ret = lookup_dirent_in_snapshot(trans, root_hash_info, root_inum,
 206                               &lostfound_str, &inum, &d_type, snapshot);
 207         if (bch2_err_matches(ret, ENOENT))
 208                 goto create_lostfound;
 209
 210         bch_err_fn(c, ret);
 211         if (ret)
 212                 return ret;
 213
 214         if (d_type != DT_DIR) {
 215                 bch_err(c, "error looking up lost+found: not a directory");
 216                 return -BCH_ERR_ENOENT_not_directory;
 217         }
 218
 219         /*
 220          * The bch2_check_dirents pass has already run, dangling dirents
 221          * shouldn't exist here:
 222          */
 223         ret = lookup_inode(trans, inum, lostfound, &snapshot);
 224         bch_err_msg(c, ret, "looking up lost+found %llu:%u in (root inode %llu, snapshot root %u)",
 225                     inum, snapshot, root_inum.inum, bch2_snapshot_root(c, snapshot));
 226         return ret;
 227
 228 create_lostfound:
 229         /*
 230          * XXX: we could have a nicer log message here  if we had a nice way to
 231          * walk backpointers to print a path
 232          */
 233         bch_notice(c, "creating lost+found in snapshot %u", le32_to_cpu(st.root_snapshot));
 234
 235         u64 now = bch2_current_time(c);
 236         struct btree_iter lostfound_iter = { NULL };
 237         u64 cpu = raw_smp_processor_id();
 238
 239         bch2_inode_init_early(c, lostfound);
 240         bch2_inode_init_late(lostfound, now, 0, 0, S_IFDIR|0700, 0, &root_inode);
 241         lostfound->bi_dir = root_inode.bi_inum;
 242
 243         root_inode.bi_nlink++;
 244
 245         ret = bch2_inode_create(trans, &lostfound_iter, lostfound, snapshot, cpu);
 246         if (ret)
 247                 goto err;
 248
 249         bch2_btree_iter_set_snapshot(&lostfound_iter, snapshot);
 250         ret = bch2_btree_iter_traverse(&lostfound_iter);
 251         if (ret)
 252                 goto err;
 253
 254         ret =   bch2_dirent_create_snapshot(trans,
 255                                 0, root_inode.bi_inum, snapshot, &root_hash_info,
 256                                 mode_to_type(lostfound->bi_mode),
 257                                 &lostfound_str,
 258                                 lostfound->bi_inum,
 259                                 &lostfound->bi_dir_offset,
 260                                 BCH_HASH_SET_MUST_CREATE) ?:
 261                 bch2_inode_write_flags(trans, &lostfound_iter, lostfound,
 262                                        BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 263 err:
 264         bch_err_msg(c, ret, "creating lost+found");
 265         bch2_trans_iter_exit(trans, &lostfound_iter);
 266         return ret;
 267 }
 268
 269 static int reattach_inode(struct btree_trans *trans,
 270                           struct bch_inode_unpacked *inode,
 271                           u32 inode_snapshot)
 272 {
 273         struct bch_hash_info dir_hash;
 274         struct bch_inode_unpacked lostfound;
 275         char name_buf[20];
 276         struct qstr name;
 277         u64 dir_offset = 0;
 278         u32 dirent_snapshot = inode_snapshot;
 279         int ret;
 280
 281         if (inode->bi_subvol) {
 282                 inode->bi_parent_subvol = BCACHEFS_ROOT_SUBVOL;
 283
 284                 u64 root_inum;
 285                 ret = subvol_lookup(trans, inode->bi_parent_subvol,
 286                                     &dirent_snapshot, &root_inum);
 287                 if (ret)
 288                         return ret;
 289
 290                 snprintf(name_buf, sizeof(name_buf), "subvol-%u", inode->bi_subvol);
 291         } else {
 292                 snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
 293         }
 294
 295         ret = lookup_lostfound(trans, dirent_snapshot, &lostfound);
 296         if (ret)
 297                 return ret;
 298
 299         if (S_ISDIR(inode->bi_mode)) {
 300                 lostfound.bi_nlink++;
 301
 302                 ret = __bch2_fsck_write_inode(trans, &lostfound, U32_MAX);
 303                 if (ret)
 304                         return ret;
 305         }
 306
 307         dir_hash = bch2_hash_info_init(trans->c, &lostfound);
 308
 309         name = (struct qstr) QSTR(name_buf);
 310
 311         ret = bch2_dirent_create_snapshot(trans,
 312                                 inode->bi_parent_subvol, lostfound.bi_inum,
 313                                 dirent_snapshot,
 314                                 &dir_hash,
 315                                 inode_d_type(inode),
 316                                 &name,
 317                                 inode->bi_subvol ?: inode->bi_inum,
 318                                 &dir_offset,
 319                                 BCH_HASH_SET_MUST_CREATE);
 320         if (ret)
 321                 return ret;
 322
 323         inode->bi_dir           = lostfound.bi_inum;
 324         inode->bi_dir_offset    = dir_offset;
 325
 326         return __bch2_fsck_write_inode(trans, inode, inode_snapshot);
 327 }
 328
 329 static int remove_backpointer(struct btree_trans *trans,
 330                               struct bch_inode_unpacked *inode)
 331 {
 332         struct btree_iter iter;
 333         struct bkey_s_c_dirent d;
 334         int ret;
 335
 336         d = bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_dirents,
 337                                      POS(inode->bi_dir, inode->bi_dir_offset), 0,
 338                                      dirent);
 339         ret =   bkey_err(d) ?:
 340                 __remove_dirent(trans, d.k->p);
 341         bch2_trans_iter_exit(trans, &iter);
 342         return ret;
 343 }
 344
 345 static int reattach_subvol(struct btree_trans *trans, struct bkey_s_c_subvolume s)
 346 {
 347         struct bch_fs *c = trans->c;
 348
 349         struct bch_inode_unpacked inode;
 350         int ret = bch2_inode_find_by_inum_trans(trans,
 351                                 (subvol_inum) { s.k->p.offset, le64_to_cpu(s.v->inode) },
 352                                 &inode);
 353         if (ret)
 354                 return ret;
 355
 356         ret = remove_backpointer(trans, &inode);
 357         bch_err_msg(c, ret, "removing dirent");
 358         if (ret)
 359                 return ret;
 360
 361         ret = reattach_inode(trans, &inode, le32_to_cpu(s.v->snapshot));
 362         bch_err_msg(c, ret, "reattaching inode %llu", inode.bi_inum);
 363         return ret;
 364 }
 365
 366 struct snapshots_seen_entry {
 367         u32                             id;
 368         u32                             equiv;
 369 };
 370
 371 struct snapshots_seen {
 372         struct bpos                     pos;
 373         DARRAY(struct snapshots_seen_entry) ids;
 374 };
 375
 376 static inline void snapshots_seen_exit(struct snapshots_seen *s)
 377 {
 378         darray_exit(&s->ids);
 379 }
 380
 381 static inline void snapshots_seen_init(struct snapshots_seen *s)
 382 {
 383         memset(s, 0, sizeof(*s));
 384 }
 385
 386 static int snapshots_seen_add_inorder(struct bch_fs *c, struct snapshots_seen *s, u32 id)
 387 {
 388         struct snapshots_seen_entry *i, n = {
 389                 .id     = id,
 390                 .equiv  = bch2_snapshot_equiv(c, id),
 391         };
 392         int ret = 0;
 393
 394         __darray_for_each(s->ids, i) {
 395                 if (i->id == id)
 396                         return 0;
 397                 if (i->id > id)
 398                         break;
 399         }
 400
 401         ret = darray_insert_item(&s->ids, i - s->ids.data, n);
 402         if (ret)
 403                 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
 404                         s->ids.size);
 405         return ret;
 406 }
 407
 408 static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s,
 409                                  enum btree_id btree_id, struct bpos pos)
 410 {
 411         struct snapshots_seen_entry n = {
 412                 .id     = pos.snapshot,
 413                 .equiv  = bch2_snapshot_equiv(c, pos.snapshot),
 414         };
 415         int ret = 0;
 416
 417         if (!bkey_eq(s->pos, pos))
 418                 s->ids.nr = 0;
 419
 420         s->pos = pos;
 421         s->pos.snapshot = n.equiv;
 422
 423         darray_for_each(s->ids, i) {
 424                 if (i->id == n.id)
 425                         return 0;
 426
 427                 /*
 428                  * We currently don't rigorously track for snapshot cleanup
 429                  * needing to be run, so it shouldn't be a fsck error yet:
 430                  */
 431                 if (i->equiv == n.equiv) {
 432                         bch_err(c, "snapshot deletion did not finish:\n"
 433                                 "  duplicate keys in btree %s at %llu:%llu snapshots %u, %u (equiv %u)\n",
 434                                 bch2_btree_id_str(btree_id),
 435                                 pos.inode, pos.offset,
 436                                 i->id, n.id, n.equiv);
 437                         set_bit(BCH_FS_need_delete_dead_snapshots, &c->flags);
 438                         return bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_delete_dead_snapshots);
 439                 }
 440         }
 441
 442         ret = darray_push(&s->ids, n);
 443         if (ret)
 444                 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
 445                         s->ids.size);
 446         return ret;
 447 }
 448
 449 /**
 450  * key_visible_in_snapshot - returns true if @id is a descendent of @ancestor,
 451  * and @ancestor hasn't been overwritten in @seen
 452  *
 453  * @c:          filesystem handle
 454  * @seen:       list of snapshot ids already seen at current position
 455  * @id:         descendent snapshot id
 456  * @ancestor:   ancestor snapshot id
 457  *
 458  * Returns:     whether key in @ancestor snapshot is visible in @id snapshot
 459  */
 460 static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *seen,
 461                                     u32 id, u32 ancestor)
 462 {
 463         ssize_t i;
 464
 465         EBUG_ON(id > ancestor);
 466         EBUG_ON(!bch2_snapshot_is_equiv(c, id));
 467         EBUG_ON(!bch2_snapshot_is_equiv(c, ancestor));
 468
 469         /* @ancestor should be the snapshot most recently added to @seen */
 470         EBUG_ON(ancestor != seen->pos.snapshot);
 471         EBUG_ON(ancestor != seen->ids.data[seen->ids.nr - 1].equiv);
 472
 473         if (id == ancestor)
 474                 return true;
 475
 476         if (!bch2_snapshot_is_ancestor(c, id, ancestor))
 477                 return false;
 478
 479         /*
 480          * We know that @id is a descendant of @ancestor, we're checking if
 481          * we've seen a key that overwrote @ancestor - i.e. also a descendent of
 482          * @ascestor and with @id as a descendent.
 483          *
 484          * But we already know that we're scanning IDs between @id and @ancestor
 485          * numerically, since snapshot ID lists are kept sorted, so if we find
 486          * an id that's an ancestor of @id we're done:
 487          */
 488
 489         for (i = seen->ids.nr - 2;
 490              i >= 0 && seen->ids.data[i].equiv >= id;
 491              --i)
 492                 if (bch2_snapshot_is_ancestor(c, id, seen->ids.data[i].equiv))
 493                         return false;
 494
 495         return true;
 496 }
 497
 498 /**
 499  * ref_visible - given a key with snapshot id @src that points to a key with
 500  * snapshot id @dst, test whether there is some snapshot in which @dst is
 501  * visible.
 502  *
 503  * @c:          filesystem handle
 504  * @s:          list of snapshot IDs already seen at @src
 505  * @src:        snapshot ID of src key
 506  * @dst:        snapshot ID of dst key
 507  * Returns:     true if there is some snapshot in which @dst is visible
 508  *
 509  * Assumes we're visiting @src keys in natural key order
 510  */
 511 static bool ref_visible(struct bch_fs *c, struct snapshots_seen *s,
 512                         u32 src, u32 dst)
 513 {
 514         return dst <= src
 515                 ? key_visible_in_snapshot(c, s, dst, src)
 516                 : bch2_snapshot_is_ancestor(c, src, dst);
 517 }
 518
 519 static int ref_visible2(struct bch_fs *c,
 520                         u32 src, struct snapshots_seen *src_seen,
 521                         u32 dst, struct snapshots_seen *dst_seen)
 522 {
 523         src = bch2_snapshot_equiv(c, src);
 524         dst = bch2_snapshot_equiv(c, dst);
 525
 526         if (dst > src) {
 527                 swap(dst, src);
 528                 swap(dst_seen, src_seen);
 529         }
 530         return key_visible_in_snapshot(c, src_seen, dst, src);
 531 }
 532
 533 #define for_each_visible_inode(_c, _s, _w, _snapshot, _i)                               \
 534         for (_i = (_w)->inodes.data; _i < (_w)->inodes.data + (_w)->inodes.nr &&        \
 535              (_i)->snapshot <= (_snapshot); _i++)                                       \
 536                 if (key_visible_in_snapshot(_c, _s, _i->snapshot, _snapshot))
 537
 538 struct inode_walker_entry {
 539         struct bch_inode_unpacked inode;
 540         u32                     snapshot;
 541         bool                    seen_this_pos;
 542         u64                     count;
 543 };
 544
 545 struct inode_walker {
 546         bool                            first_this_inode;
 547         bool                            recalculate_sums;
 548         struct bpos                     last_pos;
 549
 550         DARRAY(struct inode_walker_entry) inodes;
 551 };
 552
 553 static void inode_walker_exit(struct inode_walker *w)
 554 {
 555         darray_exit(&w->inodes);
 556 }
 557
 558 static struct inode_walker inode_walker_init(void)
 559 {
 560         return (struct inode_walker) { 0, };
 561 }
 562
 563 static int add_inode(struct bch_fs *c, struct inode_walker *w,
 564                      struct bkey_s_c inode)
 565 {
 566         struct bch_inode_unpacked u;
 567
 568         BUG_ON(bch2_inode_unpack(inode, &u));
 569
 570         return darray_push(&w->inodes, ((struct inode_walker_entry) {
 571                 .inode          = u,
 572                 .snapshot       = bch2_snapshot_equiv(c, inode.k->p.snapshot),
 573         }));
 574 }
 575
 576 static int get_inodes_all_snapshots(struct btree_trans *trans,
 577                                     struct inode_walker *w, u64 inum)
 578 {
 579         struct bch_fs *c = trans->c;
 580         struct btree_iter iter;
 581         struct bkey_s_c k;
 582         int ret;
 583
 584         w->recalculate_sums = false;
 585         w->inodes.nr = 0;
 586
 587         for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum),
 588                                      BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
 589                 if (k.k->p.offset != inum)
 590                         break;
 591
 592                 if (bkey_is_inode(k.k))
 593                         add_inode(c, w, k);
 594         }
 595         bch2_trans_iter_exit(trans, &iter);
 596
 597         if (ret)
 598                 return ret;
 599
 600         w->first_this_inode = true;
 601         return 0;
 602 }
 603
 604 static struct inode_walker_entry *
 605 lookup_inode_for_snapshot(struct bch_fs *c, struct inode_walker *w, struct bkey_s_c k)
 606 {
 607         bool is_whiteout = k.k->type == KEY_TYPE_whiteout;
 608         u32 snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
 609
 610         struct inode_walker_entry *i;
 611         __darray_for_each(w->inodes, i)
 612                 if (bch2_snapshot_is_ancestor(c, snapshot, i->snapshot))
 613                         goto found;
 614
 615         return NULL;
 616 found:
 617         BUG_ON(snapshot > i->snapshot);
 618
 619         if (snapshot != i->snapshot && !is_whiteout) {
 620                 struct inode_walker_entry new = *i;
 621
 622                 new.snapshot = snapshot;
 623                 new.count = 0;
 624
 625                 struct printbuf buf = PRINTBUF;
 626                 bch2_bkey_val_to_text(&buf, c, k);
 627
 628                 bch_info(c, "have key for inode %llu:%u but have inode in ancestor snapshot %u\n"
 629                          "unexpected because we should always update the inode when we update a key in that inode\n"
 630                          "%s",
 631                          w->last_pos.inode, snapshot, i->snapshot, buf.buf);
 632                 printbuf_exit(&buf);
 633
 634                 while (i > w->inodes.data && i[-1].snapshot > snapshot)
 635                         --i;
 636
 637                 size_t pos = i - w->inodes.data;
 638                 int ret = darray_insert_item(&w->inodes, pos, new);
 639                 if (ret)
 640                         return ERR_PTR(ret);
 641
 642                 i = w->inodes.data + pos;
 643         }
 644
 645         return i;
 646 }
 647
 648 static struct inode_walker_entry *walk_inode(struct btree_trans *trans,
 649                                              struct inode_walker *w,
 650                                              struct bkey_s_c k)
 651 {
 652         if (w->last_pos.inode != k.k->p.inode) {
 653                 int ret = get_inodes_all_snapshots(trans, w, k.k->p.inode);
 654                 if (ret)
 655                         return ERR_PTR(ret);
 656         } else if (bkey_cmp(w->last_pos, k.k->p)) {
 657                 darray_for_each(w->inodes, i)
 658                         i->seen_this_pos = false;
 659         }
 660
 661         w->last_pos = k.k->p;
 662
 663         return lookup_inode_for_snapshot(trans->c, w, k);
 664 }
 665
 666 static int __get_visible_inodes(struct btree_trans *trans,
 667                                 struct inode_walker *w,
 668                                 struct snapshots_seen *s,
 669                                 u64 inum)
 670 {
 671         struct bch_fs *c = trans->c;
 672         struct btree_iter iter;
 673         struct bkey_s_c k;
 674         int ret;
 675
 676         w->inodes.nr = 0;
 677
 678         for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum),
 679                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
 680                 u32 equiv = bch2_snapshot_equiv(c, k.k->p.snapshot);
 681
 682                 if (k.k->p.offset != inum)
 683                         break;
 684
 685                 if (!ref_visible(c, s, s->pos.snapshot, equiv))
 686                         continue;
 687
 688                 if (bkey_is_inode(k.k))
 689                         add_inode(c, w, k);
 690
 691                 if (equiv >= s->pos.snapshot)
 692                         break;
 693         }
 694         bch2_trans_iter_exit(trans, &iter);
 695
 696         return ret;
 697 }
 698
 699 static int check_key_has_snapshot(struct btree_trans *trans,
 700                                   struct btree_iter *iter,
 701                                   struct bkey_s_c k)
 702 {
 703         struct bch_fs *c = trans->c;
 704         struct printbuf buf = PRINTBUF;
 705         int ret = 0;
 706
 707         if (mustfix_fsck_err_on(!bch2_snapshot_equiv(c, k.k->p.snapshot), c,
 708                                 bkey_in_missing_snapshot,
 709                                 "key in missing snapshot: %s",
 710                                 (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
 711                 ret = bch2_btree_delete_at(trans, iter,
 712                                             BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?: 1;
 713 fsck_err:
 714         printbuf_exit(&buf);
 715         return ret;
 716 }
 717
 718 static int hash_redo_key(struct btree_trans *trans,
 719                          const struct bch_hash_desc desc,
 720                          struct bch_hash_info *hash_info,
 721                          struct btree_iter *k_iter, struct bkey_s_c k)
 722 {
 723         struct bkey_i *delete;
 724         struct bkey_i *tmp;
 725
 726         delete = bch2_trans_kmalloc(trans, sizeof(*delete));
 727         if (IS_ERR(delete))
 728                 return PTR_ERR(delete);
 729
 730         tmp = bch2_bkey_make_mut_noupdate(trans, k);
 731         if (IS_ERR(tmp))
 732                 return PTR_ERR(tmp);
 733
 734         bkey_init(&delete->k);
 735         delete->k.p = k_iter->pos;
 736         return  bch2_btree_iter_traverse(k_iter) ?:
 737                 bch2_trans_update(trans, k_iter, delete, 0) ?:
 738                 bch2_hash_set_in_snapshot(trans, desc, hash_info,
 739                                        (subvol_inum) { 0, k.k->p.inode },
 740                                        k.k->p.snapshot, tmp,
 741                                        BCH_HASH_SET_MUST_CREATE,
 742                                        BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
 743                 bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
 744 }
 745
 746 static int hash_check_key(struct btree_trans *trans,
 747                           const struct bch_hash_desc desc,
 748                           struct bch_hash_info *hash_info,
 749                           struct btree_iter *k_iter, struct bkey_s_c hash_k)
 750 {
 751         struct bch_fs *c = trans->c;
 752         struct btree_iter iter = { NULL };
 753         struct printbuf buf = PRINTBUF;
 754         struct bkey_s_c k;
 755         u64 hash;
 756         int ret = 0;
 757
 758         if (hash_k.k->type != desc.key_type)
 759                 return 0;
 760
 761         hash = desc.hash_bkey(hash_info, hash_k);
 762
 763         if (likely(hash == hash_k.k->p.offset))
 764                 return 0;
 765
 766         if (hash_k.k->p.offset < hash)
 767                 goto bad_hash;
 768
 769         for_each_btree_key_norestart(trans, iter, desc.btree_id,
 770                                      SPOS(hash_k.k->p.inode, hash, hash_k.k->p.snapshot),
 771                                      BTREE_ITER_SLOTS, k, ret) {
 772                 if (bkey_eq(k.k->p, hash_k.k->p))
 773                         break;
 774
 775                 if (fsck_err_on(k.k->type == desc.key_type &&
 776                                 !desc.cmp_bkey(k, hash_k), c,
 777                                 hash_table_key_duplicate,
 778                                 "duplicate hash table keys:\n%s",
 779                                 (printbuf_reset(&buf),
 780                                  bch2_bkey_val_to_text(&buf, c, hash_k),
 781                                  buf.buf))) {
 782                         ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0) ?: 1;
 783                         break;
 784                 }
 785
 786                 if (bkey_deleted(k.k)) {
 787                         bch2_trans_iter_exit(trans, &iter);
 788                         goto bad_hash;
 789                 }
 790         }
 791 out:
 792         bch2_trans_iter_exit(trans, &iter);
 793         printbuf_exit(&buf);
 794         return ret;
 795 bad_hash:
 796         if (fsck_err(c, hash_table_key_wrong_offset,
 797                      "hash table key at wrong offset: btree %s inode %llu offset %llu, hashed to %llu\n%s",
 798                      bch2_btree_id_str(desc.btree_id), hash_k.k->p.inode, hash_k.k->p.offset, hash,
 799                      (printbuf_reset(&buf),
 800                       bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf))) {
 801                 ret = hash_redo_key(trans, desc, hash_info, k_iter, hash_k);
 802                 bch_err_fn(c, ret);
 803                 if (ret)
 804                         return ret;
 805                 ret = -BCH_ERR_transaction_restart_nested;
 806         }
 807 fsck_err:
 808         goto out;
 809 }
 810
 811 static struct bkey_s_c_dirent dirent_get_by_pos(struct btree_trans *trans,
 812                                                 struct btree_iter *iter,
 813                                                 struct bpos pos)
 814 {
 815         return bch2_bkey_get_iter_typed(trans, iter, BTREE_ID_dirents, pos, 0, dirent);
 816 }
 817
 818 static struct bkey_s_c_dirent inode_get_dirent(struct btree_trans *trans,
 819                                                struct btree_iter *iter,
 820                                                struct bch_inode_unpacked *inode,
 821                                                u32 *snapshot)
 822 {
 823         if (inode->bi_subvol) {
 824                 u64 inum;
 825                 int ret = subvol_lookup(trans, inode->bi_parent_subvol, snapshot, &inum);
 826                 if (ret)
 827                         return ((struct bkey_s_c_dirent) { .k = ERR_PTR(ret) });
 828         }
 829
 830         return dirent_get_by_pos(trans, iter, SPOS(inode->bi_dir, inode->bi_dir_offset, *snapshot));
 831 }
 832
 833 static bool inode_points_to_dirent(struct bch_inode_unpacked *inode,
 834                                    struct bkey_s_c_dirent d)
 835 {
 836         return  inode->bi_dir           == d.k->p.inode &&
 837                 inode->bi_dir_offset    == d.k->p.offset;
 838 }
 839
 840 static bool dirent_points_to_inode(struct bkey_s_c_dirent d,
 841                                    struct bch_inode_unpacked *inode)
 842 {
 843         return d.v->d_type == DT_SUBVOL
 844                 ? le32_to_cpu(d.v->d_child_subvol)      == inode->bi_subvol
 845                 : le64_to_cpu(d.v->d_inum)              == inode->bi_inum;
 846 }
 847
 848 static int check_inode_deleted_list(struct btree_trans *trans, struct bpos p)
 849 {
 850         struct btree_iter iter;
 851         struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_deleted_inodes, p, 0);
 852         int ret = bkey_err(k) ?: k.k->type == KEY_TYPE_set;
 853         bch2_trans_iter_exit(trans, &iter);
 854         return ret;
 855 }
 856
 857 static int check_inode_dirent_inode(struct btree_trans *trans, struct bkey_s_c inode_k,
 858                                     struct bch_inode_unpacked *inode,
 859                                     u32 inode_snapshot, bool *write_inode)
 860 {
 861         struct bch_fs *c = trans->c;
 862         struct printbuf buf = PRINTBUF;
 863
 864         struct btree_iter dirent_iter = {};
 865         struct bkey_s_c_dirent d = inode_get_dirent(trans, &dirent_iter, inode, &inode_snapshot);
 866         int ret = bkey_err(d);
 867         if (ret && !bch2_err_matches(ret, ENOENT))
 868                 return ret;
 869
 870         if (fsck_err_on(ret,
 871                         c, inode_points_to_missing_dirent,
 872                         "inode points to missing dirent\n%s",
 873                         (bch2_bkey_val_to_text(&buf, c, inode_k), buf.buf)) ||
 874             fsck_err_on(!ret && !dirent_points_to_inode(d, inode),
 875                         c, inode_points_to_wrong_dirent,
 876                         "inode points to dirent that does not point back:\n%s",
 877                         (bch2_bkey_val_to_text(&buf, c, inode_k),
 878                          prt_newline(&buf),
 879                          bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
 880                 /*
 881                  * We just clear the backpointer fields for now. If we find a
 882                  * dirent that points to this inode in check_dirents(), we'll
 883                  * update it then; then when we get to check_path() if the
 884                  * backpointer is still 0 we'll reattach it.
 885                  */
 886                 inode->bi_dir = 0;
 887                 inode->bi_dir_offset = 0;
 888                 inode->bi_flags &= ~BCH_INODE_backptr_untrusted;
 889                 *write_inode = true;
 890         }
 891
 892         ret = 0;
 893 fsck_err:
 894         bch2_trans_iter_exit(trans, &dirent_iter);
 895         printbuf_exit(&buf);
 896         bch_err_fn(c, ret);
 897         return ret;
 898 }
 899
 900 static int check_inode(struct btree_trans *trans,
 901                        struct btree_iter *iter,
 902                        struct bkey_s_c k,
 903                        struct bch_inode_unpacked *prev,
 904                        struct snapshots_seen *s,
 905                        bool full)
 906 {
 907         struct bch_fs *c = trans->c;
 908         struct bch_inode_unpacked u;
 909         bool do_update = false;
 910         int ret;
 911
 912         ret = check_key_has_snapshot(trans, iter, k);
 913         if (ret < 0)
 914                 goto err;
 915         if (ret)
 916                 return 0;
 917
 918         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
 919         if (ret)
 920                 goto err;
 921
 922         if (!bkey_is_inode(k.k))
 923                 return 0;
 924
 925         BUG_ON(bch2_inode_unpack(k, &u));
 926
 927         if (!full &&
 928             !(u.bi_flags & (BCH_INODE_i_size_dirty|
 929                             BCH_INODE_i_sectors_dirty|
 930                             BCH_INODE_unlinked)))
 931                 return 0;
 932
 933         if (prev->bi_inum != u.bi_inum)
 934                 *prev = u;
 935
 936         if (fsck_err_on(prev->bi_hash_seed      != u.bi_hash_seed ||
 937                         inode_d_type(prev)      != inode_d_type(&u),
 938                         c, inode_snapshot_mismatch,
 939                         "inodes in different snapshots don't match")) {
 940                 bch_err(c, "repair not implemented yet");
 941                 return -BCH_ERR_fsck_repair_unimplemented;
 942         }
 943
 944         if ((u.bi_flags & (BCH_INODE_i_size_dirty|BCH_INODE_unlinked)) &&
 945             bch2_key_has_snapshot_overwrites(trans, BTREE_ID_inodes, k.k->p)) {
 946                 struct bpos new_min_pos;
 947
 948                 ret = bch2_propagate_key_to_snapshot_leaves(trans, iter->btree_id, k, &new_min_pos);
 949                 if (ret)
 950                         goto err;
 951
 952                 u.bi_flags &= ~BCH_INODE_i_size_dirty|BCH_INODE_unlinked;
 953
 954                 ret = __bch2_fsck_write_inode(trans, &u, iter->pos.snapshot);
 955
 956                 bch_err_msg(c, ret, "in fsck updating inode");
 957                 if (ret)
 958                         return ret;
 959
 960                 if (!bpos_eq(new_min_pos, POS_MIN))
 961                         bch2_btree_iter_set_pos(iter, bpos_predecessor(new_min_pos));
 962                 return 0;
 963         }
 964
 965         if (u.bi_flags & BCH_INODE_unlinked) {
 966                 ret = check_inode_deleted_list(trans, k.k->p);
 967                 if (ret < 0)
 968                         return ret;
 969
 970                 fsck_err_on(!ret, c, unlinked_inode_not_on_deleted_list,
 971                             "inode %llu:%u unlinked, but not on deleted list",
 972                             u.bi_inum, k.k->p.snapshot);
 973                 ret = 0;
 974         }
 975
 976         if (u.bi_flags & BCH_INODE_unlinked &&
 977             (!c->sb.clean ||
 978              fsck_err(c, inode_unlinked_but_clean,
 979                       "filesystem marked clean, but inode %llu unlinked",
 980                       u.bi_inum))) {
 981                 ret = bch2_inode_rm_snapshot(trans, u.bi_inum, iter->pos.snapshot);
 982                 bch_err_msg(c, ret, "in fsck deleting inode");
 983                 return ret;
 984         }
 985
 986         if (u.bi_flags & BCH_INODE_i_size_dirty &&
 987             (!c->sb.clean ||
 988              fsck_err(c, inode_i_size_dirty_but_clean,
 989                       "filesystem marked clean, but inode %llu has i_size dirty",
 990                       u.bi_inum))) {
 991                 bch_verbose(c, "truncating inode %llu", u.bi_inum);
 992
 993                 /*
 994                  * XXX: need to truncate partial blocks too here - or ideally
 995                  * just switch units to bytes and that issue goes away
 996                  */
 997                 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
 998                                 SPOS(u.bi_inum, round_up(u.bi_size, block_bytes(c)) >> 9,
 999                                      iter->pos.snapshot),
1000                                 POS(u.bi_inum, U64_MAX),
1001                                 0, NULL);
1002                 bch_err_msg(c, ret, "in fsck truncating inode");
1003                 if (ret)
1004                         return ret;
1005
1006                 /*
1007                  * We truncated without our normal sector accounting hook, just
1008                  * make sure we recalculate it:
1009                  */
1010                 u.bi_flags |= BCH_INODE_i_sectors_dirty;
1011
1012                 u.bi_flags &= ~BCH_INODE_i_size_dirty;
1013                 do_update = true;
1014         }
1015
1016         if (u.bi_flags & BCH_INODE_i_sectors_dirty &&
1017             (!c->sb.clean ||
1018              fsck_err(c, inode_i_sectors_dirty_but_clean,
1019                       "filesystem marked clean, but inode %llu has i_sectors dirty",
1020                       u.bi_inum))) {
1021                 s64 sectors;
1022
1023                 bch_verbose(c, "recounting sectors for inode %llu",
1024                             u.bi_inum);
1025
1026                 sectors = bch2_count_inode_sectors(trans, u.bi_inum, iter->pos.snapshot);
1027                 if (sectors < 0) {
1028                         bch_err_msg(c, sectors, "in fsck recounting inode sectors");
1029                         return sectors;
1030                 }
1031
1032                 u.bi_sectors = sectors;
1033                 u.bi_flags &= ~BCH_INODE_i_sectors_dirty;
1034                 do_update = true;
1035         }
1036
1037         if (u.bi_flags & BCH_INODE_backptr_untrusted) {
1038                 u.bi_dir = 0;
1039                 u.bi_dir_offset = 0;
1040                 u.bi_flags &= ~BCH_INODE_backptr_untrusted;
1041                 do_update = true;
1042         }
1043
1044         if (u.bi_dir || u.bi_dir_offset) {
1045                 ret = check_inode_dirent_inode(trans, k, &u, k.k->p.snapshot, &do_update);
1046                 if (ret)
1047                         goto err;
1048         }
1049
1050         if (fsck_err_on(u.bi_parent_subvol &&
1051                         (u.bi_subvol == 0 ||
1052                          u.bi_subvol == BCACHEFS_ROOT_SUBVOL),
1053                         c, inode_bi_parent_nonzero,
1054                         "inode %llu:%u has subvol %u but nonzero parent subvol %u",
1055                         u.bi_inum, k.k->p.snapshot, u.bi_subvol, u.bi_parent_subvol)) {
1056                 u.bi_parent_subvol = 0;
1057                 do_update = true;
1058         }
1059
1060         if (u.bi_subvol) {
1061                 struct bch_subvolume s;
1062
1063                 ret = bch2_subvolume_get(trans, u.bi_subvol, false, 0, &s);
1064                 if (ret && !bch2_err_matches(ret, ENOENT))
1065                         goto err;
1066
1067                 if (fsck_err_on(ret,
1068                                 c, inode_bi_subvol_missing,
1069                                 "inode %llu:%u bi_subvol points to missing subvolume %u",
1070                                 u.bi_inum, k.k->p.snapshot, u.bi_subvol) ||
1071                     fsck_err_on(le64_to_cpu(s.inode) != u.bi_inum ||
1072                                 !bch2_snapshot_is_ancestor(c, le32_to_cpu(s.snapshot),
1073                                                            k.k->p.snapshot),
1074                                 c, inode_bi_subvol_wrong,
1075                                 "inode %llu:%u points to subvol %u, but subvol points to %llu:%u",
1076                                 u.bi_inum, k.k->p.snapshot, u.bi_subvol,
1077                                 le64_to_cpu(s.inode),
1078                                 le32_to_cpu(s.snapshot))) {
1079                         u.bi_subvol = 0;
1080                         u.bi_parent_subvol = 0;
1081                         do_update = true;
1082                 }
1083         }
1084
1085         if (do_update) {
1086                 ret = __bch2_fsck_write_inode(trans, &u, iter->pos.snapshot);
1087                 bch_err_msg(c, ret, "in fsck updating inode");
1088                 if (ret)
1089                         return ret;
1090         }
1091 err:
1092 fsck_err:
1093         bch_err_fn(c, ret);
1094         return ret;
1095 }
1096
1097 int bch2_check_inodes(struct bch_fs *c)
1098 {
1099         bool full = c->opts.fsck;
1100         struct bch_inode_unpacked prev = { 0 };
1101         struct snapshots_seen s;
1102
1103         snapshots_seen_init(&s);
1104
1105         int ret = bch2_trans_run(c,
1106                 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
1107                                 POS_MIN,
1108                                 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1109                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1110                         check_inode(trans, &iter, k, &prev, &s, full)));
1111
1112         snapshots_seen_exit(&s);
1113         bch_err_fn(c, ret);
1114         return ret;
1115 }
1116
1117 static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
1118 {
1119         struct bch_fs *c = trans->c;
1120         u32 restart_count = trans->restart_count;
1121         int ret = 0;
1122         s64 count2;
1123
1124         darray_for_each(w->inodes, i) {
1125                 if (i->inode.bi_sectors == i->count)
1126                         continue;
1127
1128                 count2 = bch2_count_inode_sectors(trans, w->last_pos.inode, i->snapshot);
1129
1130                 if (w->recalculate_sums)
1131                         i->count = count2;
1132
1133                 if (i->count != count2) {
1134                         bch_err(c, "fsck counted i_sectors wrong for inode %llu:%u: got %llu should be %llu",
1135                                 w->last_pos.inode, i->snapshot, i->count, count2);
1136                         return -BCH_ERR_internal_fsck_err;
1137                 }
1138
1139                 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_sectors_dirty),
1140                                 c, inode_i_sectors_wrong,
1141                                 "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
1142                                 w->last_pos.inode, i->snapshot,
1143                                 i->inode.bi_sectors, i->count)) {
1144                         i->inode.bi_sectors = i->count;
1145                         ret = bch2_fsck_write_inode(trans, &i->inode, i->snapshot);
1146                         if (ret)
1147                                 break;
1148                 }
1149         }
1150 fsck_err:
1151         bch_err_fn(c, ret);
1152         return ret ?: trans_was_restarted(trans, restart_count);
1153 }
1154
1155 struct extent_end {
1156         u32                     snapshot;
1157         u64                     offset;
1158         struct snapshots_seen   seen;
1159 };
1160
1161 struct extent_ends {
1162         struct bpos                     last_pos;
1163         DARRAY(struct extent_end)       e;
1164 };
1165
1166 static void extent_ends_reset(struct extent_ends *extent_ends)
1167 {
1168         darray_for_each(extent_ends->e, i)
1169                 snapshots_seen_exit(&i->seen);
1170         extent_ends->e.nr = 0;
1171 }
1172
1173 static void extent_ends_exit(struct extent_ends *extent_ends)
1174 {
1175         extent_ends_reset(extent_ends);
1176         darray_exit(&extent_ends->e);
1177 }
1178
1179 static void extent_ends_init(struct extent_ends *extent_ends)
1180 {
1181         memset(extent_ends, 0, sizeof(*extent_ends));
1182 }
1183
1184 static int extent_ends_at(struct bch_fs *c,
1185                           struct extent_ends *extent_ends,
1186                           struct snapshots_seen *seen,
1187                           struct bkey_s_c k)
1188 {
1189         struct extent_end *i, n = (struct extent_end) {
1190                 .offset         = k.k->p.offset,
1191                 .snapshot       = k.k->p.snapshot,
1192                 .seen           = *seen,
1193         };
1194
1195         n.seen.ids.data = kmemdup(seen->ids.data,
1196                               sizeof(seen->ids.data[0]) * seen->ids.size,
1197                               GFP_KERNEL);
1198         if (!n.seen.ids.data)
1199                 return -BCH_ERR_ENOMEM_fsck_extent_ends_at;
1200
1201         __darray_for_each(extent_ends->e, i) {
1202                 if (i->snapshot == k.k->p.snapshot) {
1203                         snapshots_seen_exit(&i->seen);
1204                         *i = n;
1205                         return 0;
1206                 }
1207
1208                 if (i->snapshot >= k.k->p.snapshot)
1209                         break;
1210         }
1211
1212         return darray_insert_item(&extent_ends->e, i - extent_ends->e.data, n);
1213 }
1214
1215 static int overlapping_extents_found(struct btree_trans *trans,
1216                                      enum btree_id btree,
1217                                      struct bpos pos1, struct snapshots_seen *pos1_seen,
1218                                      struct bkey pos2,
1219                                      bool *fixed,
1220                                      struct extent_end *extent_end)
1221 {
1222         struct bch_fs *c = trans->c;
1223         struct printbuf buf = PRINTBUF;
1224         struct btree_iter iter1, iter2 = { NULL };
1225         struct bkey_s_c k1, k2;
1226         int ret;
1227
1228         BUG_ON(bkey_le(pos1, bkey_start_pos(&pos2)));
1229
1230         bch2_trans_iter_init(trans, &iter1, btree, pos1,
1231                              BTREE_ITER_ALL_SNAPSHOTS|
1232                              BTREE_ITER_NOT_EXTENTS);
1233         k1 = bch2_btree_iter_peek_upto(&iter1, POS(pos1.inode, U64_MAX));
1234         ret = bkey_err(k1);
1235         if (ret)
1236                 goto err;
1237
1238         prt_str(&buf, "\n  ");
1239         bch2_bkey_val_to_text(&buf, c, k1);
1240
1241         if (!bpos_eq(pos1, k1.k->p)) {
1242                 prt_str(&buf, "\n  wanted\n  ");
1243                 bch2_bpos_to_text(&buf, pos1);
1244                 prt_str(&buf, "\n  ");
1245                 bch2_bkey_to_text(&buf, &pos2);
1246
1247                 bch_err(c, "%s: error finding first overlapping extent when repairing, got%s",
1248                         __func__, buf.buf);
1249                 ret = -BCH_ERR_internal_fsck_err;
1250                 goto err;
1251         }
1252
1253         bch2_trans_copy_iter(&iter2, &iter1);
1254
1255         while (1) {
1256                 bch2_btree_iter_advance(&iter2);
1257
1258                 k2 = bch2_btree_iter_peek_upto(&iter2, POS(pos1.inode, U64_MAX));
1259                 ret = bkey_err(k2);
1260                 if (ret)
1261                         goto err;
1262
1263                 if (bpos_ge(k2.k->p, pos2.p))
1264                         break;
1265         }
1266
1267         prt_str(&buf, "\n  ");
1268         bch2_bkey_val_to_text(&buf, c, k2);
1269
1270         if (bpos_gt(k2.k->p, pos2.p) ||
1271             pos2.size != k2.k->size) {
1272                 bch_err(c, "%s: error finding seconding overlapping extent when repairing%s",
1273                         __func__, buf.buf);
1274                 ret = -BCH_ERR_internal_fsck_err;
1275                 goto err;
1276         }
1277
1278         prt_printf(&buf, "\n  overwriting %s extent",
1279                    pos1.snapshot >= pos2.p.snapshot ? "first" : "second");
1280
1281         if (fsck_err(c, extent_overlapping,
1282                      "overlapping extents%s", buf.buf)) {
1283                 struct btree_iter *old_iter = &iter1;
1284                 struct disk_reservation res = { 0 };
1285
1286                 if (pos1.snapshot < pos2.p.snapshot) {
1287                         old_iter = &iter2;
1288                         swap(k1, k2);
1289                 }
1290
1291                 trans->extra_disk_res += bch2_bkey_sectors_compressed(k2);
1292
1293                 ret =   bch2_trans_update_extent_overwrite(trans, old_iter,
1294                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE,
1295                                 k1, k2) ?:
1296                         bch2_trans_commit(trans, &res, NULL, BCH_TRANS_COMMIT_no_enospc);
1297                 bch2_disk_reservation_put(c, &res);
1298
1299                 if (ret)
1300                         goto err;
1301
1302                 *fixed = true;
1303
1304                 if (pos1.snapshot == pos2.p.snapshot) {
1305                         /*
1306                          * We overwrote the first extent, and did the overwrite
1307                          * in the same snapshot:
1308                          */
1309                         extent_end->offset = bkey_start_offset(&pos2);
1310                 } else if (pos1.snapshot > pos2.p.snapshot) {
1311                         /*
1312                          * We overwrote the first extent in pos2's snapshot:
1313                          */
1314                         ret = snapshots_seen_add_inorder(c, pos1_seen, pos2.p.snapshot);
1315                 } else {
1316                         /*
1317                          * We overwrote the second extent - restart
1318                          * check_extent() from the top:
1319                          */
1320                         ret = -BCH_ERR_transaction_restart_nested;
1321                 }
1322         }
1323 fsck_err:
1324 err:
1325         bch2_trans_iter_exit(trans, &iter2);
1326         bch2_trans_iter_exit(trans, &iter1);
1327         printbuf_exit(&buf);
1328         return ret;
1329 }
1330
1331 static int check_overlapping_extents(struct btree_trans *trans,
1332                               struct snapshots_seen *seen,
1333                               struct extent_ends *extent_ends,
1334                               struct bkey_s_c k,
1335                               u32 equiv,
1336                               struct btree_iter *iter,
1337                               bool *fixed)
1338 {
1339         struct bch_fs *c = trans->c;
1340         int ret = 0;
1341
1342         /* transaction restart, running again */
1343         if (bpos_eq(extent_ends->last_pos, k.k->p))
1344                 return 0;
1345
1346         if (extent_ends->last_pos.inode != k.k->p.inode)
1347                 extent_ends_reset(extent_ends);
1348
1349         darray_for_each(extent_ends->e, i) {
1350                 if (i->offset <= bkey_start_offset(k.k))
1351                         continue;
1352
1353                 if (!ref_visible2(c,
1354                                   k.k->p.snapshot, seen,
1355                                   i->snapshot, &i->seen))
1356                         continue;
1357
1358                 ret = overlapping_extents_found(trans, iter->btree_id,
1359                                                 SPOS(iter->pos.inode,
1360                                                      i->offset,
1361                                                      i->snapshot),
1362                                                 &i->seen,
1363                                                 *k.k, fixed, i);
1364                 if (ret)
1365                         goto err;
1366         }
1367
1368         ret = extent_ends_at(c, extent_ends, seen, k);
1369         if (ret)
1370                 goto err;
1371
1372         extent_ends->last_pos = k.k->p;
1373 err:
1374         return ret;
1375 }
1376
1377 static int check_extent_overbig(struct btree_trans *trans, struct btree_iter *iter,
1378                                 struct bkey_s_c k)
1379 {
1380         struct bch_fs *c = trans->c;
1381         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1382         struct bch_extent_crc_unpacked crc;
1383         const union bch_extent_entry *i;
1384         unsigned encoded_extent_max_sectors = c->opts.encoded_extent_max >> 9;
1385
1386         bkey_for_each_crc(k.k, ptrs, crc, i)
1387                 if (crc_is_encoded(crc) &&
1388                     crc.uncompressed_size > encoded_extent_max_sectors) {
1389                         struct printbuf buf = PRINTBUF;
1390
1391                         bch2_bkey_val_to_text(&buf, c, k);
1392                         bch_err(c, "overbig encoded extent, please report this:\n  %s", buf.buf);
1393                         printbuf_exit(&buf);
1394                 }
1395
1396         return 0;
1397 }
1398
1399 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
1400                         struct bkey_s_c k,
1401                         struct inode_walker *inode,
1402                         struct snapshots_seen *s,
1403                         struct extent_ends *extent_ends)
1404 {
1405         struct bch_fs *c = trans->c;
1406         struct inode_walker_entry *i;
1407         struct printbuf buf = PRINTBUF;
1408         struct bpos equiv = k.k->p;
1409         int ret = 0;
1410
1411         equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1412
1413         ret = check_key_has_snapshot(trans, iter, k);
1414         if (ret) {
1415                 ret = ret < 0 ? ret : 0;
1416                 goto out;
1417         }
1418
1419         if (inode->last_pos.inode != k.k->p.inode) {
1420                 ret = check_i_sectors(trans, inode);
1421                 if (ret)
1422                         goto err;
1423         }
1424
1425         i = walk_inode(trans, inode, k);
1426         ret = PTR_ERR_OR_ZERO(i);
1427         if (ret)
1428                 goto err;
1429
1430         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1431         if (ret)
1432                 goto err;
1433
1434         if (k.k->type != KEY_TYPE_whiteout) {
1435                 if (fsck_err_on(!i, c, extent_in_missing_inode,
1436                                 "extent in missing inode:\n  %s",
1437                                 (printbuf_reset(&buf),
1438                                  bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1439                         goto delete;
1440
1441                 if (fsck_err_on(i &&
1442                                 !S_ISREG(i->inode.bi_mode) &&
1443                                 !S_ISLNK(i->inode.bi_mode),
1444                                 c, extent_in_non_reg_inode,
1445                                 "extent in non regular inode mode %o:\n  %s",
1446                                 i->inode.bi_mode,
1447                                 (printbuf_reset(&buf),
1448                                  bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1449                         goto delete;
1450
1451                 ret = check_overlapping_extents(trans, s, extent_ends, k,
1452                                                 equiv.snapshot, iter,
1453                                                 &inode->recalculate_sums);
1454                 if (ret)
1455                         goto err;
1456         }
1457
1458         /*
1459          * Check inodes in reverse order, from oldest snapshots to newest,
1460          * starting from the inode that matches this extent's snapshot. If we
1461          * didn't have one, iterate over all inodes:
1462          */
1463         if (!i)
1464                 i = inode->inodes.data + inode->inodes.nr - 1;
1465
1466         for (;
1467              inode->inodes.data && i >= inode->inodes.data;
1468              --i) {
1469                 if (i->snapshot > equiv.snapshot ||
1470                     !key_visible_in_snapshot(c, s, i->snapshot, equiv.snapshot))
1471                         continue;
1472
1473                 if (k.k->type != KEY_TYPE_whiteout) {
1474                         if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_size_dirty) &&
1475                                         k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9 &&
1476                                         !bkey_extent_is_reservation(k),
1477                                         c, extent_past_end_of_inode,
1478                                         "extent type past end of inode %llu:%u, i_size %llu\n  %s",
1479                                         i->inode.bi_inum, i->snapshot, i->inode.bi_size,
1480                                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1481                                 struct btree_iter iter2;
1482
1483                                 bch2_trans_copy_iter(&iter2, iter);
1484                                 bch2_btree_iter_set_snapshot(&iter2, i->snapshot);
1485                                 ret =   bch2_btree_iter_traverse(&iter2) ?:
1486                                         bch2_btree_delete_at(trans, &iter2,
1487                                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1488                                 bch2_trans_iter_exit(trans, &iter2);
1489                                 if (ret)
1490                                         goto err;
1491
1492                                 iter->k.type = KEY_TYPE_whiteout;
1493                         }
1494
1495                         if (bkey_extent_is_allocation(k.k))
1496                                 i->count += k.k->size;
1497                 }
1498
1499                 i->seen_this_pos = true;
1500         }
1501 out:
1502 err:
1503 fsck_err:
1504         printbuf_exit(&buf);
1505         bch_err_fn(c, ret);
1506         return ret;
1507 delete:
1508         ret = bch2_btree_delete_at(trans, iter, BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1509         goto out;
1510 }
1511
1512 /*
1513  * Walk extents: verify that extents have a corresponding S_ISREG inode, and
1514  * that i_size an i_sectors are consistent
1515  */
1516 int bch2_check_extents(struct bch_fs *c)
1517 {
1518         struct inode_walker w = inode_walker_init();
1519         struct snapshots_seen s;
1520         struct extent_ends extent_ends;
1521         struct disk_reservation res = { 0 };
1522
1523         snapshots_seen_init(&s);
1524         extent_ends_init(&extent_ends);
1525
1526         int ret = bch2_trans_run(c,
1527                 for_each_btree_key_commit(trans, iter, BTREE_ID_extents,
1528                                 POS(BCACHEFS_ROOT_INO, 0),
1529                                 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1530                                 &res, NULL,
1531                                 BCH_TRANS_COMMIT_no_enospc, ({
1532                         bch2_disk_reservation_put(c, &res);
1533                         check_extent(trans, &iter, k, &w, &s, &extent_ends) ?:
1534                         check_extent_overbig(trans, &iter, k);
1535                 })) ?:
1536                 check_i_sectors(trans, &w));
1537
1538         bch2_disk_reservation_put(c, &res);
1539         extent_ends_exit(&extent_ends);
1540         inode_walker_exit(&w);
1541         snapshots_seen_exit(&s);
1542
1543         bch_err_fn(c, ret);
1544         return ret;
1545 }
1546
1547 int bch2_check_indirect_extents(struct bch_fs *c)
1548 {
1549         struct disk_reservation res = { 0 };
1550
1551         int ret = bch2_trans_run(c,
1552                 for_each_btree_key_commit(trans, iter, BTREE_ID_reflink,
1553                                 POS_MIN,
1554                                 BTREE_ITER_PREFETCH, k,
1555                                 &res, NULL,
1556                                 BCH_TRANS_COMMIT_no_enospc, ({
1557                         bch2_disk_reservation_put(c, &res);
1558                         check_extent_overbig(trans, &iter, k);
1559                 })));
1560
1561         bch2_disk_reservation_put(c, &res);
1562         bch_err_fn(c, ret);
1563         return ret;
1564 }
1565
1566 static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
1567 {
1568         struct bch_fs *c = trans->c;
1569         u32 restart_count = trans->restart_count;
1570         int ret = 0;
1571         s64 count2;
1572
1573         darray_for_each(w->inodes, i) {
1574                 if (i->inode.bi_nlink == i->count)
1575                         continue;
1576
1577                 count2 = bch2_count_subdirs(trans, w->last_pos.inode, i->snapshot);
1578                 if (count2 < 0)
1579                         return count2;
1580
1581                 if (i->count != count2) {
1582                         bch_err(c, "fsck counted subdirectories wrong: got %llu should be %llu",
1583                                 i->count, count2);
1584                         i->count = count2;
1585                         if (i->inode.bi_nlink == i->count)
1586                                 continue;
1587                 }
1588
1589                 if (fsck_err_on(i->inode.bi_nlink != i->count,
1590                                 c, inode_dir_wrong_nlink,
1591                                 "directory %llu:%u with wrong i_nlink: got %u, should be %llu",
1592                                 w->last_pos.inode, i->snapshot, i->inode.bi_nlink, i->count)) {
1593                         i->inode.bi_nlink = i->count;
1594                         ret = bch2_fsck_write_inode(trans, &i->inode, i->snapshot);
1595                         if (ret)
1596                                 break;
1597                 }
1598         }
1599 fsck_err:
1600         bch_err_fn(c, ret);
1601         return ret ?: trans_was_restarted(trans, restart_count);
1602 }
1603
1604 static int check_dirent_inode_dirent(struct btree_trans *trans,
1605                                    struct btree_iter *iter,
1606                                    struct bkey_s_c_dirent d,
1607                                    struct bch_inode_unpacked *target,
1608                                    u32 target_snapshot)
1609 {
1610         struct bch_fs *c = trans->c;
1611         struct printbuf buf = PRINTBUF;
1612         int ret = 0;
1613
1614         if (inode_points_to_dirent(target, d))
1615                 return 0;
1616
1617         if (!target->bi_dir &&
1618             !target->bi_dir_offset) {
1619                 target->bi_dir          = d.k->p.inode;
1620                 target->bi_dir_offset   = d.k->p.offset;
1621                 return __bch2_fsck_write_inode(trans, target, target_snapshot);
1622         }
1623
1624         struct btree_iter bp_iter = { NULL };
1625         struct bkey_s_c_dirent bp_dirent = dirent_get_by_pos(trans, &bp_iter,
1626                               SPOS(target->bi_dir, target->bi_dir_offset, target_snapshot));
1627         ret = bkey_err(bp_dirent);
1628         if (ret && !bch2_err_matches(ret, ENOENT))
1629                 goto err;
1630
1631         bool backpointer_exists = !ret;
1632         ret = 0;
1633
1634         if (fsck_err_on(!backpointer_exists,
1635                         c, inode_wrong_backpointer,
1636                         "inode %llu:%u has wrong backpointer:\n"
1637                         "got       %llu:%llu\n"
1638                         "should be %llu:%llu",
1639                         target->bi_inum, target_snapshot,
1640                         target->bi_dir,
1641                         target->bi_dir_offset,
1642                         d.k->p.inode,
1643                         d.k->p.offset)) {
1644                 target->bi_dir          = d.k->p.inode;
1645                 target->bi_dir_offset   = d.k->p.offset;
1646                 ret = __bch2_fsck_write_inode(trans, target, target_snapshot);
1647                 goto out;
1648         }
1649
1650         bch2_bkey_val_to_text(&buf, c, d.s_c);
1651         prt_newline(&buf);
1652         if (backpointer_exists)
1653                 bch2_bkey_val_to_text(&buf, c, bp_dirent.s_c);
1654
1655         if (fsck_err_on(backpointer_exists &&
1656                         (S_ISDIR(target->bi_mode) ||
1657                          target->bi_subvol),
1658                         c, inode_dir_multiple_links,
1659                         "%s %llu:%u with multiple links\n%s",
1660                         S_ISDIR(target->bi_mode) ? "directory" : "subvolume",
1661                         target->bi_inum, target_snapshot, buf.buf)) {
1662                 ret = __remove_dirent(trans, d.k->p);
1663                 goto out;
1664         }
1665
1666         /*
1667          * hardlinked file with nlink 0:
1668          * We're just adjusting nlink here so check_nlinks() will pick
1669          * it up, it ignores inodes with nlink 0
1670          */
1671         if (fsck_err_on(backpointer_exists && !target->bi_nlink,
1672                         c, inode_multiple_links_but_nlink_0,
1673                         "inode %llu:%u type %s has multiple links but i_nlink 0\n%s",
1674                         target->bi_inum, target_snapshot, bch2_d_types[d.v->d_type], buf.buf)) {
1675                 target->bi_nlink++;
1676                 target->bi_flags &= ~BCH_INODE_unlinked;
1677                 ret = __bch2_fsck_write_inode(trans, target, target_snapshot);
1678                 if (ret)
1679                         goto err;
1680         }
1681 out:
1682 err:
1683 fsck_err:
1684         bch2_trans_iter_exit(trans, &bp_iter);
1685         printbuf_exit(&buf);
1686         bch_err_fn(c, ret);
1687         return ret;
1688 }
1689
1690 static int check_dirent_target(struct btree_trans *trans,
1691                                struct btree_iter *iter,
1692                                struct bkey_s_c_dirent d,
1693                                struct bch_inode_unpacked *target,
1694                                u32 target_snapshot)
1695 {
1696         struct bch_fs *c = trans->c;
1697         struct bkey_i_dirent *n;
1698         struct printbuf buf = PRINTBUF;
1699         int ret = 0;
1700
1701         ret = check_dirent_inode_dirent(trans, iter, d, target, target_snapshot);
1702         if (ret)
1703                 goto err;
1704
1705         if (fsck_err_on(d.v->d_type != inode_d_type(target),
1706                         c, dirent_d_type_wrong,
1707                         "incorrect d_type: got %s, should be %s:\n%s",
1708                         bch2_d_type_str(d.v->d_type),
1709                         bch2_d_type_str(inode_d_type(target)),
1710                         (printbuf_reset(&buf),
1711                          bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
1712                 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1713                 ret = PTR_ERR_OR_ZERO(n);
1714                 if (ret)
1715                         goto err;
1716
1717                 bkey_reassemble(&n->k_i, d.s_c);
1718                 n->v.d_type = inode_d_type(target);
1719                 if (n->v.d_type == DT_SUBVOL) {
1720                         n->v.d_parent_subvol = cpu_to_le32(target->bi_parent_subvol);
1721                         n->v.d_child_subvol = cpu_to_le32(target->bi_subvol);
1722                 } else {
1723                         n->v.d_inum = cpu_to_le64(target->bi_inum);
1724                 }
1725
1726                 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1727                 if (ret)
1728                         goto err;
1729
1730                 d = dirent_i_to_s_c(n);
1731         }
1732 err:
1733 fsck_err:
1734         printbuf_exit(&buf);
1735         bch_err_fn(c, ret);
1736         return ret;
1737 }
1738
1739 /* find a subvolume that's a descendent of @snapshot: */
1740 static int find_snapshot_subvol(struct btree_trans *trans, u32 snapshot, u32 *subvolid)
1741 {
1742         struct btree_iter iter;
1743         struct bkey_s_c k;
1744         int ret;
1745
1746         for_each_btree_key_norestart(trans, iter, BTREE_ID_subvolumes, POS_MIN, 0, k, ret) {
1747                 if (k.k->type != KEY_TYPE_subvolume)
1748                         continue;
1749
1750                 struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
1751                 if (bch2_snapshot_is_ancestor(trans->c, le32_to_cpu(s.v->snapshot), snapshot)) {
1752                         bch2_trans_iter_exit(trans, &iter);
1753                         *subvolid = k.k->p.offset;
1754                         goto found;
1755                 }
1756         }
1757         if (!ret)
1758                 ret = -ENOENT;
1759 found:
1760         bch2_trans_iter_exit(trans, &iter);
1761         return ret;
1762 }
1763
1764 static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter *iter,
1765                                   struct bkey_s_c_dirent d)
1766 {
1767         struct bch_fs *c = trans->c;
1768         struct btree_iter subvol_iter = {};
1769         struct bch_inode_unpacked subvol_root;
1770         u32 parent_subvol = le32_to_cpu(d.v->d_parent_subvol);
1771         u32 target_subvol = le32_to_cpu(d.v->d_child_subvol);
1772         u32 parent_snapshot;
1773         u64 parent_inum;
1774         struct printbuf buf = PRINTBUF;
1775         int ret = 0;
1776
1777         ret = subvol_lookup(trans, parent_subvol, &parent_snapshot, &parent_inum);
1778         if (ret && !bch2_err_matches(ret, ENOENT))
1779                 return ret;
1780
1781         if (fsck_err_on(ret, c, dirent_to_missing_parent_subvol,
1782                         "dirent parent_subvol points to missing subvolume\n%s",
1783                         (bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf)) ||
1784             fsck_err_on(!ret && !bch2_snapshot_is_ancestor(c, parent_snapshot, d.k->p.snapshot),
1785                         c, dirent_not_visible_in_parent_subvol,
1786                         "dirent not visible in parent_subvol (not an ancestor of subvol snap %u)\n%s",
1787                         parent_snapshot,
1788                         (bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
1789                 u32 new_parent_subvol;
1790                 ret = find_snapshot_subvol(trans, d.k->p.snapshot, &new_parent_subvol);
1791                 if (ret)
1792                         goto err;
1793
1794                 struct bkey_i_dirent *new_dirent = bch2_bkey_make_mut_typed(trans, iter, &d.s_c, 0, dirent);
1795                 ret = PTR_ERR_OR_ZERO(new_dirent);
1796                 if (ret)
1797                         goto err;
1798
1799                 new_dirent->v.d_parent_subvol = cpu_to_le32(new_parent_subvol);
1800         }
1801
1802         struct bkey_s_c_subvolume s =
1803                 bch2_bkey_get_iter_typed(trans, &subvol_iter,
1804                                          BTREE_ID_subvolumes, POS(0, target_subvol),
1805                                          0, subvolume);
1806         ret = bkey_err(s.s_c);
1807         if (ret && !bch2_err_matches(ret, ENOENT))
1808                 return ret;
1809
1810         if (ret) {
1811                 if (fsck_err(c, dirent_to_missing_subvol,
1812                              "dirent points to missing subvolume\n%s",
1813                              (bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf)))
1814                         return __remove_dirent(trans, d.k->p);
1815                 ret = 0;
1816                 goto out;
1817         }
1818
1819         if (fsck_err_on(le32_to_cpu(s.v->fs_path_parent) != parent_subvol,
1820                         c, subvol_fs_path_parent_wrong,
1821                         "subvol with wrong fs_path_parent, should be be %u\n%s",
1822                         parent_subvol,
1823                         (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
1824                 struct bkey_i_subvolume *n =
1825                         bch2_bkey_make_mut_typed(trans, &subvol_iter, &s.s_c, 0, subvolume);
1826                 ret = PTR_ERR_OR_ZERO(n);
1827                 if (ret)
1828                         goto err;
1829
1830                 n->v.fs_path_parent = cpu_to_le32(parent_subvol);
1831         }
1832
1833         u64 target_inum = le64_to_cpu(s.v->inode);
1834         u32 target_snapshot = le32_to_cpu(s.v->snapshot);
1835
1836         ret = lookup_inode(trans, target_inum, &subvol_root, &target_snapshot);
1837         if (ret && !bch2_err_matches(ret, ENOENT))
1838                 return ret;
1839
1840         if (fsck_err_on(parent_subvol != subvol_root.bi_parent_subvol,
1841                         c, inode_bi_parent_wrong,
1842                         "subvol root %llu has wrong bi_parent_subvol: got %u, should be %u",
1843                         target_inum,
1844                         subvol_root.bi_parent_subvol, parent_subvol)) {
1845                 subvol_root.bi_parent_subvol = parent_subvol;
1846                 ret = __bch2_fsck_write_inode(trans, &subvol_root, target_snapshot);
1847                 if (ret)
1848                         return ret;
1849         }
1850
1851         ret = check_dirent_target(trans, iter, d, &subvol_root,
1852                                   target_snapshot);
1853         if (ret)
1854                 return ret;
1855 out:
1856 err:
1857 fsck_err:
1858         bch2_trans_iter_exit(trans, &subvol_iter);
1859         printbuf_exit(&buf);
1860         return ret;
1861 }
1862
1863 static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
1864                         struct bkey_s_c k,
1865                         struct bch_hash_info *hash_info,
1866                         struct inode_walker *dir,
1867                         struct inode_walker *target,
1868                         struct snapshots_seen *s)
1869 {
1870         struct bch_fs *c = trans->c;
1871         struct bkey_s_c_dirent d;
1872         struct inode_walker_entry *i;
1873         struct printbuf buf = PRINTBUF;
1874         struct bpos equiv;
1875         int ret = 0;
1876
1877         ret = check_key_has_snapshot(trans, iter, k);
1878         if (ret) {
1879                 ret = ret < 0 ? ret : 0;
1880                 goto out;
1881         }
1882
1883         equiv = k.k->p;
1884         equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1885
1886         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1887         if (ret)
1888                 goto err;
1889
1890         if (k.k->type == KEY_TYPE_whiteout)
1891                 goto out;
1892
1893         if (dir->last_pos.inode != k.k->p.inode) {
1894                 ret = check_subdir_count(trans, dir);
1895                 if (ret)
1896                         goto err;
1897         }
1898
1899         BUG_ON(!btree_iter_path(trans, iter)->should_be_locked);
1900
1901         i = walk_inode(trans, dir, k);
1902         ret = PTR_ERR_OR_ZERO(i);
1903         if (ret < 0)
1904                 goto err;
1905
1906         if (dir->first_this_inode && dir->inodes.nr)
1907                 *hash_info = bch2_hash_info_init(c, &dir->inodes.data[0].inode);
1908         dir->first_this_inode = false;
1909
1910         if (fsck_err_on(!i, c, dirent_in_missing_dir_inode,
1911                         "dirent in nonexisting directory:\n%s",
1912                         (printbuf_reset(&buf),
1913                          bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1914                 ret = bch2_btree_delete_at(trans, iter,
1915                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1916                 goto out;
1917         }
1918
1919         if (!i)
1920                 goto out;
1921
1922         if (fsck_err_on(!S_ISDIR(i->inode.bi_mode),
1923                         c, dirent_in_non_dir_inode,
1924                         "dirent in non directory inode type %s:\n%s",
1925                         bch2_d_type_str(inode_d_type(&i->inode)),
1926                         (printbuf_reset(&buf),
1927                          bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1928                 ret = bch2_btree_delete_at(trans, iter, 0);
1929                 goto out;
1930         }
1931
1932         ret = hash_check_key(trans, bch2_dirent_hash_desc, hash_info, iter, k);
1933         if (ret < 0)
1934                 goto err;
1935         if (ret) {
1936                 /* dirent has been deleted */
1937                 ret = 0;
1938                 goto out;
1939         }
1940
1941         if (k.k->type != KEY_TYPE_dirent)
1942                 goto out;
1943
1944         d = bkey_s_c_to_dirent(k);
1945
1946         if (d.v->d_type == DT_SUBVOL) {
1947                 ret = check_dirent_to_subvol(trans, iter, d);
1948                 if (ret)
1949                         goto err;
1950         } else {
1951                 ret = __get_visible_inodes(trans, target, s, le64_to_cpu(d.v->d_inum));
1952                 if (ret)
1953                         goto err;
1954
1955                 if (fsck_err_on(!target->inodes.nr,
1956                                 c, dirent_to_missing_inode,
1957                                 "dirent points to missing inode: (equiv %u)\n%s",
1958                                 equiv.snapshot,
1959                                 (printbuf_reset(&buf),
1960                                  bch2_bkey_val_to_text(&buf, c, k),
1961                                  buf.buf))) {
1962                         ret = __remove_dirent(trans, d.k->p);
1963                         if (ret)
1964                                 goto err;
1965                 }
1966
1967                 darray_for_each(target->inodes, i) {
1968                         ret = check_dirent_target(trans, iter, d,
1969                                                   &i->inode, i->snapshot);
1970                         if (ret)
1971                                 goto err;
1972                 }
1973
1974                 if (d.v->d_type == DT_DIR)
1975                         for_each_visible_inode(c, s, dir, equiv.snapshot, i)
1976                                 i->count++;
1977         }
1978 out:
1979 err:
1980 fsck_err:
1981         printbuf_exit(&buf);
1982         bch_err_fn(c, ret);
1983         return ret;
1984 }
1985
1986 /*
1987  * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
1988  * validate d_type
1989  */
1990 int bch2_check_dirents(struct bch_fs *c)
1991 {
1992         struct inode_walker dir = inode_walker_init();
1993         struct inode_walker target = inode_walker_init();
1994         struct snapshots_seen s;
1995         struct bch_hash_info hash_info;
1996
1997         snapshots_seen_init(&s);
1998
1999         int ret = bch2_trans_run(c,
2000                 for_each_btree_key_commit(trans, iter, BTREE_ID_dirents,
2001                                 POS(BCACHEFS_ROOT_INO, 0),
2002                                 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
2003                                 k,
2004                                 NULL, NULL,
2005                                 BCH_TRANS_COMMIT_no_enospc,
2006                         check_dirent(trans, &iter, k, &hash_info, &dir, &target, &s)));
2007
2008         snapshots_seen_exit(&s);
2009         inode_walker_exit(&dir);
2010         inode_walker_exit(&target);
2011         bch_err_fn(c, ret);
2012         return ret;
2013 }
2014
2015 static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
2016                        struct bkey_s_c k,
2017                        struct bch_hash_info *hash_info,
2018                        struct inode_walker *inode)
2019 {
2020         struct bch_fs *c = trans->c;
2021         struct inode_walker_entry *i;
2022         int ret;
2023
2024         ret = check_key_has_snapshot(trans, iter, k);
2025         if (ret)
2026                 return ret;
2027
2028         i = walk_inode(trans, inode, k);
2029         ret = PTR_ERR_OR_ZERO(i);
2030         if (ret)
2031                 return ret;
2032
2033         if (inode->first_this_inode && inode->inodes.nr)
2034                 *hash_info = bch2_hash_info_init(c, &inode->inodes.data[0].inode);
2035         inode->first_this_inode = false;
2036
2037         if (fsck_err_on(!i, c, xattr_in_missing_inode,
2038                         "xattr for missing inode %llu",
2039                         k.k->p.inode))
2040                 return bch2_btree_delete_at(trans, iter, 0);
2041
2042         if (!i)
2043                 return 0;
2044
2045         ret = hash_check_key(trans, bch2_xattr_hash_desc, hash_info, iter, k);
2046 fsck_err:
2047         bch_err_fn(c, ret);
2048         return ret;
2049 }
2050
2051 /*
2052  * Walk xattrs: verify that they all have a corresponding inode
2053  */
2054 int bch2_check_xattrs(struct bch_fs *c)
2055 {
2056         struct inode_walker inode = inode_walker_init();
2057         struct bch_hash_info hash_info;
2058         int ret = 0;
2059
2060         ret = bch2_trans_run(c,
2061                 for_each_btree_key_commit(trans, iter, BTREE_ID_xattrs,
2062                         POS(BCACHEFS_ROOT_INO, 0),
2063                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
2064                         k,
2065                         NULL, NULL,
2066                         BCH_TRANS_COMMIT_no_enospc,
2067                 check_xattr(trans, &iter, k, &hash_info, &inode)));
2068         bch_err_fn(c, ret);
2069         return ret;
2070 }
2071
2072 static int check_root_trans(struct btree_trans *trans)
2073 {
2074         struct bch_fs *c = trans->c;
2075         struct bch_inode_unpacked root_inode;
2076         u32 snapshot;
2077         u64 inum;
2078         int ret;
2079
2080         ret = subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
2081         if (ret && !bch2_err_matches(ret, ENOENT))
2082                 return ret;
2083
2084         if (mustfix_fsck_err_on(ret, c, root_subvol_missing,
2085                                 "root subvol missing")) {
2086                 struct bkey_i_subvolume root_subvol;
2087
2088                 snapshot        = U32_MAX;
2089                 inum            = BCACHEFS_ROOT_INO;
2090
2091                 bkey_subvolume_init(&root_subvol.k_i);
2092                 root_subvol.k.p.offset = BCACHEFS_ROOT_SUBVOL;
2093                 root_subvol.v.flags     = 0;
2094                 root_subvol.v.snapshot  = cpu_to_le32(snapshot);
2095                 root_subvol.v.inode     = cpu_to_le64(inum);
2096                 ret = bch2_btree_insert_trans(trans, BTREE_ID_subvolumes, &root_subvol.k_i, 0);
2097                 bch_err_msg(c, ret, "writing root subvol");
2098                 if (ret)
2099                         goto err;
2100         }
2101
2102         ret = lookup_inode(trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
2103         if (ret && !bch2_err_matches(ret, ENOENT))
2104                 return ret;
2105
2106         if (mustfix_fsck_err_on(ret, c, root_dir_missing,
2107                                 "root directory missing") ||
2108             mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode),
2109                                 c, root_inode_not_dir,
2110                                 "root inode not a directory")) {
2111                 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755,
2112                                 0, NULL);
2113                 root_inode.bi_inum = inum;
2114
2115                 ret = __bch2_fsck_write_inode(trans, &root_inode, snapshot);
2116                 bch_err_msg(c, ret, "writing root inode");
2117         }
2118 err:
2119 fsck_err:
2120         return ret;
2121 }
2122
2123 /* Get root directory, create if it doesn't exist: */
2124 int bch2_check_root(struct bch_fs *c)
2125 {
2126         int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2127                 check_root_trans(trans));
2128         bch_err_fn(c, ret);
2129         return ret;
2130 }
2131
2132 typedef DARRAY(u32) darray_u32;
2133
2134 static bool darray_u32_has(darray_u32 *d, u32 v)
2135 {
2136         darray_for_each(*d, i)
2137                 if (*i == v)
2138                         return true;
2139         return false;
2140 }
2141
2142 /*
2143  * We've checked that inode backpointers point to valid dirents; here, it's
2144  * sufficient to check that the subvolume root has a dirent:
2145  */
2146 static int subvol_has_dirent(struct btree_trans *trans, struct bkey_s_c_subvolume s)
2147 {
2148         struct bch_inode_unpacked inode;
2149         int ret = bch2_inode_find_by_inum_trans(trans,
2150                                 (subvol_inum) { s.k->p.offset, le64_to_cpu(s.v->inode) },
2151                                 &inode);
2152         if (ret)
2153                 return ret;
2154
2155         return inode.bi_dir != 0;
2156 }
2157
2158 static int check_subvol_path(struct btree_trans *trans, struct btree_iter *iter, struct bkey_s_c k)
2159 {
2160         struct bch_fs *c = trans->c;
2161         struct btree_iter parent_iter = {};
2162         darray_u32 subvol_path = {};
2163         struct printbuf buf = PRINTBUF;
2164         int ret = 0;
2165
2166         if (k.k->type != KEY_TYPE_subvolume)
2167                 return 0;
2168
2169         while (k.k->p.offset != BCACHEFS_ROOT_SUBVOL) {
2170                 ret = darray_push(&subvol_path, k.k->p.offset);
2171                 if (ret)
2172                         goto err;
2173
2174                 struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
2175
2176                 ret = subvol_has_dirent(trans, s);
2177                 if (ret < 0)
2178                         break;
2179
2180                 if (fsck_err_on(!ret,
2181                                 c, subvol_unreachable,
2182                                 "unreachable subvolume %s",
2183                                 (bch2_bkey_val_to_text(&buf, c, s.s_c),
2184                                  buf.buf))) {
2185                         ret = reattach_subvol(trans, s);
2186                         break;
2187                 }
2188
2189                 u32 parent = le32_to_cpu(s.v->fs_path_parent);
2190
2191                 if (darray_u32_has(&subvol_path, parent)) {
2192                         if (fsck_err(c, subvol_loop, "subvolume loop"))
2193                                 ret = reattach_subvol(trans, s);
2194                         break;
2195                 }
2196
2197                 bch2_trans_iter_exit(trans, &parent_iter);
2198                 bch2_trans_iter_init(trans, &parent_iter,
2199                                      BTREE_ID_subvolumes, POS(0, parent), 0);
2200                 k = bch2_btree_iter_peek_slot(&parent_iter);
2201                 ret = bkey_err(k);
2202                 if (ret)
2203                         goto err;
2204
2205                 if (fsck_err_on(k.k->type != KEY_TYPE_subvolume,
2206                                 c, subvol_unreachable,
2207                                 "unreachable subvolume %s",
2208                                 (bch2_bkey_val_to_text(&buf, c, s.s_c),
2209                                  buf.buf))) {
2210                         ret = reattach_subvol(trans, s);
2211                         break;
2212                 }
2213         }
2214 fsck_err:
2215 err:
2216         printbuf_exit(&buf);
2217         darray_exit(&subvol_path);
2218         bch2_trans_iter_exit(trans, &parent_iter);
2219         return ret;
2220 }
2221
2222 int bch2_check_subvolume_structure(struct bch_fs *c)
2223 {
2224         int ret = bch2_trans_run(c,
2225                 for_each_btree_key_commit(trans, iter,
2226                                 BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k,
2227                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2228                         check_subvol_path(trans, &iter, k)));
2229         bch_err_fn(c, ret);
2230         return ret;
2231 }
2232
2233 struct pathbuf_entry {
2234         u64     inum;
2235         u32     snapshot;
2236 };
2237
2238 typedef DARRAY(struct pathbuf_entry) pathbuf;
2239
2240 static bool path_is_dup(pathbuf *p, u64 inum, u32 snapshot)
2241 {
2242         darray_for_each(*p, i)
2243                 if (i->inum     == inum &&
2244                     i->snapshot == snapshot)
2245                         return true;
2246         return false;
2247 }
2248
2249 /*
2250  * Check that a given inode is reachable from its subvolume root - we already
2251  * verified subvolume connectivity:
2252  *
2253  * XXX: we should also be verifying that inodes are in the right subvolumes
2254  */
2255 static int check_path(struct btree_trans *trans, pathbuf *p, struct bkey_s_c inode_k)
2256 {
2257         struct bch_fs *c = trans->c;
2258         struct btree_iter inode_iter = {};
2259         struct bch_inode_unpacked inode;
2260         struct printbuf buf = PRINTBUF;
2261         u32 snapshot = bch2_snapshot_equiv(c, inode_k.k->p.snapshot);
2262         int ret = 0;
2263
2264         p->nr = 0;
2265
2266         BUG_ON(bch2_inode_unpack(inode_k, &inode));
2267
2268         while (!inode.bi_subvol) {
2269                 struct btree_iter dirent_iter;
2270                 struct bkey_s_c_dirent d;
2271                 u32 parent_snapshot = snapshot;
2272
2273                 d = inode_get_dirent(trans, &dirent_iter, &inode, &parent_snapshot);
2274                 ret = bkey_err(d.s_c);
2275                 if (ret && !bch2_err_matches(ret, ENOENT))
2276                         break;
2277
2278                 if (!ret && !dirent_points_to_inode(d, &inode)) {
2279                         bch2_trans_iter_exit(trans, &dirent_iter);
2280                         ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
2281                 }
2282
2283                 if (bch2_err_matches(ret, ENOENT)) {
2284                         ret = 0;
2285                         if (fsck_err(c, inode_unreachable,
2286                                      "unreachable inode\n%s",
2287                                      (printbuf_reset(&buf),
2288                                       bch2_bkey_val_to_text(&buf, c, inode_k),
2289                                       buf.buf)))
2290                                 ret = reattach_inode(trans, &inode, snapshot);
2291                         goto out;
2292                 }
2293
2294                 bch2_trans_iter_exit(trans, &dirent_iter);
2295
2296                 if (!S_ISDIR(inode.bi_mode))
2297                         break;
2298
2299                 ret = darray_push(p, ((struct pathbuf_entry) {
2300                         .inum           = inode.bi_inum,
2301                         .snapshot       = snapshot,
2302                 }));
2303                 if (ret)
2304                         return ret;
2305
2306                 snapshot = parent_snapshot;
2307
2308                 bch2_trans_iter_exit(trans, &inode_iter);
2309                 inode_k = bch2_bkey_get_iter(trans, &inode_iter, BTREE_ID_inodes,
2310                                              SPOS(0, inode.bi_dir, snapshot), 0);
2311                 ret = bkey_err(inode_k) ?:
2312                         !bkey_is_inode(inode_k.k) ? -BCH_ERR_ENOENT_inode
2313                         : bch2_inode_unpack(inode_k, &inode);
2314                 if (ret) {
2315                         /* Should have been caught in dirents pass */
2316                         if (!bch2_err_matches(ret, BCH_ERR_transaction_restart))
2317                                 bch_err(c, "error looking up parent directory: %i", ret);
2318                         break;
2319                 }
2320
2321                 snapshot = inode_k.k->p.snapshot;
2322
2323                 if (path_is_dup(p, inode.bi_inum, snapshot)) {
2324                         /* XXX print path */
2325                         bch_err(c, "directory structure loop");
2326
2327                         darray_for_each(*p, i)
2328                                 pr_err("%llu:%u", i->inum, i->snapshot);
2329                         pr_err("%llu:%u", inode.bi_inum, snapshot);
2330
2331                         if (fsck_err(c, dir_loop, "directory structure loop")) {
2332                                 ret = remove_backpointer(trans, &inode);
2333                                 bch_err_msg(c, ret, "removing dirent");
2334                                 if (ret)
2335                                         break;
2336
2337                                 ret = reattach_inode(trans, &inode, snapshot);
2338                                 bch_err_msg(c, ret, "reattaching inode %llu", inode.bi_inum);
2339                         }
2340                         break;
2341                 }
2342         }
2343 out:
2344 fsck_err:
2345         bch2_trans_iter_exit(trans, &inode_iter);
2346         printbuf_exit(&buf);
2347         bch_err_fn(c, ret);
2348         return ret;
2349 }
2350
2351 /*
2352  * Check for unreachable inodes, as well as loops in the directory structure:
2353  * After bch2_check_dirents(), if an inode backpointer doesn't exist that means it's
2354  * unreachable:
2355  */
2356 int bch2_check_directory_structure(struct bch_fs *c)
2357 {
2358         pathbuf path = { 0, };
2359         int ret;
2360
2361         ret = bch2_trans_run(c,
2362                 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes, POS_MIN,
2363                                           BTREE_ITER_INTENT|
2364                                           BTREE_ITER_PREFETCH|
2365                                           BTREE_ITER_ALL_SNAPSHOTS, k,
2366                                           NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
2367                         if (!bkey_is_inode(k.k))
2368                                 continue;
2369
2370                         if (bch2_inode_flags(k) & BCH_INODE_unlinked)
2371                                 continue;
2372
2373                         check_path(trans, &path, k);
2374                 })));
2375         darray_exit(&path);
2376
2377         bch_err_fn(c, ret);
2378         return ret;
2379 }
2380
2381 struct nlink_table {
2382         size_t          nr;
2383         size_t          size;
2384
2385         struct nlink {
2386                 u64     inum;
2387                 u32     snapshot;
2388                 u32     count;
2389         }               *d;
2390 };
2391
2392 static int add_nlink(struct bch_fs *c, struct nlink_table *t,
2393                      u64 inum, u32 snapshot)
2394 {
2395         if (t->nr == t->size) {
2396                 size_t new_size = max_t(size_t, 128UL, t->size * 2);
2397                 void *d = kvmalloc_array(new_size, sizeof(t->d[0]), GFP_KERNEL);
2398
2399                 if (!d) {
2400                         bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
2401                                 new_size);
2402                         return -BCH_ERR_ENOMEM_fsck_add_nlink;
2403                 }
2404
2405                 if (t->d)
2406                         memcpy(d, t->d, t->size * sizeof(t->d[0]));
2407                 kvfree(t->d);
2408
2409                 t->d = d;
2410                 t->size = new_size;
2411         }
2412
2413
2414         t->d[t->nr++] = (struct nlink) {
2415                 .inum           = inum,
2416                 .snapshot       = snapshot,
2417         };
2418
2419         return 0;
2420 }
2421
2422 static int nlink_cmp(const void *_l, const void *_r)
2423 {
2424         const struct nlink *l = _l;
2425         const struct nlink *r = _r;
2426
2427         return cmp_int(l->inum, r->inum);
2428 }
2429
2430 static void inc_link(struct bch_fs *c, struct snapshots_seen *s,
2431                      struct nlink_table *links,
2432                      u64 range_start, u64 range_end, u64 inum, u32 snapshot)
2433 {
2434         struct nlink *link, key = {
2435                 .inum = inum, .snapshot = U32_MAX,
2436         };
2437
2438         if (inum < range_start || inum >= range_end)
2439                 return;
2440
2441         link = __inline_bsearch(&key, links->d, links->nr,
2442                                 sizeof(links->d[0]), nlink_cmp);
2443         if (!link)
2444                 return;
2445
2446         while (link > links->d && link[0].inum == link[-1].inum)
2447                 --link;
2448
2449         for (; link < links->d + links->nr && link->inum == inum; link++)
2450                 if (ref_visible(c, s, snapshot, link->snapshot)) {
2451                         link->count++;
2452                         if (link->snapshot >= snapshot)
2453                                 break;
2454                 }
2455 }
2456
2457 noinline_for_stack
2458 static int check_nlinks_find_hardlinks(struct bch_fs *c,
2459                                        struct nlink_table *t,
2460                                        u64 start, u64 *end)
2461 {
2462         int ret = bch2_trans_run(c,
2463                 for_each_btree_key(trans, iter, BTREE_ID_inodes,
2464                                    POS(0, start),
2465                                    BTREE_ITER_INTENT|
2466                                    BTREE_ITER_PREFETCH|
2467                                    BTREE_ITER_ALL_SNAPSHOTS, k, ({
2468                         if (!bkey_is_inode(k.k))
2469                                 continue;
2470
2471                         /* Should never fail, checked by bch2_inode_invalid: */
2472                         struct bch_inode_unpacked u;
2473                         BUG_ON(bch2_inode_unpack(k, &u));
2474
2475                         /*
2476                          * Backpointer and directory structure checks are sufficient for
2477                          * directories, since they can't have hardlinks:
2478                          */
2479                         if (S_ISDIR(u.bi_mode))
2480                                 continue;
2481
2482                         if (!u.bi_nlink)
2483                                 continue;
2484
2485                         ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot);
2486                         if (ret) {
2487                                 *end = k.k->p.offset;
2488                                 ret = 0;
2489                                 break;
2490                         }
2491                         0;
2492                 })));
2493
2494         bch_err_fn(c, ret);
2495         return ret;
2496 }
2497
2498 noinline_for_stack
2499 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
2500                                      u64 range_start, u64 range_end)
2501 {
2502         struct snapshots_seen s;
2503
2504         snapshots_seen_init(&s);
2505
2506         int ret = bch2_trans_run(c,
2507                 for_each_btree_key(trans, iter, BTREE_ID_dirents, POS_MIN,
2508                                    BTREE_ITER_INTENT|
2509                                    BTREE_ITER_PREFETCH|
2510                                    BTREE_ITER_ALL_SNAPSHOTS, k, ({
2511                         ret = snapshots_seen_update(c, &s, iter.btree_id, k.k->p);
2512                         if (ret)
2513                                 break;
2514
2515                         if (k.k->type == KEY_TYPE_dirent) {
2516                                 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
2517
2518                                 if (d.v->d_type != DT_DIR &&
2519                                     d.v->d_type != DT_SUBVOL)
2520                                         inc_link(c, &s, links, range_start, range_end,
2521                                                  le64_to_cpu(d.v->d_inum),
2522                                                  bch2_snapshot_equiv(c, d.k->p.snapshot));
2523                         }
2524                         0;
2525                 })));
2526
2527         snapshots_seen_exit(&s);
2528
2529         bch_err_fn(c, ret);
2530         return ret;
2531 }
2532
2533 static int check_nlinks_update_inode(struct btree_trans *trans, struct btree_iter *iter,
2534                                      struct bkey_s_c k,
2535                                      struct nlink_table *links,
2536                                      size_t *idx, u64 range_end)
2537 {
2538         struct bch_fs *c = trans->c;
2539         struct bch_inode_unpacked u;
2540         struct nlink *link = &links->d[*idx];
2541         int ret = 0;
2542
2543         if (k.k->p.offset >= range_end)
2544                 return 1;
2545
2546         if (!bkey_is_inode(k.k))
2547                 return 0;
2548
2549         BUG_ON(bch2_inode_unpack(k, &u));
2550
2551         if (S_ISDIR(u.bi_mode))
2552                 return 0;
2553
2554         if (!u.bi_nlink)
2555                 return 0;
2556
2557         while ((cmp_int(link->inum, k.k->p.offset) ?:
2558                 cmp_int(link->snapshot, k.k->p.snapshot)) < 0) {
2559                 BUG_ON(*idx == links->nr);
2560                 link = &links->d[++*idx];
2561         }
2562
2563         if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count,
2564                         c, inode_wrong_nlink,
2565                         "inode %llu type %s has wrong i_nlink (%u, should be %u)",
2566                         u.bi_inum, bch2_d_types[mode_to_type(u.bi_mode)],
2567                         bch2_inode_nlink_get(&u), link->count)) {
2568                 bch2_inode_nlink_set(&u, link->count);
2569                 ret = __bch2_fsck_write_inode(trans, &u, k.k->p.snapshot);
2570         }
2571 fsck_err:
2572         return ret;
2573 }
2574
2575 noinline_for_stack
2576 static int check_nlinks_update_hardlinks(struct bch_fs *c,
2577                                struct nlink_table *links,
2578                                u64 range_start, u64 range_end)
2579 {
2580         size_t idx = 0;
2581
2582         int ret = bch2_trans_run(c,
2583                 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
2584                                 POS(0, range_start),
2585                                 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
2586                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2587                         check_nlinks_update_inode(trans, &iter, k, links, &idx, range_end)));
2588         if (ret < 0) {
2589                 bch_err(c, "error in fsck walking inodes: %s", bch2_err_str(ret));
2590                 return ret;
2591         }
2592
2593         return 0;
2594 }
2595
2596 int bch2_check_nlinks(struct bch_fs *c)
2597 {
2598         struct nlink_table links = { 0 };
2599         u64 this_iter_range_start, next_iter_range_start = 0;
2600         int ret = 0;
2601
2602         do {
2603                 this_iter_range_start = next_iter_range_start;
2604                 next_iter_range_start = U64_MAX;
2605
2606                 ret = check_nlinks_find_hardlinks(c, &links,
2607                                                   this_iter_range_start,
2608                                                   &next_iter_range_start);
2609
2610                 ret = check_nlinks_walk_dirents(c, &links,
2611                                           this_iter_range_start,
2612                                           next_iter_range_start);
2613                 if (ret)
2614                         break;
2615
2616                 ret = check_nlinks_update_hardlinks(c, &links,
2617                                          this_iter_range_start,
2618                                          next_iter_range_start);
2619                 if (ret)
2620                         break;
2621
2622                 links.nr = 0;
2623         } while (next_iter_range_start != U64_MAX);
2624
2625         kvfree(links.d);
2626         bch_err_fn(c, ret);
2627         return ret;
2628 }
2629
2630 static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter,
2631                              struct bkey_s_c k)
2632 {
2633         struct bkey_s_c_reflink_p p;
2634         struct bkey_i_reflink_p *u;
2635
2636         if (k.k->type != KEY_TYPE_reflink_p)
2637                 return 0;
2638
2639         p = bkey_s_c_to_reflink_p(k);
2640
2641         if (!p.v->front_pad && !p.v->back_pad)
2642                 return 0;
2643
2644         u = bch2_trans_kmalloc(trans, sizeof(*u));
2645         int ret = PTR_ERR_OR_ZERO(u);
2646         if (ret)
2647                 return ret;
2648
2649         bkey_reassemble(&u->k_i, k);
2650         u->v.front_pad  = 0;
2651         u->v.back_pad   = 0;
2652
2653         return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_NORUN);
2654 }
2655
2656 int bch2_fix_reflink_p(struct bch_fs *c)
2657 {
2658         if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
2659                 return 0;
2660
2661         int ret = bch2_trans_run(c,
2662                 for_each_btree_key_commit(trans, iter,
2663                                 BTREE_ID_extents, POS_MIN,
2664                                 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|
2665                                 BTREE_ITER_ALL_SNAPSHOTS, k,
2666                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2667                         fix_reflink_p_key(trans, &iter, k)));
2668         bch_err_fn(c, ret);
2669         return ret;
2670 }