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_notnested(struct btree_trans *trans, struct inode_walker *w)
1118 {
1119         struct bch_fs *c = trans->c;
1120         int ret = 0;
1121         s64 count2;
1122
1123         darray_for_each(w->inodes, i) {
1124                 if (i->inode.bi_sectors == i->count)
1125                         continue;
1126
1127                 count2 = bch2_count_inode_sectors(trans, w->last_pos.inode, i->snapshot);
1128
1129                 if (w->recalculate_sums)
1130                         i->count = count2;
1131
1132                 if (i->count != count2) {
1133                         bch_err(c, "fsck counted i_sectors wrong for inode %llu:%u: got %llu should be %llu",
1134                                 w->last_pos.inode, i->snapshot, i->count, count2);
1135                         return -BCH_ERR_internal_fsck_err;
1136                 }
1137
1138                 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_sectors_dirty),
1139                                 c, inode_i_sectors_wrong,
1140                                 "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
1141                                 w->last_pos.inode, i->snapshot,
1142                                 i->inode.bi_sectors, i->count)) {
1143                         i->inode.bi_sectors = i->count;
1144                         ret = bch2_fsck_write_inode(trans, &i->inode, i->snapshot);
1145                         if (ret)
1146                                 break;
1147                 }
1148         }
1149 fsck_err:
1150         bch_err_fn(c, ret);
1151         return ret;
1152 }
1153
1154 static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
1155 {
1156         u32 restart_count = trans->restart_count;
1157         return check_i_sectors_notnested(trans, w) ?:
1158                 trans_was_restarted(trans, restart_count);
1159 }
1160
1161 struct extent_end {
1162         u32                     snapshot;
1163         u64                     offset;
1164         struct snapshots_seen   seen;
1165 };
1166
1167 struct extent_ends {
1168         struct bpos                     last_pos;
1169         DARRAY(struct extent_end)       e;
1170 };
1171
1172 static void extent_ends_reset(struct extent_ends *extent_ends)
1173 {
1174         darray_for_each(extent_ends->e, i)
1175                 snapshots_seen_exit(&i->seen);
1176         extent_ends->e.nr = 0;
1177 }
1178
1179 static void extent_ends_exit(struct extent_ends *extent_ends)
1180 {
1181         extent_ends_reset(extent_ends);
1182         darray_exit(&extent_ends->e);
1183 }
1184
1185 static void extent_ends_init(struct extent_ends *extent_ends)
1186 {
1187         memset(extent_ends, 0, sizeof(*extent_ends));
1188 }
1189
1190 static int extent_ends_at(struct bch_fs *c,
1191                           struct extent_ends *extent_ends,
1192                           struct snapshots_seen *seen,
1193                           struct bkey_s_c k)
1194 {
1195         struct extent_end *i, n = (struct extent_end) {
1196                 .offset         = k.k->p.offset,
1197                 .snapshot       = k.k->p.snapshot,
1198                 .seen           = *seen,
1199         };
1200
1201         n.seen.ids.data = kmemdup(seen->ids.data,
1202                               sizeof(seen->ids.data[0]) * seen->ids.size,
1203                               GFP_KERNEL);
1204         if (!n.seen.ids.data)
1205                 return -BCH_ERR_ENOMEM_fsck_extent_ends_at;
1206
1207         __darray_for_each(extent_ends->e, i) {
1208                 if (i->snapshot == k.k->p.snapshot) {
1209                         snapshots_seen_exit(&i->seen);
1210                         *i = n;
1211                         return 0;
1212                 }
1213
1214                 if (i->snapshot >= k.k->p.snapshot)
1215                         break;
1216         }
1217
1218         return darray_insert_item(&extent_ends->e, i - extent_ends->e.data, n);
1219 }
1220
1221 static int overlapping_extents_found(struct btree_trans *trans,
1222                                      enum btree_id btree,
1223                                      struct bpos pos1, struct snapshots_seen *pos1_seen,
1224                                      struct bkey pos2,
1225                                      bool *fixed,
1226                                      struct extent_end *extent_end)
1227 {
1228         struct bch_fs *c = trans->c;
1229         struct printbuf buf = PRINTBUF;
1230         struct btree_iter iter1, iter2 = { NULL };
1231         struct bkey_s_c k1, k2;
1232         int ret;
1233
1234         BUG_ON(bkey_le(pos1, bkey_start_pos(&pos2)));
1235
1236         bch2_trans_iter_init(trans, &iter1, btree, pos1,
1237                              BTREE_ITER_ALL_SNAPSHOTS|
1238                              BTREE_ITER_NOT_EXTENTS);
1239         k1 = bch2_btree_iter_peek_upto(&iter1, POS(pos1.inode, U64_MAX));
1240         ret = bkey_err(k1);
1241         if (ret)
1242                 goto err;
1243
1244         prt_str(&buf, "\n  ");
1245         bch2_bkey_val_to_text(&buf, c, k1);
1246
1247         if (!bpos_eq(pos1, k1.k->p)) {
1248                 prt_str(&buf, "\n  wanted\n  ");
1249                 bch2_bpos_to_text(&buf, pos1);
1250                 prt_str(&buf, "\n  ");
1251                 bch2_bkey_to_text(&buf, &pos2);
1252
1253                 bch_err(c, "%s: error finding first overlapping extent when repairing, got%s",
1254                         __func__, buf.buf);
1255                 ret = -BCH_ERR_internal_fsck_err;
1256                 goto err;
1257         }
1258
1259         bch2_trans_copy_iter(&iter2, &iter1);
1260
1261         while (1) {
1262                 bch2_btree_iter_advance(&iter2);
1263
1264                 k2 = bch2_btree_iter_peek_upto(&iter2, POS(pos1.inode, U64_MAX));
1265                 ret = bkey_err(k2);
1266                 if (ret)
1267                         goto err;
1268
1269                 if (bpos_ge(k2.k->p, pos2.p))
1270                         break;
1271         }
1272
1273         prt_str(&buf, "\n  ");
1274         bch2_bkey_val_to_text(&buf, c, k2);
1275
1276         if (bpos_gt(k2.k->p, pos2.p) ||
1277             pos2.size != k2.k->size) {
1278                 bch_err(c, "%s: error finding seconding overlapping extent when repairing%s",
1279                         __func__, buf.buf);
1280                 ret = -BCH_ERR_internal_fsck_err;
1281                 goto err;
1282         }
1283
1284         prt_printf(&buf, "\n  overwriting %s extent",
1285                    pos1.snapshot >= pos2.p.snapshot ? "first" : "second");
1286
1287         if (fsck_err(c, extent_overlapping,
1288                      "overlapping extents%s", buf.buf)) {
1289                 struct btree_iter *old_iter = &iter1;
1290                 struct disk_reservation res = { 0 };
1291
1292                 if (pos1.snapshot < pos2.p.snapshot) {
1293                         old_iter = &iter2;
1294                         swap(k1, k2);
1295                 }
1296
1297                 trans->extra_disk_res += bch2_bkey_sectors_compressed(k2);
1298
1299                 ret =   bch2_trans_update_extent_overwrite(trans, old_iter,
1300                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE,
1301                                 k1, k2) ?:
1302                         bch2_trans_commit(trans, &res, NULL, BCH_TRANS_COMMIT_no_enospc);
1303                 bch2_disk_reservation_put(c, &res);
1304
1305                 if (ret)
1306                         goto err;
1307
1308                 *fixed = true;
1309
1310                 if (pos1.snapshot == pos2.p.snapshot) {
1311                         /*
1312                          * We overwrote the first extent, and did the overwrite
1313                          * in the same snapshot:
1314                          */
1315                         extent_end->offset = bkey_start_offset(&pos2);
1316                 } else if (pos1.snapshot > pos2.p.snapshot) {
1317                         /*
1318                          * We overwrote the first extent in pos2's snapshot:
1319                          */
1320                         ret = snapshots_seen_add_inorder(c, pos1_seen, pos2.p.snapshot);
1321                 } else {
1322                         /*
1323                          * We overwrote the second extent - restart
1324                          * check_extent() from the top:
1325                          */
1326                         ret = -BCH_ERR_transaction_restart_nested;
1327                 }
1328         }
1329 fsck_err:
1330 err:
1331         bch2_trans_iter_exit(trans, &iter2);
1332         bch2_trans_iter_exit(trans, &iter1);
1333         printbuf_exit(&buf);
1334         return ret;
1335 }
1336
1337 static int check_overlapping_extents(struct btree_trans *trans,
1338                               struct snapshots_seen *seen,
1339                               struct extent_ends *extent_ends,
1340                               struct bkey_s_c k,
1341                               u32 equiv,
1342                               struct btree_iter *iter,
1343                               bool *fixed)
1344 {
1345         struct bch_fs *c = trans->c;
1346         int ret = 0;
1347
1348         /* transaction restart, running again */
1349         if (bpos_eq(extent_ends->last_pos, k.k->p))
1350                 return 0;
1351
1352         if (extent_ends->last_pos.inode != k.k->p.inode)
1353                 extent_ends_reset(extent_ends);
1354
1355         darray_for_each(extent_ends->e, i) {
1356                 if (i->offset <= bkey_start_offset(k.k))
1357                         continue;
1358
1359                 if (!ref_visible2(c,
1360                                   k.k->p.snapshot, seen,
1361                                   i->snapshot, &i->seen))
1362                         continue;
1363
1364                 ret = overlapping_extents_found(trans, iter->btree_id,
1365                                                 SPOS(iter->pos.inode,
1366                                                      i->offset,
1367                                                      i->snapshot),
1368                                                 &i->seen,
1369                                                 *k.k, fixed, i);
1370                 if (ret)
1371                         goto err;
1372         }
1373
1374         ret = extent_ends_at(c, extent_ends, seen, k);
1375         if (ret)
1376                 goto err;
1377
1378         extent_ends->last_pos = k.k->p;
1379 err:
1380         return ret;
1381 }
1382
1383 static int check_extent_overbig(struct btree_trans *trans, struct btree_iter *iter,
1384                                 struct bkey_s_c k)
1385 {
1386         struct bch_fs *c = trans->c;
1387         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1388         struct bch_extent_crc_unpacked crc;
1389         const union bch_extent_entry *i;
1390         unsigned encoded_extent_max_sectors = c->opts.encoded_extent_max >> 9;
1391
1392         bkey_for_each_crc(k.k, ptrs, crc, i)
1393                 if (crc_is_encoded(crc) &&
1394                     crc.uncompressed_size > encoded_extent_max_sectors) {
1395                         struct printbuf buf = PRINTBUF;
1396
1397                         bch2_bkey_val_to_text(&buf, c, k);
1398                         bch_err(c, "overbig encoded extent, please report this:\n  %s", buf.buf);
1399                         printbuf_exit(&buf);
1400                 }
1401
1402         return 0;
1403 }
1404
1405 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
1406                         struct bkey_s_c k,
1407                         struct inode_walker *inode,
1408                         struct snapshots_seen *s,
1409                         struct extent_ends *extent_ends)
1410 {
1411         struct bch_fs *c = trans->c;
1412         struct inode_walker_entry *i;
1413         struct printbuf buf = PRINTBUF;
1414         struct bpos equiv = k.k->p;
1415         int ret = 0;
1416
1417         equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1418
1419         ret = check_key_has_snapshot(trans, iter, k);
1420         if (ret) {
1421                 ret = ret < 0 ? ret : 0;
1422                 goto out;
1423         }
1424
1425         if (inode->last_pos.inode != k.k->p.inode) {
1426                 ret = check_i_sectors(trans, inode);
1427                 if (ret)
1428                         goto err;
1429         }
1430
1431         i = walk_inode(trans, inode, k);
1432         ret = PTR_ERR_OR_ZERO(i);
1433         if (ret)
1434                 goto err;
1435
1436         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1437         if (ret)
1438                 goto err;
1439
1440         if (k.k->type != KEY_TYPE_whiteout) {
1441                 if (fsck_err_on(!i, c, extent_in_missing_inode,
1442                                 "extent in missing inode:\n  %s",
1443                                 (printbuf_reset(&buf),
1444                                  bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1445                         goto delete;
1446
1447                 if (fsck_err_on(i &&
1448                                 !S_ISREG(i->inode.bi_mode) &&
1449                                 !S_ISLNK(i->inode.bi_mode),
1450                                 c, extent_in_non_reg_inode,
1451                                 "extent in non regular inode mode %o:\n  %s",
1452                                 i->inode.bi_mode,
1453                                 (printbuf_reset(&buf),
1454                                  bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1455                         goto delete;
1456
1457                 ret = check_overlapping_extents(trans, s, extent_ends, k,
1458                                                 equiv.snapshot, iter,
1459                                                 &inode->recalculate_sums);
1460                 if (ret)
1461                         goto err;
1462         }
1463
1464         /*
1465          * Check inodes in reverse order, from oldest snapshots to newest,
1466          * starting from the inode that matches this extent's snapshot. If we
1467          * didn't have one, iterate over all inodes:
1468          */
1469         if (!i)
1470                 i = inode->inodes.data + inode->inodes.nr - 1;
1471
1472         for (;
1473              inode->inodes.data && i >= inode->inodes.data;
1474              --i) {
1475                 if (i->snapshot > equiv.snapshot ||
1476                     !key_visible_in_snapshot(c, s, i->snapshot, equiv.snapshot))
1477                         continue;
1478
1479                 if (k.k->type != KEY_TYPE_whiteout) {
1480                         if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_size_dirty) &&
1481                                         k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9 &&
1482                                         !bkey_extent_is_reservation(k),
1483                                         c, extent_past_end_of_inode,
1484                                         "extent type past end of inode %llu:%u, i_size %llu\n  %s",
1485                                         i->inode.bi_inum, i->snapshot, i->inode.bi_size,
1486                                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1487                                 struct btree_iter iter2;
1488
1489                                 bch2_trans_copy_iter(&iter2, iter);
1490                                 bch2_btree_iter_set_snapshot(&iter2, i->snapshot);
1491                                 ret =   bch2_btree_iter_traverse(&iter2) ?:
1492                                         bch2_btree_delete_at(trans, &iter2,
1493                                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1494                                 bch2_trans_iter_exit(trans, &iter2);
1495                                 if (ret)
1496                                         goto err;
1497
1498                                 iter->k.type = KEY_TYPE_whiteout;
1499                         }
1500
1501                         if (bkey_extent_is_allocation(k.k))
1502                                 i->count += k.k->size;
1503                 }
1504
1505                 i->seen_this_pos = true;
1506         }
1507 out:
1508 err:
1509 fsck_err:
1510         printbuf_exit(&buf);
1511         bch_err_fn(c, ret);
1512         return ret;
1513 delete:
1514         ret = bch2_btree_delete_at(trans, iter, BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1515         goto out;
1516 }
1517
1518 /*
1519  * Walk extents: verify that extents have a corresponding S_ISREG inode, and
1520  * that i_size an i_sectors are consistent
1521  */
1522 int bch2_check_extents(struct bch_fs *c)
1523 {
1524         struct inode_walker w = inode_walker_init();
1525         struct snapshots_seen s;
1526         struct extent_ends extent_ends;
1527         struct disk_reservation res = { 0 };
1528
1529         snapshots_seen_init(&s);
1530         extent_ends_init(&extent_ends);
1531
1532         int ret = bch2_trans_run(c,
1533                 for_each_btree_key_commit(trans, iter, BTREE_ID_extents,
1534                                 POS(BCACHEFS_ROOT_INO, 0),
1535                                 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1536                                 &res, NULL,
1537                                 BCH_TRANS_COMMIT_no_enospc, ({
1538                         bch2_disk_reservation_put(c, &res);
1539                         check_extent(trans, &iter, k, &w, &s, &extent_ends) ?:
1540                         check_extent_overbig(trans, &iter, k);
1541                 })) ?:
1542                 check_i_sectors_notnested(trans, &w));
1543
1544         bch2_disk_reservation_put(c, &res);
1545         extent_ends_exit(&extent_ends);
1546         inode_walker_exit(&w);
1547         snapshots_seen_exit(&s);
1548
1549         bch_err_fn(c, ret);
1550         return ret;
1551 }
1552
1553 int bch2_check_indirect_extents(struct bch_fs *c)
1554 {
1555         struct disk_reservation res = { 0 };
1556
1557         int ret = bch2_trans_run(c,
1558                 for_each_btree_key_commit(trans, iter, BTREE_ID_reflink,
1559                                 POS_MIN,
1560                                 BTREE_ITER_PREFETCH, k,
1561                                 &res, NULL,
1562                                 BCH_TRANS_COMMIT_no_enospc, ({
1563                         bch2_disk_reservation_put(c, &res);
1564                         check_extent_overbig(trans, &iter, k);
1565                 })));
1566
1567         bch2_disk_reservation_put(c, &res);
1568         bch_err_fn(c, ret);
1569         return ret;
1570 }
1571
1572 static int check_subdir_count_notnested(struct btree_trans *trans, struct inode_walker *w)
1573 {
1574         struct bch_fs *c = trans->c;
1575         int ret = 0;
1576         s64 count2;
1577
1578         darray_for_each(w->inodes, i) {
1579                 if (i->inode.bi_nlink == i->count)
1580                         continue;
1581
1582                 count2 = bch2_count_subdirs(trans, w->last_pos.inode, i->snapshot);
1583                 if (count2 < 0)
1584                         return count2;
1585
1586                 if (i->count != count2) {
1587                         bch_err(c, "fsck counted subdirectories wrong: got %llu should be %llu",
1588                                 i->count, count2);
1589                         i->count = count2;
1590                         if (i->inode.bi_nlink == i->count)
1591                                 continue;
1592                 }
1593
1594                 if (fsck_err_on(i->inode.bi_nlink != i->count,
1595                                 c, inode_dir_wrong_nlink,
1596                                 "directory %llu:%u with wrong i_nlink: got %u, should be %llu",
1597                                 w->last_pos.inode, i->snapshot, i->inode.bi_nlink, i->count)) {
1598                         i->inode.bi_nlink = i->count;
1599                         ret = bch2_fsck_write_inode(trans, &i->inode, i->snapshot);
1600                         if (ret)
1601                                 break;
1602                 }
1603         }
1604 fsck_err:
1605         bch_err_fn(c, ret);
1606         return ret;
1607 }
1608
1609 static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
1610 {
1611         u32 restart_count = trans->restart_count;
1612         return check_subdir_count_notnested(trans, w) ?:
1613                 trans_was_restarted(trans, restart_count);
1614 }
1615
1616 static int check_dirent_inode_dirent(struct btree_trans *trans,
1617                                    struct btree_iter *iter,
1618                                    struct bkey_s_c_dirent d,
1619                                    struct bch_inode_unpacked *target,
1620                                    u32 target_snapshot)
1621 {
1622         struct bch_fs *c = trans->c;
1623         struct printbuf buf = PRINTBUF;
1624         int ret = 0;
1625
1626         if (inode_points_to_dirent(target, d))
1627                 return 0;
1628
1629         if (!target->bi_dir &&
1630             !target->bi_dir_offset) {
1631                 target->bi_dir          = d.k->p.inode;
1632                 target->bi_dir_offset   = d.k->p.offset;
1633                 return __bch2_fsck_write_inode(trans, target, target_snapshot);
1634         }
1635
1636         struct btree_iter bp_iter = { NULL };
1637         struct bkey_s_c_dirent bp_dirent = dirent_get_by_pos(trans, &bp_iter,
1638                               SPOS(target->bi_dir, target->bi_dir_offset, target_snapshot));
1639         ret = bkey_err(bp_dirent);
1640         if (ret && !bch2_err_matches(ret, ENOENT))
1641                 goto err;
1642
1643         bool backpointer_exists = !ret;
1644         ret = 0;
1645
1646         if (fsck_err_on(!backpointer_exists,
1647                         c, inode_wrong_backpointer,
1648                         "inode %llu:%u has wrong backpointer:\n"
1649                         "got       %llu:%llu\n"
1650                         "should be %llu:%llu",
1651                         target->bi_inum, target_snapshot,
1652                         target->bi_dir,
1653                         target->bi_dir_offset,
1654                         d.k->p.inode,
1655                         d.k->p.offset)) {
1656                 target->bi_dir          = d.k->p.inode;
1657                 target->bi_dir_offset   = d.k->p.offset;
1658                 ret = __bch2_fsck_write_inode(trans, target, target_snapshot);
1659                 goto out;
1660         }
1661
1662         bch2_bkey_val_to_text(&buf, c, d.s_c);
1663         prt_newline(&buf);
1664         if (backpointer_exists)
1665                 bch2_bkey_val_to_text(&buf, c, bp_dirent.s_c);
1666
1667         if (fsck_err_on(backpointer_exists &&
1668                         (S_ISDIR(target->bi_mode) ||
1669                          target->bi_subvol),
1670                         c, inode_dir_multiple_links,
1671                         "%s %llu:%u with multiple links\n%s",
1672                         S_ISDIR(target->bi_mode) ? "directory" : "subvolume",
1673                         target->bi_inum, target_snapshot, buf.buf)) {
1674                 ret = __remove_dirent(trans, d.k->p);
1675                 goto out;
1676         }
1677
1678         /*
1679          * hardlinked file with nlink 0:
1680          * We're just adjusting nlink here so check_nlinks() will pick
1681          * it up, it ignores inodes with nlink 0
1682          */
1683         if (fsck_err_on(backpointer_exists && !target->bi_nlink,
1684                         c, inode_multiple_links_but_nlink_0,
1685                         "inode %llu:%u type %s has multiple links but i_nlink 0\n%s",
1686                         target->bi_inum, target_snapshot, bch2_d_types[d.v->d_type], buf.buf)) {
1687                 target->bi_nlink++;
1688                 target->bi_flags &= ~BCH_INODE_unlinked;
1689                 ret = __bch2_fsck_write_inode(trans, target, target_snapshot);
1690                 if (ret)
1691                         goto err;
1692         }
1693 out:
1694 err:
1695 fsck_err:
1696         bch2_trans_iter_exit(trans, &bp_iter);
1697         printbuf_exit(&buf);
1698         bch_err_fn(c, ret);
1699         return ret;
1700 }
1701
1702 static int check_dirent_target(struct btree_trans *trans,
1703                                struct btree_iter *iter,
1704                                struct bkey_s_c_dirent d,
1705                                struct bch_inode_unpacked *target,
1706                                u32 target_snapshot)
1707 {
1708         struct bch_fs *c = trans->c;
1709         struct bkey_i_dirent *n;
1710         struct printbuf buf = PRINTBUF;
1711         int ret = 0;
1712
1713         ret = check_dirent_inode_dirent(trans, iter, d, target, target_snapshot);
1714         if (ret)
1715                 goto err;
1716
1717         if (fsck_err_on(d.v->d_type != inode_d_type(target),
1718                         c, dirent_d_type_wrong,
1719                         "incorrect d_type: got %s, should be %s:\n%s",
1720                         bch2_d_type_str(d.v->d_type),
1721                         bch2_d_type_str(inode_d_type(target)),
1722                         (printbuf_reset(&buf),
1723                          bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
1724                 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1725                 ret = PTR_ERR_OR_ZERO(n);
1726                 if (ret)
1727                         goto err;
1728
1729                 bkey_reassemble(&n->k_i, d.s_c);
1730                 n->v.d_type = inode_d_type(target);
1731                 if (n->v.d_type == DT_SUBVOL) {
1732                         n->v.d_parent_subvol = cpu_to_le32(target->bi_parent_subvol);
1733                         n->v.d_child_subvol = cpu_to_le32(target->bi_subvol);
1734                 } else {
1735                         n->v.d_inum = cpu_to_le64(target->bi_inum);
1736                 }
1737
1738                 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1739                 if (ret)
1740                         goto err;
1741
1742                 d = dirent_i_to_s_c(n);
1743         }
1744 err:
1745 fsck_err:
1746         printbuf_exit(&buf);
1747         bch_err_fn(c, ret);
1748         return ret;
1749 }
1750
1751 /* find a subvolume that's a descendent of @snapshot: */
1752 static int find_snapshot_subvol(struct btree_trans *trans, u32 snapshot, u32 *subvolid)
1753 {
1754         struct btree_iter iter;
1755         struct bkey_s_c k;
1756         int ret;
1757
1758         for_each_btree_key_norestart(trans, iter, BTREE_ID_subvolumes, POS_MIN, 0, k, ret) {
1759                 if (k.k->type != KEY_TYPE_subvolume)
1760                         continue;
1761
1762                 struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
1763                 if (bch2_snapshot_is_ancestor(trans->c, le32_to_cpu(s.v->snapshot), snapshot)) {
1764                         bch2_trans_iter_exit(trans, &iter);
1765                         *subvolid = k.k->p.offset;
1766                         goto found;
1767                 }
1768         }
1769         if (!ret)
1770                 ret = -ENOENT;
1771 found:
1772         bch2_trans_iter_exit(trans, &iter);
1773         return ret;
1774 }
1775
1776 static int check_dirent_to_subvol(struct btree_trans *trans, struct btree_iter *iter,
1777                                   struct bkey_s_c_dirent d)
1778 {
1779         struct bch_fs *c = trans->c;
1780         struct btree_iter subvol_iter = {};
1781         struct bch_inode_unpacked subvol_root;
1782         u32 parent_subvol = le32_to_cpu(d.v->d_parent_subvol);
1783         u32 target_subvol = le32_to_cpu(d.v->d_child_subvol);
1784         u32 parent_snapshot;
1785         u64 parent_inum;
1786         struct printbuf buf = PRINTBUF;
1787         int ret = 0;
1788
1789         ret = subvol_lookup(trans, parent_subvol, &parent_snapshot, &parent_inum);
1790         if (ret && !bch2_err_matches(ret, ENOENT))
1791                 return ret;
1792
1793         if (fsck_err_on(ret, c, dirent_to_missing_parent_subvol,
1794                         "dirent parent_subvol points to missing subvolume\n%s",
1795                         (bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf)) ||
1796             fsck_err_on(!ret && !bch2_snapshot_is_ancestor(c, parent_snapshot, d.k->p.snapshot),
1797                         c, dirent_not_visible_in_parent_subvol,
1798                         "dirent not visible in parent_subvol (not an ancestor of subvol snap %u)\n%s",
1799                         parent_snapshot,
1800                         (bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
1801                 u32 new_parent_subvol;
1802                 ret = find_snapshot_subvol(trans, d.k->p.snapshot, &new_parent_subvol);
1803                 if (ret)
1804                         goto err;
1805
1806                 struct bkey_i_dirent *new_dirent = bch2_bkey_make_mut_typed(trans, iter, &d.s_c, 0, dirent);
1807                 ret = PTR_ERR_OR_ZERO(new_dirent);
1808                 if (ret)
1809                         goto err;
1810
1811                 new_dirent->v.d_parent_subvol = cpu_to_le32(new_parent_subvol);
1812         }
1813
1814         struct bkey_s_c_subvolume s =
1815                 bch2_bkey_get_iter_typed(trans, &subvol_iter,
1816                                          BTREE_ID_subvolumes, POS(0, target_subvol),
1817                                          0, subvolume);
1818         ret = bkey_err(s.s_c);
1819         if (ret && !bch2_err_matches(ret, ENOENT))
1820                 return ret;
1821
1822         if (ret) {
1823                 if (fsck_err(c, dirent_to_missing_subvol,
1824                              "dirent points to missing subvolume\n%s",
1825                              (bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf)))
1826                         return __remove_dirent(trans, d.k->p);
1827                 ret = 0;
1828                 goto out;
1829         }
1830
1831         if (fsck_err_on(le32_to_cpu(s.v->fs_path_parent) != parent_subvol,
1832                         c, subvol_fs_path_parent_wrong,
1833                         "subvol with wrong fs_path_parent, should be be %u\n%s",
1834                         parent_subvol,
1835                         (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
1836                 struct bkey_i_subvolume *n =
1837                         bch2_bkey_make_mut_typed(trans, &subvol_iter, &s.s_c, 0, subvolume);
1838                 ret = PTR_ERR_OR_ZERO(n);
1839                 if (ret)
1840                         goto err;
1841
1842                 n->v.fs_path_parent = cpu_to_le32(parent_subvol);
1843         }
1844
1845         u64 target_inum = le64_to_cpu(s.v->inode);
1846         u32 target_snapshot = le32_to_cpu(s.v->snapshot);
1847
1848         ret = lookup_inode(trans, target_inum, &subvol_root, &target_snapshot);
1849         if (ret && !bch2_err_matches(ret, ENOENT))
1850                 return ret;
1851
1852         if (fsck_err_on(parent_subvol != subvol_root.bi_parent_subvol,
1853                         c, inode_bi_parent_wrong,
1854                         "subvol root %llu has wrong bi_parent_subvol: got %u, should be %u",
1855                         target_inum,
1856                         subvol_root.bi_parent_subvol, parent_subvol)) {
1857                 subvol_root.bi_parent_subvol = parent_subvol;
1858                 ret = __bch2_fsck_write_inode(trans, &subvol_root, target_snapshot);
1859                 if (ret)
1860                         return ret;
1861         }
1862
1863         ret = check_dirent_target(trans, iter, d, &subvol_root,
1864                                   target_snapshot);
1865         if (ret)
1866                 return ret;
1867 out:
1868 err:
1869 fsck_err:
1870         bch2_trans_iter_exit(trans, &subvol_iter);
1871         printbuf_exit(&buf);
1872         return ret;
1873 }
1874
1875 static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
1876                         struct bkey_s_c k,
1877                         struct bch_hash_info *hash_info,
1878                         struct inode_walker *dir,
1879                         struct inode_walker *target,
1880                         struct snapshots_seen *s)
1881 {
1882         struct bch_fs *c = trans->c;
1883         struct bkey_s_c_dirent d;
1884         struct inode_walker_entry *i;
1885         struct printbuf buf = PRINTBUF;
1886         struct bpos equiv;
1887         int ret = 0;
1888
1889         ret = check_key_has_snapshot(trans, iter, k);
1890         if (ret) {
1891                 ret = ret < 0 ? ret : 0;
1892                 goto out;
1893         }
1894
1895         equiv = k.k->p;
1896         equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1897
1898         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1899         if (ret)
1900                 goto err;
1901
1902         if (k.k->type == KEY_TYPE_whiteout)
1903                 goto out;
1904
1905         if (dir->last_pos.inode != k.k->p.inode) {
1906                 ret = check_subdir_count(trans, dir);
1907                 if (ret)
1908                         goto err;
1909         }
1910
1911         BUG_ON(!btree_iter_path(trans, iter)->should_be_locked);
1912
1913         i = walk_inode(trans, dir, k);
1914         ret = PTR_ERR_OR_ZERO(i);
1915         if (ret < 0)
1916                 goto err;
1917
1918         if (dir->first_this_inode && dir->inodes.nr)
1919                 *hash_info = bch2_hash_info_init(c, &dir->inodes.data[0].inode);
1920         dir->first_this_inode = false;
1921
1922         if (fsck_err_on(!i, c, dirent_in_missing_dir_inode,
1923                         "dirent in nonexisting directory:\n%s",
1924                         (printbuf_reset(&buf),
1925                          bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1926                 ret = bch2_btree_delete_at(trans, iter,
1927                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1928                 goto out;
1929         }
1930
1931         if (!i)
1932                 goto out;
1933
1934         if (fsck_err_on(!S_ISDIR(i->inode.bi_mode),
1935                         c, dirent_in_non_dir_inode,
1936                         "dirent in non directory inode type %s:\n%s",
1937                         bch2_d_type_str(inode_d_type(&i->inode)),
1938                         (printbuf_reset(&buf),
1939                          bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1940                 ret = bch2_btree_delete_at(trans, iter, 0);
1941                 goto out;
1942         }
1943
1944         ret = hash_check_key(trans, bch2_dirent_hash_desc, hash_info, iter, k);
1945         if (ret < 0)
1946                 goto err;
1947         if (ret) {
1948                 /* dirent has been deleted */
1949                 ret = 0;
1950                 goto out;
1951         }
1952
1953         if (k.k->type != KEY_TYPE_dirent)
1954                 goto out;
1955
1956         d = bkey_s_c_to_dirent(k);
1957
1958         if (d.v->d_type == DT_SUBVOL) {
1959                 ret = check_dirent_to_subvol(trans, iter, d);
1960                 if (ret)
1961                         goto err;
1962         } else {
1963                 ret = __get_visible_inodes(trans, target, s, le64_to_cpu(d.v->d_inum));
1964                 if (ret)
1965                         goto err;
1966
1967                 if (fsck_err_on(!target->inodes.nr,
1968                                 c, dirent_to_missing_inode,
1969                                 "dirent points to missing inode: (equiv %u)\n%s",
1970                                 equiv.snapshot,
1971                                 (printbuf_reset(&buf),
1972                                  bch2_bkey_val_to_text(&buf, c, k),
1973                                  buf.buf))) {
1974                         ret = __remove_dirent(trans, d.k->p);
1975                         if (ret)
1976                                 goto err;
1977                 }
1978
1979                 darray_for_each(target->inodes, i) {
1980                         ret = check_dirent_target(trans, iter, d,
1981                                                   &i->inode, i->snapshot);
1982                         if (ret)
1983                                 goto err;
1984                 }
1985
1986                 if (d.v->d_type == DT_DIR)
1987                         for_each_visible_inode(c, s, dir, equiv.snapshot, i)
1988                                 i->count++;
1989         }
1990 out:
1991 err:
1992 fsck_err:
1993         printbuf_exit(&buf);
1994         bch_err_fn(c, ret);
1995         return ret;
1996 }
1997
1998 /*
1999  * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
2000  * validate d_type
2001  */
2002 int bch2_check_dirents(struct bch_fs *c)
2003 {
2004         struct inode_walker dir = inode_walker_init();
2005         struct inode_walker target = inode_walker_init();
2006         struct snapshots_seen s;
2007         struct bch_hash_info hash_info;
2008
2009         snapshots_seen_init(&s);
2010
2011         int ret = bch2_trans_run(c,
2012                 for_each_btree_key_commit(trans, iter, BTREE_ID_dirents,
2013                                 POS(BCACHEFS_ROOT_INO, 0),
2014                                 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
2015                                 k,
2016                                 NULL, NULL,
2017                                 BCH_TRANS_COMMIT_no_enospc,
2018                         check_dirent(trans, &iter, k, &hash_info, &dir, &target, &s)) ?:
2019                 check_subdir_count_notnested(trans, &dir));
2020
2021         snapshots_seen_exit(&s);
2022         inode_walker_exit(&dir);
2023         inode_walker_exit(&target);
2024         bch_err_fn(c, ret);
2025         return ret;
2026 }
2027
2028 static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
2029                        struct bkey_s_c k,
2030                        struct bch_hash_info *hash_info,
2031                        struct inode_walker *inode)
2032 {
2033         struct bch_fs *c = trans->c;
2034         struct inode_walker_entry *i;
2035         int ret;
2036
2037         ret = check_key_has_snapshot(trans, iter, k);
2038         if (ret < 0)
2039                 return ret;
2040         if (ret)
2041                 return 0;
2042
2043         i = walk_inode(trans, inode, k);
2044         ret = PTR_ERR_OR_ZERO(i);
2045         if (ret)
2046                 return ret;
2047
2048         if (inode->first_this_inode && inode->inodes.nr)
2049                 *hash_info = bch2_hash_info_init(c, &inode->inodes.data[0].inode);
2050         inode->first_this_inode = false;
2051
2052         if (fsck_err_on(!i, c, xattr_in_missing_inode,
2053                         "xattr for missing inode %llu",
2054                         k.k->p.inode))
2055                 return bch2_btree_delete_at(trans, iter, 0);
2056
2057         if (!i)
2058                 return 0;
2059
2060         ret = hash_check_key(trans, bch2_xattr_hash_desc, hash_info, iter, k);
2061 fsck_err:
2062         bch_err_fn(c, ret);
2063         return ret;
2064 }
2065
2066 /*
2067  * Walk xattrs: verify that they all have a corresponding inode
2068  */
2069 int bch2_check_xattrs(struct bch_fs *c)
2070 {
2071         struct inode_walker inode = inode_walker_init();
2072         struct bch_hash_info hash_info;
2073         int ret = 0;
2074
2075         ret = bch2_trans_run(c,
2076                 for_each_btree_key_commit(trans, iter, BTREE_ID_xattrs,
2077                         POS(BCACHEFS_ROOT_INO, 0),
2078                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
2079                         k,
2080                         NULL, NULL,
2081                         BCH_TRANS_COMMIT_no_enospc,
2082                 check_xattr(trans, &iter, k, &hash_info, &inode)));
2083         bch_err_fn(c, ret);
2084         return ret;
2085 }
2086
2087 static int check_root_trans(struct btree_trans *trans)
2088 {
2089         struct bch_fs *c = trans->c;
2090         struct bch_inode_unpacked root_inode;
2091         u32 snapshot;
2092         u64 inum;
2093         int ret;
2094
2095         ret = subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
2096         if (ret && !bch2_err_matches(ret, ENOENT))
2097                 return ret;
2098
2099         if (mustfix_fsck_err_on(ret, c, root_subvol_missing,
2100                                 "root subvol missing")) {
2101                 struct bkey_i_subvolume root_subvol;
2102
2103                 snapshot        = U32_MAX;
2104                 inum            = BCACHEFS_ROOT_INO;
2105
2106                 bkey_subvolume_init(&root_subvol.k_i);
2107                 root_subvol.k.p.offset = BCACHEFS_ROOT_SUBVOL;
2108                 root_subvol.v.flags     = 0;
2109                 root_subvol.v.snapshot  = cpu_to_le32(snapshot);
2110                 root_subvol.v.inode     = cpu_to_le64(inum);
2111                 ret = bch2_btree_insert_trans(trans, BTREE_ID_subvolumes, &root_subvol.k_i, 0);
2112                 bch_err_msg(c, ret, "writing root subvol");
2113                 if (ret)
2114                         goto err;
2115         }
2116
2117         ret = lookup_inode(trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
2118         if (ret && !bch2_err_matches(ret, ENOENT))
2119                 return ret;
2120
2121         if (mustfix_fsck_err_on(ret, c, root_dir_missing,
2122                                 "root directory missing") ||
2123             mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode),
2124                                 c, root_inode_not_dir,
2125                                 "root inode not a directory")) {
2126                 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755,
2127                                 0, NULL);
2128                 root_inode.bi_inum = inum;
2129
2130                 ret = __bch2_fsck_write_inode(trans, &root_inode, snapshot);
2131                 bch_err_msg(c, ret, "writing root inode");
2132         }
2133 err:
2134 fsck_err:
2135         return ret;
2136 }
2137
2138 /* Get root directory, create if it doesn't exist: */
2139 int bch2_check_root(struct bch_fs *c)
2140 {
2141         int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2142                 check_root_trans(trans));
2143         bch_err_fn(c, ret);
2144         return ret;
2145 }
2146
2147 typedef DARRAY(u32) darray_u32;
2148
2149 static bool darray_u32_has(darray_u32 *d, u32 v)
2150 {
2151         darray_for_each(*d, i)
2152                 if (*i == v)
2153                         return true;
2154         return false;
2155 }
2156
2157 /*
2158  * We've checked that inode backpointers point to valid dirents; here, it's
2159  * sufficient to check that the subvolume root has a dirent:
2160  */
2161 static int subvol_has_dirent(struct btree_trans *trans, struct bkey_s_c_subvolume s)
2162 {
2163         struct bch_inode_unpacked inode;
2164         int ret = bch2_inode_find_by_inum_trans(trans,
2165                                 (subvol_inum) { s.k->p.offset, le64_to_cpu(s.v->inode) },
2166                                 &inode);
2167         if (ret)
2168                 return ret;
2169
2170         return inode.bi_dir != 0;
2171 }
2172
2173 static int check_subvol_path(struct btree_trans *trans, struct btree_iter *iter, struct bkey_s_c k)
2174 {
2175         struct bch_fs *c = trans->c;
2176         struct btree_iter parent_iter = {};
2177         darray_u32 subvol_path = {};
2178         struct printbuf buf = PRINTBUF;
2179         int ret = 0;
2180
2181         if (k.k->type != KEY_TYPE_subvolume)
2182                 return 0;
2183
2184         while (k.k->p.offset != BCACHEFS_ROOT_SUBVOL) {
2185                 ret = darray_push(&subvol_path, k.k->p.offset);
2186                 if (ret)
2187                         goto err;
2188
2189                 struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
2190
2191                 ret = subvol_has_dirent(trans, s);
2192                 if (ret < 0)
2193                         break;
2194
2195                 if (fsck_err_on(!ret,
2196                                 c, subvol_unreachable,
2197                                 "unreachable subvolume %s",
2198                                 (bch2_bkey_val_to_text(&buf, c, s.s_c),
2199                                  buf.buf))) {
2200                         ret = reattach_subvol(trans, s);
2201                         break;
2202                 }
2203
2204                 u32 parent = le32_to_cpu(s.v->fs_path_parent);
2205
2206                 if (darray_u32_has(&subvol_path, parent)) {
2207                         if (fsck_err(c, subvol_loop, "subvolume loop"))
2208                                 ret = reattach_subvol(trans, s);
2209                         break;
2210                 }
2211
2212                 bch2_trans_iter_exit(trans, &parent_iter);
2213                 bch2_trans_iter_init(trans, &parent_iter,
2214                                      BTREE_ID_subvolumes, POS(0, parent), 0);
2215                 k = bch2_btree_iter_peek_slot(&parent_iter);
2216                 ret = bkey_err(k);
2217                 if (ret)
2218                         goto err;
2219
2220                 if (fsck_err_on(k.k->type != KEY_TYPE_subvolume,
2221                                 c, subvol_unreachable,
2222                                 "unreachable subvolume %s",
2223                                 (bch2_bkey_val_to_text(&buf, c, s.s_c),
2224                                  buf.buf))) {
2225                         ret = reattach_subvol(trans, s);
2226                         break;
2227                 }
2228         }
2229 fsck_err:
2230 err:
2231         printbuf_exit(&buf);
2232         darray_exit(&subvol_path);
2233         bch2_trans_iter_exit(trans, &parent_iter);
2234         return ret;
2235 }
2236
2237 int bch2_check_subvolume_structure(struct bch_fs *c)
2238 {
2239         int ret = bch2_trans_run(c,
2240                 for_each_btree_key_commit(trans, iter,
2241                                 BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k,
2242                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2243                         check_subvol_path(trans, &iter, k)));
2244         bch_err_fn(c, ret);
2245         return ret;
2246 }
2247
2248 struct pathbuf_entry {
2249         u64     inum;
2250         u32     snapshot;
2251 };
2252
2253 typedef DARRAY(struct pathbuf_entry) pathbuf;
2254
2255 static bool path_is_dup(pathbuf *p, u64 inum, u32 snapshot)
2256 {
2257         darray_for_each(*p, i)
2258                 if (i->inum     == inum &&
2259                     i->snapshot == snapshot)
2260                         return true;
2261         return false;
2262 }
2263
2264 /*
2265  * Check that a given inode is reachable from its subvolume root - we already
2266  * verified subvolume connectivity:
2267  *
2268  * XXX: we should also be verifying that inodes are in the right subvolumes
2269  */
2270 static int check_path(struct btree_trans *trans, pathbuf *p, struct bkey_s_c inode_k)
2271 {
2272         struct bch_fs *c = trans->c;
2273         struct btree_iter inode_iter = {};
2274         struct bch_inode_unpacked inode;
2275         struct printbuf buf = PRINTBUF;
2276         u32 snapshot = bch2_snapshot_equiv(c, inode_k.k->p.snapshot);
2277         int ret = 0;
2278
2279         p->nr = 0;
2280
2281         BUG_ON(bch2_inode_unpack(inode_k, &inode));
2282
2283         while (!inode.bi_subvol) {
2284                 struct btree_iter dirent_iter;
2285                 struct bkey_s_c_dirent d;
2286                 u32 parent_snapshot = snapshot;
2287
2288                 d = inode_get_dirent(trans, &dirent_iter, &inode, &parent_snapshot);
2289                 ret = bkey_err(d.s_c);
2290                 if (ret && !bch2_err_matches(ret, ENOENT))
2291                         break;
2292
2293                 if (!ret && !dirent_points_to_inode(d, &inode)) {
2294                         bch2_trans_iter_exit(trans, &dirent_iter);
2295                         ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
2296                 }
2297
2298                 if (bch2_err_matches(ret, ENOENT)) {
2299                         ret = 0;
2300                         if (fsck_err(c, inode_unreachable,
2301                                      "unreachable inode\n%s",
2302                                      (printbuf_reset(&buf),
2303                                       bch2_bkey_val_to_text(&buf, c, inode_k),
2304                                       buf.buf)))
2305                                 ret = reattach_inode(trans, &inode, snapshot);
2306                         goto out;
2307                 }
2308
2309                 bch2_trans_iter_exit(trans, &dirent_iter);
2310
2311                 if (!S_ISDIR(inode.bi_mode))
2312                         break;
2313
2314                 ret = darray_push(p, ((struct pathbuf_entry) {
2315                         .inum           = inode.bi_inum,
2316                         .snapshot       = snapshot,
2317                 }));
2318                 if (ret)
2319                         return ret;
2320
2321                 snapshot = parent_snapshot;
2322
2323                 bch2_trans_iter_exit(trans, &inode_iter);
2324                 inode_k = bch2_bkey_get_iter(trans, &inode_iter, BTREE_ID_inodes,
2325                                              SPOS(0, inode.bi_dir, snapshot), 0);
2326                 ret = bkey_err(inode_k) ?:
2327                         !bkey_is_inode(inode_k.k) ? -BCH_ERR_ENOENT_inode
2328                         : bch2_inode_unpack(inode_k, &inode);
2329                 if (ret) {
2330                         /* Should have been caught in dirents pass */
2331                         if (!bch2_err_matches(ret, BCH_ERR_transaction_restart))
2332                                 bch_err(c, "error looking up parent directory: %i", ret);
2333                         break;
2334                 }
2335
2336                 snapshot = inode_k.k->p.snapshot;
2337
2338                 if (path_is_dup(p, inode.bi_inum, snapshot)) {
2339                         /* XXX print path */
2340                         bch_err(c, "directory structure loop");
2341
2342                         darray_for_each(*p, i)
2343                                 pr_err("%llu:%u", i->inum, i->snapshot);
2344                         pr_err("%llu:%u", inode.bi_inum, snapshot);
2345
2346                         if (fsck_err(c, dir_loop, "directory structure loop")) {
2347                                 ret = remove_backpointer(trans, &inode);
2348                                 bch_err_msg(c, ret, "removing dirent");
2349                                 if (ret)
2350                                         break;
2351
2352                                 ret = reattach_inode(trans, &inode, snapshot);
2353                                 bch_err_msg(c, ret, "reattaching inode %llu", inode.bi_inum);
2354                         }
2355                         break;
2356                 }
2357         }
2358 out:
2359 fsck_err:
2360         bch2_trans_iter_exit(trans, &inode_iter);
2361         printbuf_exit(&buf);
2362         bch_err_fn(c, ret);
2363         return ret;
2364 }
2365
2366 /*
2367  * Check for unreachable inodes, as well as loops in the directory structure:
2368  * After bch2_check_dirents(), if an inode backpointer doesn't exist that means it's
2369  * unreachable:
2370  */
2371 int bch2_check_directory_structure(struct bch_fs *c)
2372 {
2373         pathbuf path = { 0, };
2374         int ret;
2375
2376         ret = bch2_trans_run(c,
2377                 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes, POS_MIN,
2378                                           BTREE_ITER_INTENT|
2379                                           BTREE_ITER_PREFETCH|
2380                                           BTREE_ITER_ALL_SNAPSHOTS, k,
2381                                           NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
2382                         if (!bkey_is_inode(k.k))
2383                                 continue;
2384
2385                         if (bch2_inode_flags(k) & BCH_INODE_unlinked)
2386                                 continue;
2387
2388                         check_path(trans, &path, k);
2389                 })));
2390         darray_exit(&path);
2391
2392         bch_err_fn(c, ret);
2393         return ret;
2394 }
2395
2396 struct nlink_table {
2397         size_t          nr;
2398         size_t          size;
2399
2400         struct nlink {
2401                 u64     inum;
2402                 u32     snapshot;
2403                 u32     count;
2404         }               *d;
2405 };
2406
2407 static int add_nlink(struct bch_fs *c, struct nlink_table *t,
2408                      u64 inum, u32 snapshot)
2409 {
2410         if (t->nr == t->size) {
2411                 size_t new_size = max_t(size_t, 128UL, t->size * 2);
2412                 void *d = kvmalloc_array(new_size, sizeof(t->d[0]), GFP_KERNEL);
2413
2414                 if (!d) {
2415                         bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
2416                                 new_size);
2417                         return -BCH_ERR_ENOMEM_fsck_add_nlink;
2418                 }
2419
2420                 if (t->d)
2421                         memcpy(d, t->d, t->size * sizeof(t->d[0]));
2422                 kvfree(t->d);
2423
2424                 t->d = d;
2425                 t->size = new_size;
2426         }
2427
2428
2429         t->d[t->nr++] = (struct nlink) {
2430                 .inum           = inum,
2431                 .snapshot       = snapshot,
2432         };
2433
2434         return 0;
2435 }
2436
2437 static int nlink_cmp(const void *_l, const void *_r)
2438 {
2439         const struct nlink *l = _l;
2440         const struct nlink *r = _r;
2441
2442         return cmp_int(l->inum, r->inum);
2443 }
2444
2445 static void inc_link(struct bch_fs *c, struct snapshots_seen *s,
2446                      struct nlink_table *links,
2447                      u64 range_start, u64 range_end, u64 inum, u32 snapshot)
2448 {
2449         struct nlink *link, key = {
2450                 .inum = inum, .snapshot = U32_MAX,
2451         };
2452
2453         if (inum < range_start || inum >= range_end)
2454                 return;
2455
2456         link = __inline_bsearch(&key, links->d, links->nr,
2457                                 sizeof(links->d[0]), nlink_cmp);
2458         if (!link)
2459                 return;
2460
2461         while (link > links->d && link[0].inum == link[-1].inum)
2462                 --link;
2463
2464         for (; link < links->d + links->nr && link->inum == inum; link++)
2465                 if (ref_visible(c, s, snapshot, link->snapshot)) {
2466                         link->count++;
2467                         if (link->snapshot >= snapshot)
2468                                 break;
2469                 }
2470 }
2471
2472 noinline_for_stack
2473 static int check_nlinks_find_hardlinks(struct bch_fs *c,
2474                                        struct nlink_table *t,
2475                                        u64 start, u64 *end)
2476 {
2477         int ret = bch2_trans_run(c,
2478                 for_each_btree_key(trans, iter, BTREE_ID_inodes,
2479                                    POS(0, start),
2480                                    BTREE_ITER_INTENT|
2481                                    BTREE_ITER_PREFETCH|
2482                                    BTREE_ITER_ALL_SNAPSHOTS, k, ({
2483                         if (!bkey_is_inode(k.k))
2484                                 continue;
2485
2486                         /* Should never fail, checked by bch2_inode_invalid: */
2487                         struct bch_inode_unpacked u;
2488                         BUG_ON(bch2_inode_unpack(k, &u));
2489
2490                         /*
2491                          * Backpointer and directory structure checks are sufficient for
2492                          * directories, since they can't have hardlinks:
2493                          */
2494                         if (S_ISDIR(u.bi_mode))
2495                                 continue;
2496
2497                         if (!u.bi_nlink)
2498                                 continue;
2499
2500                         ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot);
2501                         if (ret) {
2502                                 *end = k.k->p.offset;
2503                                 ret = 0;
2504                                 break;
2505                         }
2506                         0;
2507                 })));
2508
2509         bch_err_fn(c, ret);
2510         return ret;
2511 }
2512
2513 noinline_for_stack
2514 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
2515                                      u64 range_start, u64 range_end)
2516 {
2517         struct snapshots_seen s;
2518
2519         snapshots_seen_init(&s);
2520
2521         int ret = bch2_trans_run(c,
2522                 for_each_btree_key(trans, iter, BTREE_ID_dirents, POS_MIN,
2523                                    BTREE_ITER_INTENT|
2524                                    BTREE_ITER_PREFETCH|
2525                                    BTREE_ITER_ALL_SNAPSHOTS, k, ({
2526                         ret = snapshots_seen_update(c, &s, iter.btree_id, k.k->p);
2527                         if (ret)
2528                                 break;
2529
2530                         if (k.k->type == KEY_TYPE_dirent) {
2531                                 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
2532
2533                                 if (d.v->d_type != DT_DIR &&
2534                                     d.v->d_type != DT_SUBVOL)
2535                                         inc_link(c, &s, links, range_start, range_end,
2536                                                  le64_to_cpu(d.v->d_inum),
2537                                                  bch2_snapshot_equiv(c, d.k->p.snapshot));
2538                         }
2539                         0;
2540                 })));
2541
2542         snapshots_seen_exit(&s);
2543
2544         bch_err_fn(c, ret);
2545         return ret;
2546 }
2547
2548 static int check_nlinks_update_inode(struct btree_trans *trans, struct btree_iter *iter,
2549                                      struct bkey_s_c k,
2550                                      struct nlink_table *links,
2551                                      size_t *idx, u64 range_end)
2552 {
2553         struct bch_fs *c = trans->c;
2554         struct bch_inode_unpacked u;
2555         struct nlink *link = &links->d[*idx];
2556         int ret = 0;
2557
2558         if (k.k->p.offset >= range_end)
2559                 return 1;
2560
2561         if (!bkey_is_inode(k.k))
2562                 return 0;
2563
2564         BUG_ON(bch2_inode_unpack(k, &u));
2565
2566         if (S_ISDIR(u.bi_mode))
2567                 return 0;
2568
2569         if (!u.bi_nlink)
2570                 return 0;
2571
2572         while ((cmp_int(link->inum, k.k->p.offset) ?:
2573                 cmp_int(link->snapshot, k.k->p.snapshot)) < 0) {
2574                 BUG_ON(*idx == links->nr);
2575                 link = &links->d[++*idx];
2576         }
2577
2578         if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count,
2579                         c, inode_wrong_nlink,
2580                         "inode %llu type %s has wrong i_nlink (%u, should be %u)",
2581                         u.bi_inum, bch2_d_types[mode_to_type(u.bi_mode)],
2582                         bch2_inode_nlink_get(&u), link->count)) {
2583                 bch2_inode_nlink_set(&u, link->count);
2584                 ret = __bch2_fsck_write_inode(trans, &u, k.k->p.snapshot);
2585         }
2586 fsck_err:
2587         return ret;
2588 }
2589
2590 noinline_for_stack
2591 static int check_nlinks_update_hardlinks(struct bch_fs *c,
2592                                struct nlink_table *links,
2593                                u64 range_start, u64 range_end)
2594 {
2595         size_t idx = 0;
2596
2597         int ret = bch2_trans_run(c,
2598                 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
2599                                 POS(0, range_start),
2600                                 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
2601                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2602                         check_nlinks_update_inode(trans, &iter, k, links, &idx, range_end)));
2603         if (ret < 0) {
2604                 bch_err(c, "error in fsck walking inodes: %s", bch2_err_str(ret));
2605                 return ret;
2606         }
2607
2608         return 0;
2609 }
2610
2611 int bch2_check_nlinks(struct bch_fs *c)
2612 {
2613         struct nlink_table links = { 0 };
2614         u64 this_iter_range_start, next_iter_range_start = 0;
2615         int ret = 0;
2616
2617         do {
2618                 this_iter_range_start = next_iter_range_start;
2619                 next_iter_range_start = U64_MAX;
2620
2621                 ret = check_nlinks_find_hardlinks(c, &links,
2622                                                   this_iter_range_start,
2623                                                   &next_iter_range_start);
2624
2625                 ret = check_nlinks_walk_dirents(c, &links,
2626                                           this_iter_range_start,
2627                                           next_iter_range_start);
2628                 if (ret)
2629                         break;
2630
2631                 ret = check_nlinks_update_hardlinks(c, &links,
2632                                          this_iter_range_start,
2633                                          next_iter_range_start);
2634                 if (ret)
2635                         break;
2636
2637                 links.nr = 0;
2638         } while (next_iter_range_start != U64_MAX);
2639
2640         kvfree(links.d);
2641         bch_err_fn(c, ret);
2642         return ret;
2643 }
2644
2645 static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter,
2646                              struct bkey_s_c k)
2647 {
2648         struct bkey_s_c_reflink_p p;
2649         struct bkey_i_reflink_p *u;
2650
2651         if (k.k->type != KEY_TYPE_reflink_p)
2652                 return 0;
2653
2654         p = bkey_s_c_to_reflink_p(k);
2655
2656         if (!p.v->front_pad && !p.v->back_pad)
2657                 return 0;
2658
2659         u = bch2_trans_kmalloc(trans, sizeof(*u));
2660         int ret = PTR_ERR_OR_ZERO(u);
2661         if (ret)
2662                 return ret;
2663
2664         bkey_reassemble(&u->k_i, k);
2665         u->v.front_pad  = 0;
2666         u->v.back_pad   = 0;
2667
2668         return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_NORUN);
2669 }
2670
2671 int bch2_fix_reflink_p(struct bch_fs *c)
2672 {
2673         if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
2674                 return 0;
2675
2676         int ret = bch2_trans_run(c,
2677                 for_each_btree_key_commit(trans, iter,
2678                                 BTREE_ID_extents, POS_MIN,
2679                                 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|
2680                                 BTREE_ITER_ALL_SNAPSHOTS, k,
2681                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2682                         fix_reflink_p_key(trans, &iter, k)));
2683         bch_err_fn(c, ret);
2684         return ret;
2685 }