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,
 211                                   BCH_TRANS_COMMIT_no_enospc|
 212                                   BCH_TRANS_COMMIT_lazy_rw,
 213                                   __write_inode(trans, inode, snapshot));
 214         if (ret)
 215                 bch_err_fn(trans->c, ret);
 216         return ret;
 217 }
 218
 219 static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
 220 {
 221         struct bch_fs *c = trans->c;
 222         struct btree_iter iter;
 223         struct bch_inode_unpacked dir_inode;
 224         struct bch_hash_info dir_hash_info;
 225         int ret;
 226
 227         ret = lookup_first_inode(trans, pos.inode, &dir_inode);
 228         if (ret)
 229                 goto err;
 230
 231         dir_hash_info = bch2_hash_info_init(c, &dir_inode);
 232
 233         bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_INTENT);
 234
 235         ret = bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
 236                                   &dir_hash_info, &iter,
 237                                   BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 238         bch2_trans_iter_exit(trans, &iter);
 239 err:
 240         bch_err_fn(c, ret);
 241         return ret;
 242 }
 243
 244 /* Get lost+found, create if it doesn't exist: */
 245 static int lookup_lostfound(struct btree_trans *trans, u32 subvol,
 246                             struct bch_inode_unpacked *lostfound)
 247 {
 248         struct bch_fs *c = trans->c;
 249         struct bch_inode_unpacked root;
 250         struct bch_hash_info root_hash_info;
 251         struct qstr lostfound_str = QSTR("lost+found");
 252         subvol_inum root_inum = { .subvol = subvol };
 253         u64 inum = 0;
 254         unsigned d_type = 0;
 255         u32 snapshot;
 256         int ret;
 257
 258         ret = __subvol_lookup(trans, subvol, &snapshot, &root_inum.inum);
 259         if (ret)
 260                 return ret;
 261
 262         ret = __lookup_inode(trans, root_inum.inum, &root, &snapshot);
 263         if (ret)
 264                 return ret;
 265
 266         root_hash_info = bch2_hash_info_init(c, &root);
 267
 268         ret = __lookup_dirent(trans, root_hash_info, root_inum,
 269                             &lostfound_str, &inum, &d_type);
 270         if (bch2_err_matches(ret, ENOENT)) {
 271                 bch_notice(c, "creating lost+found");
 272                 goto create_lostfound;
 273         }
 274
 275         bch_err_fn(c, ret);
 276         if (ret)
 277                 return ret;
 278
 279         if (d_type != DT_DIR) {
 280                 bch_err(c, "error looking up lost+found: not a directory");
 281                 return -BCH_ERR_ENOENT_not_directory;
 282         }
 283
 284         /*
 285          * The bch2_check_dirents pass has already run, dangling dirents
 286          * shouldn't exist here:
 287          */
 288         return __lookup_inode(trans, inum, lostfound, &snapshot);
 289
 290 create_lostfound:
 291         bch2_inode_init_early(c, lostfound);
 292
 293         ret = bch2_create_trans(trans, root_inum, &root,
 294                                 lostfound, &lostfound_str,
 295                                 0, 0, S_IFDIR|0700, 0, NULL, NULL,
 296                                 (subvol_inum) { }, 0);
 297         bch_err_msg(c, ret, "creating lost+found");
 298         return ret;
 299 }
 300
 301 static int __reattach_inode(struct btree_trans *trans,
 302                           struct bch_inode_unpacked *inode,
 303                           u32 inode_snapshot)
 304 {
 305         struct bch_hash_info dir_hash;
 306         struct bch_inode_unpacked lostfound;
 307         char name_buf[20];
 308         struct qstr name;
 309         u64 dir_offset = 0;
 310         u32 subvol;
 311         int ret;
 312
 313         ret = __snapshot_lookup_subvol(trans, inode_snapshot, &subvol);
 314         if (ret)
 315                 return ret;
 316
 317         ret = lookup_lostfound(trans, subvol, &lostfound);
 318         if (ret)
 319                 return ret;
 320
 321         if (S_ISDIR(inode->bi_mode)) {
 322                 lostfound.bi_nlink++;
 323
 324                 ret = __write_inode(trans, &lostfound, U32_MAX);
 325                 if (ret)
 326                         return ret;
 327         }
 328
 329         dir_hash = bch2_hash_info_init(trans->c, &lostfound);
 330
 331         snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
 332         name = (struct qstr) QSTR(name_buf);
 333
 334         ret = bch2_dirent_create(trans,
 335                                  (subvol_inum) {
 336                                         .subvol = subvol,
 337                                         .inum = lostfound.bi_inum,
 338                                  },
 339                                  &dir_hash,
 340                                  inode_d_type(inode),
 341                                  &name, inode->bi_inum, &dir_offset,
 342                                  BCH_HASH_SET_MUST_CREATE);
 343         if (ret)
 344                 return ret;
 345
 346         inode->bi_dir           = lostfound.bi_inum;
 347         inode->bi_dir_offset    = dir_offset;
 348
 349         return __write_inode(trans, inode, inode_snapshot);
 350 }
 351
 352 static int reattach_inode(struct btree_trans *trans,
 353                           struct bch_inode_unpacked *inode,
 354                           u32 inode_snapshot)
 355 {
 356         int ret = commit_do(trans, NULL, NULL,
 357                                   BCH_TRANS_COMMIT_lazy_rw|
 358                                   BCH_TRANS_COMMIT_no_enospc,
 359                         __reattach_inode(trans, inode, inode_snapshot));
 360         bch_err_msg(trans->c, ret, "reattaching inode %llu", inode->bi_inum);
 361         return ret;
 362 }
 363
 364 static int remove_backpointer(struct btree_trans *trans,
 365                               struct bch_inode_unpacked *inode)
 366 {
 367         struct btree_iter iter;
 368         struct bkey_s_c_dirent d;
 369         int ret;
 370
 371         d = bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_dirents,
 372                                      POS(inode->bi_dir, inode->bi_dir_offset), 0,
 373                                      dirent);
 374         ret =   bkey_err(d) ?:
 375                 __remove_dirent(trans, d.k->p);
 376         bch2_trans_iter_exit(trans, &iter);
 377         return ret;
 378 }
 379
 380 struct snapshots_seen_entry {
 381         u32                             id;
 382         u32                             equiv;
 383 };
 384
 385 struct snapshots_seen {
 386         struct bpos                     pos;
 387         DARRAY(struct snapshots_seen_entry) ids;
 388 };
 389
 390 static inline void snapshots_seen_exit(struct snapshots_seen *s)
 391 {
 392         darray_exit(&s->ids);
 393 }
 394
 395 static inline void snapshots_seen_init(struct snapshots_seen *s)
 396 {
 397         memset(s, 0, sizeof(*s));
 398 }
 399
 400 static int snapshots_seen_add_inorder(struct bch_fs *c, struct snapshots_seen *s, u32 id)
 401 {
 402         struct snapshots_seen_entry *i, n = {
 403                 .id     = id,
 404                 .equiv  = bch2_snapshot_equiv(c, id),
 405         };
 406         int ret = 0;
 407
 408         darray_for_each(s->ids, i) {
 409                 if (i->id == id)
 410                         return 0;
 411                 if (i->id > id)
 412                         break;
 413         }
 414
 415         ret = darray_insert_item(&s->ids, i - s->ids.data, n);
 416         if (ret)
 417                 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
 418                         s->ids.size);
 419         return ret;
 420 }
 421
 422 static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s,
 423                                  enum btree_id btree_id, struct bpos pos)
 424 {
 425         struct snapshots_seen_entry *i, n = {
 426                 .id     = pos.snapshot,
 427                 .equiv  = bch2_snapshot_equiv(c, pos.snapshot),
 428         };
 429         int ret = 0;
 430
 431         if (!bkey_eq(s->pos, pos))
 432                 s->ids.nr = 0;
 433
 434         s->pos = pos;
 435         s->pos.snapshot = n.equiv;
 436
 437         darray_for_each(s->ids, i) {
 438                 if (i->id == n.id)
 439                         return 0;
 440
 441                 /*
 442                  * We currently don't rigorously track for snapshot cleanup
 443                  * needing to be run, so it shouldn't be a fsck error yet:
 444                  */
 445                 if (i->equiv == n.equiv) {
 446                         bch_err(c, "snapshot deletion did not finish:\n"
 447                                 "  duplicate keys in btree %s at %llu:%llu snapshots %u, %u (equiv %u)\n",
 448                                 bch2_btree_id_str(btree_id),
 449                                 pos.inode, pos.offset,
 450                                 i->id, n.id, n.equiv);
 451                         set_bit(BCH_FS_NEED_DELETE_DEAD_SNAPSHOTS, &c->flags);
 452                         return bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_delete_dead_snapshots);
 453                 }
 454         }
 455
 456         ret = darray_push(&s->ids, n);
 457         if (ret)
 458                 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
 459                         s->ids.size);
 460         return ret;
 461 }
 462
 463 /**
 464  * key_visible_in_snapshot - returns true if @id is a descendent of @ancestor,
 465  * and @ancestor hasn't been overwritten in @seen
 466  *
 467  * @c:          filesystem handle
 468  * @seen:       list of snapshot ids already seen at current position
 469  * @id:         descendent snapshot id
 470  * @ancestor:   ancestor snapshot id
 471  *
 472  * Returns:     whether key in @ancestor snapshot is visible in @id snapshot
 473  */
 474 static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *seen,
 475                                     u32 id, u32 ancestor)
 476 {
 477         ssize_t i;
 478
 479         EBUG_ON(id > ancestor);
 480         EBUG_ON(!bch2_snapshot_is_equiv(c, id));
 481         EBUG_ON(!bch2_snapshot_is_equiv(c, ancestor));
 482
 483         /* @ancestor should be the snapshot most recently added to @seen */
 484         EBUG_ON(ancestor != seen->pos.snapshot);
 485         EBUG_ON(ancestor != seen->ids.data[seen->ids.nr - 1].equiv);
 486
 487         if (id == ancestor)
 488                 return true;
 489
 490         if (!bch2_snapshot_is_ancestor(c, id, ancestor))
 491                 return false;
 492
 493         /*
 494          * We know that @id is a descendant of @ancestor, we're checking if
 495          * we've seen a key that overwrote @ancestor - i.e. also a descendent of
 496          * @ascestor and with @id as a descendent.
 497          *
 498          * But we already know that we're scanning IDs between @id and @ancestor
 499          * numerically, since snapshot ID lists are kept sorted, so if we find
 500          * an id that's an ancestor of @id we're done:
 501          */
 502
 503         for (i = seen->ids.nr - 2;
 504              i >= 0 && seen->ids.data[i].equiv >= id;
 505              --i)
 506                 if (bch2_snapshot_is_ancestor(c, id, seen->ids.data[i].equiv))
 507                         return false;
 508
 509         return true;
 510 }
 511
 512 /**
 513  * ref_visible - given a key with snapshot id @src that points to a key with
 514  * snapshot id @dst, test whether there is some snapshot in which @dst is
 515  * visible.
 516  *
 517  * @c:          filesystem handle
 518  * @s:          list of snapshot IDs already seen at @src
 519  * @src:        snapshot ID of src key
 520  * @dst:        snapshot ID of dst key
 521  * Returns:     true if there is some snapshot in which @dst is visible
 522  *
 523  * Assumes we're visiting @src keys in natural key order
 524  */
 525 static bool ref_visible(struct bch_fs *c, struct snapshots_seen *s,
 526                         u32 src, u32 dst)
 527 {
 528         return dst <= src
 529                 ? key_visible_in_snapshot(c, s, dst, src)
 530                 : bch2_snapshot_is_ancestor(c, src, dst);
 531 }
 532
 533 static int ref_visible2(struct bch_fs *c,
 534                         u32 src, struct snapshots_seen *src_seen,
 535                         u32 dst, struct snapshots_seen *dst_seen)
 536 {
 537         src = bch2_snapshot_equiv(c, src);
 538         dst = bch2_snapshot_equiv(c, dst);
 539
 540         if (dst > src) {
 541                 swap(dst, src);
 542                 swap(dst_seen, src_seen);
 543         }
 544         return key_visible_in_snapshot(c, src_seen, dst, src);
 545 }
 546
 547 #define for_each_visible_inode(_c, _s, _w, _snapshot, _i)                               \
 548         for (_i = (_w)->inodes.data; _i < (_w)->inodes.data + (_w)->inodes.nr &&        \
 549              (_i)->snapshot <= (_snapshot); _i++)                                       \
 550                 if (key_visible_in_snapshot(_c, _s, _i->snapshot, _snapshot))
 551
 552 struct inode_walker_entry {
 553         struct bch_inode_unpacked inode;
 554         u32                     snapshot;
 555         bool                    seen_this_pos;
 556         u64                     count;
 557 };
 558
 559 struct inode_walker {
 560         bool                            first_this_inode;
 561         bool                            recalculate_sums;
 562         struct bpos                     last_pos;
 563
 564         DARRAY(struct inode_walker_entry) inodes;
 565 };
 566
 567 static void inode_walker_exit(struct inode_walker *w)
 568 {
 569         darray_exit(&w->inodes);
 570 }
 571
 572 static struct inode_walker inode_walker_init(void)
 573 {
 574         return (struct inode_walker) { 0, };
 575 }
 576
 577 static int add_inode(struct bch_fs *c, struct inode_walker *w,
 578                      struct bkey_s_c inode)
 579 {
 580         struct bch_inode_unpacked u;
 581
 582         BUG_ON(bch2_inode_unpack(inode, &u));
 583
 584         return darray_push(&w->inodes, ((struct inode_walker_entry) {
 585                 .inode          = u,
 586                 .snapshot       = bch2_snapshot_equiv(c, inode.k->p.snapshot),
 587         }));
 588 }
 589
 590 static int get_inodes_all_snapshots(struct btree_trans *trans,
 591                                     struct inode_walker *w, u64 inum)
 592 {
 593         struct bch_fs *c = trans->c;
 594         struct btree_iter iter;
 595         struct bkey_s_c k;
 596         u32 restart_count = trans->restart_count;
 597         int ret;
 598
 599         w->recalculate_sums = false;
 600         w->inodes.nr = 0;
 601
 602         for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, inum),
 603                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
 604                 if (k.k->p.offset != inum)
 605                         break;
 606
 607                 if (bkey_is_inode(k.k))
 608                         add_inode(c, w, k);
 609         }
 610         bch2_trans_iter_exit(trans, &iter);
 611
 612         if (ret)
 613                 return ret;
 614
 615         w->first_this_inode = true;
 616
 617         return trans_was_restarted(trans, restart_count);
 618 }
 619
 620 static struct inode_walker_entry *
 621 lookup_inode_for_snapshot(struct bch_fs *c, struct inode_walker *w,
 622                           u32 snapshot, bool is_whiteout)
 623 {
 624         struct inode_walker_entry *i;
 625
 626         snapshot = bch2_snapshot_equiv(c, snapshot);
 627
 628         darray_for_each(w->inodes, i)
 629                 if (bch2_snapshot_is_ancestor(c, snapshot, i->snapshot))
 630                         goto found;
 631
 632         return NULL;
 633 found:
 634         BUG_ON(snapshot > i->snapshot);
 635
 636         if (snapshot != i->snapshot && !is_whiteout) {
 637                 struct inode_walker_entry new = *i;
 638                 size_t pos;
 639                 int ret;
 640
 641                 new.snapshot = snapshot;
 642                 new.count = 0;
 643
 644                 bch_info(c, "have key for inode %llu:%u but have inode in ancestor snapshot %u",
 645                          w->last_pos.inode, snapshot, i->snapshot);
 646
 647                 while (i > w->inodes.data && i[-1].snapshot > snapshot)
 648                         --i;
 649
 650                 pos = i - w->inodes.data;
 651                 ret = darray_insert_item(&w->inodes, pos, new);
 652                 if (ret)
 653                         return ERR_PTR(ret);
 654
 655                 i = w->inodes.data + pos;
 656         }
 657
 658         return i;
 659 }
 660
 661 static struct inode_walker_entry *walk_inode(struct btree_trans *trans,
 662                                              struct inode_walker *w, struct bpos pos,
 663                                              bool is_whiteout)
 664 {
 665         if (w->last_pos.inode != pos.inode) {
 666                 int ret = get_inodes_all_snapshots(trans, w, pos.inode);
 667                 if (ret)
 668                         return ERR_PTR(ret);
 669         } else if (bkey_cmp(w->last_pos, pos)) {
 670                 struct inode_walker_entry *i;
 671
 672                 darray_for_each(w->inodes, i)
 673                         i->seen_this_pos = false;
 674
 675         }
 676
 677         w->last_pos = pos;
 678
 679         return lookup_inode_for_snapshot(trans->c, w, pos.snapshot, is_whiteout);
 680 }
 681
 682 static int __get_visible_inodes(struct btree_trans *trans,
 683                                 struct inode_walker *w,
 684                                 struct snapshots_seen *s,
 685                                 u64 inum)
 686 {
 687         struct bch_fs *c = trans->c;
 688         struct btree_iter iter;
 689         struct bkey_s_c k;
 690         int ret;
 691
 692         w->inodes.nr = 0;
 693
 694         for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum),
 695                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
 696                 u32 equiv = bch2_snapshot_equiv(c, k.k->p.snapshot);
 697
 698                 if (k.k->p.offset != inum)
 699                         break;
 700
 701                 if (!ref_visible(c, s, s->pos.snapshot, equiv))
 702                         continue;
 703
 704                 if (bkey_is_inode(k.k))
 705                         add_inode(c, w, k);
 706
 707                 if (equiv >= s->pos.snapshot)
 708                         break;
 709         }
 710         bch2_trans_iter_exit(trans, &iter);
 711
 712         return ret;
 713 }
 714
 715 static int check_key_has_snapshot(struct btree_trans *trans,
 716                                   struct btree_iter *iter,
 717                                   struct bkey_s_c k)
 718 {
 719         struct bch_fs *c = trans->c;
 720         struct printbuf buf = PRINTBUF;
 721         int ret = 0;
 722
 723         if (mustfix_fsck_err_on(!bch2_snapshot_equiv(c, k.k->p.snapshot), c,
 724                                 bkey_in_missing_snapshot,
 725                                 "key in missing snapshot: %s",
 726                                 (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
 727                 ret = bch2_btree_delete_at(trans, iter,
 728                                             BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?: 1;
 729 fsck_err:
 730         printbuf_exit(&buf);
 731         return ret;
 732 }
 733
 734 static int hash_redo_key(struct btree_trans *trans,
 735                          const struct bch_hash_desc desc,
 736                          struct bch_hash_info *hash_info,
 737                          struct btree_iter *k_iter, struct bkey_s_c k)
 738 {
 739         struct bkey_i *delete;
 740         struct bkey_i *tmp;
 741
 742         delete = bch2_trans_kmalloc(trans, sizeof(*delete));
 743         if (IS_ERR(delete))
 744                 return PTR_ERR(delete);
 745
 746         tmp = bch2_bkey_make_mut_noupdate(trans, k);
 747         if (IS_ERR(tmp))
 748                 return PTR_ERR(tmp);
 749
 750         bkey_init(&delete->k);
 751         delete->k.p = k_iter->pos;
 752         return  bch2_btree_iter_traverse(k_iter) ?:
 753                 bch2_trans_update(trans, k_iter, delete, 0) ?:
 754                 bch2_hash_set_snapshot(trans, desc, hash_info,
 755                                        (subvol_inum) { 0, k.k->p.inode },
 756                                        k.k->p.snapshot, tmp,
 757                                        BCH_HASH_SET_MUST_CREATE,
 758                                        BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
 759                 bch2_trans_commit(trans, NULL, NULL,
 760                                   BCH_TRANS_COMMIT_no_enospc|
 761                                   BCH_TRANS_COMMIT_lazy_rw);
 762 }
 763
 764 static int hash_check_key(struct btree_trans *trans,
 765                           const struct bch_hash_desc desc,
 766                           struct bch_hash_info *hash_info,
 767                           struct btree_iter *k_iter, struct bkey_s_c hash_k)
 768 {
 769         struct bch_fs *c = trans->c;
 770         struct btree_iter iter = { NULL };
 771         struct printbuf buf = PRINTBUF;
 772         struct bkey_s_c k;
 773         u64 hash;
 774         int ret = 0;
 775
 776         if (hash_k.k->type != desc.key_type)
 777                 return 0;
 778
 779         hash = desc.hash_bkey(hash_info, hash_k);
 780
 781         if (likely(hash == hash_k.k->p.offset))
 782                 return 0;
 783
 784         if (hash_k.k->p.offset < hash)
 785                 goto bad_hash;
 786
 787         for_each_btree_key_norestart(trans, iter, desc.btree_id,
 788                                      SPOS(hash_k.k->p.inode, hash, hash_k.k->p.snapshot),
 789                                      BTREE_ITER_SLOTS, k, ret) {
 790                 if (bkey_eq(k.k->p, hash_k.k->p))
 791                         break;
 792
 793                 if (fsck_err_on(k.k->type == desc.key_type &&
 794                                 !desc.cmp_bkey(k, hash_k), c,
 795                                 hash_table_key_duplicate,
 796                                 "duplicate hash table keys:\n%s",
 797                                 (printbuf_reset(&buf),
 798                                  bch2_bkey_val_to_text(&buf, c, hash_k),
 799                                  buf.buf))) {
 800                         ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0) ?: 1;
 801                         break;
 802                 }
 803
 804                 if (bkey_deleted(k.k)) {
 805                         bch2_trans_iter_exit(trans, &iter);
 806                         goto bad_hash;
 807                 }
 808         }
 809 out:
 810         bch2_trans_iter_exit(trans, &iter);
 811         printbuf_exit(&buf);
 812         return ret;
 813 bad_hash:
 814         if (fsck_err(c, hash_table_key_wrong_offset,
 815                      "hash table key at wrong offset: btree %s inode %llu offset %llu, hashed to %llu\n%s",
 816                      bch2_btree_id_str(desc.btree_id), hash_k.k->p.inode, hash_k.k->p.offset, hash,
 817                      (printbuf_reset(&buf),
 818                       bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf))) {
 819                 ret = hash_redo_key(trans, desc, hash_info, k_iter, hash_k);
 820                 bch_err_fn(c, ret);
 821                 if (ret)
 822                         return ret;
 823                 ret = -BCH_ERR_transaction_restart_nested;
 824         }
 825 fsck_err:
 826         goto out;
 827 }
 828
 829 static int check_inode(struct btree_trans *trans,
 830                        struct btree_iter *iter,
 831                        struct bkey_s_c k,
 832                        struct bch_inode_unpacked *prev,
 833                        struct snapshots_seen *s,
 834                        bool full)
 835 {
 836         struct bch_fs *c = trans->c;
 837         struct bch_inode_unpacked u;
 838         bool do_update = false;
 839         int ret;
 840
 841         ret = check_key_has_snapshot(trans, iter, k);
 842         if (ret < 0)
 843                 goto err;
 844         if (ret)
 845                 return 0;
 846
 847         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
 848         if (ret)
 849                 goto err;
 850
 851         if (!bkey_is_inode(k.k))
 852                 return 0;
 853
 854         BUG_ON(bch2_inode_unpack(k, &u));
 855
 856         if (!full &&
 857             !(u.bi_flags & (BCH_INODE_i_size_dirty|
 858                             BCH_INODE_i_sectors_dirty|
 859                             BCH_INODE_unlinked)))
 860                 return 0;
 861
 862         if (prev->bi_inum != u.bi_inum)
 863                 *prev = u;
 864
 865         if (fsck_err_on(prev->bi_hash_seed      != u.bi_hash_seed ||
 866                         inode_d_type(prev)      != inode_d_type(&u),
 867                         c, inode_snapshot_mismatch,
 868                         "inodes in different snapshots don't match")) {
 869                 bch_err(c, "repair not implemented yet");
 870                 return -EINVAL;
 871         }
 872
 873         if ((u.bi_flags & (BCH_INODE_i_size_dirty|BCH_INODE_unlinked)) &&
 874             bch2_key_has_snapshot_overwrites(trans, BTREE_ID_inodes, k.k->p)) {
 875                 struct bpos new_min_pos;
 876
 877                 ret = bch2_propagate_key_to_snapshot_leaves(trans, iter->btree_id, k, &new_min_pos);
 878                 if (ret)
 879                         goto err;
 880
 881                 u.bi_flags &= ~BCH_INODE_i_size_dirty|BCH_INODE_unlinked;
 882
 883                 ret = __write_inode(trans, &u, iter->pos.snapshot);
 884                 bch_err_msg(c, ret, "in fsck updating inode");
 885                 if (ret)
 886                         return ret;
 887
 888                 if (!bpos_eq(new_min_pos, POS_MIN))
 889                         bch2_btree_iter_set_pos(iter, bpos_predecessor(new_min_pos));
 890                 return 0;
 891         }
 892
 893         if (u.bi_flags & BCH_INODE_unlinked &&
 894             (!c->sb.clean ||
 895              fsck_err(c, inode_unlinked_but_clean,
 896                       "filesystem marked clean, but inode %llu unlinked",
 897                       u.bi_inum))) {
 898                 bch2_trans_unlock(trans);
 899                 bch2_fs_lazy_rw(c);
 900
 901                 ret = bch2_inode_rm_snapshot(trans, u.bi_inum, iter->pos.snapshot);
 902                 bch_err_msg(c, ret, "in fsck deleting inode");
 903                 return ret;
 904         }
 905
 906         if (u.bi_flags & BCH_INODE_i_size_dirty &&
 907             (!c->sb.clean ||
 908              fsck_err(c, inode_i_size_dirty_but_clean,
 909                       "filesystem marked clean, but inode %llu has i_size dirty",
 910                       u.bi_inum))) {
 911                 bch_verbose(c, "truncating inode %llu", u.bi_inum);
 912
 913                 bch2_trans_unlock(trans);
 914                 bch2_fs_lazy_rw(c);
 915
 916                 /*
 917                  * XXX: need to truncate partial blocks too here - or ideally
 918                  * just switch units to bytes and that issue goes away
 919                  */
 920                 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
 921                                 SPOS(u.bi_inum, round_up(u.bi_size, block_bytes(c)) >> 9,
 922                                      iter->pos.snapshot),
 923                                 POS(u.bi_inum, U64_MAX),
 924                                 0, NULL);
 925                 bch_err_msg(c, ret, "in fsck truncating inode");
 926                 if (ret)
 927                         return ret;
 928
 929                 /*
 930                  * We truncated without our normal sector accounting hook, just
 931                  * make sure we recalculate it:
 932                  */
 933                 u.bi_flags |= BCH_INODE_i_sectors_dirty;
 934
 935                 u.bi_flags &= ~BCH_INODE_i_size_dirty;
 936                 do_update = true;
 937         }
 938
 939         if (u.bi_flags & BCH_INODE_i_sectors_dirty &&
 940             (!c->sb.clean ||
 941              fsck_err(c, inode_i_sectors_dirty_but_clean,
 942                       "filesystem marked clean, but inode %llu has i_sectors dirty",
 943                       u.bi_inum))) {
 944                 s64 sectors;
 945
 946                 bch_verbose(c, "recounting sectors for inode %llu",
 947                             u.bi_inum);
 948
 949                 sectors = bch2_count_inode_sectors(trans, u.bi_inum, iter->pos.snapshot);
 950                 if (sectors < 0) {
 951                         bch_err_msg(c, sectors, "in fsck recounting inode sectors");
 952                         return sectors;
 953                 }
 954
 955                 u.bi_sectors = sectors;
 956                 u.bi_flags &= ~BCH_INODE_i_sectors_dirty;
 957                 do_update = true;
 958         }
 959
 960         if (u.bi_flags & BCH_INODE_backptr_untrusted) {
 961                 u.bi_dir = 0;
 962                 u.bi_dir_offset = 0;
 963                 u.bi_flags &= ~BCH_INODE_backptr_untrusted;
 964                 do_update = true;
 965         }
 966
 967         if (do_update) {
 968                 ret = __write_inode(trans, &u, iter->pos.snapshot);
 969                 bch_err_msg(c, ret, "in fsck updating inode");
 970                 if (ret)
 971                         return ret;
 972         }
 973 err:
 974 fsck_err:
 975         bch_err_fn(c, ret);
 976         return ret;
 977 }
 978
 979 noinline_for_stack
 980 int bch2_check_inodes(struct bch_fs *c)
 981 {
 982         bool full = c->opts.fsck;
 983         struct btree_trans *trans = bch2_trans_get(c);
 984         struct btree_iter iter;
 985         struct bch_inode_unpacked prev = { 0 };
 986         struct snapshots_seen s;
 987         struct bkey_s_c k;
 988         int ret;
 989
 990         snapshots_seen_init(&s);
 991
 992         ret = for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
 993                         POS_MIN,
 994                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
 995                         NULL, NULL, BCH_TRANS_COMMIT_lazy_rw|BCH_TRANS_COMMIT_no_enospc,
 996                 check_inode(trans, &iter, k, &prev, &s, full));
 997
 998         snapshots_seen_exit(&s);
 999         bch2_trans_put(trans);
1000         bch_err_fn(c, ret);
1001         return ret;
1002 }
1003
1004 static struct bkey_s_c_dirent dirent_get_by_pos(struct btree_trans *trans,
1005                                                 struct btree_iter *iter,
1006                                                 struct bpos pos)
1007 {
1008         return bch2_bkey_get_iter_typed(trans, iter, BTREE_ID_dirents, pos, 0, dirent);
1009 }
1010
1011 static bool inode_points_to_dirent(struct bch_inode_unpacked *inode,
1012                                    struct bkey_s_c_dirent d)
1013 {
1014         return  inode->bi_dir           == d.k->p.inode &&
1015                 inode->bi_dir_offset    == d.k->p.offset;
1016 }
1017
1018 static bool dirent_points_to_inode(struct bkey_s_c_dirent d,
1019                                    struct bch_inode_unpacked *inode)
1020 {
1021         return d.v->d_type == DT_SUBVOL
1022                 ? le32_to_cpu(d.v->d_child_subvol)      == inode->bi_subvol
1023                 : le64_to_cpu(d.v->d_inum)              == inode->bi_inum;
1024 }
1025
1026 static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
1027 {
1028         struct bch_fs *c = trans->c;
1029         struct inode_walker_entry *i;
1030         u32 restart_count = trans->restart_count;
1031         int ret = 0;
1032         s64 count2;
1033
1034         darray_for_each(w->inodes, i) {
1035                 if (i->inode.bi_sectors == i->count)
1036                         continue;
1037
1038                 count2 = bch2_count_inode_sectors(trans, w->last_pos.inode, i->snapshot);
1039
1040                 if (w->recalculate_sums)
1041                         i->count = count2;
1042
1043                 if (i->count != count2) {
1044                         bch_err(c, "fsck counted i_sectors wrong for inode %llu:%u: got %llu should be %llu",
1045                                 w->last_pos.inode, i->snapshot, i->count, count2);
1046                         return -BCH_ERR_internal_fsck_err;
1047                 }
1048
1049                 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_sectors_dirty),
1050                                 c, inode_i_sectors_wrong,
1051                                 "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
1052                                 w->last_pos.inode, i->snapshot,
1053                                 i->inode.bi_sectors, i->count)) {
1054                         i->inode.bi_sectors = i->count;
1055                         ret = fsck_write_inode(trans, &i->inode, i->snapshot);
1056                         if (ret)
1057                                 break;
1058                 }
1059         }
1060 fsck_err:
1061         bch_err_fn(c, ret);
1062         return ret ?: trans_was_restarted(trans, restart_count);
1063 }
1064
1065 struct extent_end {
1066         u32                     snapshot;
1067         u64                     offset;
1068         struct snapshots_seen   seen;
1069 };
1070
1071 struct extent_ends {
1072         struct bpos                     last_pos;
1073         DARRAY(struct extent_end)       e;
1074 };
1075
1076 static void extent_ends_reset(struct extent_ends *extent_ends)
1077 {
1078         struct extent_end *i;
1079
1080         darray_for_each(extent_ends->e, i)
1081                 snapshots_seen_exit(&i->seen);
1082
1083         extent_ends->e.nr = 0;
1084 }
1085
1086 static void extent_ends_exit(struct extent_ends *extent_ends)
1087 {
1088         extent_ends_reset(extent_ends);
1089         darray_exit(&extent_ends->e);
1090 }
1091
1092 static void extent_ends_init(struct extent_ends *extent_ends)
1093 {
1094         memset(extent_ends, 0, sizeof(*extent_ends));
1095 }
1096
1097 static int extent_ends_at(struct bch_fs *c,
1098                           struct extent_ends *extent_ends,
1099                           struct snapshots_seen *seen,
1100                           struct bkey_s_c k)
1101 {
1102         struct extent_end *i, n = (struct extent_end) {
1103                 .offset         = k.k->p.offset,
1104                 .snapshot       = k.k->p.snapshot,
1105                 .seen           = *seen,
1106         };
1107
1108         n.seen.ids.data = kmemdup(seen->ids.data,
1109                               sizeof(seen->ids.data[0]) * seen->ids.size,
1110                               GFP_KERNEL);
1111         if (!n.seen.ids.data)
1112                 return -BCH_ERR_ENOMEM_fsck_extent_ends_at;
1113
1114         darray_for_each(extent_ends->e, i) {
1115                 if (i->snapshot == k.k->p.snapshot) {
1116                         snapshots_seen_exit(&i->seen);
1117                         *i = n;
1118                         return 0;
1119                 }
1120
1121                 if (i->snapshot >= k.k->p.snapshot)
1122                         break;
1123         }
1124
1125         return darray_insert_item(&extent_ends->e, i - extent_ends->e.data, n);
1126 }
1127
1128 static int overlapping_extents_found(struct btree_trans *trans,
1129                                      enum btree_id btree,
1130                                      struct bpos pos1, struct snapshots_seen *pos1_seen,
1131                                      struct bkey pos2,
1132                                      bool *fixed,
1133                                      struct extent_end *extent_end)
1134 {
1135         struct bch_fs *c = trans->c;
1136         struct printbuf buf = PRINTBUF;
1137         struct btree_iter iter1, iter2 = { NULL };
1138         struct bkey_s_c k1, k2;
1139         int ret;
1140
1141         BUG_ON(bkey_le(pos1, bkey_start_pos(&pos2)));
1142
1143         bch2_trans_iter_init(trans, &iter1, btree, pos1,
1144                              BTREE_ITER_ALL_SNAPSHOTS|
1145                              BTREE_ITER_NOT_EXTENTS);
1146         k1 = bch2_btree_iter_peek_upto(&iter1, POS(pos1.inode, U64_MAX));
1147         ret = bkey_err(k1);
1148         if (ret)
1149                 goto err;
1150
1151         prt_str(&buf, "\n  ");
1152         bch2_bkey_val_to_text(&buf, c, k1);
1153
1154         if (!bpos_eq(pos1, k1.k->p)) {
1155                 prt_str(&buf, "\n  wanted\n  ");
1156                 bch2_bpos_to_text(&buf, pos1);
1157                 prt_str(&buf, "\n  ");
1158                 bch2_bkey_to_text(&buf, &pos2);
1159
1160                 bch_err(c, "%s: error finding first overlapping extent when repairing, got%s",
1161                         __func__, buf.buf);
1162                 ret = -BCH_ERR_internal_fsck_err;
1163                 goto err;
1164         }
1165
1166         bch2_trans_copy_iter(&iter2, &iter1);
1167
1168         while (1) {
1169                 bch2_btree_iter_advance(&iter2);
1170
1171                 k2 = bch2_btree_iter_peek_upto(&iter2, POS(pos1.inode, U64_MAX));
1172                 ret = bkey_err(k2);
1173                 if (ret)
1174                         goto err;
1175
1176                 if (bpos_ge(k2.k->p, pos2.p))
1177                         break;
1178         }
1179
1180         prt_str(&buf, "\n  ");
1181         bch2_bkey_val_to_text(&buf, c, k2);
1182
1183         if (bpos_gt(k2.k->p, pos2.p) ||
1184             pos2.size != k2.k->size) {
1185                 bch_err(c, "%s: error finding seconding overlapping extent when repairing%s",
1186                         __func__, buf.buf);
1187                 ret = -BCH_ERR_internal_fsck_err;
1188                 goto err;
1189         }
1190
1191         prt_printf(&buf, "\n  overwriting %s extent",
1192                    pos1.snapshot >= pos2.p.snapshot ? "first" : "second");
1193
1194         if (fsck_err(c, extent_overlapping,
1195                      "overlapping extents%s", buf.buf)) {
1196                 struct btree_iter *old_iter = &iter1;
1197                 struct disk_reservation res = { 0 };
1198
1199                 if (pos1.snapshot < pos2.p.snapshot) {
1200                         old_iter = &iter2;
1201                         swap(k1, k2);
1202                 }
1203
1204                 trans->extra_journal_res += bch2_bkey_sectors_compressed(k2);
1205
1206                 ret =   bch2_trans_update_extent_overwrite(trans, old_iter,
1207                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE,
1208                                 k1, k2) ?:
1209                         bch2_trans_commit(trans, &res, NULL,
1210                                 BCH_TRANS_COMMIT_lazy_rw|BCH_TRANS_COMMIT_no_enospc);
1211                 bch2_disk_reservation_put(c, &res);
1212
1213                 if (ret)
1214                         goto err;
1215
1216                 *fixed = true;
1217
1218                 if (pos1.snapshot == pos2.p.snapshot) {
1219                         /*
1220                          * We overwrote the first extent, and did the overwrite
1221                          * in the same snapshot:
1222                          */
1223                         extent_end->offset = bkey_start_offset(&pos2);
1224                 } else if (pos1.snapshot > pos2.p.snapshot) {
1225                         /*
1226                          * We overwrote the first extent in pos2's snapshot:
1227                          */
1228                         ret = snapshots_seen_add_inorder(c, pos1_seen, pos2.p.snapshot);
1229                 } else {
1230                         /*
1231                          * We overwrote the second extent - restart
1232                          * check_extent() from the top:
1233                          */
1234                         ret = -BCH_ERR_transaction_restart_nested;
1235                 }
1236         }
1237 fsck_err:
1238 err:
1239         bch2_trans_iter_exit(trans, &iter2);
1240         bch2_trans_iter_exit(trans, &iter1);
1241         printbuf_exit(&buf);
1242         return ret;
1243 }
1244
1245 static int check_overlapping_extents(struct btree_trans *trans,
1246                               struct snapshots_seen *seen,
1247                               struct extent_ends *extent_ends,
1248                               struct bkey_s_c k,
1249                               u32 equiv,
1250                               struct btree_iter *iter,
1251                               bool *fixed)
1252 {
1253         struct bch_fs *c = trans->c;
1254         struct extent_end *i;
1255         int ret = 0;
1256
1257         /* transaction restart, running again */
1258         if (bpos_eq(extent_ends->last_pos, k.k->p))
1259                 return 0;
1260
1261         if (extent_ends->last_pos.inode != k.k->p.inode)
1262                 extent_ends_reset(extent_ends);
1263
1264         darray_for_each(extent_ends->e, i) {
1265                 if (i->offset <= bkey_start_offset(k.k))
1266                         continue;
1267
1268                 if (!ref_visible2(c,
1269                                   k.k->p.snapshot, seen,
1270                                   i->snapshot, &i->seen))
1271                         continue;
1272
1273                 ret = overlapping_extents_found(trans, iter->btree_id,
1274                                                 SPOS(iter->pos.inode,
1275                                                      i->offset,
1276                                                      i->snapshot),
1277                                                 &i->seen,
1278                                                 *k.k, fixed, i);
1279                 if (ret)
1280                         goto err;
1281         }
1282
1283         ret = extent_ends_at(c, extent_ends, seen, k);
1284         if (ret)
1285                 goto err;
1286
1287         extent_ends->last_pos = k.k->p;
1288 err:
1289         return ret;
1290 }
1291
1292 static int check_extent_overbig(struct btree_trans *trans, struct btree_iter *iter,
1293                                 struct bkey_s_c k)
1294 {
1295         struct bch_fs *c = trans->c;
1296         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1297         struct bch_extent_crc_unpacked crc;
1298         const union bch_extent_entry *i;
1299         unsigned encoded_extent_max_sectors = c->opts.encoded_extent_max >> 9;
1300
1301         bkey_for_each_crc(k.k, ptrs, crc, i)
1302                 if (crc_is_encoded(crc) &&
1303                     crc.uncompressed_size > encoded_extent_max_sectors) {
1304                         struct printbuf buf = PRINTBUF;
1305
1306                         bch2_bkey_val_to_text(&buf, c, k);
1307                         bch_err(c, "overbig encoded extent, please report this:\n  %s", buf.buf);
1308                         printbuf_exit(&buf);
1309                 }
1310
1311         return 0;
1312 }
1313
1314 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
1315                         struct bkey_s_c k,
1316                         struct inode_walker *inode,
1317                         struct snapshots_seen *s,
1318                         struct extent_ends *extent_ends)
1319 {
1320         struct bch_fs *c = trans->c;
1321         struct inode_walker_entry *i;
1322         struct printbuf buf = PRINTBUF;
1323         struct bpos equiv = k.k->p;
1324         int ret = 0;
1325
1326         equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1327
1328         ret = check_key_has_snapshot(trans, iter, k);
1329         if (ret) {
1330                 ret = ret < 0 ? ret : 0;
1331                 goto out;
1332         }
1333
1334         if (inode->last_pos.inode != k.k->p.inode) {
1335                 ret = check_i_sectors(trans, inode);
1336                 if (ret)
1337                         goto err;
1338         }
1339
1340         i = walk_inode(trans, inode, equiv, k.k->type == KEY_TYPE_whiteout);
1341         ret = PTR_ERR_OR_ZERO(i);
1342         if (ret)
1343                 goto err;
1344
1345         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1346         if (ret)
1347                 goto err;
1348
1349         if (k.k->type != KEY_TYPE_whiteout) {
1350                 if (fsck_err_on(!i, c, extent_in_missing_inode,
1351                                 "extent in missing inode:\n  %s",
1352                                 (printbuf_reset(&buf),
1353                                  bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1354                         goto delete;
1355
1356                 if (fsck_err_on(i &&
1357                                 !S_ISREG(i->inode.bi_mode) &&
1358                                 !S_ISLNK(i->inode.bi_mode),
1359                                 c, extent_in_non_reg_inode,
1360                                 "extent in non regular inode mode %o:\n  %s",
1361                                 i->inode.bi_mode,
1362                                 (printbuf_reset(&buf),
1363                                  bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1364                         goto delete;
1365
1366                 ret = check_overlapping_extents(trans, s, extent_ends, k,
1367                                                 equiv.snapshot, iter,
1368                                                 &inode->recalculate_sums);
1369                 if (ret)
1370                         goto err;
1371         }
1372
1373         /*
1374          * Check inodes in reverse order, from oldest snapshots to newest,
1375          * starting from the inode that matches this extent's snapshot. If we
1376          * didn't have one, iterate over all inodes:
1377          */
1378         if (!i)
1379                 i = inode->inodes.data + inode->inodes.nr - 1;
1380
1381         for (;
1382              inode->inodes.data && i >= inode->inodes.data;
1383              --i) {
1384                 if (i->snapshot > equiv.snapshot ||
1385                     !key_visible_in_snapshot(c, s, i->snapshot, equiv.snapshot))
1386                         continue;
1387
1388                 if (k.k->type != KEY_TYPE_whiteout) {
1389                         if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_size_dirty) &&
1390                                         k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9 &&
1391                                         !bkey_extent_is_reservation(k),
1392                                         c, extent_past_end_of_inode,
1393                                         "extent type past end of inode %llu:%u, i_size %llu\n  %s",
1394                                         i->inode.bi_inum, i->snapshot, i->inode.bi_size,
1395                                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1396                                 struct btree_iter iter2;
1397
1398                                 bch2_trans_copy_iter(&iter2, iter);
1399                                 bch2_btree_iter_set_snapshot(&iter2, i->snapshot);
1400                                 ret =   bch2_btree_iter_traverse(&iter2) ?:
1401                                         bch2_btree_delete_at(trans, &iter2,
1402                                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1403                                 bch2_trans_iter_exit(trans, &iter2);
1404                                 if (ret)
1405                                         goto err;
1406
1407                                 iter->k.type = KEY_TYPE_whiteout;
1408                         }
1409
1410                         if (bkey_extent_is_allocation(k.k))
1411                                 i->count += k.k->size;
1412                 }
1413
1414                 i->seen_this_pos = true;
1415         }
1416 out:
1417 err:
1418 fsck_err:
1419         printbuf_exit(&buf);
1420         bch_err_fn(c, ret);
1421         return ret;
1422 delete:
1423         ret = bch2_btree_delete_at(trans, iter, BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1424         goto out;
1425 }
1426
1427 /*
1428  * Walk extents: verify that extents have a corresponding S_ISREG inode, and
1429  * that i_size an i_sectors are consistent
1430  */
1431 int bch2_check_extents(struct bch_fs *c)
1432 {
1433         struct inode_walker w = inode_walker_init();
1434         struct snapshots_seen s;
1435         struct btree_trans *trans = bch2_trans_get(c);
1436         struct btree_iter iter;
1437         struct bkey_s_c k;
1438         struct extent_ends extent_ends;
1439         struct disk_reservation res = { 0 };
1440         int ret = 0;
1441
1442         snapshots_seen_init(&s);
1443         extent_ends_init(&extent_ends);
1444
1445         ret = for_each_btree_key_commit(trans, iter, BTREE_ID_extents,
1446                         POS(BCACHEFS_ROOT_INO, 0),
1447                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1448                         &res, NULL,
1449                         BCH_TRANS_COMMIT_lazy_rw|BCH_TRANS_COMMIT_no_enospc, ({
1450                 bch2_disk_reservation_put(c, &res);
1451                 check_extent(trans, &iter, k, &w, &s, &extent_ends) ?:
1452                 check_extent_overbig(trans, &iter, k);
1453         })) ?:
1454         check_i_sectors(trans, &w);
1455
1456         bch2_disk_reservation_put(c, &res);
1457         extent_ends_exit(&extent_ends);
1458         inode_walker_exit(&w);
1459         snapshots_seen_exit(&s);
1460         bch2_trans_put(trans);
1461
1462         bch_err_fn(c, ret);
1463         return ret;
1464 }
1465
1466 int bch2_check_indirect_extents(struct bch_fs *c)
1467 {
1468         struct btree_trans *trans = bch2_trans_get(c);
1469         struct btree_iter iter;
1470         struct bkey_s_c k;
1471         struct disk_reservation res = { 0 };
1472         int ret = 0;
1473
1474         ret = for_each_btree_key_commit(trans, iter, BTREE_ID_reflink,
1475                         POS_MIN,
1476                         BTREE_ITER_PREFETCH, k,
1477                         &res, NULL,
1478                         BCH_TRANS_COMMIT_lazy_rw|BCH_TRANS_COMMIT_no_enospc, ({
1479                 bch2_disk_reservation_put(c, &res);
1480                 check_extent_overbig(trans, &iter, k);
1481         }));
1482
1483         bch2_disk_reservation_put(c, &res);
1484         bch2_trans_put(trans);
1485
1486         bch_err_fn(c, ret);
1487         return ret;
1488 }
1489
1490 static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
1491 {
1492         struct bch_fs *c = trans->c;
1493         struct inode_walker_entry *i;
1494         u32 restart_count = trans->restart_count;
1495         int ret = 0;
1496         s64 count2;
1497
1498         darray_for_each(w->inodes, i) {
1499                 if (i->inode.bi_nlink == i->count)
1500                         continue;
1501
1502                 count2 = bch2_count_subdirs(trans, w->last_pos.inode, i->snapshot);
1503                 if (count2 < 0)
1504                         return count2;
1505
1506                 if (i->count != count2) {
1507                         bch_err(c, "fsck counted subdirectories wrong: got %llu should be %llu",
1508                                 i->count, count2);
1509                         i->count = count2;
1510                         if (i->inode.bi_nlink == i->count)
1511                                 continue;
1512                 }
1513
1514                 if (fsck_err_on(i->inode.bi_nlink != i->count,
1515                                 c, inode_dir_wrong_nlink,
1516                                 "directory %llu:%u with wrong i_nlink: got %u, should be %llu",
1517                                 w->last_pos.inode, i->snapshot, i->inode.bi_nlink, i->count)) {
1518                         i->inode.bi_nlink = i->count;
1519                         ret = fsck_write_inode(trans, &i->inode, i->snapshot);
1520                         if (ret)
1521                                 break;
1522                 }
1523         }
1524 fsck_err:
1525         bch_err_fn(c, ret);
1526         return ret ?: trans_was_restarted(trans, restart_count);
1527 }
1528
1529 static int check_dirent_target(struct btree_trans *trans,
1530                                struct btree_iter *iter,
1531                                struct bkey_s_c_dirent d,
1532                                struct bch_inode_unpacked *target,
1533                                u32 target_snapshot)
1534 {
1535         struct bch_fs *c = trans->c;
1536         struct bkey_i_dirent *n;
1537         struct printbuf buf = PRINTBUF;
1538         struct btree_iter bp_iter = { NULL };
1539         int ret = 0;
1540
1541         if (!target->bi_dir &&
1542             !target->bi_dir_offset) {
1543                 target->bi_dir          = d.k->p.inode;
1544                 target->bi_dir_offset   = d.k->p.offset;
1545
1546                 ret = __write_inode(trans, target, target_snapshot);
1547                 if (ret)
1548                         goto err;
1549         }
1550
1551         if (!inode_points_to_dirent(target, d)) {
1552                 struct bkey_s_c_dirent bp_dirent = dirent_get_by_pos(trans, &bp_iter,
1553                                       SPOS(target->bi_dir, target->bi_dir_offset, target_snapshot));
1554                 ret = bkey_err(bp_dirent);
1555                 if (ret && !bch2_err_matches(ret, ENOENT))
1556                         goto err;
1557
1558                 bool backpointer_exists = !ret;
1559                 ret = 0;
1560
1561                 bch2_bkey_val_to_text(&buf, c, d.s_c);
1562                 prt_newline(&buf);
1563                 if (backpointer_exists)
1564                         bch2_bkey_val_to_text(&buf, c, bp_dirent.s_c);
1565
1566                 if (fsck_err_on(S_ISDIR(target->bi_mode) && backpointer_exists,
1567                                 c, inode_dir_multiple_links,
1568                                 "directory %llu:%u with multiple links\n%s",
1569                                 target->bi_inum, target_snapshot, buf.buf)) {
1570                         ret = __remove_dirent(trans, d.k->p);
1571                         goto out;
1572                 }
1573
1574                 /*
1575                  * hardlinked file with nlink 0:
1576                  * We're just adjusting nlink here so check_nlinks() will pick
1577                  * it up, it ignores inodes with nlink 0
1578                  */
1579                 if (fsck_err_on(backpointer_exists && !target->bi_nlink,
1580                                 c, inode_multiple_links_but_nlink_0,
1581                                 "inode %llu:%u type %s has multiple links but i_nlink 0\n%s",
1582                                 target->bi_inum, target_snapshot, bch2_d_types[d.v->d_type], buf.buf)) {
1583                         target->bi_nlink++;
1584                         target->bi_flags &= ~BCH_INODE_unlinked;
1585
1586                         ret = __write_inode(trans, target, target_snapshot);
1587                         if (ret)
1588                                 goto err;
1589                 }
1590
1591                 if (fsck_err_on(!backpointer_exists,
1592                                 c, inode_wrong_backpointer,
1593                                 "inode %llu:%u has wrong backpointer:\n"
1594                                 "got       %llu:%llu\n"
1595                                 "should be %llu:%llu",
1596                                 target->bi_inum, target_snapshot,
1597                                 target->bi_dir,
1598                                 target->bi_dir_offset,
1599                                 d.k->p.inode,
1600                                 d.k->p.offset)) {
1601                         target->bi_dir          = d.k->p.inode;
1602                         target->bi_dir_offset   = d.k->p.offset;
1603
1604                         ret = __write_inode(trans, target, target_snapshot);
1605                         if (ret)
1606                                 goto err;
1607                 }
1608         }
1609
1610         if (fsck_err_on(d.v->d_type != inode_d_type(target),
1611                         c, dirent_d_type_wrong,
1612                         "incorrect d_type: got %s, should be %s:\n%s",
1613                         bch2_d_type_str(d.v->d_type),
1614                         bch2_d_type_str(inode_d_type(target)),
1615                         (printbuf_reset(&buf),
1616                          bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
1617                 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1618                 ret = PTR_ERR_OR_ZERO(n);
1619                 if (ret)
1620                         goto err;
1621
1622                 bkey_reassemble(&n->k_i, d.s_c);
1623                 n->v.d_type = inode_d_type(target);
1624
1625                 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1626                 if (ret)
1627                         goto err;
1628
1629                 d = dirent_i_to_s_c(n);
1630         }
1631
1632         if (d.v->d_type == DT_SUBVOL &&
1633             target->bi_parent_subvol != le32_to_cpu(d.v->d_parent_subvol) &&
1634             (c->sb.version < bcachefs_metadata_version_subvol_dirent ||
1635              fsck_err(c, dirent_d_parent_subvol_wrong,
1636                       "dirent has wrong d_parent_subvol field: got %u, should be %u",
1637                       le32_to_cpu(d.v->d_parent_subvol),
1638                       target->bi_parent_subvol))) {
1639                 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1640                 ret = PTR_ERR_OR_ZERO(n);
1641                 if (ret)
1642                         goto err;
1643
1644                 bkey_reassemble(&n->k_i, d.s_c);
1645                 n->v.d_parent_subvol = cpu_to_le32(target->bi_parent_subvol);
1646
1647                 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1648                 if (ret)
1649                         goto err;
1650
1651                 d = dirent_i_to_s_c(n);
1652         }
1653 out:
1654 err:
1655 fsck_err:
1656         bch2_trans_iter_exit(trans, &bp_iter);
1657         printbuf_exit(&buf);
1658         bch_err_fn(c, ret);
1659         return ret;
1660 }
1661
1662 static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
1663                         struct bkey_s_c k,
1664                         struct bch_hash_info *hash_info,
1665                         struct inode_walker *dir,
1666                         struct inode_walker *target,
1667                         struct snapshots_seen *s)
1668 {
1669         struct bch_fs *c = trans->c;
1670         struct bkey_s_c_dirent d;
1671         struct inode_walker_entry *i;
1672         struct printbuf buf = PRINTBUF;
1673         struct bpos equiv;
1674         int ret = 0;
1675
1676         ret = check_key_has_snapshot(trans, iter, k);
1677         if (ret) {
1678                 ret = ret < 0 ? ret : 0;
1679                 goto out;
1680         }
1681
1682         equiv = k.k->p;
1683         equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1684
1685         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1686         if (ret)
1687                 goto err;
1688
1689         if (k.k->type == KEY_TYPE_whiteout)
1690                 goto out;
1691
1692         if (dir->last_pos.inode != k.k->p.inode) {
1693                 ret = check_subdir_count(trans, dir);
1694                 if (ret)
1695                         goto err;
1696         }
1697
1698         BUG_ON(!iter->path->should_be_locked);
1699
1700         i = walk_inode(trans, dir, equiv, k.k->type == KEY_TYPE_whiteout);
1701         ret = PTR_ERR_OR_ZERO(i);
1702         if (ret < 0)
1703                 goto err;
1704
1705         if (dir->first_this_inode && dir->inodes.nr)
1706                 *hash_info = bch2_hash_info_init(c, &dir->inodes.data[0].inode);
1707         dir->first_this_inode = false;
1708
1709         if (fsck_err_on(!i, c, dirent_in_missing_dir_inode,
1710                         "dirent in nonexisting directory:\n%s",
1711                         (printbuf_reset(&buf),
1712                          bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1713                 ret = bch2_btree_delete_at(trans, iter,
1714                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1715                 goto out;
1716         }
1717
1718         if (!i)
1719                 goto out;
1720
1721         if (fsck_err_on(!S_ISDIR(i->inode.bi_mode),
1722                         c, dirent_in_non_dir_inode,
1723                         "dirent in non directory inode type %s:\n%s",
1724                         bch2_d_type_str(inode_d_type(&i->inode)),
1725                         (printbuf_reset(&buf),
1726                          bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1727                 ret = bch2_btree_delete_at(trans, iter, 0);
1728                 goto out;
1729         }
1730
1731         ret = hash_check_key(trans, bch2_dirent_hash_desc, hash_info, iter, k);
1732         if (ret < 0)
1733                 goto err;
1734         if (ret) {
1735                 /* dirent has been deleted */
1736                 ret = 0;
1737                 goto out;
1738         }
1739
1740         if (k.k->type != KEY_TYPE_dirent)
1741                 goto out;
1742
1743         d = bkey_s_c_to_dirent(k);
1744
1745         if (d.v->d_type == DT_SUBVOL) {
1746                 struct bch_inode_unpacked subvol_root;
1747                 u32 target_subvol = le32_to_cpu(d.v->d_child_subvol);
1748                 u32 target_snapshot;
1749                 u64 target_inum;
1750
1751                 ret = __subvol_lookup(trans, target_subvol,
1752                                       &target_snapshot, &target_inum);
1753                 if (ret && !bch2_err_matches(ret, ENOENT))
1754                         goto err;
1755
1756                 if (fsck_err_on(ret, c, dirent_to_missing_subvol,
1757                                 "dirent points to missing subvolume %u",
1758                                 le32_to_cpu(d.v->d_child_subvol))) {
1759                         ret = __remove_dirent(trans, d.k->p);
1760                         goto err;
1761                 }
1762
1763                 ret = __lookup_inode(trans, target_inum,
1764                                    &subvol_root, &target_snapshot);
1765                 if (ret && !bch2_err_matches(ret, ENOENT))
1766                         goto err;
1767
1768                 if (fsck_err_on(ret, c, subvol_to_missing_root,
1769                                 "subvolume %u points to missing subvolume root %llu",
1770                                 target_subvol,
1771                                 target_inum)) {
1772                         bch_err(c, "repair not implemented yet");
1773                         ret = -EINVAL;
1774                         goto err;
1775                 }
1776
1777                 if (fsck_err_on(subvol_root.bi_subvol != target_subvol,
1778                                 c, subvol_root_wrong_bi_subvol,
1779                                 "subvol root %llu has wrong bi_subvol field: got %u, should be %u",
1780                                 target_inum,
1781                                 subvol_root.bi_subvol, target_subvol)) {
1782                         subvol_root.bi_subvol = target_subvol;
1783                         ret = __write_inode(trans, &subvol_root, target_snapshot);
1784                         if (ret)
1785                                 goto err;
1786                 }
1787
1788                 ret = check_dirent_target(trans, iter, d, &subvol_root,
1789                                           target_snapshot);
1790                 if (ret)
1791                         goto err;
1792         } else {
1793                 ret = __get_visible_inodes(trans, target, s, le64_to_cpu(d.v->d_inum));
1794                 if (ret)
1795                         goto err;
1796
1797                 if (fsck_err_on(!target->inodes.nr,
1798                                 c, dirent_to_missing_inode,
1799                                 "dirent points to missing inode: (equiv %u)\n%s",
1800                                 equiv.snapshot,
1801                                 (printbuf_reset(&buf),
1802                                  bch2_bkey_val_to_text(&buf, c, k),
1803                                  buf.buf))) {
1804                         ret = __remove_dirent(trans, d.k->p);
1805                         if (ret)
1806                                 goto err;
1807                 }
1808
1809                 darray_for_each(target->inodes, i) {
1810                         ret = check_dirent_target(trans, iter, d,
1811                                                   &i->inode, i->snapshot);
1812                         if (ret)
1813                                 goto err;
1814                 }
1815         }
1816
1817         if (d.v->d_type == DT_DIR)
1818                 for_each_visible_inode(c, s, dir, equiv.snapshot, i)
1819                         i->count++;
1820
1821 out:
1822 err:
1823 fsck_err:
1824         printbuf_exit(&buf);
1825         bch_err_fn(c, ret);
1826         return ret;
1827 }
1828
1829 /*
1830  * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
1831  * validate d_type
1832  */
1833 int bch2_check_dirents(struct bch_fs *c)
1834 {
1835         struct inode_walker dir = inode_walker_init();
1836         struct inode_walker target = inode_walker_init();
1837         struct snapshots_seen s;
1838         struct bch_hash_info hash_info;
1839         struct btree_trans *trans = bch2_trans_get(c);
1840         struct btree_iter iter;
1841         struct bkey_s_c k;
1842         int ret = 0;
1843
1844         snapshots_seen_init(&s);
1845
1846         ret = for_each_btree_key_commit(trans, iter, BTREE_ID_dirents,
1847                         POS(BCACHEFS_ROOT_INO, 0),
1848                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
1849                         k,
1850                         NULL, NULL,
1851                         BCH_TRANS_COMMIT_lazy_rw|BCH_TRANS_COMMIT_no_enospc,
1852                 check_dirent(trans, &iter, k, &hash_info, &dir, &target, &s));
1853
1854         bch2_trans_put(trans);
1855         snapshots_seen_exit(&s);
1856         inode_walker_exit(&dir);
1857         inode_walker_exit(&target);
1858         bch_err_fn(c, ret);
1859         return ret;
1860 }
1861
1862 static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
1863                        struct bkey_s_c k,
1864                        struct bch_hash_info *hash_info,
1865                        struct inode_walker *inode)
1866 {
1867         struct bch_fs *c = trans->c;
1868         struct inode_walker_entry *i;
1869         int ret;
1870
1871         ret = check_key_has_snapshot(trans, iter, k);
1872         if (ret)
1873                 return ret;
1874
1875         i = walk_inode(trans, inode, k.k->p, k.k->type == KEY_TYPE_whiteout);
1876         ret = PTR_ERR_OR_ZERO(i);
1877         if (ret)
1878                 return ret;
1879
1880         if (inode->first_this_inode && inode->inodes.nr)
1881                 *hash_info = bch2_hash_info_init(c, &inode->inodes.data[0].inode);
1882         inode->first_this_inode = false;
1883
1884         if (fsck_err_on(!i, c, xattr_in_missing_inode,
1885                         "xattr for missing inode %llu",
1886                         k.k->p.inode))
1887                 return bch2_btree_delete_at(trans, iter, 0);
1888
1889         if (!i)
1890                 return 0;
1891
1892         ret = hash_check_key(trans, bch2_xattr_hash_desc, hash_info, iter, k);
1893 fsck_err:
1894         bch_err_fn(c, ret);
1895         return ret;
1896 }
1897
1898 /*
1899  * Walk xattrs: verify that they all have a corresponding inode
1900  */
1901 int bch2_check_xattrs(struct bch_fs *c)
1902 {
1903         struct inode_walker inode = inode_walker_init();
1904         struct bch_hash_info hash_info;
1905         struct btree_iter iter;
1906         struct bkey_s_c k;
1907         int ret = 0;
1908
1909         ret = bch2_trans_run(c,
1910                 for_each_btree_key_commit(trans, iter, BTREE_ID_xattrs,
1911                         POS(BCACHEFS_ROOT_INO, 0),
1912                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
1913                         k,
1914                         NULL, NULL,
1915                         BCH_TRANS_COMMIT_lazy_rw|BCH_TRANS_COMMIT_no_enospc,
1916                 check_xattr(trans, &iter, k, &hash_info, &inode)));
1917         bch_err_fn(c, ret);
1918         return ret;
1919 }
1920
1921 static int check_root_trans(struct btree_trans *trans)
1922 {
1923         struct bch_fs *c = trans->c;
1924         struct bch_inode_unpacked root_inode;
1925         u32 snapshot;
1926         u64 inum;
1927         int ret;
1928
1929         ret = __subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
1930         if (ret && !bch2_err_matches(ret, ENOENT))
1931                 return ret;
1932
1933         if (mustfix_fsck_err_on(ret, c, root_subvol_missing,
1934                                 "root subvol missing")) {
1935                 struct bkey_i_subvolume root_subvol;
1936
1937                 snapshot        = U32_MAX;
1938                 inum            = BCACHEFS_ROOT_INO;
1939
1940                 bkey_subvolume_init(&root_subvol.k_i);
1941                 root_subvol.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1942                 root_subvol.v.flags     = 0;
1943                 root_subvol.v.snapshot  = cpu_to_le32(snapshot);
1944                 root_subvol.v.inode     = cpu_to_le64(inum);
1945                 ret = commit_do(trans, NULL, NULL,
1946                                       BCH_TRANS_COMMIT_no_enospc|
1947                                       BCH_TRANS_COMMIT_lazy_rw,
1948                         bch2_btree_insert_trans(trans, BTREE_ID_subvolumes,
1949                                             &root_subvol.k_i, 0));
1950                 bch_err_msg(c, ret, "writing root subvol");
1951                 if (ret)
1952                         goto err;
1953
1954         }
1955
1956         ret = __lookup_inode(trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
1957         if (ret && !bch2_err_matches(ret, ENOENT))
1958                 return ret;
1959
1960         if (mustfix_fsck_err_on(ret, c, root_dir_missing,
1961                                 "root directory missing") ||
1962             mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode),
1963                                 c, root_inode_not_dir,
1964                                 "root inode not a directory")) {
1965                 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755,
1966                                 0, NULL);
1967                 root_inode.bi_inum = inum;
1968
1969                 ret = __write_inode(trans, &root_inode, snapshot);
1970                 bch_err_msg(c, ret, "writing root inode");
1971         }
1972 err:
1973 fsck_err:
1974         return ret;
1975 }
1976
1977 /* Get root directory, create if it doesn't exist: */
1978 int bch2_check_root(struct bch_fs *c)
1979 {
1980         int ret;
1981
1982         ret = bch2_trans_do(c, NULL, NULL,
1983                              BCH_TRANS_COMMIT_no_enospc|
1984                              BCH_TRANS_COMMIT_lazy_rw,
1985                 check_root_trans(trans));
1986         bch_err_fn(c, ret);
1987         return ret;
1988 }
1989
1990 struct pathbuf_entry {
1991         u64     inum;
1992         u32     snapshot;
1993 };
1994
1995 typedef DARRAY(struct pathbuf_entry) pathbuf;
1996
1997 static bool path_is_dup(pathbuf *p, u64 inum, u32 snapshot)
1998 {
1999         struct pathbuf_entry *i;
2000
2001         darray_for_each(*p, i)
2002                 if (i->inum     == inum &&
2003                     i->snapshot == snapshot)
2004                         return true;
2005
2006         return false;
2007 }
2008
2009 static int path_down(struct bch_fs *c, pathbuf *p,
2010                      u64 inum, u32 snapshot)
2011 {
2012         int ret = darray_push(p, ((struct pathbuf_entry) {
2013                 .inum           = inum,
2014                 .snapshot       = snapshot,
2015         }));
2016
2017         if (ret)
2018                 bch_err(c, "fsck: error allocating memory for pathbuf, size %zu",
2019                         p->size);
2020         return ret;
2021 }
2022
2023 /*
2024  * Check that a given inode is reachable from the root:
2025  *
2026  * XXX: we should also be verifying that inodes are in the right subvolumes
2027  */
2028 static int check_path(struct btree_trans *trans,
2029                       pathbuf *p,
2030                       struct bch_inode_unpacked *inode,
2031                       u32 snapshot)
2032 {
2033         struct bch_fs *c = trans->c;
2034         int ret = 0;
2035
2036         snapshot = bch2_snapshot_equiv(c, snapshot);
2037         p->nr = 0;
2038
2039         while (!(inode->bi_inum == BCACHEFS_ROOT_INO &&
2040                  inode->bi_subvol == BCACHEFS_ROOT_SUBVOL)) {
2041                 struct btree_iter dirent_iter;
2042                 struct bkey_s_c_dirent d;
2043                 u32 parent_snapshot = snapshot;
2044
2045                 if (inode->bi_subvol) {
2046                         u64 inum;
2047
2048                         ret = subvol_lookup(trans, inode->bi_parent_subvol,
2049                                             &parent_snapshot, &inum);
2050                         if (ret)
2051                                 break;
2052                 }
2053
2054                 ret = lockrestart_do(trans,
2055                         PTR_ERR_OR_ZERO((d = dirent_get_by_pos(trans, &dirent_iter,
2056                                           SPOS(inode->bi_dir, inode->bi_dir_offset,
2057                                                parent_snapshot))).k));
2058                 if (ret && !bch2_err_matches(ret, ENOENT))
2059                         break;
2060
2061                 if (!ret && !dirent_points_to_inode(d, inode)) {
2062                         bch2_trans_iter_exit(trans, &dirent_iter);
2063                         ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
2064                 }
2065
2066                 if (bch2_err_matches(ret, ENOENT)) {
2067                         if (fsck_err(c,  inode_unreachable,
2068                                      "unreachable inode %llu:%u, type %s nlink %u backptr %llu:%llu",
2069                                      inode->bi_inum, snapshot,
2070                                      bch2_d_type_str(inode_d_type(inode)),
2071                                      inode->bi_nlink,
2072                                      inode->bi_dir,
2073                                      inode->bi_dir_offset))
2074                                 ret = reattach_inode(trans, inode, snapshot);
2075                         break;
2076                 }
2077
2078                 bch2_trans_iter_exit(trans, &dirent_iter);
2079
2080                 if (!S_ISDIR(inode->bi_mode))
2081                         break;
2082
2083                 ret = path_down(c, p, inode->bi_inum, snapshot);
2084                 if (ret) {
2085                         bch_err(c, "memory allocation failure");
2086                         return ret;
2087                 }
2088
2089                 snapshot = parent_snapshot;
2090
2091                 ret = lookup_inode(trans, inode->bi_dir, inode, &snapshot);
2092                 if (ret) {
2093                         /* Should have been caught in dirents pass */
2094                         bch_err(c, "error looking up parent directory: %i", ret);
2095                         break;
2096                 }
2097
2098                 if (path_is_dup(p, inode->bi_inum, snapshot)) {
2099                         struct pathbuf_entry *i;
2100
2101                         /* XXX print path */
2102                         bch_err(c, "directory structure loop");
2103
2104                         darray_for_each(*p, i)
2105                                 pr_err("%llu:%u", i->inum, i->snapshot);
2106                         pr_err("%llu:%u", inode->bi_inum, snapshot);
2107
2108                         if (!fsck_err(c, dir_loop,
2109                                       "directory structure loop"))
2110                                 return 0;
2111
2112                         ret = commit_do(trans, NULL, NULL,
2113                                               BCH_TRANS_COMMIT_no_enospc|
2114                                               BCH_TRANS_COMMIT_lazy_rw,
2115                                         remove_backpointer(trans, inode));
2116                         if (ret) {
2117                                 bch_err(c, "error removing dirent: %i", ret);
2118                                 break;
2119                         }
2120
2121                         ret = reattach_inode(trans, inode, snapshot);
2122                 }
2123         }
2124 fsck_err:
2125         bch_err_fn(c, ret);
2126         return ret;
2127 }
2128
2129 /*
2130  * Check for unreachable inodes, as well as loops in the directory structure:
2131  * After bch2_check_dirents(), if an inode backpointer doesn't exist that means it's
2132  * unreachable:
2133  */
2134 int bch2_check_directory_structure(struct bch_fs *c)
2135 {
2136         struct btree_trans *trans = bch2_trans_get(c);
2137         struct btree_iter iter;
2138         struct bkey_s_c k;
2139         struct bch_inode_unpacked u;
2140         pathbuf path = { 0, };
2141         int ret;
2142
2143         for_each_btree_key(trans, iter, BTREE_ID_inodes, POS_MIN,
2144                            BTREE_ITER_INTENT|
2145                            BTREE_ITER_PREFETCH|
2146                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2147                 if (!bkey_is_inode(k.k))
2148                         continue;
2149
2150                 ret = bch2_inode_unpack(k, &u);
2151                 if (ret) {
2152                         /* Should have been caught earlier in fsck: */
2153                         bch_err(c, "error unpacking inode %llu: %i", k.k->p.offset, ret);
2154                         break;
2155                 }
2156
2157                 if (u.bi_flags & BCH_INODE_unlinked)
2158                         continue;
2159
2160                 ret = check_path(trans, &path, &u, iter.pos.snapshot);
2161                 if (ret)
2162                         break;
2163         }
2164         bch2_trans_iter_exit(trans, &iter);
2165         bch2_trans_put(trans);
2166         darray_exit(&path);
2167         bch_err_fn(c, ret);
2168         return ret;
2169 }
2170
2171 struct nlink_table {
2172         size_t          nr;
2173         size_t          size;
2174
2175         struct nlink {
2176                 u64     inum;
2177                 u32     snapshot;
2178                 u32     count;
2179         }               *d;
2180 };
2181
2182 static int add_nlink(struct bch_fs *c, struct nlink_table *t,
2183                      u64 inum, u32 snapshot)
2184 {
2185         if (t->nr == t->size) {
2186                 size_t new_size = max_t(size_t, 128UL, t->size * 2);
2187                 void *d = kvmalloc_array(new_size, sizeof(t->d[0]), GFP_KERNEL);
2188
2189                 if (!d) {
2190                         bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
2191                                 new_size);
2192                         return -BCH_ERR_ENOMEM_fsck_add_nlink;
2193                 }
2194
2195                 if (t->d)
2196                         memcpy(d, t->d, t->size * sizeof(t->d[0]));
2197                 kvfree(t->d);
2198
2199                 t->d = d;
2200                 t->size = new_size;
2201         }
2202
2203
2204         t->d[t->nr++] = (struct nlink) {
2205                 .inum           = inum,
2206                 .snapshot       = snapshot,
2207         };
2208
2209         return 0;
2210 }
2211
2212 static int nlink_cmp(const void *_l, const void *_r)
2213 {
2214         const struct nlink *l = _l;
2215         const struct nlink *r = _r;
2216
2217         return cmp_int(l->inum, r->inum);
2218 }
2219
2220 static void inc_link(struct bch_fs *c, struct snapshots_seen *s,
2221                      struct nlink_table *links,
2222                      u64 range_start, u64 range_end, u64 inum, u32 snapshot)
2223 {
2224         struct nlink *link, key = {
2225                 .inum = inum, .snapshot = U32_MAX,
2226         };
2227
2228         if (inum < range_start || inum >= range_end)
2229                 return;
2230
2231         link = __inline_bsearch(&key, links->d, links->nr,
2232                                 sizeof(links->d[0]), nlink_cmp);
2233         if (!link)
2234                 return;
2235
2236         while (link > links->d && link[0].inum == link[-1].inum)
2237                 --link;
2238
2239         for (; link < links->d + links->nr && link->inum == inum; link++)
2240                 if (ref_visible(c, s, snapshot, link->snapshot)) {
2241                         link->count++;
2242                         if (link->snapshot >= snapshot)
2243                                 break;
2244                 }
2245 }
2246
2247 noinline_for_stack
2248 static int check_nlinks_find_hardlinks(struct bch_fs *c,
2249                                        struct nlink_table *t,
2250                                        u64 start, u64 *end)
2251 {
2252         struct btree_trans *trans = bch2_trans_get(c);
2253         struct btree_iter iter;
2254         struct bkey_s_c k;
2255         struct bch_inode_unpacked u;
2256         int ret = 0;
2257
2258         for_each_btree_key(trans, iter, BTREE_ID_inodes,
2259                            POS(0, start),
2260                            BTREE_ITER_INTENT|
2261                            BTREE_ITER_PREFETCH|
2262                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2263                 if (!bkey_is_inode(k.k))
2264                         continue;
2265
2266                 /* Should never fail, checked by bch2_inode_invalid: */
2267                 BUG_ON(bch2_inode_unpack(k, &u));
2268
2269                 /*
2270                  * Backpointer and directory structure checks are sufficient for
2271                  * directories, since they can't have hardlinks:
2272                  */
2273                 if (S_ISDIR(u.bi_mode))
2274                         continue;
2275
2276                 if (!u.bi_nlink)
2277                         continue;
2278
2279                 ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot);
2280                 if (ret) {
2281                         *end = k.k->p.offset;
2282                         ret = 0;
2283                         break;
2284                 }
2285
2286         }
2287         bch2_trans_iter_exit(trans, &iter);
2288         bch2_trans_put(trans);
2289
2290         if (ret)
2291                 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
2292
2293         return ret;
2294 }
2295
2296 noinline_for_stack
2297 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
2298                                      u64 range_start, u64 range_end)
2299 {
2300         struct btree_trans *trans = bch2_trans_get(c);
2301         struct snapshots_seen s;
2302         struct btree_iter iter;
2303         struct bkey_s_c k;
2304         struct bkey_s_c_dirent d;
2305         int ret;
2306
2307         snapshots_seen_init(&s);
2308
2309         for_each_btree_key(trans, iter, BTREE_ID_dirents, POS_MIN,
2310                            BTREE_ITER_INTENT|
2311                            BTREE_ITER_PREFETCH|
2312                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2313                 ret = snapshots_seen_update(c, &s, iter.btree_id, k.k->p);
2314                 if (ret)
2315                         break;
2316
2317                 switch (k.k->type) {
2318                 case KEY_TYPE_dirent:
2319                         d = bkey_s_c_to_dirent(k);
2320
2321                         if (d.v->d_type != DT_DIR &&
2322                             d.v->d_type != DT_SUBVOL)
2323                                 inc_link(c, &s, links, range_start, range_end,
2324                                          le64_to_cpu(d.v->d_inum),
2325                                          bch2_snapshot_equiv(c, d.k->p.snapshot));
2326                         break;
2327                 }
2328         }
2329         bch2_trans_iter_exit(trans, &iter);
2330
2331         if (ret)
2332                 bch_err(c, "error in fsck: btree error %i while walking dirents", ret);
2333
2334         bch2_trans_put(trans);
2335         snapshots_seen_exit(&s);
2336         return ret;
2337 }
2338
2339 static int check_nlinks_update_inode(struct btree_trans *trans, struct btree_iter *iter,
2340                                      struct bkey_s_c k,
2341                                      struct nlink_table *links,
2342                                      size_t *idx, u64 range_end)
2343 {
2344         struct bch_fs *c = trans->c;
2345         struct bch_inode_unpacked u;
2346         struct nlink *link = &links->d[*idx];
2347         int ret = 0;
2348
2349         if (k.k->p.offset >= range_end)
2350                 return 1;
2351
2352         if (!bkey_is_inode(k.k))
2353                 return 0;
2354
2355         BUG_ON(bch2_inode_unpack(k, &u));
2356
2357         if (S_ISDIR(u.bi_mode))
2358                 return 0;
2359
2360         if (!u.bi_nlink)
2361                 return 0;
2362
2363         while ((cmp_int(link->inum, k.k->p.offset) ?:
2364                 cmp_int(link->snapshot, k.k->p.snapshot)) < 0) {
2365                 BUG_ON(*idx == links->nr);
2366                 link = &links->d[++*idx];
2367         }
2368
2369         if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count,
2370                         c, inode_wrong_nlink,
2371                         "inode %llu type %s has wrong i_nlink (%u, should be %u)",
2372                         u.bi_inum, bch2_d_types[mode_to_type(u.bi_mode)],
2373                         bch2_inode_nlink_get(&u), link->count)) {
2374                 bch2_inode_nlink_set(&u, link->count);
2375                 ret = __write_inode(trans, &u, k.k->p.snapshot);
2376         }
2377 fsck_err:
2378         return ret;
2379 }
2380
2381 noinline_for_stack
2382 static int check_nlinks_update_hardlinks(struct bch_fs *c,
2383                                struct nlink_table *links,
2384                                u64 range_start, u64 range_end)
2385 {
2386         struct btree_iter iter;
2387         struct bkey_s_c k;
2388         size_t idx = 0;
2389         int ret = 0;
2390
2391         ret = bch2_trans_run(c,
2392                 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
2393                                 POS(0, range_start),
2394                                 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
2395                                 NULL, NULL, BCH_TRANS_COMMIT_lazy_rw|BCH_TRANS_COMMIT_no_enospc,
2396                         check_nlinks_update_inode(trans, &iter, k, links, &idx, range_end)));
2397         if (ret < 0) {
2398                 bch_err(c, "error in fsck walking inodes: %s", bch2_err_str(ret));
2399                 return ret;
2400         }
2401
2402         return 0;
2403 }
2404
2405 int bch2_check_nlinks(struct bch_fs *c)
2406 {
2407         struct nlink_table links = { 0 };
2408         u64 this_iter_range_start, next_iter_range_start = 0;
2409         int ret = 0;
2410
2411         do {
2412                 this_iter_range_start = next_iter_range_start;
2413                 next_iter_range_start = U64_MAX;
2414
2415                 ret = check_nlinks_find_hardlinks(c, &links,
2416                                                   this_iter_range_start,
2417                                                   &next_iter_range_start);
2418
2419                 ret = check_nlinks_walk_dirents(c, &links,
2420                                           this_iter_range_start,
2421                                           next_iter_range_start);
2422                 if (ret)
2423                         break;
2424
2425                 ret = check_nlinks_update_hardlinks(c, &links,
2426                                          this_iter_range_start,
2427                                          next_iter_range_start);
2428                 if (ret)
2429                         break;
2430
2431                 links.nr = 0;
2432         } while (next_iter_range_start != U64_MAX);
2433
2434         kvfree(links.d);
2435         bch_err_fn(c, ret);
2436         return ret;
2437 }
2438
2439 static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter,
2440                              struct bkey_s_c k)
2441 {
2442         struct bkey_s_c_reflink_p p;
2443         struct bkey_i_reflink_p *u;
2444         int ret;
2445
2446         if (k.k->type != KEY_TYPE_reflink_p)
2447                 return 0;
2448
2449         p = bkey_s_c_to_reflink_p(k);
2450
2451         if (!p.v->front_pad && !p.v->back_pad)
2452                 return 0;
2453
2454         u = bch2_trans_kmalloc(trans, sizeof(*u));
2455         ret = PTR_ERR_OR_ZERO(u);
2456         if (ret)
2457                 return ret;
2458
2459         bkey_reassemble(&u->k_i, k);
2460         u->v.front_pad  = 0;
2461         u->v.back_pad   = 0;
2462
2463         return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_NORUN);
2464 }
2465
2466 int bch2_fix_reflink_p(struct bch_fs *c)
2467 {
2468         struct btree_iter iter;
2469         struct bkey_s_c k;
2470         int ret;
2471
2472         if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
2473                 return 0;
2474
2475         ret = bch2_trans_run(c,
2476                 for_each_btree_key_commit(trans, iter,
2477                                 BTREE_ID_extents, POS_MIN,
2478                                 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|
2479                                 BTREE_ITER_ALL_SNAPSHOTS, k,
2480                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc|BCH_TRANS_COMMIT_lazy_rw,
2481                         fix_reflink_p_key(trans, &iter, k)));
2482         bch_err_fn(c, ret);
2483         return ret;
2484 }