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

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "bkey_buf.h"
   5 #include "btree_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 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  24
  25 /*
  26  * XXX: this is handling transaction restarts without returning
  27  * -BCH_ERR_transaction_restart_nested, this is not how we do things anymore:
  28  */
  29 static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum,
  30                                     u32 snapshot)
  31 {
  32         struct btree_iter iter;
  33         struct bkey_s_c k;
  34         u64 sectors = 0;
  35         int ret;
  36
  37         for_each_btree_key_upto(trans, iter, BTREE_ID_extents,
  38                                 SPOS(inum, 0, snapshot),
  39                                 POS(inum, U64_MAX),
  40                                 0, k, ret)
  41                 if (bkey_extent_is_allocation(k.k))
  42                         sectors += k.k->size;
  43
  44         bch2_trans_iter_exit(trans, &iter);
  45
  46         return ret ?: sectors;
  47 }
  48
  49 static s64 bch2_count_subdirs(struct btree_trans *trans, u64 inum,
  50                                     u32 snapshot)
  51 {
  52         struct btree_iter iter;
  53         struct bkey_s_c k;
  54         struct bkey_s_c_dirent d;
  55         u64 subdirs = 0;
  56         int ret;
  57
  58         for_each_btree_key_upto(trans, iter, BTREE_ID_dirents,
  59                                 SPOS(inum, 0, snapshot),
  60                                 POS(inum, U64_MAX),
  61                                 0, k, ret) {
  62                 if (k.k->type != KEY_TYPE_dirent)
  63                         continue;
  64
  65                 d = bkey_s_c_to_dirent(k);
  66                 if (d.v->d_type == DT_DIR)
  67                         subdirs++;
  68         }
  69         bch2_trans_iter_exit(trans, &iter);
  70
  71         return ret ?: subdirs;
  72 }
  73
  74 static int __snapshot_lookup_subvol(struct btree_trans *trans, u32 snapshot,
  75                                     u32 *subvol)
  76 {
  77         struct bch_snapshot s;
  78         int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_snapshots,
  79                                           POS(0, snapshot), 0,
  80                                           snapshot, &s);
  81         if (!ret)
  82                 *subvol = le32_to_cpu(s.subvol);
  83         else if (bch2_err_matches(ret, ENOENT))
  84                 bch_err(trans->c, "snapshot %u not found", snapshot);
  85         return ret;
  86
  87 }
  88
  89 static int __subvol_lookup(struct btree_trans *trans, u32 subvol,
  90                            u32 *snapshot, u64 *inum)
  91 {
  92         struct bch_subvolume s;
  93         int ret;
  94
  95         ret = bch2_subvolume_get(trans, subvol, false, 0, &s);
  96
  97         *snapshot = le32_to_cpu(s.snapshot);
  98         *inum = le64_to_cpu(s.inode);
  99         return ret;
 100 }
 101
 102 static int subvol_lookup(struct btree_trans *trans, u32 subvol,
 103                          u32 *snapshot, u64 *inum)
 104 {
 105         return lockrestart_do(trans, __subvol_lookup(trans, subvol, snapshot, inum));
 106 }
 107
 108 static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr,
 109                               struct bch_inode_unpacked *inode)
 110 {
 111         struct btree_iter iter;
 112         struct bkey_s_c k;
 113         int ret;
 114
 115         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
 116                              POS(0, inode_nr),
 117                              BTREE_ITER_ALL_SNAPSHOTS);
 118         k = bch2_btree_iter_peek(&iter);
 119         ret = bkey_err(k);
 120         if (ret)
 121                 goto err;
 122
 123         if (!k.k || !bkey_eq(k.k->p, POS(0, inode_nr))) {
 124                 ret = -BCH_ERR_ENOENT_inode;
 125                 goto err;
 126         }
 127
 128         ret = bch2_inode_unpack(k, inode);
 129 err:
 130         bch_err_msg(trans->c, ret, "fetching inode %llu", inode_nr);
 131         bch2_trans_iter_exit(trans, &iter);
 132         return ret;
 133 }
 134
 135 static int __lookup_inode(struct btree_trans *trans, u64 inode_nr,
 136                           struct bch_inode_unpacked *inode,
 137                           u32 *snapshot)
 138 {
 139         struct btree_iter iter;
 140         struct bkey_s_c k;
 141         int ret;
 142
 143         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
 144                                SPOS(0, inode_nr, *snapshot), 0);
 145         ret = bkey_err(k);
 146         if (ret)
 147                 goto err;
 148
 149         ret = bkey_is_inode(k.k)
 150                 ? bch2_inode_unpack(k, inode)
 151                 : -BCH_ERR_ENOENT_inode;
 152         if (!ret)
 153                 *snapshot = iter.pos.snapshot;
 154 err:
 155         bch_err_msg(trans->c, ret, "fetching inode %llu:%u", inode_nr, *snapshot);
 156         bch2_trans_iter_exit(trans, &iter);
 157         return ret;
 158 }
 159
 160 static int lookup_inode(struct btree_trans *trans, u64 inode_nr,
 161                         struct bch_inode_unpacked *inode,
 162                         u32 *snapshot)
 163 {
 164         return lockrestart_do(trans, __lookup_inode(trans, inode_nr, inode, snapshot));
 165 }
 166
 167 static int __lookup_dirent(struct btree_trans *trans,
 168                            struct bch_hash_info hash_info,
 169                            subvol_inum dir, struct qstr *name,
 170                            u64 *target, unsigned *type)
 171 {
 172         struct btree_iter iter;
 173         struct bkey_s_c_dirent d;
 174         int ret;
 175
 176         ret = bch2_hash_lookup(trans, &iter, bch2_dirent_hash_desc,
 177                                &hash_info, dir, name, 0);
 178         if (ret)
 179                 return ret;
 180
 181         d = bkey_s_c_to_dirent(bch2_btree_iter_peek_slot(&iter));
 182         *target = le64_to_cpu(d.v->d_inum);
 183         *type = d.v->d_type;
 184         bch2_trans_iter_exit(trans, &iter);
 185         return 0;
 186 }
 187
 188 static int __write_inode(struct btree_trans *trans,
 189                          struct bch_inode_unpacked *inode,
 190                          u32 snapshot)
 191 {
 192         struct bkey_inode_buf *inode_p =
 193                 bch2_trans_kmalloc(trans, sizeof(*inode_p));
 194
 195         if (IS_ERR(inode_p))
 196                 return PTR_ERR(inode_p);
 197
 198         bch2_inode_pack(inode_p, inode);
 199         inode_p->inode.k.p.snapshot = snapshot;
 200
 201         return bch2_btree_insert_nonextent(trans, BTREE_ID_inodes,
 202                                 &inode_p->inode.k_i,
 203                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 204 }
 205
 206 static int fsck_write_inode(struct btree_trans *trans,
 207                             struct bch_inode_unpacked *inode,
 208                             u32 snapshot)
 209 {
 210         int ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
 211                             __write_inode(trans, inode, snapshot));
 212         if (ret)
 213                 bch_err_fn(trans->c, ret);
 214         return ret;
 215 }
 216
 217 static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
 218 {
 219         struct bch_fs *c = trans->c;
 220         struct btree_iter iter;
 221         struct bch_inode_unpacked dir_inode;
 222         struct bch_hash_info dir_hash_info;
 223         int ret;
 224
 225         ret = lookup_first_inode(trans, pos.inode, &dir_inode);
 226         if (ret)
 227                 goto err;
 228
 229         dir_hash_info = bch2_hash_info_init(c, &dir_inode);
 230
 231         bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_INTENT);
 232
 233         ret = bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
 234                                   &dir_hash_info, &iter,
 235                                   BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 236         bch2_trans_iter_exit(trans, &iter);
 237 err:
 238         bch_err_fn(c, ret);
 239         return ret;
 240 }
 241
 242 /* Get lost+found, create if it doesn't exist: */
 243 static int lookup_lostfound(struct btree_trans *trans, u32 subvol,
 244                             struct bch_inode_unpacked *lostfound)
 245 {
 246         struct bch_fs *c = trans->c;
 247         struct bch_inode_unpacked root;
 248         struct bch_hash_info root_hash_info;
 249         struct qstr lostfound_str = QSTR("lost+found");
 250         subvol_inum root_inum = { .subvol = subvol };
 251         u64 inum = 0;
 252         unsigned d_type = 0;
 253         u32 snapshot;
 254         int ret;
 255
 256         ret = __subvol_lookup(trans, subvol, &snapshot, &root_inum.inum);
 257         if (ret)
 258                 return ret;
 259
 260         ret = __lookup_inode(trans, root_inum.inum, &root, &snapshot);
 261         if (ret)
 262                 return ret;
 263
 264         root_hash_info = bch2_hash_info_init(c, &root);
 265
 266         ret = __lookup_dirent(trans, root_hash_info, root_inum,
 267                             &lostfound_str, &inum, &d_type);
 268         if (bch2_err_matches(ret, ENOENT)) {
 269                 bch_notice(c, "creating lost+found");
 270                 goto create_lostfound;
 271         }
 272
 273         bch_err_fn(c, ret);
 274         if (ret)
 275                 return ret;
 276
 277         if (d_type != DT_DIR) {
 278                 bch_err(c, "error looking up lost+found: not a directory");
 279                 return -BCH_ERR_ENOENT_not_directory;
 280         }
 281
 282         /*
 283          * The bch2_check_dirents pass has already run, dangling dirents
 284          * shouldn't exist here:
 285          */
 286         return __lookup_inode(trans, inum, lostfound, &snapshot);
 287
 288 create_lostfound:
 289         bch2_inode_init_early(c, lostfound);
 290
 291         ret = bch2_create_trans(trans, root_inum, &root,
 292                                 lostfound, &lostfound_str,
 293                                 0, 0, S_IFDIR|0700, 0, NULL, NULL,
 294                                 (subvol_inum) { }, 0);
 295         bch_err_msg(c, ret, "creating lost+found");
 296         return ret;
 297 }
 298
 299 static int __reattach_inode(struct btree_trans *trans,
 300                           struct bch_inode_unpacked *inode,
 301                           u32 inode_snapshot)
 302 {
 303         struct bch_hash_info dir_hash;
 304         struct bch_inode_unpacked lostfound;
 305         char name_buf[20];
 306         struct qstr name;
 307         u64 dir_offset = 0;
 308         u32 subvol;
 309         int ret;
 310
 311         ret = __snapshot_lookup_subvol(trans, inode_snapshot, &subvol);
 312         if (ret)
 313                 return ret;
 314
 315         ret = lookup_lostfound(trans, subvol, &lostfound);
 316         if (ret)
 317                 return ret;
 318
 319         if (S_ISDIR(inode->bi_mode)) {
 320                 lostfound.bi_nlink++;
 321
 322                 ret = __write_inode(trans, &lostfound, U32_MAX);
 323                 if (ret)
 324                         return ret;
 325         }
 326
 327         dir_hash = bch2_hash_info_init(trans->c, &lostfound);
 328
 329         snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
 330         name = (struct qstr) QSTR(name_buf);
 331
 332         ret = bch2_dirent_create(trans,
 333                                  (subvol_inum) {
 334                                         .subvol = subvol,
 335                                         .inum = lostfound.bi_inum,
 336                                  },
 337                                  &dir_hash,
 338                                  inode_d_type(inode),
 339                                  &name, inode->bi_inum, &dir_offset,
 340                                  BCH_HASH_SET_MUST_CREATE);
 341         if (ret)
 342                 return ret;
 343
 344         inode->bi_dir           = lostfound.bi_inum;
 345         inode->bi_dir_offset    = dir_offset;
 346
 347         return __write_inode(trans, inode, inode_snapshot);
 348 }
 349
 350 static int reattach_inode(struct btree_trans *trans,
 351                           struct bch_inode_unpacked *inode,
 352                           u32 inode_snapshot)
 353 {
 354         int ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
 355                         __reattach_inode(trans, inode, inode_snapshot));
 356         bch_err_msg(trans->c, ret, "reattaching inode %llu", inode->bi_inum);
 357         return ret;
 358 }
 359
 360 static int remove_backpointer(struct btree_trans *trans,
 361                               struct bch_inode_unpacked *inode)
 362 {
 363         struct btree_iter iter;
 364         struct bkey_s_c_dirent d;
 365         int ret;
 366
 367         d = bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_dirents,
 368                                      POS(inode->bi_dir, inode->bi_dir_offset), 0,
 369                                      dirent);
 370         ret =   bkey_err(d) ?:
 371                 __remove_dirent(trans, d.k->p);
 372         bch2_trans_iter_exit(trans, &iter);
 373         return ret;
 374 }
 375
 376 struct snapshots_seen_entry {
 377         u32                             id;
 378         u32                             equiv;
 379 };
 380
 381 struct snapshots_seen {
 382         struct bpos                     pos;
 383         DARRAY(struct snapshots_seen_entry) ids;
 384 };
 385
 386 static inline void snapshots_seen_exit(struct snapshots_seen *s)
 387 {
 388         darray_exit(&s->ids);
 389 }
 390
 391 static inline void snapshots_seen_init(struct snapshots_seen *s)
 392 {
 393         memset(s, 0, sizeof(*s));
 394 }
 395
 396 static int snapshots_seen_add_inorder(struct bch_fs *c, struct snapshots_seen *s, u32 id)
 397 {
 398         struct snapshots_seen_entry *i, n = {
 399                 .id     = id,
 400                 .equiv  = bch2_snapshot_equiv(c, id),
 401         };
 402         int ret = 0;
 403
 404         darray_for_each(s->ids, i) {
 405                 if (i->id == id)
 406                         return 0;
 407                 if (i->id > id)
 408                         break;
 409         }
 410
 411         ret = darray_insert_item(&s->ids, i - s->ids.data, n);
 412         if (ret)
 413                 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
 414                         s->ids.size);
 415         return ret;
 416 }
 417
 418 static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s,
 419                                  enum btree_id btree_id, struct bpos pos)
 420 {
 421         struct snapshots_seen_entry *i, n = {
 422                 .id     = pos.snapshot,
 423                 .equiv  = bch2_snapshot_equiv(c, pos.snapshot),
 424         };
 425         int ret = 0;
 426
 427         if (!bkey_eq(s->pos, pos))
 428                 s->ids.nr = 0;
 429
 430         s->pos = pos;
 431         s->pos.snapshot = n.equiv;
 432
 433         darray_for_each(s->ids, i) {
 434                 if (i->id == n.id)
 435                         return 0;
 436
 437                 /*
 438                  * We currently don't rigorously track for snapshot cleanup
 439                  * needing to be run, so it shouldn't be a fsck error yet:
 440                  */
 441                 if (i->equiv == n.equiv) {
 442                         bch_err(c, "snapshot deletion did not finish:\n"
 443                                 "  duplicate keys in btree %s at %llu:%llu snapshots %u, %u (equiv %u)\n",
 444                                 bch2_btree_id_str(btree_id),
 445                                 pos.inode, pos.offset,
 446                                 i->id, n.id, n.equiv);
 447                         set_bit(BCH_FS_need_delete_dead_snapshots, &c->flags);
 448                         return bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_delete_dead_snapshots);
 449                 }
 450         }
 451
 452         ret = darray_push(&s->ids, n);
 453         if (ret)
 454                 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
 455                         s->ids.size);
 456         return ret;
 457 }
 458
 459 /**
 460  * key_visible_in_snapshot - returns true if @id is a descendent of @ancestor,
 461  * and @ancestor hasn't been overwritten in @seen
 462  *
 463  * @c:          filesystem handle
 464  * @seen:       list of snapshot ids already seen at current position
 465  * @id:         descendent snapshot id
 466  * @ancestor:   ancestor snapshot id
 467  *
 468  * Returns:     whether key in @ancestor snapshot is visible in @id snapshot
 469  */
 470 static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *seen,
 471                                     u32 id, u32 ancestor)
 472 {
 473         ssize_t i;
 474
 475         EBUG_ON(id > ancestor);
 476         EBUG_ON(!bch2_snapshot_is_equiv(c, id));
 477         EBUG_ON(!bch2_snapshot_is_equiv(c, ancestor));
 478
 479         /* @ancestor should be the snapshot most recently added to @seen */
 480         EBUG_ON(ancestor != seen->pos.snapshot);
 481         EBUG_ON(ancestor != seen->ids.data[seen->ids.nr - 1].equiv);
 482
 483         if (id == ancestor)
 484                 return true;
 485
 486         if (!bch2_snapshot_is_ancestor(c, id, ancestor))
 487                 return false;
 488
 489         /*
 490          * We know that @id is a descendant of @ancestor, we're checking if
 491          * we've seen a key that overwrote @ancestor - i.e. also a descendent of
 492          * @ascestor and with @id as a descendent.
 493          *
 494          * But we already know that we're scanning IDs between @id and @ancestor
 495          * numerically, since snapshot ID lists are kept sorted, so if we find
 496          * an id that's an ancestor of @id we're done:
 497          */
 498
 499         for (i = seen->ids.nr - 2;
 500              i >= 0 && seen->ids.data[i].equiv >= id;
 501              --i)
 502                 if (bch2_snapshot_is_ancestor(c, id, seen->ids.data[i].equiv))
 503                         return false;
 504
 505         return true;
 506 }
 507
 508 /**
 509  * ref_visible - given a key with snapshot id @src that points to a key with
 510  * snapshot id @dst, test whether there is some snapshot in which @dst is
 511  * visible.
 512  *
 513  * @c:          filesystem handle
 514  * @s:          list of snapshot IDs already seen at @src
 515  * @src:        snapshot ID of src key
 516  * @dst:        snapshot ID of dst key
 517  * Returns:     true if there is some snapshot in which @dst is visible
 518  *
 519  * Assumes we're visiting @src keys in natural key order
 520  */
 521 static bool ref_visible(struct bch_fs *c, struct snapshots_seen *s,
 522                         u32 src, u32 dst)
 523 {
 524         return dst <= src
 525                 ? key_visible_in_snapshot(c, s, dst, src)
 526                 : bch2_snapshot_is_ancestor(c, src, dst);
 527 }
 528
 529 static int ref_visible2(struct bch_fs *c,
 530                         u32 src, struct snapshots_seen *src_seen,
 531                         u32 dst, struct snapshots_seen *dst_seen)
 532 {
 533         src = bch2_snapshot_equiv(c, src);
 534         dst = bch2_snapshot_equiv(c, dst);
 535
 536         if (dst > src) {
 537                 swap(dst, src);
 538                 swap(dst_seen, src_seen);
 539         }
 540         return key_visible_in_snapshot(c, src_seen, dst, src);
 541 }
 542
 543 #define for_each_visible_inode(_c, _s, _w, _snapshot, _i)                               \
 544         for (_i = (_w)->inodes.data; _i < (_w)->inodes.data + (_w)->inodes.nr &&        \
 545              (_i)->snapshot <= (_snapshot); _i++)                                       \
 546                 if (key_visible_in_snapshot(_c, _s, _i->snapshot, _snapshot))
 547
 548 struct inode_walker_entry {
 549         struct bch_inode_unpacked inode;
 550         u32                     snapshot;
 551         bool                    seen_this_pos;
 552         u64                     count;
 553 };
 554
 555 struct inode_walker {
 556         bool                            first_this_inode;
 557         bool                            recalculate_sums;
 558         struct bpos                     last_pos;
 559
 560         DARRAY(struct inode_walker_entry) inodes;
 561 };
 562
 563 static void inode_walker_exit(struct inode_walker *w)
 564 {
 565         darray_exit(&w->inodes);
 566 }
 567
 568 static struct inode_walker inode_walker_init(void)
 569 {
 570         return (struct inode_walker) { 0, };
 571 }
 572
 573 static int add_inode(struct bch_fs *c, struct inode_walker *w,
 574                      struct bkey_s_c inode)
 575 {
 576         struct bch_inode_unpacked u;
 577
 578         BUG_ON(bch2_inode_unpack(inode, &u));
 579
 580         return darray_push(&w->inodes, ((struct inode_walker_entry) {
 581                 .inode          = u,
 582                 .snapshot       = bch2_snapshot_equiv(c, inode.k->p.snapshot),
 583         }));
 584 }
 585
 586 static int get_inodes_all_snapshots(struct btree_trans *trans,
 587                                     struct inode_walker *w, u64 inum)
 588 {
 589         struct bch_fs *c = trans->c;
 590         struct btree_iter iter;
 591         struct bkey_s_c k;
 592         u32 restart_count = trans->restart_count;
 593         int ret;
 594
 595         w->recalculate_sums = false;
 596         w->inodes.nr = 0;
 597
 598         for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, inum),
 599                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
 600                 if (k.k->p.offset != inum)
 601                         break;
 602
 603                 if (bkey_is_inode(k.k))
 604                         add_inode(c, w, k);
 605         }
 606         bch2_trans_iter_exit(trans, &iter);
 607
 608         if (ret)
 609                 return ret;
 610
 611         w->first_this_inode = true;
 612
 613         return trans_was_restarted(trans, restart_count);
 614 }
 615
 616 static struct inode_walker_entry *
 617 lookup_inode_for_snapshot(struct bch_fs *c, struct inode_walker *w,
 618                           u32 snapshot, bool is_whiteout)
 619 {
 620         struct inode_walker_entry *i;
 621
 622         snapshot = bch2_snapshot_equiv(c, snapshot);
 623
 624         darray_for_each(w->inodes, i)
 625                 if (bch2_snapshot_is_ancestor(c, snapshot, i->snapshot))
 626                         goto found;
 627
 628         return NULL;
 629 found:
 630         BUG_ON(snapshot > i->snapshot);
 631
 632         if (snapshot != i->snapshot && !is_whiteout) {
 633                 struct inode_walker_entry new = *i;
 634                 size_t pos;
 635                 int ret;
 636
 637                 new.snapshot = snapshot;
 638                 new.count = 0;
 639
 640                 bch_info(c, "have key for inode %llu:%u but have inode in ancestor snapshot %u",
 641                          w->last_pos.inode, snapshot, i->snapshot);
 642
 643                 while (i > w->inodes.data && i[-1].snapshot > snapshot)
 644                         --i;
 645
 646                 pos = i - w->inodes.data;
 647                 ret = darray_insert_item(&w->inodes, pos, new);
 648                 if (ret)
 649                         return ERR_PTR(ret);
 650
 651                 i = w->inodes.data + pos;
 652         }
 653
 654         return i;
 655 }
 656
 657 static struct inode_walker_entry *walk_inode(struct btree_trans *trans,
 658                                              struct inode_walker *w, struct bpos pos,
 659                                              bool is_whiteout)
 660 {
 661         if (w->last_pos.inode != pos.inode) {
 662                 int ret = get_inodes_all_snapshots(trans, w, pos.inode);
 663                 if (ret)
 664                         return ERR_PTR(ret);
 665         } else if (bkey_cmp(w->last_pos, pos)) {
 666                 struct inode_walker_entry *i;
 667
 668                 darray_for_each(w->inodes, i)
 669                         i->seen_this_pos = false;
 670
 671         }
 672
 673         w->last_pos = pos;
 674
 675         return lookup_inode_for_snapshot(trans->c, w, pos.snapshot, is_whiteout);
 676 }
 677
 678 static int __get_visible_inodes(struct btree_trans *trans,
 679                                 struct inode_walker *w,
 680                                 struct snapshots_seen *s,
 681                                 u64 inum)
 682 {
 683         struct bch_fs *c = trans->c;
 684         struct btree_iter iter;
 685         struct bkey_s_c k;
 686         int ret;
 687
 688         w->inodes.nr = 0;
 689
 690         for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum),
 691                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
 692                 u32 equiv = bch2_snapshot_equiv(c, k.k->p.snapshot);
 693
 694                 if (k.k->p.offset != inum)
 695                         break;
 696
 697                 if (!ref_visible(c, s, s->pos.snapshot, equiv))
 698                         continue;
 699
 700                 if (bkey_is_inode(k.k))
 701                         add_inode(c, w, k);
 702
 703                 if (equiv >= s->pos.snapshot)
 704                         break;
 705         }
 706         bch2_trans_iter_exit(trans, &iter);
 707
 708         return ret;
 709 }
 710
 711 static int check_key_has_snapshot(struct btree_trans *trans,
 712                                   struct btree_iter *iter,
 713                                   struct bkey_s_c k)
 714 {
 715         struct bch_fs *c = trans->c;
 716         struct printbuf buf = PRINTBUF;
 717         int ret = 0;
 718
 719         if (mustfix_fsck_err_on(!bch2_snapshot_equiv(c, k.k->p.snapshot), c,
 720                                 bkey_in_missing_snapshot,
 721                                 "key in missing snapshot: %s",
 722                                 (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
 723                 ret = bch2_btree_delete_at(trans, iter,
 724                                             BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?: 1;
 725 fsck_err:
 726         printbuf_exit(&buf);
 727         return ret;
 728 }
 729
 730 static int hash_redo_key(struct btree_trans *trans,
 731                          const struct bch_hash_desc desc,
 732                          struct bch_hash_info *hash_info,
 733                          struct btree_iter *k_iter, struct bkey_s_c k)
 734 {
 735         struct bkey_i *delete;
 736         struct bkey_i *tmp;
 737
 738         delete = bch2_trans_kmalloc(trans, sizeof(*delete));
 739         if (IS_ERR(delete))
 740                 return PTR_ERR(delete);
 741
 742         tmp = bch2_bkey_make_mut_noupdate(trans, k);
 743         if (IS_ERR(tmp))
 744                 return PTR_ERR(tmp);
 745
 746         bkey_init(&delete->k);
 747         delete->k.p = k_iter->pos;
 748         return  bch2_btree_iter_traverse(k_iter) ?:
 749                 bch2_trans_update(trans, k_iter, delete, 0) ?:
 750                 bch2_hash_set_snapshot(trans, desc, hash_info,
 751                                        (subvol_inum) { 0, k.k->p.inode },
 752                                        k.k->p.snapshot, tmp,
 753                                        BCH_HASH_SET_MUST_CREATE,
 754                                        BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
 755                 bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
 756 }
 757
 758 static int hash_check_key(struct btree_trans *trans,
 759                           const struct bch_hash_desc desc,
 760                           struct bch_hash_info *hash_info,
 761                           struct btree_iter *k_iter, struct bkey_s_c hash_k)
 762 {
 763         struct bch_fs *c = trans->c;
 764         struct btree_iter iter = { NULL };
 765         struct printbuf buf = PRINTBUF;
 766         struct bkey_s_c k;
 767         u64 hash;
 768         int ret = 0;
 769
 770         if (hash_k.k->type != desc.key_type)
 771                 return 0;
 772
 773         hash = desc.hash_bkey(hash_info, hash_k);
 774
 775         if (likely(hash == hash_k.k->p.offset))
 776                 return 0;
 777
 778         if (hash_k.k->p.offset < hash)
 779                 goto bad_hash;
 780
 781         for_each_btree_key_norestart(trans, iter, desc.btree_id,
 782                                      SPOS(hash_k.k->p.inode, hash, hash_k.k->p.snapshot),
 783                                      BTREE_ITER_SLOTS, k, ret) {
 784                 if (bkey_eq(k.k->p, hash_k.k->p))
 785                         break;
 786
 787                 if (fsck_err_on(k.k->type == desc.key_type &&
 788                                 !desc.cmp_bkey(k, hash_k), c,
 789                                 hash_table_key_duplicate,
 790                                 "duplicate hash table keys:\n%s",
 791                                 (printbuf_reset(&buf),
 792                                  bch2_bkey_val_to_text(&buf, c, hash_k),
 793                                  buf.buf))) {
 794                         ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0) ?: 1;
 795                         break;
 796                 }
 797
 798                 if (bkey_deleted(k.k)) {
 799                         bch2_trans_iter_exit(trans, &iter);
 800                         goto bad_hash;
 801                 }
 802         }
 803 out:
 804         bch2_trans_iter_exit(trans, &iter);
 805         printbuf_exit(&buf);
 806         return ret;
 807 bad_hash:
 808         if (fsck_err(c, hash_table_key_wrong_offset,
 809                      "hash table key at wrong offset: btree %s inode %llu offset %llu, hashed to %llu\n%s",
 810                      bch2_btree_id_str(desc.btree_id), hash_k.k->p.inode, hash_k.k->p.offset, hash,
 811                      (printbuf_reset(&buf),
 812                       bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf))) {
 813                 ret = hash_redo_key(trans, desc, hash_info, k_iter, hash_k);
 814                 bch_err_fn(c, ret);
 815                 if (ret)
 816                         return ret;
 817                 ret = -BCH_ERR_transaction_restart_nested;
 818         }
 819 fsck_err:
 820         goto out;
 821 }
 822
 823 static int check_inode_deleted_list(struct btree_trans *trans, struct bpos p)
 824 {
 825         struct btree_iter iter;
 826         struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_deleted_inodes, p, 0);
 827         int ret = bkey_err(k);
 828         if (ret)
 829                 return ret;
 830
 831         bch2_trans_iter_exit(trans, &iter);
 832         return k.k->type == KEY_TYPE_set;
 833 }
 834
 835 static int check_inode(struct btree_trans *trans,
 836                        struct btree_iter *iter,
 837                        struct bkey_s_c k,
 838                        struct bch_inode_unpacked *prev,
 839                        struct snapshots_seen *s,
 840                        bool full)
 841 {
 842         struct bch_fs *c = trans->c;
 843         struct bch_inode_unpacked u;
 844         bool do_update = false;
 845         int ret;
 846
 847         ret = check_key_has_snapshot(trans, iter, k);
 848         if (ret < 0)
 849                 goto err;
 850         if (ret)
 851                 return 0;
 852
 853         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
 854         if (ret)
 855                 goto err;
 856
 857         if (!bkey_is_inode(k.k))
 858                 return 0;
 859
 860         BUG_ON(bch2_inode_unpack(k, &u));
 861
 862         if (!full &&
 863             !(u.bi_flags & (BCH_INODE_i_size_dirty|
 864                             BCH_INODE_i_sectors_dirty|
 865                             BCH_INODE_unlinked)))
 866                 return 0;
 867
 868         if (prev->bi_inum != u.bi_inum)
 869                 *prev = u;
 870
 871         if (fsck_err_on(prev->bi_hash_seed      != u.bi_hash_seed ||
 872                         inode_d_type(prev)      != inode_d_type(&u),
 873                         c, inode_snapshot_mismatch,
 874                         "inodes in different snapshots don't match")) {
 875                 bch_err(c, "repair not implemented yet");
 876                 return -BCH_ERR_fsck_repair_unimplemented;
 877         }
 878
 879         if ((u.bi_flags & (BCH_INODE_i_size_dirty|BCH_INODE_unlinked)) &&
 880             bch2_key_has_snapshot_overwrites(trans, BTREE_ID_inodes, k.k->p)) {
 881                 struct bpos new_min_pos;
 882
 883                 ret = bch2_propagate_key_to_snapshot_leaves(trans, iter->btree_id, k, &new_min_pos);
 884                 if (ret)
 885                         goto err;
 886
 887                 u.bi_flags &= ~BCH_INODE_i_size_dirty|BCH_INODE_unlinked;
 888
 889                 ret = __write_inode(trans, &u, iter->pos.snapshot);
 890                 bch_err_msg(c, ret, "in fsck updating inode");
 891                 if (ret)
 892                         return ret;
 893
 894                 if (!bpos_eq(new_min_pos, POS_MIN))
 895                         bch2_btree_iter_set_pos(iter, bpos_predecessor(new_min_pos));
 896                 return 0;
 897         }
 898
 899         if (u.bi_flags & BCH_INODE_unlinked &&
 900             c->sb.version >= bcachefs_metadata_version_deleted_inodes) {
 901                 ret = check_inode_deleted_list(trans, k.k->p);
 902                 if (ret < 0)
 903                         return ret;
 904
 905                 fsck_err_on(ret, c, unlinked_inode_not_on_deleted_list,
 906                             "inode %llu:%u unlinked, but not on deleted list",
 907                             u.bi_inum, k.k->p.snapshot);
 908                 ret = 0;
 909         }
 910
 911         if (u.bi_flags & BCH_INODE_unlinked &&
 912             (!c->sb.clean ||
 913              fsck_err(c, inode_unlinked_but_clean,
 914                       "filesystem marked clean, but inode %llu unlinked",
 915                       u.bi_inum))) {
 916                 ret = bch2_inode_rm_snapshot(trans, u.bi_inum, iter->pos.snapshot);
 917                 bch_err_msg(c, ret, "in fsck deleting inode");
 918                 return ret;
 919         }
 920
 921         if (u.bi_flags & BCH_INODE_i_size_dirty &&
 922             (!c->sb.clean ||
 923              fsck_err(c, inode_i_size_dirty_but_clean,
 924                       "filesystem marked clean, but inode %llu has i_size dirty",
 925                       u.bi_inum))) {
 926                 bch_verbose(c, "truncating inode %llu", u.bi_inum);
 927
 928                 /*
 929                  * XXX: need to truncate partial blocks too here - or ideally
 930                  * just switch units to bytes and that issue goes away
 931                  */
 932                 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
 933                                 SPOS(u.bi_inum, round_up(u.bi_size, block_bytes(c)) >> 9,
 934                                      iter->pos.snapshot),
 935                                 POS(u.bi_inum, U64_MAX),
 936                                 0, NULL);
 937                 bch_err_msg(c, ret, "in fsck truncating inode");
 938                 if (ret)
 939                         return ret;
 940
 941                 /*
 942                  * We truncated without our normal sector accounting hook, just
 943                  * make sure we recalculate it:
 944                  */
 945                 u.bi_flags |= BCH_INODE_i_sectors_dirty;
 946
 947                 u.bi_flags &= ~BCH_INODE_i_size_dirty;
 948                 do_update = true;
 949         }
 950
 951         if (u.bi_flags & BCH_INODE_i_sectors_dirty &&
 952             (!c->sb.clean ||
 953              fsck_err(c, inode_i_sectors_dirty_but_clean,
 954                       "filesystem marked clean, but inode %llu has i_sectors dirty",
 955                       u.bi_inum))) {
 956                 s64 sectors;
 957
 958                 bch_verbose(c, "recounting sectors for inode %llu",
 959                             u.bi_inum);
 960
 961                 sectors = bch2_count_inode_sectors(trans, u.bi_inum, iter->pos.snapshot);
 962                 if (sectors < 0) {
 963                         bch_err_msg(c, sectors, "in fsck recounting inode sectors");
 964                         return sectors;
 965                 }
 966
 967                 u.bi_sectors = sectors;
 968                 u.bi_flags &= ~BCH_INODE_i_sectors_dirty;
 969                 do_update = true;
 970         }
 971
 972         if (u.bi_flags & BCH_INODE_backptr_untrusted) {
 973                 u.bi_dir = 0;
 974                 u.bi_dir_offset = 0;
 975                 u.bi_flags &= ~BCH_INODE_backptr_untrusted;
 976                 do_update = true;
 977         }
 978
 979         if (do_update) {
 980                 ret = __write_inode(trans, &u, iter->pos.snapshot);
 981                 bch_err_msg(c, ret, "in fsck updating inode");
 982                 if (ret)
 983                         return ret;
 984         }
 985 err:
 986 fsck_err:
 987         bch_err_fn(c, ret);
 988         return ret;
 989 }
 990
 991 int bch2_check_inodes(struct bch_fs *c)
 992 {
 993         bool full = c->opts.fsck;
 994         struct btree_trans *trans = bch2_trans_get(c);
 995         struct btree_iter iter;
 996         struct bch_inode_unpacked prev = { 0 };
 997         struct snapshots_seen s;
 998         struct bkey_s_c k;
 999         int ret;
1000
1001         snapshots_seen_init(&s);
1002
1003         ret = for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
1004                         POS_MIN,
1005                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1006                         NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1007                 check_inode(trans, &iter, k, &prev, &s, full));
1008
1009         snapshots_seen_exit(&s);
1010         bch2_trans_put(trans);
1011         bch_err_fn(c, ret);
1012         return ret;
1013 }
1014
1015 static struct bkey_s_c_dirent dirent_get_by_pos(struct btree_trans *trans,
1016                                                 struct btree_iter *iter,
1017                                                 struct bpos pos)
1018 {
1019         return bch2_bkey_get_iter_typed(trans, iter, BTREE_ID_dirents, pos, 0, dirent);
1020 }
1021
1022 static bool inode_points_to_dirent(struct bch_inode_unpacked *inode,
1023                                    struct bkey_s_c_dirent d)
1024 {
1025         return  inode->bi_dir           == d.k->p.inode &&
1026                 inode->bi_dir_offset    == d.k->p.offset;
1027 }
1028
1029 static bool dirent_points_to_inode(struct bkey_s_c_dirent d,
1030                                    struct bch_inode_unpacked *inode)
1031 {
1032         return d.v->d_type == DT_SUBVOL
1033                 ? le32_to_cpu(d.v->d_child_subvol)      == inode->bi_subvol
1034                 : le64_to_cpu(d.v->d_inum)              == inode->bi_inum;
1035 }
1036
1037 static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
1038 {
1039         struct bch_fs *c = trans->c;
1040         struct inode_walker_entry *i;
1041         u32 restart_count = trans->restart_count;
1042         int ret = 0;
1043         s64 count2;
1044
1045         darray_for_each(w->inodes, i) {
1046                 if (i->inode.bi_sectors == i->count)
1047                         continue;
1048
1049                 count2 = bch2_count_inode_sectors(trans, w->last_pos.inode, i->snapshot);
1050
1051                 if (w->recalculate_sums)
1052                         i->count = count2;
1053
1054                 if (i->count != count2) {
1055                         bch_err(c, "fsck counted i_sectors wrong for inode %llu:%u: got %llu should be %llu",
1056                                 w->last_pos.inode, i->snapshot, i->count, count2);
1057                         return -BCH_ERR_internal_fsck_err;
1058                 }
1059
1060                 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_sectors_dirty),
1061                                 c, inode_i_sectors_wrong,
1062                                 "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
1063                                 w->last_pos.inode, i->snapshot,
1064                                 i->inode.bi_sectors, i->count)) {
1065                         i->inode.bi_sectors = i->count;
1066                         ret = fsck_write_inode(trans, &i->inode, i->snapshot);
1067                         if (ret)
1068                                 break;
1069                 }
1070         }
1071 fsck_err:
1072         bch_err_fn(c, ret);
1073         return ret ?: trans_was_restarted(trans, restart_count);
1074 }
1075
1076 struct extent_end {
1077         u32                     snapshot;
1078         u64                     offset;
1079         struct snapshots_seen   seen;
1080 };
1081
1082 struct extent_ends {
1083         struct bpos                     last_pos;
1084         DARRAY(struct extent_end)       e;
1085 };
1086
1087 static void extent_ends_reset(struct extent_ends *extent_ends)
1088 {
1089         struct extent_end *i;
1090
1091         darray_for_each(extent_ends->e, i)
1092                 snapshots_seen_exit(&i->seen);
1093
1094         extent_ends->e.nr = 0;
1095 }
1096
1097 static void extent_ends_exit(struct extent_ends *extent_ends)
1098 {
1099         extent_ends_reset(extent_ends);
1100         darray_exit(&extent_ends->e);
1101 }
1102
1103 static void extent_ends_init(struct extent_ends *extent_ends)
1104 {
1105         memset(extent_ends, 0, sizeof(*extent_ends));
1106 }
1107
1108 static int extent_ends_at(struct bch_fs *c,
1109                           struct extent_ends *extent_ends,
1110                           struct snapshots_seen *seen,
1111                           struct bkey_s_c k)
1112 {
1113         struct extent_end *i, n = (struct extent_end) {
1114                 .offset         = k.k->p.offset,
1115                 .snapshot       = k.k->p.snapshot,
1116                 .seen           = *seen,
1117         };
1118
1119         n.seen.ids.data = kmemdup(seen->ids.data,
1120                               sizeof(seen->ids.data[0]) * seen->ids.size,
1121                               GFP_KERNEL);
1122         if (!n.seen.ids.data)
1123                 return -BCH_ERR_ENOMEM_fsck_extent_ends_at;
1124
1125         darray_for_each(extent_ends->e, i) {
1126                 if (i->snapshot == k.k->p.snapshot) {
1127                         snapshots_seen_exit(&i->seen);
1128                         *i = n;
1129                         return 0;
1130                 }
1131
1132                 if (i->snapshot >= k.k->p.snapshot)
1133                         break;
1134         }
1135
1136         return darray_insert_item(&extent_ends->e, i - extent_ends->e.data, n);
1137 }
1138
1139 static int overlapping_extents_found(struct btree_trans *trans,
1140                                      enum btree_id btree,
1141                                      struct bpos pos1, struct snapshots_seen *pos1_seen,
1142                                      struct bkey pos2,
1143                                      bool *fixed,
1144                                      struct extent_end *extent_end)
1145 {
1146         struct bch_fs *c = trans->c;
1147         struct printbuf buf = PRINTBUF;
1148         struct btree_iter iter1, iter2 = { NULL };
1149         struct bkey_s_c k1, k2;
1150         int ret;
1151
1152         BUG_ON(bkey_le(pos1, bkey_start_pos(&pos2)));
1153
1154         bch2_trans_iter_init(trans, &iter1, btree, pos1,
1155                              BTREE_ITER_ALL_SNAPSHOTS|
1156                              BTREE_ITER_NOT_EXTENTS);
1157         k1 = bch2_btree_iter_peek_upto(&iter1, POS(pos1.inode, U64_MAX));
1158         ret = bkey_err(k1);
1159         if (ret)
1160                 goto err;
1161
1162         prt_str(&buf, "\n  ");
1163         bch2_bkey_val_to_text(&buf, c, k1);
1164
1165         if (!bpos_eq(pos1, k1.k->p)) {
1166                 prt_str(&buf, "\n  wanted\n  ");
1167                 bch2_bpos_to_text(&buf, pos1);
1168                 prt_str(&buf, "\n  ");
1169                 bch2_bkey_to_text(&buf, &pos2);
1170
1171                 bch_err(c, "%s: error finding first overlapping extent when repairing, got%s",
1172                         __func__, buf.buf);
1173                 ret = -BCH_ERR_internal_fsck_err;
1174                 goto err;
1175         }
1176
1177         bch2_trans_copy_iter(&iter2, &iter1);
1178
1179         while (1) {
1180                 bch2_btree_iter_advance(&iter2);
1181
1182                 k2 = bch2_btree_iter_peek_upto(&iter2, POS(pos1.inode, U64_MAX));
1183                 ret = bkey_err(k2);
1184                 if (ret)
1185                         goto err;
1186
1187                 if (bpos_ge(k2.k->p, pos2.p))
1188                         break;
1189         }
1190
1191         prt_str(&buf, "\n  ");
1192         bch2_bkey_val_to_text(&buf, c, k2);
1193
1194         if (bpos_gt(k2.k->p, pos2.p) ||
1195             pos2.size != k2.k->size) {
1196                 bch_err(c, "%s: error finding seconding overlapping extent when repairing%s",
1197                         __func__, buf.buf);
1198                 ret = -BCH_ERR_internal_fsck_err;
1199                 goto err;
1200         }
1201
1202         prt_printf(&buf, "\n  overwriting %s extent",
1203                    pos1.snapshot >= pos2.p.snapshot ? "first" : "second");
1204
1205         if (fsck_err(c, extent_overlapping,
1206                      "overlapping extents%s", buf.buf)) {
1207                 struct btree_iter *old_iter = &iter1;
1208                 struct disk_reservation res = { 0 };
1209
1210                 if (pos1.snapshot < pos2.p.snapshot) {
1211                         old_iter = &iter2;
1212                         swap(k1, k2);
1213                 }
1214
1215                 trans->extra_journal_res += bch2_bkey_sectors_compressed(k2);
1216
1217                 ret =   bch2_trans_update_extent_overwrite(trans, old_iter,
1218                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE,
1219                                 k1, k2) ?:
1220                         bch2_trans_commit(trans, &res, NULL, BCH_TRANS_COMMIT_no_enospc);
1221                 bch2_disk_reservation_put(c, &res);
1222
1223                 if (ret)
1224                         goto err;
1225
1226                 *fixed = true;
1227
1228                 if (pos1.snapshot == pos2.p.snapshot) {
1229                         /*
1230                          * We overwrote the first extent, and did the overwrite
1231                          * in the same snapshot:
1232                          */
1233                         extent_end->offset = bkey_start_offset(&pos2);
1234                 } else if (pos1.snapshot > pos2.p.snapshot) {
1235                         /*
1236                          * We overwrote the first extent in pos2's snapshot:
1237                          */
1238                         ret = snapshots_seen_add_inorder(c, pos1_seen, pos2.p.snapshot);
1239                 } else {
1240                         /*
1241                          * We overwrote the second extent - restart
1242                          * check_extent() from the top:
1243                          */
1244                         ret = -BCH_ERR_transaction_restart_nested;
1245                 }
1246         }
1247 fsck_err:
1248 err:
1249         bch2_trans_iter_exit(trans, &iter2);
1250         bch2_trans_iter_exit(trans, &iter1);
1251         printbuf_exit(&buf);
1252         return ret;
1253 }
1254
1255 static int check_overlapping_extents(struct btree_trans *trans,
1256                               struct snapshots_seen *seen,
1257                               struct extent_ends *extent_ends,
1258                               struct bkey_s_c k,
1259                               u32 equiv,
1260                               struct btree_iter *iter,
1261                               bool *fixed)
1262 {
1263         struct bch_fs *c = trans->c;
1264         struct extent_end *i;
1265         int ret = 0;
1266
1267         /* transaction restart, running again */
1268         if (bpos_eq(extent_ends->last_pos, k.k->p))
1269                 return 0;
1270
1271         if (extent_ends->last_pos.inode != k.k->p.inode)
1272                 extent_ends_reset(extent_ends);
1273
1274         darray_for_each(extent_ends->e, i) {
1275                 if (i->offset <= bkey_start_offset(k.k))
1276                         continue;
1277
1278                 if (!ref_visible2(c,
1279                                   k.k->p.snapshot, seen,
1280                                   i->snapshot, &i->seen))
1281                         continue;
1282
1283                 ret = overlapping_extents_found(trans, iter->btree_id,
1284                                                 SPOS(iter->pos.inode,
1285                                                      i->offset,
1286                                                      i->snapshot),
1287                                                 &i->seen,
1288                                                 *k.k, fixed, i);
1289                 if (ret)
1290                         goto err;
1291         }
1292
1293         ret = extent_ends_at(c, extent_ends, seen, k);
1294         if (ret)
1295                 goto err;
1296
1297         extent_ends->last_pos = k.k->p;
1298 err:
1299         return ret;
1300 }
1301
1302 static int check_extent_overbig(struct btree_trans *trans, struct btree_iter *iter,
1303                                 struct bkey_s_c k)
1304 {
1305         struct bch_fs *c = trans->c;
1306         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1307         struct bch_extent_crc_unpacked crc;
1308         const union bch_extent_entry *i;
1309         unsigned encoded_extent_max_sectors = c->opts.encoded_extent_max >> 9;
1310
1311         bkey_for_each_crc(k.k, ptrs, crc, i)
1312                 if (crc_is_encoded(crc) &&
1313                     crc.uncompressed_size > encoded_extent_max_sectors) {
1314                         struct printbuf buf = PRINTBUF;
1315
1316                         bch2_bkey_val_to_text(&buf, c, k);
1317                         bch_err(c, "overbig encoded extent, please report this:\n  %s", buf.buf);
1318                         printbuf_exit(&buf);
1319                 }
1320
1321         return 0;
1322 }
1323
1324 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
1325                         struct bkey_s_c k,
1326                         struct inode_walker *inode,
1327                         struct snapshots_seen *s,
1328                         struct extent_ends *extent_ends)
1329 {
1330         struct bch_fs *c = trans->c;
1331         struct inode_walker_entry *i;
1332         struct printbuf buf = PRINTBUF;
1333         struct bpos equiv = k.k->p;
1334         int ret = 0;
1335
1336         equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1337
1338         ret = check_key_has_snapshot(trans, iter, k);
1339         if (ret) {
1340                 ret = ret < 0 ? ret : 0;
1341                 goto out;
1342         }
1343
1344         if (inode->last_pos.inode != k.k->p.inode) {
1345                 ret = check_i_sectors(trans, inode);
1346                 if (ret)
1347                         goto err;
1348         }
1349
1350         i = walk_inode(trans, inode, equiv, k.k->type == KEY_TYPE_whiteout);
1351         ret = PTR_ERR_OR_ZERO(i);
1352         if (ret)
1353                 goto err;
1354
1355         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1356         if (ret)
1357                 goto err;
1358
1359         if (k.k->type != KEY_TYPE_whiteout) {
1360                 if (fsck_err_on(!i, c, extent_in_missing_inode,
1361                                 "extent in missing inode:\n  %s",
1362                                 (printbuf_reset(&buf),
1363                                  bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1364                         goto delete;
1365
1366                 if (fsck_err_on(i &&
1367                                 !S_ISREG(i->inode.bi_mode) &&
1368                                 !S_ISLNK(i->inode.bi_mode),
1369                                 c, extent_in_non_reg_inode,
1370                                 "extent in non regular inode mode %o:\n  %s",
1371                                 i->inode.bi_mode,
1372                                 (printbuf_reset(&buf),
1373                                  bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1374                         goto delete;
1375
1376                 ret = check_overlapping_extents(trans, s, extent_ends, k,
1377                                                 equiv.snapshot, iter,
1378                                                 &inode->recalculate_sums);
1379                 if (ret)
1380                         goto err;
1381         }
1382
1383         /*
1384          * Check inodes in reverse order, from oldest snapshots to newest,
1385          * starting from the inode that matches this extent's snapshot. If we
1386          * didn't have one, iterate over all inodes:
1387          */
1388         if (!i)
1389                 i = inode->inodes.data + inode->inodes.nr - 1;
1390
1391         for (;
1392              inode->inodes.data && i >= inode->inodes.data;
1393              --i) {
1394                 if (i->snapshot > equiv.snapshot ||
1395                     !key_visible_in_snapshot(c, s, i->snapshot, equiv.snapshot))
1396                         continue;
1397
1398                 if (k.k->type != KEY_TYPE_whiteout) {
1399                         if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_size_dirty) &&
1400                                         k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9 &&
1401                                         !bkey_extent_is_reservation(k),
1402                                         c, extent_past_end_of_inode,
1403                                         "extent type past end of inode %llu:%u, i_size %llu\n  %s",
1404                                         i->inode.bi_inum, i->snapshot, i->inode.bi_size,
1405                                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1406                                 struct btree_iter iter2;
1407
1408                                 bch2_trans_copy_iter(&iter2, iter);
1409                                 bch2_btree_iter_set_snapshot(&iter2, i->snapshot);
1410                                 ret =   bch2_btree_iter_traverse(&iter2) ?:
1411                                         bch2_btree_delete_at(trans, &iter2,
1412                                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1413                                 bch2_trans_iter_exit(trans, &iter2);
1414                                 if (ret)
1415                                         goto err;
1416
1417                                 iter->k.type = KEY_TYPE_whiteout;
1418                         }
1419
1420                         if (bkey_extent_is_allocation(k.k))
1421                                 i->count += k.k->size;
1422                 }
1423
1424                 i->seen_this_pos = true;
1425         }
1426 out:
1427 err:
1428 fsck_err:
1429         printbuf_exit(&buf);
1430         bch_err_fn(c, ret);
1431         return ret;
1432 delete:
1433         ret = bch2_btree_delete_at(trans, iter, BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1434         goto out;
1435 }
1436
1437 /*
1438  * Walk extents: verify that extents have a corresponding S_ISREG inode, and
1439  * that i_size an i_sectors are consistent
1440  */
1441 int bch2_check_extents(struct bch_fs *c)
1442 {
1443         struct inode_walker w = inode_walker_init();
1444         struct snapshots_seen s;
1445         struct btree_trans *trans = bch2_trans_get(c);
1446         struct btree_iter iter;
1447         struct bkey_s_c k;
1448         struct extent_ends extent_ends;
1449         struct disk_reservation res = { 0 };
1450         int ret = 0;
1451
1452         snapshots_seen_init(&s);
1453         extent_ends_init(&extent_ends);
1454
1455         ret = for_each_btree_key_commit(trans, iter, BTREE_ID_extents,
1456                         POS(BCACHEFS_ROOT_INO, 0),
1457                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1458                         &res, NULL,
1459                         BCH_TRANS_COMMIT_no_enospc, ({
1460                 bch2_disk_reservation_put(c, &res);
1461                 check_extent(trans, &iter, k, &w, &s, &extent_ends) ?:
1462                 check_extent_overbig(trans, &iter, k);
1463         })) ?:
1464         check_i_sectors(trans, &w);
1465
1466         bch2_disk_reservation_put(c, &res);
1467         extent_ends_exit(&extent_ends);
1468         inode_walker_exit(&w);
1469         snapshots_seen_exit(&s);
1470         bch2_trans_put(trans);
1471
1472         bch_err_fn(c, ret);
1473         return ret;
1474 }
1475
1476 int bch2_check_indirect_extents(struct bch_fs *c)
1477 {
1478         struct btree_trans *trans = bch2_trans_get(c);
1479         struct btree_iter iter;
1480         struct bkey_s_c k;
1481         struct disk_reservation res = { 0 };
1482         int ret = 0;
1483
1484         ret = for_each_btree_key_commit(trans, iter, BTREE_ID_reflink,
1485                         POS_MIN,
1486                         BTREE_ITER_PREFETCH, k,
1487                         &res, NULL,
1488                         BCH_TRANS_COMMIT_no_enospc, ({
1489                 bch2_disk_reservation_put(c, &res);
1490                 check_extent_overbig(trans, &iter, k);
1491         }));
1492
1493         bch2_disk_reservation_put(c, &res);
1494         bch2_trans_put(trans);
1495
1496         bch_err_fn(c, ret);
1497         return ret;
1498 }
1499
1500 static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
1501 {
1502         struct bch_fs *c = trans->c;
1503         struct inode_walker_entry *i;
1504         u32 restart_count = trans->restart_count;
1505         int ret = 0;
1506         s64 count2;
1507
1508         darray_for_each(w->inodes, i) {
1509                 if (i->inode.bi_nlink == i->count)
1510                         continue;
1511
1512                 count2 = bch2_count_subdirs(trans, w->last_pos.inode, i->snapshot);
1513                 if (count2 < 0)
1514                         return count2;
1515
1516                 if (i->count != count2) {
1517                         bch_err(c, "fsck counted subdirectories wrong: got %llu should be %llu",
1518                                 i->count, count2);
1519                         i->count = count2;
1520                         if (i->inode.bi_nlink == i->count)
1521                                 continue;
1522                 }
1523
1524                 if (fsck_err_on(i->inode.bi_nlink != i->count,
1525                                 c, inode_dir_wrong_nlink,
1526                                 "directory %llu:%u with wrong i_nlink: got %u, should be %llu",
1527                                 w->last_pos.inode, i->snapshot, i->inode.bi_nlink, i->count)) {
1528                         i->inode.bi_nlink = i->count;
1529                         ret = fsck_write_inode(trans, &i->inode, i->snapshot);
1530                         if (ret)
1531                                 break;
1532                 }
1533         }
1534 fsck_err:
1535         bch_err_fn(c, ret);
1536         return ret ?: trans_was_restarted(trans, restart_count);
1537 }
1538
1539 static int check_dirent_target(struct btree_trans *trans,
1540                                struct btree_iter *iter,
1541                                struct bkey_s_c_dirent d,
1542                                struct bch_inode_unpacked *target,
1543                                u32 target_snapshot)
1544 {
1545         struct bch_fs *c = trans->c;
1546         struct bkey_i_dirent *n;
1547         struct printbuf buf = PRINTBUF;
1548         struct btree_iter bp_iter = { NULL };
1549         int ret = 0;
1550
1551         if (!target->bi_dir &&
1552             !target->bi_dir_offset) {
1553                 target->bi_dir          = d.k->p.inode;
1554                 target->bi_dir_offset   = d.k->p.offset;
1555
1556                 ret = __write_inode(trans, target, target_snapshot);
1557                 if (ret)
1558                         goto err;
1559         }
1560
1561         if (!inode_points_to_dirent(target, d)) {
1562                 struct bkey_s_c_dirent bp_dirent = dirent_get_by_pos(trans, &bp_iter,
1563                                       SPOS(target->bi_dir, target->bi_dir_offset, target_snapshot));
1564                 ret = bkey_err(bp_dirent);
1565                 if (ret && !bch2_err_matches(ret, ENOENT))
1566                         goto err;
1567
1568                 bool backpointer_exists = !ret;
1569                 ret = 0;
1570
1571                 bch2_bkey_val_to_text(&buf, c, d.s_c);
1572                 prt_newline(&buf);
1573                 if (backpointer_exists)
1574                         bch2_bkey_val_to_text(&buf, c, bp_dirent.s_c);
1575
1576                 if (fsck_err_on(S_ISDIR(target->bi_mode) && backpointer_exists,
1577                                 c, inode_dir_multiple_links,
1578                                 "directory %llu:%u with multiple links\n%s",
1579                                 target->bi_inum, target_snapshot, buf.buf)) {
1580                         ret = __remove_dirent(trans, d.k->p);
1581                         goto out;
1582                 }
1583
1584                 /*
1585                  * hardlinked file with nlink 0:
1586                  * We're just adjusting nlink here so check_nlinks() will pick
1587                  * it up, it ignores inodes with nlink 0
1588                  */
1589                 if (fsck_err_on(backpointer_exists && !target->bi_nlink,
1590                                 c, inode_multiple_links_but_nlink_0,
1591                                 "inode %llu:%u type %s has multiple links but i_nlink 0\n%s",
1592                                 target->bi_inum, target_snapshot, bch2_d_types[d.v->d_type], buf.buf)) {
1593                         target->bi_nlink++;
1594                         target->bi_flags &= ~BCH_INODE_unlinked;
1595
1596                         ret = __write_inode(trans, target, target_snapshot);
1597                         if (ret)
1598                                 goto err;
1599                 }
1600
1601                 if (fsck_err_on(!backpointer_exists,
1602                                 c, inode_wrong_backpointer,
1603                                 "inode %llu:%u has wrong backpointer:\n"
1604                                 "got       %llu:%llu\n"
1605                                 "should be %llu:%llu",
1606                                 target->bi_inum, target_snapshot,
1607                                 target->bi_dir,
1608                                 target->bi_dir_offset,
1609                                 d.k->p.inode,
1610                                 d.k->p.offset)) {
1611                         target->bi_dir          = d.k->p.inode;
1612                         target->bi_dir_offset   = d.k->p.offset;
1613
1614                         ret = __write_inode(trans, target, target_snapshot);
1615                         if (ret)
1616                                 goto err;
1617                 }
1618         }
1619
1620         if (fsck_err_on(d.v->d_type != inode_d_type(target),
1621                         c, dirent_d_type_wrong,
1622                         "incorrect d_type: got %s, should be %s:\n%s",
1623                         bch2_d_type_str(d.v->d_type),
1624                         bch2_d_type_str(inode_d_type(target)),
1625                         (printbuf_reset(&buf),
1626                          bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
1627                 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1628                 ret = PTR_ERR_OR_ZERO(n);
1629                 if (ret)
1630                         goto err;
1631
1632                 bkey_reassemble(&n->k_i, d.s_c);
1633                 n->v.d_type = inode_d_type(target);
1634
1635                 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1636                 if (ret)
1637                         goto err;
1638
1639                 d = dirent_i_to_s_c(n);
1640         }
1641
1642         if (d.v->d_type == DT_SUBVOL &&
1643             target->bi_parent_subvol != le32_to_cpu(d.v->d_parent_subvol) &&
1644             (c->sb.version < bcachefs_metadata_version_subvol_dirent ||
1645              fsck_err(c, dirent_d_parent_subvol_wrong,
1646                       "dirent has wrong d_parent_subvol field: got %u, should be %u",
1647                       le32_to_cpu(d.v->d_parent_subvol),
1648                       target->bi_parent_subvol))) {
1649                 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1650                 ret = PTR_ERR_OR_ZERO(n);
1651                 if (ret)
1652                         goto err;
1653
1654                 bkey_reassemble(&n->k_i, d.s_c);
1655                 n->v.d_parent_subvol = cpu_to_le32(target->bi_parent_subvol);
1656
1657                 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1658                 if (ret)
1659                         goto err;
1660
1661                 d = dirent_i_to_s_c(n);
1662         }
1663 out:
1664 err:
1665 fsck_err:
1666         bch2_trans_iter_exit(trans, &bp_iter);
1667         printbuf_exit(&buf);
1668         bch_err_fn(c, ret);
1669         return ret;
1670 }
1671
1672 static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
1673                         struct bkey_s_c k,
1674                         struct bch_hash_info *hash_info,
1675                         struct inode_walker *dir,
1676                         struct inode_walker *target,
1677                         struct snapshots_seen *s)
1678 {
1679         struct bch_fs *c = trans->c;
1680         struct bkey_s_c_dirent d;
1681         struct inode_walker_entry *i;
1682         struct printbuf buf = PRINTBUF;
1683         struct bpos equiv;
1684         int ret = 0;
1685
1686         ret = check_key_has_snapshot(trans, iter, k);
1687         if (ret) {
1688                 ret = ret < 0 ? ret : 0;
1689                 goto out;
1690         }
1691
1692         equiv = k.k->p;
1693         equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1694
1695         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1696         if (ret)
1697                 goto err;
1698
1699         if (k.k->type == KEY_TYPE_whiteout)
1700                 goto out;
1701
1702         if (dir->last_pos.inode != k.k->p.inode) {
1703                 ret = check_subdir_count(trans, dir);
1704                 if (ret)
1705                         goto err;
1706         }
1707
1708         BUG_ON(!iter->path->should_be_locked);
1709
1710         i = walk_inode(trans, dir, equiv, k.k->type == KEY_TYPE_whiteout);
1711         ret = PTR_ERR_OR_ZERO(i);
1712         if (ret < 0)
1713                 goto err;
1714
1715         if (dir->first_this_inode && dir->inodes.nr)
1716                 *hash_info = bch2_hash_info_init(c, &dir->inodes.data[0].inode);
1717         dir->first_this_inode = false;
1718
1719         if (fsck_err_on(!i, c, dirent_in_missing_dir_inode,
1720                         "dirent in nonexisting directory:\n%s",
1721                         (printbuf_reset(&buf),
1722                          bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1723                 ret = bch2_btree_delete_at(trans, iter,
1724                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1725                 goto out;
1726         }
1727
1728         if (!i)
1729                 goto out;
1730
1731         if (fsck_err_on(!S_ISDIR(i->inode.bi_mode),
1732                         c, dirent_in_non_dir_inode,
1733                         "dirent in non directory inode type %s:\n%s",
1734                         bch2_d_type_str(inode_d_type(&i->inode)),
1735                         (printbuf_reset(&buf),
1736                          bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1737                 ret = bch2_btree_delete_at(trans, iter, 0);
1738                 goto out;
1739         }
1740
1741         ret = hash_check_key(trans, bch2_dirent_hash_desc, hash_info, iter, k);
1742         if (ret < 0)
1743                 goto err;
1744         if (ret) {
1745                 /* dirent has been deleted */
1746                 ret = 0;
1747                 goto out;
1748         }
1749
1750         if (k.k->type != KEY_TYPE_dirent)
1751                 goto out;
1752
1753         d = bkey_s_c_to_dirent(k);
1754
1755         if (d.v->d_type == DT_SUBVOL) {
1756                 struct bch_inode_unpacked subvol_root;
1757                 u32 target_subvol = le32_to_cpu(d.v->d_child_subvol);
1758                 u32 target_snapshot;
1759                 u64 target_inum;
1760
1761                 ret = __subvol_lookup(trans, target_subvol,
1762                                       &target_snapshot, &target_inum);
1763                 if (ret && !bch2_err_matches(ret, ENOENT))
1764                         goto err;
1765
1766                 if (fsck_err_on(ret, c, dirent_to_missing_subvol,
1767                                 "dirent points to missing subvolume %u",
1768                                 le32_to_cpu(d.v->d_child_subvol))) {
1769                         ret = __remove_dirent(trans, d.k->p);
1770                         goto err;
1771                 }
1772
1773                 ret = __lookup_inode(trans, target_inum,
1774                                    &subvol_root, &target_snapshot);
1775                 if (ret && !bch2_err_matches(ret, ENOENT))
1776                         goto err;
1777
1778                 if (fsck_err_on(ret, c, subvol_to_missing_root,
1779                                 "subvolume %u points to missing subvolume root %llu",
1780                                 target_subvol,
1781                                 target_inum)) {
1782                         bch_err(c, "repair not implemented yet");
1783                         ret = -EINVAL;
1784                         goto err;
1785                 }
1786
1787                 if (fsck_err_on(subvol_root.bi_subvol != target_subvol,
1788                                 c, subvol_root_wrong_bi_subvol,
1789                                 "subvol root %llu has wrong bi_subvol field: got %u, should be %u",
1790                                 target_inum,
1791                                 subvol_root.bi_subvol, target_subvol)) {
1792                         subvol_root.bi_subvol = target_subvol;
1793                         ret = __write_inode(trans, &subvol_root, target_snapshot);
1794                         if (ret)
1795                                 goto err;
1796                 }
1797
1798                 ret = check_dirent_target(trans, iter, d, &subvol_root,
1799                                           target_snapshot);
1800                 if (ret)
1801                         goto err;
1802         } else {
1803                 ret = __get_visible_inodes(trans, target, s, le64_to_cpu(d.v->d_inum));
1804                 if (ret)
1805                         goto err;
1806
1807                 if (fsck_err_on(!target->inodes.nr,
1808                                 c, dirent_to_missing_inode,
1809                                 "dirent points to missing inode: (equiv %u)\n%s",
1810                                 equiv.snapshot,
1811                                 (printbuf_reset(&buf),
1812                                  bch2_bkey_val_to_text(&buf, c, k),
1813                                  buf.buf))) {
1814                         ret = __remove_dirent(trans, d.k->p);
1815                         if (ret)
1816                                 goto err;
1817                 }
1818
1819                 darray_for_each(target->inodes, i) {
1820                         ret = check_dirent_target(trans, iter, d,
1821                                                   &i->inode, i->snapshot);
1822                         if (ret)
1823                                 goto err;
1824                 }
1825         }
1826
1827         if (d.v->d_type == DT_DIR)
1828                 for_each_visible_inode(c, s, dir, equiv.snapshot, i)
1829                         i->count++;
1830
1831 out:
1832 err:
1833 fsck_err:
1834         printbuf_exit(&buf);
1835         bch_err_fn(c, ret);
1836         return ret;
1837 }
1838
1839 /*
1840  * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
1841  * validate d_type
1842  */
1843 int bch2_check_dirents(struct bch_fs *c)
1844 {
1845         struct inode_walker dir = inode_walker_init();
1846         struct inode_walker target = inode_walker_init();
1847         struct snapshots_seen s;
1848         struct bch_hash_info hash_info;
1849         struct btree_trans *trans = bch2_trans_get(c);
1850         struct btree_iter iter;
1851         struct bkey_s_c k;
1852         int ret = 0;
1853
1854         snapshots_seen_init(&s);
1855
1856         ret = for_each_btree_key_commit(trans, iter, BTREE_ID_dirents,
1857                         POS(BCACHEFS_ROOT_INO, 0),
1858                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
1859                         k,
1860                         NULL, NULL,
1861                         BCH_TRANS_COMMIT_no_enospc,
1862                 check_dirent(trans, &iter, k, &hash_info, &dir, &target, &s));
1863
1864         bch2_trans_put(trans);
1865         snapshots_seen_exit(&s);
1866         inode_walker_exit(&dir);
1867         inode_walker_exit(&target);
1868         bch_err_fn(c, ret);
1869         return ret;
1870 }
1871
1872 static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
1873                        struct bkey_s_c k,
1874                        struct bch_hash_info *hash_info,
1875                        struct inode_walker *inode)
1876 {
1877         struct bch_fs *c = trans->c;
1878         struct inode_walker_entry *i;
1879         int ret;
1880
1881         ret = check_key_has_snapshot(trans, iter, k);
1882         if (ret)
1883                 return ret;
1884
1885         i = walk_inode(trans, inode, k.k->p, k.k->type == KEY_TYPE_whiteout);
1886         ret = PTR_ERR_OR_ZERO(i);
1887         if (ret)
1888                 return ret;
1889
1890         if (inode->first_this_inode && inode->inodes.nr)
1891                 *hash_info = bch2_hash_info_init(c, &inode->inodes.data[0].inode);
1892         inode->first_this_inode = false;
1893
1894         if (fsck_err_on(!i, c, xattr_in_missing_inode,
1895                         "xattr for missing inode %llu",
1896                         k.k->p.inode))
1897                 return bch2_btree_delete_at(trans, iter, 0);
1898
1899         if (!i)
1900                 return 0;
1901
1902         ret = hash_check_key(trans, bch2_xattr_hash_desc, hash_info, iter, k);
1903 fsck_err:
1904         bch_err_fn(c, ret);
1905         return ret;
1906 }
1907
1908 /*
1909  * Walk xattrs: verify that they all have a corresponding inode
1910  */
1911 int bch2_check_xattrs(struct bch_fs *c)
1912 {
1913         struct inode_walker inode = inode_walker_init();
1914         struct bch_hash_info hash_info;
1915         struct btree_iter iter;
1916         struct bkey_s_c k;
1917         int ret = 0;
1918
1919         ret = bch2_trans_run(c,
1920                 for_each_btree_key_commit(trans, iter, BTREE_ID_xattrs,
1921                         POS(BCACHEFS_ROOT_INO, 0),
1922                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
1923                         k,
1924                         NULL, NULL,
1925                         BCH_TRANS_COMMIT_no_enospc,
1926                 check_xattr(trans, &iter, k, &hash_info, &inode)));
1927         bch_err_fn(c, ret);
1928         return ret;
1929 }
1930
1931 static int check_root_trans(struct btree_trans *trans)
1932 {
1933         struct bch_fs *c = trans->c;
1934         struct bch_inode_unpacked root_inode;
1935         u32 snapshot;
1936         u64 inum;
1937         int ret;
1938
1939         ret = __subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
1940         if (ret && !bch2_err_matches(ret, ENOENT))
1941                 return ret;
1942
1943         if (mustfix_fsck_err_on(ret, c, root_subvol_missing,
1944                                 "root subvol missing")) {
1945                 struct bkey_i_subvolume root_subvol;
1946
1947                 snapshot        = U32_MAX;
1948                 inum            = BCACHEFS_ROOT_INO;
1949
1950                 bkey_subvolume_init(&root_subvol.k_i);
1951                 root_subvol.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1952                 root_subvol.v.flags     = 0;
1953                 root_subvol.v.snapshot  = cpu_to_le32(snapshot);
1954                 root_subvol.v.inode     = cpu_to_le64(inum);
1955                 ret = commit_do(trans, NULL, NULL,
1956                                       BCH_TRANS_COMMIT_no_enospc,
1957                         bch2_btree_insert_trans(trans, BTREE_ID_subvolumes,
1958                                             &root_subvol.k_i, 0));
1959                 bch_err_msg(c, ret, "writing root subvol");
1960                 if (ret)
1961                         goto err;
1962
1963         }
1964
1965         ret = __lookup_inode(trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
1966         if (ret && !bch2_err_matches(ret, ENOENT))
1967                 return ret;
1968
1969         if (mustfix_fsck_err_on(ret, c, root_dir_missing,
1970                                 "root directory missing") ||
1971             mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode),
1972                                 c, root_inode_not_dir,
1973                                 "root inode not a directory")) {
1974                 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755,
1975                                 0, NULL);
1976                 root_inode.bi_inum = inum;
1977
1978                 ret = __write_inode(trans, &root_inode, snapshot);
1979                 bch_err_msg(c, ret, "writing root inode");
1980         }
1981 err:
1982 fsck_err:
1983         return ret;
1984 }
1985
1986 /* Get root directory, create if it doesn't exist: */
1987 int bch2_check_root(struct bch_fs *c)
1988 {
1989         int ret;
1990
1991         ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1992                 check_root_trans(trans));
1993         bch_err_fn(c, ret);
1994         return ret;
1995 }
1996
1997 struct pathbuf_entry {
1998         u64     inum;
1999         u32     snapshot;
2000 };
2001
2002 typedef DARRAY(struct pathbuf_entry) pathbuf;
2003
2004 static bool path_is_dup(pathbuf *p, u64 inum, u32 snapshot)
2005 {
2006         struct pathbuf_entry *i;
2007
2008         darray_for_each(*p, i)
2009                 if (i->inum     == inum &&
2010                     i->snapshot == snapshot)
2011                         return true;
2012
2013         return false;
2014 }
2015
2016 static int path_down(struct bch_fs *c, pathbuf *p,
2017                      u64 inum, u32 snapshot)
2018 {
2019         int ret = darray_push(p, ((struct pathbuf_entry) {
2020                 .inum           = inum,
2021                 .snapshot       = snapshot,
2022         }));
2023
2024         if (ret)
2025                 bch_err(c, "fsck: error allocating memory for pathbuf, size %zu",
2026                         p->size);
2027         return ret;
2028 }
2029
2030 /*
2031  * Check that a given inode is reachable from the root:
2032  *
2033  * XXX: we should also be verifying that inodes are in the right subvolumes
2034  */
2035 static int check_path(struct btree_trans *trans,
2036                       pathbuf *p,
2037                       struct bch_inode_unpacked *inode,
2038                       u32 snapshot)
2039 {
2040         struct bch_fs *c = trans->c;
2041         int ret = 0;
2042
2043         snapshot = bch2_snapshot_equiv(c, snapshot);
2044         p->nr = 0;
2045
2046         while (!(inode->bi_inum == BCACHEFS_ROOT_INO &&
2047                  inode->bi_subvol == BCACHEFS_ROOT_SUBVOL)) {
2048                 struct btree_iter dirent_iter;
2049                 struct bkey_s_c_dirent d;
2050                 u32 parent_snapshot = snapshot;
2051
2052                 if (inode->bi_subvol) {
2053                         u64 inum;
2054
2055                         ret = subvol_lookup(trans, inode->bi_parent_subvol,
2056                                             &parent_snapshot, &inum);
2057                         if (ret)
2058                                 break;
2059                 }
2060
2061                 ret = lockrestart_do(trans,
2062                         PTR_ERR_OR_ZERO((d = dirent_get_by_pos(trans, &dirent_iter,
2063                                           SPOS(inode->bi_dir, inode->bi_dir_offset,
2064                                                parent_snapshot))).k));
2065                 if (ret && !bch2_err_matches(ret, ENOENT))
2066                         break;
2067
2068                 if (!ret && !dirent_points_to_inode(d, inode)) {
2069                         bch2_trans_iter_exit(trans, &dirent_iter);
2070                         ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
2071                 }
2072
2073                 if (bch2_err_matches(ret, ENOENT)) {
2074                         if (fsck_err(c,  inode_unreachable,
2075                                      "unreachable inode %llu:%u, type %s nlink %u backptr %llu:%llu",
2076                                      inode->bi_inum, snapshot,
2077                                      bch2_d_type_str(inode_d_type(inode)),
2078                                      inode->bi_nlink,
2079                                      inode->bi_dir,
2080                                      inode->bi_dir_offset))
2081                                 ret = reattach_inode(trans, inode, snapshot);
2082                         break;
2083                 }
2084
2085                 bch2_trans_iter_exit(trans, &dirent_iter);
2086
2087                 if (!S_ISDIR(inode->bi_mode))
2088                         break;
2089
2090                 ret = path_down(c, p, inode->bi_inum, snapshot);
2091                 if (ret) {
2092                         bch_err(c, "memory allocation failure");
2093                         return ret;
2094                 }
2095
2096                 snapshot = parent_snapshot;
2097
2098                 ret = lookup_inode(trans, inode->bi_dir, inode, &snapshot);
2099                 if (ret) {
2100                         /* Should have been caught in dirents pass */
2101                         bch_err(c, "error looking up parent directory: %i", ret);
2102                         break;
2103                 }
2104
2105                 if (path_is_dup(p, inode->bi_inum, snapshot)) {
2106                         struct pathbuf_entry *i;
2107
2108                         /* XXX print path */
2109                         bch_err(c, "directory structure loop");
2110
2111                         darray_for_each(*p, i)
2112                                 pr_err("%llu:%u", i->inum, i->snapshot);
2113                         pr_err("%llu:%u", inode->bi_inum, snapshot);
2114
2115                         if (!fsck_err(c, dir_loop,
2116                                       "directory structure loop"))
2117                                 return 0;
2118
2119                         ret = commit_do(trans, NULL, NULL,
2120                                               BCH_TRANS_COMMIT_no_enospc,
2121                                         remove_backpointer(trans, inode));
2122                         if (ret) {
2123                                 bch_err(c, "error removing dirent: %i", ret);
2124                                 break;
2125                         }
2126
2127                         ret = reattach_inode(trans, inode, snapshot);
2128                 }
2129         }
2130 fsck_err:
2131         bch_err_fn(c, ret);
2132         return ret;
2133 }
2134
2135 /*
2136  * Check for unreachable inodes, as well as loops in the directory structure:
2137  * After bch2_check_dirents(), if an inode backpointer doesn't exist that means it's
2138  * unreachable:
2139  */
2140 int bch2_check_directory_structure(struct bch_fs *c)
2141 {
2142         struct btree_trans *trans = bch2_trans_get(c);
2143         struct btree_iter iter;
2144         struct bkey_s_c k;
2145         struct bch_inode_unpacked u;
2146         pathbuf path = { 0, };
2147         int ret;
2148
2149         for_each_btree_key(trans, iter, BTREE_ID_inodes, POS_MIN,
2150                            BTREE_ITER_INTENT|
2151                            BTREE_ITER_PREFETCH|
2152                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2153                 if (!bkey_is_inode(k.k))
2154                         continue;
2155
2156                 ret = bch2_inode_unpack(k, &u);
2157                 if (ret) {
2158                         /* Should have been caught earlier in fsck: */
2159                         bch_err(c, "error unpacking inode %llu: %i", k.k->p.offset, ret);
2160                         break;
2161                 }
2162
2163                 if (u.bi_flags & BCH_INODE_unlinked)
2164                         continue;
2165
2166                 ret = check_path(trans, &path, &u, iter.pos.snapshot);
2167                 if (ret)
2168                         break;
2169         }
2170         bch2_trans_iter_exit(trans, &iter);
2171         bch2_trans_put(trans);
2172         darray_exit(&path);
2173         bch_err_fn(c, ret);
2174         return ret;
2175 }
2176
2177 struct nlink_table {
2178         size_t          nr;
2179         size_t          size;
2180
2181         struct nlink {
2182                 u64     inum;
2183                 u32     snapshot;
2184                 u32     count;
2185         }               *d;
2186 };
2187
2188 static int add_nlink(struct bch_fs *c, struct nlink_table *t,
2189                      u64 inum, u32 snapshot)
2190 {
2191         if (t->nr == t->size) {
2192                 size_t new_size = max_t(size_t, 128UL, t->size * 2);
2193                 void *d = kvmalloc_array(new_size, sizeof(t->d[0]), GFP_KERNEL);
2194
2195                 if (!d) {
2196                         bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
2197                                 new_size);
2198                         return -BCH_ERR_ENOMEM_fsck_add_nlink;
2199                 }
2200
2201                 if (t->d)
2202                         memcpy(d, t->d, t->size * sizeof(t->d[0]));
2203                 kvfree(t->d);
2204
2205                 t->d = d;
2206                 t->size = new_size;
2207         }
2208
2209
2210         t->d[t->nr++] = (struct nlink) {
2211                 .inum           = inum,
2212                 .snapshot       = snapshot,
2213         };
2214
2215         return 0;
2216 }
2217
2218 static int nlink_cmp(const void *_l, const void *_r)
2219 {
2220         const struct nlink *l = _l;
2221         const struct nlink *r = _r;
2222
2223         return cmp_int(l->inum, r->inum);
2224 }
2225
2226 static void inc_link(struct bch_fs *c, struct snapshots_seen *s,
2227                      struct nlink_table *links,
2228                      u64 range_start, u64 range_end, u64 inum, u32 snapshot)
2229 {
2230         struct nlink *link, key = {
2231                 .inum = inum, .snapshot = U32_MAX,
2232         };
2233
2234         if (inum < range_start || inum >= range_end)
2235                 return;
2236
2237         link = __inline_bsearch(&key, links->d, links->nr,
2238                                 sizeof(links->d[0]), nlink_cmp);
2239         if (!link)
2240                 return;
2241
2242         while (link > links->d && link[0].inum == link[-1].inum)
2243                 --link;
2244
2245         for (; link < links->d + links->nr && link->inum == inum; link++)
2246                 if (ref_visible(c, s, snapshot, link->snapshot)) {
2247                         link->count++;
2248                         if (link->snapshot >= snapshot)
2249                                 break;
2250                 }
2251 }
2252
2253 noinline_for_stack
2254 static int check_nlinks_find_hardlinks(struct bch_fs *c,
2255                                        struct nlink_table *t,
2256                                        u64 start, u64 *end)
2257 {
2258         struct btree_trans *trans = bch2_trans_get(c);
2259         struct btree_iter iter;
2260         struct bkey_s_c k;
2261         struct bch_inode_unpacked u;
2262         int ret = 0;
2263
2264         for_each_btree_key(trans, iter, BTREE_ID_inodes,
2265                            POS(0, start),
2266                            BTREE_ITER_INTENT|
2267                            BTREE_ITER_PREFETCH|
2268                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2269                 if (!bkey_is_inode(k.k))
2270                         continue;
2271
2272                 /* Should never fail, checked by bch2_inode_invalid: */
2273                 BUG_ON(bch2_inode_unpack(k, &u));
2274
2275                 /*
2276                  * Backpointer and directory structure checks are sufficient for
2277                  * directories, since they can't have hardlinks:
2278                  */
2279                 if (S_ISDIR(u.bi_mode))
2280                         continue;
2281
2282                 if (!u.bi_nlink)
2283                         continue;
2284
2285                 ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot);
2286                 if (ret) {
2287                         *end = k.k->p.offset;
2288                         ret = 0;
2289                         break;
2290                 }
2291
2292         }
2293         bch2_trans_iter_exit(trans, &iter);
2294         bch2_trans_put(trans);
2295
2296         if (ret)
2297                 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
2298
2299         return ret;
2300 }
2301
2302 noinline_for_stack
2303 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
2304                                      u64 range_start, u64 range_end)
2305 {
2306         struct btree_trans *trans = bch2_trans_get(c);
2307         struct snapshots_seen s;
2308         struct btree_iter iter;
2309         struct bkey_s_c k;
2310         struct bkey_s_c_dirent d;
2311         int ret;
2312
2313         snapshots_seen_init(&s);
2314
2315         for_each_btree_key(trans, iter, BTREE_ID_dirents, POS_MIN,
2316                            BTREE_ITER_INTENT|
2317                            BTREE_ITER_PREFETCH|
2318                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2319                 ret = snapshots_seen_update(c, &s, iter.btree_id, k.k->p);
2320                 if (ret)
2321                         break;
2322
2323                 switch (k.k->type) {
2324                 case KEY_TYPE_dirent:
2325                         d = bkey_s_c_to_dirent(k);
2326
2327                         if (d.v->d_type != DT_DIR &&
2328                             d.v->d_type != DT_SUBVOL)
2329                                 inc_link(c, &s, links, range_start, range_end,
2330                                          le64_to_cpu(d.v->d_inum),
2331                                          bch2_snapshot_equiv(c, d.k->p.snapshot));
2332                         break;
2333                 }
2334         }
2335         bch2_trans_iter_exit(trans, &iter);
2336
2337         if (ret)
2338                 bch_err(c, "error in fsck: btree error %i while walking dirents", ret);
2339
2340         bch2_trans_put(trans);
2341         snapshots_seen_exit(&s);
2342         return ret;
2343 }
2344
2345 static int check_nlinks_update_inode(struct btree_trans *trans, struct btree_iter *iter,
2346                                      struct bkey_s_c k,
2347                                      struct nlink_table *links,
2348                                      size_t *idx, u64 range_end)
2349 {
2350         struct bch_fs *c = trans->c;
2351         struct bch_inode_unpacked u;
2352         struct nlink *link = &links->d[*idx];
2353         int ret = 0;
2354
2355         if (k.k->p.offset >= range_end)
2356                 return 1;
2357
2358         if (!bkey_is_inode(k.k))
2359                 return 0;
2360
2361         BUG_ON(bch2_inode_unpack(k, &u));
2362
2363         if (S_ISDIR(u.bi_mode))
2364                 return 0;
2365
2366         if (!u.bi_nlink)
2367                 return 0;
2368
2369         while ((cmp_int(link->inum, k.k->p.offset) ?:
2370                 cmp_int(link->snapshot, k.k->p.snapshot)) < 0) {
2371                 BUG_ON(*idx == links->nr);
2372                 link = &links->d[++*idx];
2373         }
2374
2375         if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count,
2376                         c, inode_wrong_nlink,
2377                         "inode %llu type %s has wrong i_nlink (%u, should be %u)",
2378                         u.bi_inum, bch2_d_types[mode_to_type(u.bi_mode)],
2379                         bch2_inode_nlink_get(&u), link->count)) {
2380                 bch2_inode_nlink_set(&u, link->count);
2381                 ret = __write_inode(trans, &u, k.k->p.snapshot);
2382         }
2383 fsck_err:
2384         return ret;
2385 }
2386
2387 noinline_for_stack
2388 static int check_nlinks_update_hardlinks(struct bch_fs *c,
2389                                struct nlink_table *links,
2390                                u64 range_start, u64 range_end)
2391 {
2392         struct btree_iter iter;
2393         struct bkey_s_c k;
2394         size_t idx = 0;
2395         int ret = 0;
2396
2397         ret = bch2_trans_run(c,
2398                 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
2399                                 POS(0, range_start),
2400                                 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
2401                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2402                         check_nlinks_update_inode(trans, &iter, k, links, &idx, range_end)));
2403         if (ret < 0) {
2404                 bch_err(c, "error in fsck walking inodes: %s", bch2_err_str(ret));
2405                 return ret;
2406         }
2407
2408         return 0;
2409 }
2410
2411 int bch2_check_nlinks(struct bch_fs *c)
2412 {
2413         struct nlink_table links = { 0 };
2414         u64 this_iter_range_start, next_iter_range_start = 0;
2415         int ret = 0;
2416
2417         do {
2418                 this_iter_range_start = next_iter_range_start;
2419                 next_iter_range_start = U64_MAX;
2420
2421                 ret = check_nlinks_find_hardlinks(c, &links,
2422                                                   this_iter_range_start,
2423                                                   &next_iter_range_start);
2424
2425                 ret = check_nlinks_walk_dirents(c, &links,
2426                                           this_iter_range_start,
2427                                           next_iter_range_start);
2428                 if (ret)
2429                         break;
2430
2431                 ret = check_nlinks_update_hardlinks(c, &links,
2432                                          this_iter_range_start,
2433                                          next_iter_range_start);
2434                 if (ret)
2435                         break;
2436
2437                 links.nr = 0;
2438         } while (next_iter_range_start != U64_MAX);
2439
2440         kvfree(links.d);
2441         bch_err_fn(c, ret);
2442         return ret;
2443 }
2444
2445 static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter,
2446                              struct bkey_s_c k)
2447 {
2448         struct bkey_s_c_reflink_p p;
2449         struct bkey_i_reflink_p *u;
2450         int ret;
2451
2452         if (k.k->type != KEY_TYPE_reflink_p)
2453                 return 0;
2454
2455         p = bkey_s_c_to_reflink_p(k);
2456
2457         if (!p.v->front_pad && !p.v->back_pad)
2458                 return 0;
2459
2460         u = bch2_trans_kmalloc(trans, sizeof(*u));
2461         ret = PTR_ERR_OR_ZERO(u);
2462         if (ret)
2463                 return ret;
2464
2465         bkey_reassemble(&u->k_i, k);
2466         u->v.front_pad  = 0;
2467         u->v.back_pad   = 0;
2468
2469         return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_NORUN);
2470 }
2471
2472 int bch2_fix_reflink_p(struct bch_fs *c)
2473 {
2474         struct btree_iter iter;
2475         struct bkey_s_c k;
2476         int ret;
2477
2478         if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
2479                 return 0;
2480
2481         ret = bch2_trans_run(c,
2482                 for_each_btree_key_commit(trans, iter,
2483                                 BTREE_ID_extents, POS_MIN,
2484                                 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|
2485                                 BTREE_ITER_ALL_SNAPSHOTS, k,
2486                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2487                         fix_reflink_p_key(trans, &iter, k)));
2488         bch_err_fn(c, ret);
2489         return ret;
2490 }