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

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "bkey_buf.h"
   5 #include "btree_update.h"
   6 #include "buckets.h"
   7 #include "darray.h"
   8 #include "dirent.h"
   9 #include "error.h"
  10 #include "fs-common.h"
  11 #include "fsck.h"
  12 #include "inode.h"
  13 #include "keylist.h"
  14 #include "recovery.h"
  15 #include "snapshot.h"
  16 #include "super.h"
  17 #include "xattr.h"
  18
  19 #include <linux/bsearch.h>
  20 #include <linux/dcache.h> /* struct qstr */
  21
  22 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  23
  24 /*
  25  * XXX: this is handling transaction restarts without returning
  26  * -BCH_ERR_transaction_restart_nested, this is not how we do things anymore:
  27  */
  28 static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum,
  29                                     u32 snapshot)
  30 {
  31         struct btree_iter iter;
  32         struct bkey_s_c k;
  33         u64 sectors = 0;
  34         int ret;
  35
  36         for_each_btree_key_upto(trans, iter, BTREE_ID_extents,
  37                                 SPOS(inum, 0, snapshot),
  38                                 POS(inum, U64_MAX),
  39                                 0, k, ret)
  40                 if (bkey_extent_is_allocation(k.k))
  41                         sectors += k.k->size;
  42
  43         bch2_trans_iter_exit(trans, &iter);
  44
  45         return ret ?: sectors;
  46 }
  47
  48 static s64 bch2_count_subdirs(struct btree_trans *trans, u64 inum,
  49                                     u32 snapshot)
  50 {
  51         struct btree_iter iter;
  52         struct bkey_s_c k;
  53         struct bkey_s_c_dirent d;
  54         u64 subdirs = 0;
  55         int ret;
  56
  57         for_each_btree_key_upto(trans, iter, BTREE_ID_dirents,
  58                                 SPOS(inum, 0, snapshot),
  59                                 POS(inum, U64_MAX),
  60                                 0, k, ret) {
  61                 if (k.k->type != KEY_TYPE_dirent)
  62                         continue;
  63
  64                 d = bkey_s_c_to_dirent(k);
  65                 if (d.v->d_type == DT_DIR)
  66                         subdirs++;
  67         }
  68         bch2_trans_iter_exit(trans, &iter);
  69
  70         return ret ?: subdirs;
  71 }
  72
  73 static int __snapshot_lookup_subvol(struct btree_trans *trans, u32 snapshot,
  74                                     u32 *subvol)
  75 {
  76         struct bch_snapshot s;
  77         int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_snapshots,
  78                                           POS(0, snapshot), 0,
  79                                           snapshot, &s);
  80         if (!ret)
  81                 *subvol = le32_to_cpu(s.subvol);
  82         else if (bch2_err_matches(ret, ENOENT))
  83                 bch_err(trans->c, "snapshot %u not found", snapshot);
  84         return ret;
  85
  86 }
  87
  88 static int __subvol_lookup(struct btree_trans *trans, u32 subvol,
  89                            u32 *snapshot, u64 *inum)
  90 {
  91         struct bch_subvolume s;
  92         int ret;
  93
  94         ret = bch2_subvolume_get(trans, subvol, false, 0, &s);
  95
  96         *snapshot = le32_to_cpu(s.snapshot);
  97         *inum = le64_to_cpu(s.inode);
  98         return ret;
  99 }
 100
 101 static int subvol_lookup(struct btree_trans *trans, u32 subvol,
 102                          u32 *snapshot, u64 *inum)
 103 {
 104         return lockrestart_do(trans, __subvol_lookup(trans, subvol, snapshot, inum));
 105 }
 106
 107 static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr,
 108                               struct bch_inode_unpacked *inode)
 109 {
 110         struct btree_iter iter;
 111         struct bkey_s_c k;
 112         int ret;
 113
 114         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
 115                              POS(0, inode_nr),
 116                              BTREE_ITER_ALL_SNAPSHOTS);
 117         k = bch2_btree_iter_peek(&iter);
 118         ret = bkey_err(k);
 119         if (ret)
 120                 goto err;
 121
 122         if (!k.k || !bkey_eq(k.k->p, POS(0, inode_nr))) {
 123                 ret = -BCH_ERR_ENOENT_inode;
 124                 goto err;
 125         }
 126
 127         ret = bch2_inode_unpack(k, inode);
 128 err:
 129         bch_err_msg(trans->c, ret, "fetching inode %llu", inode_nr);
 130         bch2_trans_iter_exit(trans, &iter);
 131         return ret;
 132 }
 133
 134 static int __lookup_inode(struct btree_trans *trans, u64 inode_nr,
 135                           struct bch_inode_unpacked *inode,
 136                           u32 *snapshot)
 137 {
 138         struct btree_iter iter;
 139         struct bkey_s_c k;
 140         int ret;
 141
 142         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
 143                                SPOS(0, inode_nr, *snapshot), 0);
 144         ret = bkey_err(k);
 145         if (ret)
 146                 goto err;
 147
 148         ret = bkey_is_inode(k.k)
 149                 ? bch2_inode_unpack(k, inode)
 150                 : -BCH_ERR_ENOENT_inode;
 151         if (!ret)
 152                 *snapshot = iter.pos.snapshot;
 153 err:
 154         bch_err_msg(trans->c, ret, "fetching inode %llu:%u", inode_nr, *snapshot);
 155         bch2_trans_iter_exit(trans, &iter);
 156         return ret;
 157 }
 158
 159 static int lookup_inode(struct btree_trans *trans, u64 inode_nr,
 160                         struct bch_inode_unpacked *inode,
 161                         u32 *snapshot)
 162 {
 163         return lockrestart_do(trans, __lookup_inode(trans, inode_nr, inode, snapshot));
 164 }
 165
 166 static int __lookup_dirent(struct btree_trans *trans,
 167                            struct bch_hash_info hash_info,
 168                            subvol_inum dir, struct qstr *name,
 169                            u64 *target, unsigned *type)
 170 {
 171         struct btree_iter iter;
 172         struct bkey_s_c_dirent d;
 173         int ret;
 174
 175         ret = bch2_hash_lookup(trans, &iter, bch2_dirent_hash_desc,
 176                                &hash_info, dir, name, 0);
 177         if (ret)
 178                 return ret;
 179
 180         d = bkey_s_c_to_dirent(bch2_btree_iter_peek_slot(&iter));
 181         *target = le64_to_cpu(d.v->d_inum);
 182         *type = d.v->d_type;
 183         bch2_trans_iter_exit(trans, &iter);
 184         return 0;
 185 }
 186
 187 static int __write_inode(struct btree_trans *trans,
 188                          struct bch_inode_unpacked *inode,
 189                          u32 snapshot)
 190 {
 191         struct bkey_inode_buf *inode_p =
 192                 bch2_trans_kmalloc(trans, sizeof(*inode_p));
 193
 194         if (IS_ERR(inode_p))
 195                 return PTR_ERR(inode_p);
 196
 197         bch2_inode_pack(inode_p, inode);
 198         inode_p->inode.k.p.snapshot = snapshot;
 199
 200         return bch2_btree_insert_nonextent(trans, BTREE_ID_inodes,
 201                                 &inode_p->inode.k_i,
 202                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 203 }
 204
 205 static int fsck_write_inode(struct btree_trans *trans,
 206                             struct bch_inode_unpacked *inode,
 207                             u32 snapshot)
 208 {
 209         int ret = commit_do(trans, NULL, NULL,
 210                                   BTREE_INSERT_NOFAIL|
 211                                   BTREE_INSERT_LAZY_RW,
 212                                   __write_inode(trans, inode, snapshot));
 213         if (ret)
 214                 bch_err_fn(trans->c, ret);
 215         return ret;
 216 }
 217
 218 static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
 219 {
 220         struct bch_fs *c = trans->c;
 221         struct btree_iter iter;
 222         struct bch_inode_unpacked dir_inode;
 223         struct bch_hash_info dir_hash_info;
 224         int ret;
 225
 226         ret = lookup_first_inode(trans, pos.inode, &dir_inode);
 227         if (ret)
 228                 goto err;
 229
 230         dir_hash_info = bch2_hash_info_init(c, &dir_inode);
 231
 232         bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_INTENT);
 233
 234         ret = bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
 235                                   &dir_hash_info, &iter,
 236                                   BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 237         bch2_trans_iter_exit(trans, &iter);
 238 err:
 239         bch_err_fn(c, ret);
 240         return ret;
 241 }
 242
 243 /* Get lost+found, create if it doesn't exist: */
 244 static int lookup_lostfound(struct btree_trans *trans, u32 subvol,
 245                             struct bch_inode_unpacked *lostfound)
 246 {
 247         struct bch_fs *c = trans->c;
 248         struct bch_inode_unpacked root;
 249         struct bch_hash_info root_hash_info;
 250         struct qstr lostfound_str = QSTR("lost+found");
 251         subvol_inum root_inum = { .subvol = subvol };
 252         u64 inum = 0;
 253         unsigned d_type = 0;
 254         u32 snapshot;
 255         int ret;
 256
 257         ret = __subvol_lookup(trans, subvol, &snapshot, &root_inum.inum);
 258         if (ret)
 259                 return ret;
 260
 261         ret = __lookup_inode(trans, root_inum.inum, &root, &snapshot);
 262         if (ret)
 263                 return ret;
 264
 265         root_hash_info = bch2_hash_info_init(c, &root);
 266
 267         ret = __lookup_dirent(trans, root_hash_info, root_inum,
 268                             &lostfound_str, &inum, &d_type);
 269         if (bch2_err_matches(ret, ENOENT)) {
 270                 bch_notice(c, "creating lost+found");
 271                 goto create_lostfound;
 272         }
 273
 274         bch_err_fn(c, ret);
 275         if (ret)
 276                 return ret;
 277
 278         if (d_type != DT_DIR) {
 279                 bch_err(c, "error looking up lost+found: not a directory");
 280                 return -BCH_ERR_ENOENT_not_directory;
 281         }
 282
 283         /*
 284          * The bch2_check_dirents pass has already run, dangling dirents
 285          * shouldn't exist here:
 286          */
 287         return __lookup_inode(trans, inum, lostfound, &snapshot);
 288
 289 create_lostfound:
 290         bch2_inode_init_early(c, lostfound);
 291
 292         ret = bch2_create_trans(trans, root_inum, &root,
 293                                 lostfound, &lostfound_str,
 294                                 0, 0, S_IFDIR|0700, 0, NULL, NULL,
 295                                 (subvol_inum) { }, 0);
 296         bch_err_msg(c, ret, "creating lost+found");
 297         return ret;
 298 }
 299
 300 static int __reattach_inode(struct btree_trans *trans,
 301                           struct bch_inode_unpacked *inode,
 302                           u32 inode_snapshot)
 303 {
 304         struct bch_hash_info dir_hash;
 305         struct bch_inode_unpacked lostfound;
 306         char name_buf[20];
 307         struct qstr name;
 308         u64 dir_offset = 0;
 309         u32 subvol;
 310         int ret;
 311
 312         ret = __snapshot_lookup_subvol(trans, inode_snapshot, &subvol);
 313         if (ret)
 314                 return ret;
 315
 316         ret = lookup_lostfound(trans, subvol, &lostfound);
 317         if (ret)
 318                 return ret;
 319
 320         if (S_ISDIR(inode->bi_mode)) {
 321                 lostfound.bi_nlink++;
 322
 323                 ret = __write_inode(trans, &lostfound, U32_MAX);
 324                 if (ret)
 325                         return ret;
 326         }
 327
 328         dir_hash = bch2_hash_info_init(trans->c, &lostfound);
 329
 330         snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
 331         name = (struct qstr) QSTR(name_buf);
 332
 333         ret = bch2_dirent_create(trans,
 334                                  (subvol_inum) {
 335                                         .subvol = subvol,
 336                                         .inum = lostfound.bi_inum,
 337                                  },
 338                                  &dir_hash,
 339                                  inode_d_type(inode),
 340                                  &name, inode->bi_inum, &dir_offset,
 341                                  BCH_HASH_SET_MUST_CREATE);
 342         if (ret)
 343                 return ret;
 344
 345         inode->bi_dir           = lostfound.bi_inum;
 346         inode->bi_dir_offset    = dir_offset;
 347
 348         return __write_inode(trans, inode, inode_snapshot);
 349 }
 350
 351 static int reattach_inode(struct btree_trans *trans,
 352                           struct bch_inode_unpacked *inode,
 353                           u32 inode_snapshot)
 354 {
 355         int ret = commit_do(trans, NULL, NULL,
 356                                   BTREE_INSERT_LAZY_RW|
 357                                   BTREE_INSERT_NOFAIL,
 358                         __reattach_inode(trans, inode, inode_snapshot));
 359         bch_err_msg(trans->c, ret, "reattaching inode %llu", inode->bi_inum);
 360         return ret;
 361 }
 362
 363 static int remove_backpointer(struct btree_trans *trans,
 364                               struct bch_inode_unpacked *inode)
 365 {
 366         struct btree_iter iter;
 367         struct bkey_s_c_dirent d;
 368         int ret;
 369
 370         d = bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_dirents,
 371                                      POS(inode->bi_dir, inode->bi_dir_offset), 0,
 372                                      dirent);
 373         ret =   bkey_err(d) ?:
 374                 __remove_dirent(trans, d.k->p);
 375         bch2_trans_iter_exit(trans, &iter);
 376         return ret;
 377 }
 378
 379 struct snapshots_seen_entry {
 380         u32                             id;
 381         u32                             equiv;
 382 };
 383
 384 struct snapshots_seen {
 385         struct bpos                     pos;
 386         DARRAY(struct snapshots_seen_entry) ids;
 387 };
 388
 389 static inline void snapshots_seen_exit(struct snapshots_seen *s)
 390 {
 391         darray_exit(&s->ids);
 392 }
 393
 394 static inline void snapshots_seen_init(struct snapshots_seen *s)
 395 {
 396         memset(s, 0, sizeof(*s));
 397 }
 398
 399 static int snapshots_seen_add_inorder(struct bch_fs *c, struct snapshots_seen *s, u32 id)
 400 {
 401         struct snapshots_seen_entry *i, n = {
 402                 .id     = id,
 403                 .equiv  = bch2_snapshot_equiv(c, id),
 404         };
 405         int ret = 0;
 406
 407         darray_for_each(s->ids, i) {
 408                 if (i->id == id)
 409                         return 0;
 410                 if (i->id > id)
 411                         break;
 412         }
 413
 414         ret = darray_insert_item(&s->ids, i - s->ids.data, n);
 415         if (ret)
 416                 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
 417                         s->ids.size);
 418         return ret;
 419 }
 420
 421 static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s,
 422                                  enum btree_id btree_id, struct bpos pos)
 423 {
 424         struct snapshots_seen_entry *i, n = {
 425                 .id     = pos.snapshot,
 426                 .equiv  = bch2_snapshot_equiv(c, pos.snapshot),
 427         };
 428         int ret = 0;
 429
 430         if (!bkey_eq(s->pos, pos))
 431                 s->ids.nr = 0;
 432
 433         s->pos = pos;
 434         s->pos.snapshot = n.equiv;
 435
 436         darray_for_each(s->ids, i) {
 437                 if (i->id == n.id)
 438                         return 0;
 439
 440                 /*
 441                  * We currently don't rigorously track for snapshot cleanup
 442                  * needing to be run, so it shouldn't be a fsck error yet:
 443                  */
 444                 if (i->equiv == n.equiv) {
 445                         bch_err(c, "snapshot deletion did not finish:\n"
 446                                 "  duplicate keys in btree %s at %llu:%llu snapshots %u, %u (equiv %u)\n",
 447                                 bch2_btree_ids[btree_id],
 448                                 pos.inode, pos.offset,
 449                                 i->id, n.id, n.equiv);
 450                         return bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_delete_dead_snapshots);
 451                 }
 452         }
 453
 454         ret = darray_push(&s->ids, n);
 455         if (ret)
 456                 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
 457                         s->ids.size);
 458         return ret;
 459 }
 460
 461 /**
 462  * key_visible_in_snapshot - returns true if @id is a descendent of @ancestor,
 463  * and @ancestor hasn't been overwritten in @seen
 464  *
 465  * @c:          filesystem handle
 466  * @seen:       list of snapshot ids already seen at current position
 467  * @id:         descendent snapshot id
 468  * @ancestor:   ancestor snapshot id
 469  *
 470  * Returns:     whether key in @ancestor snapshot is visible in @id snapshot
 471  */
 472 static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *seen,
 473                                     u32 id, u32 ancestor)
 474 {
 475         ssize_t i;
 476
 477         EBUG_ON(id > ancestor);
 478         EBUG_ON(!bch2_snapshot_is_equiv(c, id));
 479         EBUG_ON(!bch2_snapshot_is_equiv(c, ancestor));
 480
 481         /* @ancestor should be the snapshot most recently added to @seen */
 482         EBUG_ON(ancestor != seen->pos.snapshot);
 483         EBUG_ON(ancestor != seen->ids.data[seen->ids.nr - 1].equiv);
 484
 485         if (id == ancestor)
 486                 return true;
 487
 488         if (!bch2_snapshot_is_ancestor(c, id, ancestor))
 489                 return false;
 490
 491         /*
 492          * We know that @id is a descendant of @ancestor, we're checking if
 493          * we've seen a key that overwrote @ancestor - i.e. also a descendent of
 494          * @ascestor and with @id as a descendent.
 495          *
 496          * But we already know that we're scanning IDs between @id and @ancestor
 497          * numerically, since snapshot ID lists are kept sorted, so if we find
 498          * an id that's an ancestor of @id we're done:
 499          */
 500
 501         for (i = seen->ids.nr - 2;
 502              i >= 0 && seen->ids.data[i].equiv >= id;
 503              --i)
 504                 if (bch2_snapshot_is_ancestor(c, id, seen->ids.data[i].equiv))
 505                         return false;
 506
 507         return true;
 508 }
 509
 510 /**
 511  * ref_visible - given a key with snapshot id @src that points to a key with
 512  * snapshot id @dst, test whether there is some snapshot in which @dst is
 513  * visible.
 514  *
 515  * @c:          filesystem handle
 516  * @s:          list of snapshot IDs already seen at @src
 517  * @src:        snapshot ID of src key
 518  * @dst:        snapshot ID of dst key
 519  * Returns:     true if there is some snapshot in which @dst is visible
 520  *
 521  * Assumes we're visiting @src keys in natural key order
 522  */
 523 static bool ref_visible(struct bch_fs *c, struct snapshots_seen *s,
 524                         u32 src, u32 dst)
 525 {
 526         return dst <= src
 527                 ? key_visible_in_snapshot(c, s, dst, src)
 528                 : bch2_snapshot_is_ancestor(c, src, dst);
 529 }
 530
 531 static int ref_visible2(struct bch_fs *c,
 532                         u32 src, struct snapshots_seen *src_seen,
 533                         u32 dst, struct snapshots_seen *dst_seen)
 534 {
 535         src = bch2_snapshot_equiv(c, src);
 536         dst = bch2_snapshot_equiv(c, dst);
 537
 538         if (dst > src) {
 539                 swap(dst, src);
 540                 swap(dst_seen, src_seen);
 541         }
 542         return key_visible_in_snapshot(c, src_seen, dst, src);
 543 }
 544
 545 #define for_each_visible_inode(_c, _s, _w, _snapshot, _i)                               \
 546         for (_i = (_w)->inodes.data; _i < (_w)->inodes.data + (_w)->inodes.nr &&        \
 547              (_i)->snapshot <= (_snapshot); _i++)                                       \
 548                 if (key_visible_in_snapshot(_c, _s, _i->snapshot, _snapshot))
 549
 550 struct inode_walker_entry {
 551         struct bch_inode_unpacked inode;
 552         u32                     snapshot;
 553         bool                    seen_this_pos;
 554         u64                     count;
 555 };
 556
 557 struct inode_walker {
 558         bool                            first_this_inode;
 559         bool                            recalculate_sums;
 560         struct bpos                     last_pos;
 561
 562         DARRAY(struct inode_walker_entry) inodes;
 563 };
 564
 565 static void inode_walker_exit(struct inode_walker *w)
 566 {
 567         darray_exit(&w->inodes);
 568 }
 569
 570 static struct inode_walker inode_walker_init(void)
 571 {
 572         return (struct inode_walker) { 0, };
 573 }
 574
 575 static int add_inode(struct bch_fs *c, struct inode_walker *w,
 576                      struct bkey_s_c inode)
 577 {
 578         struct bch_inode_unpacked u;
 579
 580         BUG_ON(bch2_inode_unpack(inode, &u));
 581
 582         return darray_push(&w->inodes, ((struct inode_walker_entry) {
 583                 .inode          = u,
 584                 .snapshot       = bch2_snapshot_equiv(c, inode.k->p.snapshot),
 585         }));
 586 }
 587
 588 static int get_inodes_all_snapshots(struct btree_trans *trans,
 589                                     struct inode_walker *w, u64 inum)
 590 {
 591         struct bch_fs *c = trans->c;
 592         struct btree_iter iter;
 593         struct bkey_s_c k;
 594         u32 restart_count = trans->restart_count;
 595         int ret;
 596
 597         w->recalculate_sums = false;
 598         w->inodes.nr = 0;
 599
 600         for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, inum),
 601                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
 602                 if (k.k->p.offset != inum)
 603                         break;
 604
 605                 if (bkey_is_inode(k.k))
 606                         add_inode(c, w, k);
 607         }
 608         bch2_trans_iter_exit(trans, &iter);
 609
 610         if (ret)
 611                 return ret;
 612
 613         w->first_this_inode = true;
 614
 615         return trans_was_restarted(trans, restart_count);
 616 }
 617
 618 static struct inode_walker_entry *
 619 lookup_inode_for_snapshot(struct bch_fs *c, struct inode_walker *w,
 620                           u32 snapshot, bool is_whiteout)
 621 {
 622         struct inode_walker_entry *i;
 623
 624         snapshot = bch2_snapshot_equiv(c, snapshot);
 625
 626         darray_for_each(w->inodes, i)
 627                 if (bch2_snapshot_is_ancestor(c, snapshot, i->snapshot))
 628                         goto found;
 629
 630         return NULL;
 631 found:
 632         BUG_ON(snapshot > i->snapshot);
 633
 634         if (snapshot != i->snapshot && !is_whiteout) {
 635                 struct inode_walker_entry new = *i;
 636                 size_t pos;
 637                 int ret;
 638
 639                 new.snapshot = snapshot;
 640                 new.count = 0;
 641
 642                 bch_info(c, "have key for inode %llu:%u but have inode in ancestor snapshot %u",
 643                          w->last_pos.inode, snapshot, i->snapshot);
 644
 645                 while (i > w->inodes.data && i[-1].snapshot > snapshot)
 646                         --i;
 647
 648                 pos = i - w->inodes.data;
 649                 ret = darray_insert_item(&w->inodes, pos, new);
 650                 if (ret)
 651                         return ERR_PTR(ret);
 652
 653                 i = w->inodes.data + pos;
 654         }
 655
 656         return i;
 657 }
 658
 659 static struct inode_walker_entry *walk_inode(struct btree_trans *trans,
 660                                              struct inode_walker *w, struct bpos pos,
 661                                              bool is_whiteout)
 662 {
 663         if (w->last_pos.inode != pos.inode) {
 664                 int ret = get_inodes_all_snapshots(trans, w, pos.inode);
 665                 if (ret)
 666                         return ERR_PTR(ret);
 667         } else if (bkey_cmp(w->last_pos, pos)) {
 668                 struct inode_walker_entry *i;
 669
 670                 darray_for_each(w->inodes, i)
 671                         i->seen_this_pos = false;
 672
 673         }
 674
 675         w->last_pos = pos;
 676
 677         return lookup_inode_for_snapshot(trans->c, w, pos.snapshot, is_whiteout);
 678 }
 679
 680 static int __get_visible_inodes(struct btree_trans *trans,
 681                                 struct inode_walker *w,
 682                                 struct snapshots_seen *s,
 683                                 u64 inum)
 684 {
 685         struct bch_fs *c = trans->c;
 686         struct btree_iter iter;
 687         struct bkey_s_c k;
 688         int ret;
 689
 690         w->inodes.nr = 0;
 691
 692         for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum),
 693                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
 694                 u32 equiv = bch2_snapshot_equiv(c, k.k->p.snapshot);
 695
 696                 if (k.k->p.offset != inum)
 697                         break;
 698
 699                 if (!ref_visible(c, s, s->pos.snapshot, equiv))
 700                         continue;
 701
 702                 if (bkey_is_inode(k.k))
 703                         add_inode(c, w, k);
 704
 705                 if (equiv >= s->pos.snapshot)
 706                         break;
 707         }
 708         bch2_trans_iter_exit(trans, &iter);
 709
 710         return ret;
 711 }
 712
 713 static int check_key_has_snapshot(struct btree_trans *trans,
 714                                   struct btree_iter *iter,
 715                                   struct bkey_s_c k)
 716 {
 717         struct bch_fs *c = trans->c;
 718         struct printbuf buf = PRINTBUF;
 719         int ret = 0;
 720
 721         if (mustfix_fsck_err_on(!bch2_snapshot_equiv(c, k.k->p.snapshot), c,
 722                         "key in missing snapshot: %s",
 723                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
 724                 ret = bch2_btree_delete_at(trans, iter,
 725                                             BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?: 1;
 726 fsck_err:
 727         printbuf_exit(&buf);
 728         return ret;
 729 }
 730
 731 static int hash_redo_key(struct btree_trans *trans,
 732                          const struct bch_hash_desc desc,
 733                          struct bch_hash_info *hash_info,
 734                          struct btree_iter *k_iter, struct bkey_s_c k)
 735 {
 736         struct bkey_i *delete;
 737         struct bkey_i *tmp;
 738
 739         delete = bch2_trans_kmalloc(trans, sizeof(*delete));
 740         if (IS_ERR(delete))
 741                 return PTR_ERR(delete);
 742
 743         tmp = bch2_bkey_make_mut_noupdate(trans, k);
 744         if (IS_ERR(tmp))
 745                 return PTR_ERR(tmp);
 746
 747         bkey_init(&delete->k);
 748         delete->k.p = k_iter->pos;
 749         return  bch2_btree_iter_traverse(k_iter) ?:
 750                 bch2_trans_update(trans, k_iter, delete, 0) ?:
 751                 bch2_hash_set_snapshot(trans, desc, hash_info,
 752                                        (subvol_inum) { 0, k.k->p.inode },
 753                                        k.k->p.snapshot, tmp,
 754                                        BCH_HASH_SET_MUST_CREATE,
 755                                        BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
 756                 bch2_trans_commit(trans, NULL, NULL,
 757                                   BTREE_INSERT_NOFAIL|
 758                                   BTREE_INSERT_LAZY_RW);
 759 }
 760
 761 static int hash_check_key(struct btree_trans *trans,
 762                           const struct bch_hash_desc desc,
 763                           struct bch_hash_info *hash_info,
 764                           struct btree_iter *k_iter, struct bkey_s_c hash_k)
 765 {
 766         struct bch_fs *c = trans->c;
 767         struct btree_iter iter = { NULL };
 768         struct printbuf buf = PRINTBUF;
 769         struct bkey_s_c k;
 770         u64 hash;
 771         int ret = 0;
 772
 773         if (hash_k.k->type != desc.key_type)
 774                 return 0;
 775
 776         hash = desc.hash_bkey(hash_info, hash_k);
 777
 778         if (likely(hash == hash_k.k->p.offset))
 779                 return 0;
 780
 781         if (hash_k.k->p.offset < hash)
 782                 goto bad_hash;
 783
 784         for_each_btree_key_norestart(trans, iter, desc.btree_id,
 785                                      SPOS(hash_k.k->p.inode, hash, hash_k.k->p.snapshot),
 786                                      BTREE_ITER_SLOTS, k, ret) {
 787                 if (bkey_eq(k.k->p, hash_k.k->p))
 788                         break;
 789
 790                 if (fsck_err_on(k.k->type == desc.key_type &&
 791                                 !desc.cmp_bkey(k, hash_k), c,
 792                                 "duplicate hash table keys:\n%s",
 793                                 (printbuf_reset(&buf),
 794                                  bch2_bkey_val_to_text(&buf, c, hash_k),
 795                                  buf.buf))) {
 796                         ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0) ?: 1;
 797                         break;
 798                 }
 799
 800                 if (bkey_deleted(k.k)) {
 801                         bch2_trans_iter_exit(trans, &iter);
 802                         goto bad_hash;
 803                 }
 804         }
 805 out:
 806         bch2_trans_iter_exit(trans, &iter);
 807         printbuf_exit(&buf);
 808         return ret;
 809 bad_hash:
 810         if (fsck_err(c, "hash table key at wrong offset: btree %s inode %llu offset %llu, hashed to %llu\n%s",
 811                      bch2_btree_ids[desc.btree_id], hash_k.k->p.inode, hash_k.k->p.offset, hash,
 812                      (printbuf_reset(&buf),
 813                       bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf))) {
 814                 ret = hash_redo_key(trans, desc, hash_info, k_iter, hash_k);
 815                 bch_err_fn(c, ret);
 816                 if (ret)
 817                         return ret;
 818                 ret = -BCH_ERR_transaction_restart_nested;
 819         }
 820 fsck_err:
 821         goto out;
 822 }
 823
 824 static int check_inode(struct btree_trans *trans,
 825                        struct btree_iter *iter,
 826                        struct bkey_s_c k,
 827                        struct bch_inode_unpacked *prev,
 828                        struct snapshots_seen *s,
 829                        bool full)
 830 {
 831         struct bch_fs *c = trans->c;
 832         struct bch_inode_unpacked u;
 833         bool do_update = false;
 834         int ret;
 835
 836         ret = check_key_has_snapshot(trans, iter, k);
 837         if (ret < 0)
 838                 goto err;
 839         if (ret)
 840                 return 0;
 841
 842         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
 843         if (ret)
 844                 goto err;
 845
 846         if (!bkey_is_inode(k.k))
 847                 return 0;
 848
 849         BUG_ON(bch2_inode_unpack(k, &u));
 850
 851         if (!full &&
 852             !(u.bi_flags & (BCH_INODE_I_SIZE_DIRTY|
 853                             BCH_INODE_I_SECTORS_DIRTY|
 854                             BCH_INODE_UNLINKED)))
 855                 return 0;
 856
 857         if (prev->bi_inum != u.bi_inum)
 858                 *prev = u;
 859
 860         if (fsck_err_on(prev->bi_hash_seed      != u.bi_hash_seed ||
 861                         inode_d_type(prev)      != inode_d_type(&u), c,
 862                         "inodes in different snapshots don't match")) {
 863                 bch_err(c, "repair not implemented yet");
 864                 return -EINVAL;
 865         }
 866
 867         if ((u.bi_flags & (BCH_INODE_I_SIZE_DIRTY|BCH_INODE_UNLINKED)) &&
 868             bch2_key_has_snapshot_overwrites(trans, BTREE_ID_inodes, k.k->p)) {
 869                 struct bpos new_min_pos;
 870
 871                 ret = bch2_propagate_key_to_snapshot_leaves(trans, iter->btree_id, k, &new_min_pos);
 872                 if (ret)
 873                         goto err;
 874
 875                 u.bi_flags &= ~BCH_INODE_I_SIZE_DIRTY|BCH_INODE_UNLINKED;
 876
 877                 ret = __write_inode(trans, &u, iter->pos.snapshot);
 878                 bch_err_msg(c, ret, "in fsck updating inode");
 879                 if (ret)
 880                         return ret;
 881
 882                 if (!bpos_eq(new_min_pos, POS_MIN))
 883                         bch2_btree_iter_set_pos(iter, bpos_predecessor(new_min_pos));
 884                 return 0;
 885         }
 886
 887         if (u.bi_flags & BCH_INODE_UNLINKED &&
 888             (!c->sb.clean ||
 889              fsck_err(c, "filesystem marked clean, but inode %llu unlinked",
 890                       u.bi_inum))) {
 891                 bch2_trans_unlock(trans);
 892                 bch2_fs_lazy_rw(c);
 893
 894                 ret = bch2_inode_rm_snapshot(trans, u.bi_inum, iter->pos.snapshot);
 895                 bch_err_msg(c, ret, "in fsck deleting inode");
 896                 return ret;
 897         }
 898
 899         if (u.bi_flags & BCH_INODE_I_SIZE_DIRTY &&
 900             (!c->sb.clean ||
 901              fsck_err(c, "filesystem marked clean, but inode %llu has i_size dirty",
 902                       u.bi_inum))) {
 903                 bch_verbose(c, "truncating inode %llu", u.bi_inum);
 904
 905                 bch2_trans_unlock(trans);
 906                 bch2_fs_lazy_rw(c);
 907
 908                 /*
 909                  * XXX: need to truncate partial blocks too here - or ideally
 910                  * just switch units to bytes and that issue goes away
 911                  */
 912                 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
 913                                 SPOS(u.bi_inum, round_up(u.bi_size, block_bytes(c)) >> 9,
 914                                      iter->pos.snapshot),
 915                                 POS(u.bi_inum, U64_MAX),
 916                                 0, NULL);
 917                 bch_err_msg(c, ret, "in fsck truncating inode");
 918                 if (ret)
 919                         return ret;
 920
 921                 /*
 922                  * We truncated without our normal sector accounting hook, just
 923                  * make sure we recalculate it:
 924                  */
 925                 u.bi_flags |= BCH_INODE_I_SECTORS_DIRTY;
 926
 927                 u.bi_flags &= ~BCH_INODE_I_SIZE_DIRTY;
 928                 do_update = true;
 929         }
 930
 931         if (u.bi_flags & BCH_INODE_I_SECTORS_DIRTY &&
 932             (!c->sb.clean ||
 933              fsck_err(c, "filesystem marked clean, but inode %llu has i_sectors dirty",
 934                       u.bi_inum))) {
 935                 s64 sectors;
 936
 937                 bch_verbose(c, "recounting sectors for inode %llu",
 938                             u.bi_inum);
 939
 940                 sectors = bch2_count_inode_sectors(trans, u.bi_inum, iter->pos.snapshot);
 941                 if (sectors < 0) {
 942                         bch_err_msg(c, sectors, "in fsck recounting inode sectors");
 943                         return sectors;
 944                 }
 945
 946                 u.bi_sectors = sectors;
 947                 u.bi_flags &= ~BCH_INODE_I_SECTORS_DIRTY;
 948                 do_update = true;
 949         }
 950
 951         if (u.bi_flags & BCH_INODE_BACKPTR_UNTRUSTED) {
 952                 u.bi_dir = 0;
 953                 u.bi_dir_offset = 0;
 954                 u.bi_flags &= ~BCH_INODE_BACKPTR_UNTRUSTED;
 955                 do_update = true;
 956         }
 957
 958         if (do_update) {
 959                 ret = __write_inode(trans, &u, iter->pos.snapshot);
 960                 bch_err_msg(c, ret, "in fsck updating inode");
 961                 if (ret)
 962                         return ret;
 963         }
 964 err:
 965 fsck_err:
 966         bch_err_fn(c, ret);
 967         return ret;
 968 }
 969
 970 noinline_for_stack
 971 int bch2_check_inodes(struct bch_fs *c)
 972 {
 973         bool full = c->opts.fsck;
 974         struct btree_trans *trans = bch2_trans_get(c);
 975         struct btree_iter iter;
 976         struct bch_inode_unpacked prev = { 0 };
 977         struct snapshots_seen s;
 978         struct bkey_s_c k;
 979         int ret;
 980
 981         snapshots_seen_init(&s);
 982
 983         ret = for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
 984                         POS_MIN,
 985                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
 986                         NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
 987                 check_inode(trans, &iter, k, &prev, &s, full));
 988
 989         snapshots_seen_exit(&s);
 990         bch2_trans_put(trans);
 991         bch_err_fn(c, ret);
 992         return ret;
 993 }
 994
 995 static struct bkey_s_c_dirent dirent_get_by_pos(struct btree_trans *trans,
 996                                                 struct btree_iter *iter,
 997                                                 struct bpos pos)
 998 {
 999         return bch2_bkey_get_iter_typed(trans, iter, BTREE_ID_dirents, pos, 0, dirent);
1000 }
1001
1002 static bool inode_points_to_dirent(struct bch_inode_unpacked *inode,
1003                                    struct bkey_s_c_dirent d)
1004 {
1005         return  inode->bi_dir           == d.k->p.inode &&
1006                 inode->bi_dir_offset    == d.k->p.offset;
1007 }
1008
1009 static bool dirent_points_to_inode(struct bkey_s_c_dirent d,
1010                                    struct bch_inode_unpacked *inode)
1011 {
1012         return d.v->d_type == DT_SUBVOL
1013                 ? le32_to_cpu(d.v->d_child_subvol)      == inode->bi_subvol
1014                 : le64_to_cpu(d.v->d_inum)              == inode->bi_inum;
1015 }
1016
1017 static int inode_backpointer_exists(struct btree_trans *trans,
1018                                     struct bch_inode_unpacked *inode,
1019                                     u32 snapshot)
1020 {
1021         struct btree_iter iter;
1022         struct bkey_s_c_dirent d;
1023         int ret;
1024
1025         d = dirent_get_by_pos(trans, &iter,
1026                         SPOS(inode->bi_dir, inode->bi_dir_offset, snapshot));
1027         ret = bkey_err(d);
1028         if (ret)
1029                 return bch2_err_matches(ret, ENOENT) ? 0 : ret;
1030
1031         ret = dirent_points_to_inode(d, inode);
1032         bch2_trans_iter_exit(trans, &iter);
1033         return ret;
1034 }
1035
1036 static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
1037 {
1038         struct bch_fs *c = trans->c;
1039         struct inode_walker_entry *i;
1040         u32 restart_count = trans->restart_count;
1041         int ret = 0;
1042         s64 count2;
1043
1044         darray_for_each(w->inodes, i) {
1045                 if (i->inode.bi_sectors == i->count)
1046                         continue;
1047
1048                 count2 = bch2_count_inode_sectors(trans, w->last_pos.inode, i->snapshot);
1049
1050                 if (w->recalculate_sums)
1051                         i->count = count2;
1052
1053                 if (i->count != count2) {
1054                         bch_err(c, "fsck counted i_sectors wrong for inode %llu:%u: got %llu should be %llu",
1055                                 w->last_pos.inode, i->snapshot, i->count, count2);
1056                         return -BCH_ERR_internal_fsck_err;
1057                 }
1058
1059                 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SECTORS_DIRTY), c,
1060                             "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
1061                             w->last_pos.inode, i->snapshot,
1062                             i->inode.bi_sectors, i->count)) {
1063                         i->inode.bi_sectors = i->count;
1064                         ret = fsck_write_inode(trans, &i->inode, i->snapshot);
1065                         if (ret)
1066                                 break;
1067                 }
1068         }
1069 fsck_err:
1070         bch_err_fn(c, ret);
1071         return ret ?: trans_was_restarted(trans, restart_count);
1072 }
1073
1074 struct extent_end {
1075         u32                     snapshot;
1076         u64                     offset;
1077         struct snapshots_seen   seen;
1078 };
1079
1080 struct extent_ends {
1081         struct bpos                     last_pos;
1082         DARRAY(struct extent_end)       e;
1083 };
1084
1085 static void extent_ends_reset(struct extent_ends *extent_ends)
1086 {
1087         struct extent_end *i;
1088
1089         darray_for_each(extent_ends->e, i)
1090                 snapshots_seen_exit(&i->seen);
1091
1092         extent_ends->e.nr = 0;
1093 }
1094
1095 static void extent_ends_exit(struct extent_ends *extent_ends)
1096 {
1097         extent_ends_reset(extent_ends);
1098         darray_exit(&extent_ends->e);
1099 }
1100
1101 static void extent_ends_init(struct extent_ends *extent_ends)
1102 {
1103         memset(extent_ends, 0, sizeof(*extent_ends));
1104 }
1105
1106 static int extent_ends_at(struct bch_fs *c,
1107                           struct extent_ends *extent_ends,
1108                           struct snapshots_seen *seen,
1109                           struct bkey_s_c k)
1110 {
1111         struct extent_end *i, n = (struct extent_end) {
1112                 .offset         = k.k->p.offset,
1113                 .snapshot       = k.k->p.snapshot,
1114                 .seen           = *seen,
1115         };
1116
1117         n.seen.ids.data = kmemdup(seen->ids.data,
1118                               sizeof(seen->ids.data[0]) * seen->ids.size,
1119                               GFP_KERNEL);
1120         if (!n.seen.ids.data)
1121                 return -BCH_ERR_ENOMEM_fsck_extent_ends_at;
1122
1123         darray_for_each(extent_ends->e, i) {
1124                 if (i->snapshot == k.k->p.snapshot) {
1125                         snapshots_seen_exit(&i->seen);
1126                         *i = n;
1127                         return 0;
1128                 }
1129
1130                 if (i->snapshot >= k.k->p.snapshot)
1131                         break;
1132         }
1133
1134         return darray_insert_item(&extent_ends->e, i - extent_ends->e.data, n);
1135 }
1136
1137 static int overlapping_extents_found(struct btree_trans *trans,
1138                                      enum btree_id btree,
1139                                      struct bpos pos1, struct snapshots_seen *pos1_seen,
1140                                      struct bkey pos2,
1141                                      bool *fixed,
1142                                      struct extent_end *extent_end)
1143 {
1144         struct bch_fs *c = trans->c;
1145         struct printbuf buf = PRINTBUF;
1146         struct btree_iter iter1, iter2 = { NULL };
1147         struct bkey_s_c k1, k2;
1148         int ret;
1149
1150         BUG_ON(bkey_le(pos1, bkey_start_pos(&pos2)));
1151
1152         bch2_trans_iter_init(trans, &iter1, btree, pos1,
1153                              BTREE_ITER_ALL_SNAPSHOTS|
1154                              BTREE_ITER_NOT_EXTENTS);
1155         k1 = bch2_btree_iter_peek_upto(&iter1, POS(pos1.inode, U64_MAX));
1156         ret = bkey_err(k1);
1157         if (ret)
1158                 goto err;
1159
1160         prt_str(&buf, "\n  ");
1161         bch2_bkey_val_to_text(&buf, c, k1);
1162
1163         if (!bpos_eq(pos1, k1.k->p)) {
1164                 prt_str(&buf, "\n  wanted\n  ");
1165                 bch2_bpos_to_text(&buf, pos1);
1166                 prt_str(&buf, "\n  ");
1167                 bch2_bkey_to_text(&buf, &pos2);
1168
1169                 bch_err(c, "%s: error finding first overlapping extent when repairing, got%s",
1170                         __func__, buf.buf);
1171                 ret = -BCH_ERR_internal_fsck_err;
1172                 goto err;
1173         }
1174
1175         bch2_trans_copy_iter(&iter2, &iter1);
1176
1177         while (1) {
1178                 bch2_btree_iter_advance(&iter2);
1179
1180                 k2 = bch2_btree_iter_peek_upto(&iter2, POS(pos1.inode, U64_MAX));
1181                 ret = bkey_err(k2);
1182                 if (ret)
1183                         goto err;
1184
1185                 if (bpos_ge(k2.k->p, pos2.p))
1186                         break;
1187         }
1188
1189         prt_str(&buf, "\n  ");
1190         bch2_bkey_val_to_text(&buf, c, k2);
1191
1192         if (bpos_gt(k2.k->p, pos2.p) ||
1193             pos2.size != k2.k->size) {
1194                 bch_err(c, "%s: error finding seconding overlapping extent when repairing%s",
1195                         __func__, buf.buf);
1196                 ret = -BCH_ERR_internal_fsck_err;
1197                 goto err;
1198         }
1199
1200         prt_printf(&buf, "\n  overwriting %s extent",
1201                    pos1.snapshot >= pos2.p.snapshot ? "first" : "second");
1202
1203         if (fsck_err(c, "overlapping extents%s", buf.buf)) {
1204                 struct btree_iter *old_iter = &iter1;
1205                 struct disk_reservation res = { 0 };
1206
1207                 if (pos1.snapshot < pos2.p.snapshot) {
1208                         old_iter = &iter2;
1209                         swap(k1, k2);
1210                 }
1211
1212                 trans->extra_journal_res += bch2_bkey_sectors_compressed(k2);
1213
1214                 ret =   bch2_trans_update_extent_overwrite(trans, old_iter,
1215                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE,
1216                                 k1, k2) ?:
1217                         bch2_trans_commit(trans, &res, NULL,
1218                                 BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL);
1219                 bch2_disk_reservation_put(c, &res);
1220
1221                 if (ret)
1222                         goto err;
1223
1224                 *fixed = true;
1225
1226                 if (pos1.snapshot == pos2.p.snapshot) {
1227                         /*
1228                          * We overwrote the first extent, and did the overwrite
1229                          * in the same snapshot:
1230                          */
1231                         extent_end->offset = bkey_start_offset(&pos2);
1232                 } else if (pos1.snapshot > pos2.p.snapshot) {
1233                         /*
1234                          * We overwrote the first extent in pos2's snapshot:
1235                          */
1236                         ret = snapshots_seen_add_inorder(c, pos1_seen, pos2.p.snapshot);
1237                 } else {
1238                         /*
1239                          * We overwrote the second extent - restart
1240                          * check_extent() from the top:
1241                          */
1242                         ret = -BCH_ERR_transaction_restart_nested;
1243                 }
1244         }
1245 fsck_err:
1246 err:
1247         bch2_trans_iter_exit(trans, &iter2);
1248         bch2_trans_iter_exit(trans, &iter1);
1249         printbuf_exit(&buf);
1250         return ret;
1251 }
1252
1253 static int check_overlapping_extents(struct btree_trans *trans,
1254                               struct snapshots_seen *seen,
1255                               struct extent_ends *extent_ends,
1256                               struct bkey_s_c k,
1257                               u32 equiv,
1258                               struct btree_iter *iter,
1259                               bool *fixed)
1260 {
1261         struct bch_fs *c = trans->c;
1262         struct extent_end *i;
1263         int ret = 0;
1264
1265         /* transaction restart, running again */
1266         if (bpos_eq(extent_ends->last_pos, k.k->p))
1267                 return 0;
1268
1269         if (extent_ends->last_pos.inode != k.k->p.inode)
1270                 extent_ends_reset(extent_ends);
1271
1272         darray_for_each(extent_ends->e, i) {
1273                 if (i->offset <= bkey_start_offset(k.k))
1274                         continue;
1275
1276                 if (!ref_visible2(c,
1277                                   k.k->p.snapshot, seen,
1278                                   i->snapshot, &i->seen))
1279                         continue;
1280
1281                 ret = overlapping_extents_found(trans, iter->btree_id,
1282                                                 SPOS(iter->pos.inode,
1283                                                      i->offset,
1284                                                      i->snapshot),
1285                                                 &i->seen,
1286                                                 *k.k, fixed, i);
1287                 if (ret)
1288                         goto err;
1289         }
1290
1291         ret = extent_ends_at(c, extent_ends, seen, k);
1292         if (ret)
1293                 goto err;
1294
1295         extent_ends->last_pos = k.k->p;
1296 err:
1297         return ret;
1298 }
1299
1300 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
1301                         struct bkey_s_c k,
1302                         struct inode_walker *inode,
1303                         struct snapshots_seen *s,
1304                         struct extent_ends *extent_ends)
1305 {
1306         struct bch_fs *c = trans->c;
1307         struct inode_walker_entry *i;
1308         struct printbuf buf = PRINTBUF;
1309         struct bpos equiv = k.k->p;
1310         int ret = 0;
1311
1312         equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1313
1314         ret = check_key_has_snapshot(trans, iter, k);
1315         if (ret) {
1316                 ret = ret < 0 ? ret : 0;
1317                 goto out;
1318         }
1319
1320         if (inode->last_pos.inode != k.k->p.inode) {
1321                 ret = check_i_sectors(trans, inode);
1322                 if (ret)
1323                         goto err;
1324         }
1325
1326         i = walk_inode(trans, inode, equiv, k.k->type == KEY_TYPE_whiteout);
1327         ret = PTR_ERR_OR_ZERO(i);
1328         if (ret)
1329                 goto err;
1330
1331         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1332         if (ret)
1333                 goto err;
1334
1335         if (k.k->type != KEY_TYPE_whiteout) {
1336                 if (fsck_err_on(!i, c,
1337                                 "extent in missing inode:\n  %s",
1338                                 (printbuf_reset(&buf),
1339                                  bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1340                         goto delete;
1341
1342                 if (fsck_err_on(i &&
1343                                 !S_ISREG(i->inode.bi_mode) &&
1344                                 !S_ISLNK(i->inode.bi_mode), c,
1345                                 "extent in non regular inode mode %o:\n  %s",
1346                                 i->inode.bi_mode,
1347                                 (printbuf_reset(&buf),
1348                                  bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1349                         goto delete;
1350
1351                 ret = check_overlapping_extents(trans, s, extent_ends, k,
1352                                                 equiv.snapshot, iter,
1353                                                 &inode->recalculate_sums);
1354                 if (ret)
1355                         goto err;
1356         }
1357
1358         /*
1359          * Check inodes in reverse order, from oldest snapshots to newest,
1360          * starting from the inode that matches this extent's snapshot. If we
1361          * didn't have one, iterate over all inodes:
1362          */
1363         if (!i)
1364                 i = inode->inodes.data + inode->inodes.nr - 1;
1365
1366         for (;
1367              inode->inodes.data && i >= inode->inodes.data;
1368              --i) {
1369                 if (i->snapshot > equiv.snapshot ||
1370                     !key_visible_in_snapshot(c, s, i->snapshot, equiv.snapshot))
1371                         continue;
1372
1373                 if (k.k->type != KEY_TYPE_whiteout) {
1374                         if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SIZE_DIRTY) &&
1375                                         k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9 &&
1376                                         !bkey_extent_is_reservation(k), c,
1377                                         "extent type past end of inode %llu:%u, i_size %llu\n  %s",
1378                                         i->inode.bi_inum, i->snapshot, i->inode.bi_size,
1379                                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1380                                 struct btree_iter iter2;
1381
1382                                 bch2_trans_copy_iter(&iter2, iter);
1383                                 bch2_btree_iter_set_snapshot(&iter2, i->snapshot);
1384                                 ret =   bch2_btree_iter_traverse(&iter2) ?:
1385                                         bch2_btree_delete_at(trans, &iter2,
1386                                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1387                                 bch2_trans_iter_exit(trans, &iter2);
1388                                 if (ret)
1389                                         goto err;
1390
1391                                 iter->k.type = KEY_TYPE_whiteout;
1392                         }
1393
1394                         if (bkey_extent_is_allocation(k.k))
1395                                 i->count += k.k->size;
1396                 }
1397
1398                 i->seen_this_pos = true;
1399         }
1400 out:
1401 err:
1402 fsck_err:
1403         printbuf_exit(&buf);
1404         bch_err_fn(c, ret);
1405         return ret;
1406 delete:
1407         ret = bch2_btree_delete_at(trans, iter, BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1408         goto out;
1409 }
1410
1411 /*
1412  * Walk extents: verify that extents have a corresponding S_ISREG inode, and
1413  * that i_size an i_sectors are consistent
1414  */
1415 int bch2_check_extents(struct bch_fs *c)
1416 {
1417         struct inode_walker w = inode_walker_init();
1418         struct snapshots_seen s;
1419         struct btree_trans *trans = bch2_trans_get(c);
1420         struct btree_iter iter;
1421         struct bkey_s_c k;
1422         struct extent_ends extent_ends;
1423         struct disk_reservation res = { 0 };
1424         int ret = 0;
1425
1426         snapshots_seen_init(&s);
1427         extent_ends_init(&extent_ends);
1428
1429         ret = for_each_btree_key_commit(trans, iter, BTREE_ID_extents,
1430                         POS(BCACHEFS_ROOT_INO, 0),
1431                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1432                         &res, NULL,
1433                         BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL, ({
1434                 bch2_disk_reservation_put(c, &res);
1435                 check_extent(trans, &iter, k, &w, &s, &extent_ends);
1436         })) ?:
1437         check_i_sectors(trans, &w);
1438
1439         bch2_disk_reservation_put(c, &res);
1440         extent_ends_exit(&extent_ends);
1441         inode_walker_exit(&w);
1442         snapshots_seen_exit(&s);
1443         bch2_trans_put(trans);
1444
1445         bch_err_fn(c, ret);
1446         return ret;
1447 }
1448
1449 static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
1450 {
1451         struct bch_fs *c = trans->c;
1452         struct inode_walker_entry *i;
1453         u32 restart_count = trans->restart_count;
1454         int ret = 0;
1455         s64 count2;
1456
1457         darray_for_each(w->inodes, i) {
1458                 if (i->inode.bi_nlink == i->count)
1459                         continue;
1460
1461                 count2 = bch2_count_subdirs(trans, w->last_pos.inode, i->snapshot);
1462                 if (count2 < 0)
1463                         return count2;
1464
1465                 if (i->count != count2) {
1466                         bch_err(c, "fsck counted subdirectories wrong: got %llu should be %llu",
1467                                 i->count, count2);
1468                         i->count = count2;
1469                         if (i->inode.bi_nlink == i->count)
1470                                 continue;
1471                 }
1472
1473                 if (fsck_err_on(i->inode.bi_nlink != i->count, c,
1474                                 "directory %llu:%u with wrong i_nlink: got %u, should be %llu",
1475                                 w->last_pos.inode, i->snapshot, i->inode.bi_nlink, i->count)) {
1476                         i->inode.bi_nlink = i->count;
1477                         ret = fsck_write_inode(trans, &i->inode, i->snapshot);
1478                         if (ret)
1479                                 break;
1480                 }
1481         }
1482 fsck_err:
1483         bch_err_fn(c, ret);
1484         return ret ?: trans_was_restarted(trans, restart_count);
1485 }
1486
1487 static int check_dirent_target(struct btree_trans *trans,
1488                                struct btree_iter *iter,
1489                                struct bkey_s_c_dirent d,
1490                                struct bch_inode_unpacked *target,
1491                                u32 target_snapshot)
1492 {
1493         struct bch_fs *c = trans->c;
1494         struct bkey_i_dirent *n;
1495         bool backpointer_exists = true;
1496         struct printbuf buf = PRINTBUF;
1497         int ret = 0;
1498
1499         if (!target->bi_dir &&
1500             !target->bi_dir_offset) {
1501                 target->bi_dir          = d.k->p.inode;
1502                 target->bi_dir_offset   = d.k->p.offset;
1503
1504                 ret = __write_inode(trans, target, target_snapshot);
1505                 if (ret)
1506                         goto err;
1507         }
1508
1509         if (!inode_points_to_dirent(target, d)) {
1510                 ret = inode_backpointer_exists(trans, target, d.k->p.snapshot);
1511                 if (ret < 0)
1512                         goto err;
1513
1514                 backpointer_exists = ret;
1515                 ret = 0;
1516
1517                 if (fsck_err_on(S_ISDIR(target->bi_mode) &&
1518                                 backpointer_exists, c,
1519                                 "directory %llu with multiple links",
1520                                 target->bi_inum)) {
1521                         ret = __remove_dirent(trans, d.k->p);
1522                         goto out;
1523                 }
1524
1525                 if (fsck_err_on(backpointer_exists &&
1526                                 !target->bi_nlink, c,
1527                                 "inode %llu type %s has multiple links but i_nlink 0",
1528                                 target->bi_inum, bch2_d_types[d.v->d_type])) {
1529                         target->bi_nlink++;
1530                         target->bi_flags &= ~BCH_INODE_UNLINKED;
1531
1532                         ret = __write_inode(trans, target, target_snapshot);
1533                         if (ret)
1534                                 goto err;
1535                 }
1536
1537                 if (fsck_err_on(!backpointer_exists, c,
1538                                 "inode %llu:%u has wrong backpointer:\n"
1539                                 "got       %llu:%llu\n"
1540                                 "should be %llu:%llu",
1541                                 target->bi_inum, target_snapshot,
1542                                 target->bi_dir,
1543                                 target->bi_dir_offset,
1544                                 d.k->p.inode,
1545                                 d.k->p.offset)) {
1546                         target->bi_dir          = d.k->p.inode;
1547                         target->bi_dir_offset   = d.k->p.offset;
1548
1549                         ret = __write_inode(trans, target, target_snapshot);
1550                         if (ret)
1551                                 goto err;
1552                 }
1553         }
1554
1555         if (fsck_err_on(d.v->d_type != inode_d_type(target), c,
1556                         "incorrect d_type: got %s, should be %s:\n%s",
1557                         bch2_d_type_str(d.v->d_type),
1558                         bch2_d_type_str(inode_d_type(target)),
1559                         (printbuf_reset(&buf),
1560                          bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
1561                 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1562                 ret = PTR_ERR_OR_ZERO(n);
1563                 if (ret)
1564                         goto err;
1565
1566                 bkey_reassemble(&n->k_i, d.s_c);
1567                 n->v.d_type = inode_d_type(target);
1568
1569                 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1570                 if (ret)
1571                         goto err;
1572
1573                 d = dirent_i_to_s_c(n);
1574         }
1575
1576         if (d.v->d_type == DT_SUBVOL &&
1577             target->bi_parent_subvol != le32_to_cpu(d.v->d_parent_subvol) &&
1578             (c->sb.version < bcachefs_metadata_version_subvol_dirent ||
1579              fsck_err(c, "dirent has wrong d_parent_subvol field: got %u, should be %u",
1580                       le32_to_cpu(d.v->d_parent_subvol),
1581                       target->bi_parent_subvol))) {
1582                 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1583                 ret = PTR_ERR_OR_ZERO(n);
1584                 if (ret)
1585                         goto err;
1586
1587                 bkey_reassemble(&n->k_i, d.s_c);
1588                 n->v.d_parent_subvol = cpu_to_le32(target->bi_parent_subvol);
1589
1590                 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1591                 if (ret)
1592                         goto err;
1593
1594                 d = dirent_i_to_s_c(n);
1595         }
1596 out:
1597 err:
1598 fsck_err:
1599         printbuf_exit(&buf);
1600         bch_err_fn(c, ret);
1601         return ret;
1602 }
1603
1604 static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
1605                         struct bkey_s_c k,
1606                         struct bch_hash_info *hash_info,
1607                         struct inode_walker *dir,
1608                         struct inode_walker *target,
1609                         struct snapshots_seen *s)
1610 {
1611         struct bch_fs *c = trans->c;
1612         struct bkey_s_c_dirent d;
1613         struct inode_walker_entry *i;
1614         struct printbuf buf = PRINTBUF;
1615         struct bpos equiv;
1616         int ret = 0;
1617
1618         ret = check_key_has_snapshot(trans, iter, k);
1619         if (ret) {
1620                 ret = ret < 0 ? ret : 0;
1621                 goto out;
1622         }
1623
1624         equiv = k.k->p;
1625         equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1626
1627         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1628         if (ret)
1629                 goto err;
1630
1631         if (k.k->type == KEY_TYPE_whiteout)
1632                 goto out;
1633
1634         if (dir->last_pos.inode != k.k->p.inode) {
1635                 ret = check_subdir_count(trans, dir);
1636                 if (ret)
1637                         goto err;
1638         }
1639
1640         BUG_ON(!iter->path->should_be_locked);
1641
1642         i = walk_inode(trans, dir, equiv, k.k->type == KEY_TYPE_whiteout);
1643         ret = PTR_ERR_OR_ZERO(i);
1644         if (ret < 0)
1645                 goto err;
1646
1647         if (dir->first_this_inode && dir->inodes.nr)
1648                 *hash_info = bch2_hash_info_init(c, &dir->inodes.data[0].inode);
1649         dir->first_this_inode = false;
1650
1651         if (fsck_err_on(!i, c,
1652                         "dirent in nonexisting directory:\n%s",
1653                         (printbuf_reset(&buf),
1654                          bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1655                 ret = bch2_btree_delete_at(trans, iter,
1656                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1657                 goto out;
1658         }
1659
1660         if (!i)
1661                 goto out;
1662
1663         if (fsck_err_on(!S_ISDIR(i->inode.bi_mode), c,
1664                         "dirent in non directory inode type %s:\n%s",
1665                         bch2_d_type_str(inode_d_type(&i->inode)),
1666                         (printbuf_reset(&buf),
1667                          bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1668                 ret = bch2_btree_delete_at(trans, iter, 0);
1669                 goto out;
1670         }
1671
1672         ret = hash_check_key(trans, bch2_dirent_hash_desc, hash_info, iter, k);
1673         if (ret < 0)
1674                 goto err;
1675         if (ret) {
1676                 /* dirent has been deleted */
1677                 ret = 0;
1678                 goto out;
1679         }
1680
1681         if (k.k->type != KEY_TYPE_dirent)
1682                 goto out;
1683
1684         d = bkey_s_c_to_dirent(k);
1685
1686         if (d.v->d_type == DT_SUBVOL) {
1687                 struct bch_inode_unpacked subvol_root;
1688                 u32 target_subvol = le32_to_cpu(d.v->d_child_subvol);
1689                 u32 target_snapshot;
1690                 u64 target_inum;
1691
1692                 ret = __subvol_lookup(trans, target_subvol,
1693                                       &target_snapshot, &target_inum);
1694                 if (ret && !bch2_err_matches(ret, ENOENT))
1695                         goto err;
1696
1697                 if (fsck_err_on(ret, c,
1698                                 "dirent points to missing subvolume %u",
1699                                 le32_to_cpu(d.v->d_child_subvol))) {
1700                         ret = __remove_dirent(trans, d.k->p);
1701                         goto err;
1702                 }
1703
1704                 ret = __lookup_inode(trans, target_inum,
1705                                    &subvol_root, &target_snapshot);
1706                 if (ret && !bch2_err_matches(ret, ENOENT))
1707                         goto err;
1708
1709                 if (fsck_err_on(ret, c,
1710                                 "subvolume %u points to missing subvolume root %llu",
1711                                 target_subvol,
1712                                 target_inum)) {
1713                         bch_err(c, "repair not implemented yet");
1714                         ret = -EINVAL;
1715                         goto err;
1716                 }
1717
1718                 if (fsck_err_on(subvol_root.bi_subvol != target_subvol, c,
1719                                 "subvol root %llu has wrong bi_subvol field: got %u, should be %u",
1720                                 target_inum,
1721                                 subvol_root.bi_subvol, target_subvol)) {
1722                         subvol_root.bi_subvol = target_subvol;
1723                         ret = __write_inode(trans, &subvol_root, target_snapshot);
1724                         if (ret)
1725                                 goto err;
1726                 }
1727
1728                 ret = check_dirent_target(trans, iter, d, &subvol_root,
1729                                           target_snapshot);
1730                 if (ret)
1731                         goto err;
1732         } else {
1733                 ret = __get_visible_inodes(trans, target, s, le64_to_cpu(d.v->d_inum));
1734                 if (ret)
1735                         goto err;
1736
1737                 if (fsck_err_on(!target->inodes.nr, c,
1738                                 "dirent points to missing inode: (equiv %u)\n%s",
1739                                 equiv.snapshot,
1740                                 (printbuf_reset(&buf),
1741                                  bch2_bkey_val_to_text(&buf, c, k),
1742                                  buf.buf))) {
1743                         ret = __remove_dirent(trans, d.k->p);
1744                         if (ret)
1745                                 goto err;
1746                 }
1747
1748                 darray_for_each(target->inodes, i) {
1749                         ret = check_dirent_target(trans, iter, d,
1750                                                   &i->inode, i->snapshot);
1751                         if (ret)
1752                                 goto err;
1753                 }
1754         }
1755
1756         if (d.v->d_type == DT_DIR)
1757                 for_each_visible_inode(c, s, dir, equiv.snapshot, i)
1758                         i->count++;
1759
1760 out:
1761 err:
1762 fsck_err:
1763         printbuf_exit(&buf);
1764         bch_err_fn(c, ret);
1765         return ret;
1766 }
1767
1768 /*
1769  * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
1770  * validate d_type
1771  */
1772 int bch2_check_dirents(struct bch_fs *c)
1773 {
1774         struct inode_walker dir = inode_walker_init();
1775         struct inode_walker target = inode_walker_init();
1776         struct snapshots_seen s;
1777         struct bch_hash_info hash_info;
1778         struct btree_trans *trans = bch2_trans_get(c);
1779         struct btree_iter iter;
1780         struct bkey_s_c k;
1781         int ret = 0;
1782
1783         snapshots_seen_init(&s);
1784
1785         ret = for_each_btree_key_commit(trans, iter, BTREE_ID_dirents,
1786                         POS(BCACHEFS_ROOT_INO, 0),
1787                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
1788                         k,
1789                         NULL, NULL,
1790                         BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
1791                 check_dirent(trans, &iter, k, &hash_info, &dir, &target, &s));
1792
1793         bch2_trans_put(trans);
1794         snapshots_seen_exit(&s);
1795         inode_walker_exit(&dir);
1796         inode_walker_exit(&target);
1797         bch_err_fn(c, ret);
1798         return ret;
1799 }
1800
1801 static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
1802                        struct bkey_s_c k,
1803                        struct bch_hash_info *hash_info,
1804                        struct inode_walker *inode)
1805 {
1806         struct bch_fs *c = trans->c;
1807         struct inode_walker_entry *i;
1808         int ret;
1809
1810         ret = check_key_has_snapshot(trans, iter, k);
1811         if (ret)
1812                 return ret;
1813
1814         i = walk_inode(trans, inode, k.k->p, k.k->type == KEY_TYPE_whiteout);
1815         ret = PTR_ERR_OR_ZERO(i);
1816         if (ret)
1817                 return ret;
1818
1819         if (inode->first_this_inode && inode->inodes.nr)
1820                 *hash_info = bch2_hash_info_init(c, &inode->inodes.data[0].inode);
1821         inode->first_this_inode = false;
1822
1823         if (fsck_err_on(!i, c,
1824                         "xattr for missing inode %llu",
1825                         k.k->p.inode))
1826                 return bch2_btree_delete_at(trans, iter, 0);
1827
1828         if (!i)
1829                 return 0;
1830
1831         ret = hash_check_key(trans, bch2_xattr_hash_desc, hash_info, iter, k);
1832 fsck_err:
1833         bch_err_fn(c, ret);
1834         return ret;
1835 }
1836
1837 /*
1838  * Walk xattrs: verify that they all have a corresponding inode
1839  */
1840 int bch2_check_xattrs(struct bch_fs *c)
1841 {
1842         struct inode_walker inode = inode_walker_init();
1843         struct bch_hash_info hash_info;
1844         struct btree_iter iter;
1845         struct bkey_s_c k;
1846         int ret = 0;
1847
1848         ret = bch2_trans_run(c,
1849                 for_each_btree_key_commit(trans, iter, BTREE_ID_xattrs,
1850                         POS(BCACHEFS_ROOT_INO, 0),
1851                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
1852                         k,
1853                         NULL, NULL,
1854                         BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
1855                 check_xattr(trans, &iter, k, &hash_info, &inode)));
1856         bch_err_fn(c, ret);
1857         return ret;
1858 }
1859
1860 static int check_root_trans(struct btree_trans *trans)
1861 {
1862         struct bch_fs *c = trans->c;
1863         struct bch_inode_unpacked root_inode;
1864         u32 snapshot;
1865         u64 inum;
1866         int ret;
1867
1868         ret = __subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
1869         if (ret && !bch2_err_matches(ret, ENOENT))
1870                 return ret;
1871
1872         if (mustfix_fsck_err_on(ret, c, "root subvol missing")) {
1873                 struct bkey_i_subvolume root_subvol;
1874
1875                 snapshot        = U32_MAX;
1876                 inum            = BCACHEFS_ROOT_INO;
1877
1878                 bkey_subvolume_init(&root_subvol.k_i);
1879                 root_subvol.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1880                 root_subvol.v.flags     = 0;
1881                 root_subvol.v.snapshot  = cpu_to_le32(snapshot);
1882                 root_subvol.v.inode     = cpu_to_le64(inum);
1883                 ret = commit_do(trans, NULL, NULL,
1884                                       BTREE_INSERT_NOFAIL|
1885                                       BTREE_INSERT_LAZY_RW,
1886                         bch2_btree_insert_trans(trans, BTREE_ID_subvolumes,
1887                                             &root_subvol.k_i, 0));
1888                 bch_err_msg(c, ret, "writing root subvol");
1889                 if (ret)
1890                         goto err;
1891
1892         }
1893
1894         ret = __lookup_inode(trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
1895         if (ret && !bch2_err_matches(ret, ENOENT))
1896                 return ret;
1897
1898         if (mustfix_fsck_err_on(ret, c, "root directory missing") ||
1899             mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode), c,
1900                                 "root inode not a directory")) {
1901                 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755,
1902                                 0, NULL);
1903                 root_inode.bi_inum = inum;
1904
1905                 ret = __write_inode(trans, &root_inode, snapshot);
1906                 bch_err_msg(c, ret, "writing root inode");
1907         }
1908 err:
1909 fsck_err:
1910         return ret;
1911 }
1912
1913 /* Get root directory, create if it doesn't exist: */
1914 int bch2_check_root(struct bch_fs *c)
1915 {
1916         int ret;
1917
1918         ret = bch2_trans_do(c, NULL, NULL,
1919                              BTREE_INSERT_NOFAIL|
1920                              BTREE_INSERT_LAZY_RW,
1921                 check_root_trans(trans));
1922         bch_err_fn(c, ret);
1923         return ret;
1924 }
1925
1926 struct pathbuf_entry {
1927         u64     inum;
1928         u32     snapshot;
1929 };
1930
1931 typedef DARRAY(struct pathbuf_entry) pathbuf;
1932
1933 static bool path_is_dup(pathbuf *p, u64 inum, u32 snapshot)
1934 {
1935         struct pathbuf_entry *i;
1936
1937         darray_for_each(*p, i)
1938                 if (i->inum     == inum &&
1939                     i->snapshot == snapshot)
1940                         return true;
1941
1942         return false;
1943 }
1944
1945 static int path_down(struct bch_fs *c, pathbuf *p,
1946                      u64 inum, u32 snapshot)
1947 {
1948         int ret = darray_push(p, ((struct pathbuf_entry) {
1949                 .inum           = inum,
1950                 .snapshot       = snapshot,
1951         }));
1952
1953         if (ret)
1954                 bch_err(c, "fsck: error allocating memory for pathbuf, size %zu",
1955                         p->size);
1956         return ret;
1957 }
1958
1959 /*
1960  * Check that a given inode is reachable from the root:
1961  *
1962  * XXX: we should also be verifying that inodes are in the right subvolumes
1963  */
1964 static int check_path(struct btree_trans *trans,
1965                       pathbuf *p,
1966                       struct bch_inode_unpacked *inode,
1967                       u32 snapshot)
1968 {
1969         struct bch_fs *c = trans->c;
1970         int ret = 0;
1971
1972         snapshot = bch2_snapshot_equiv(c, snapshot);
1973         p->nr = 0;
1974
1975         while (!(inode->bi_inum == BCACHEFS_ROOT_INO &&
1976                  inode->bi_subvol == BCACHEFS_ROOT_SUBVOL)) {
1977                 struct btree_iter dirent_iter;
1978                 struct bkey_s_c_dirent d;
1979                 u32 parent_snapshot = snapshot;
1980
1981                 if (inode->bi_subvol) {
1982                         u64 inum;
1983
1984                         ret = subvol_lookup(trans, inode->bi_parent_subvol,
1985                                             &parent_snapshot, &inum);
1986                         if (ret)
1987                                 break;
1988                 }
1989
1990                 ret = lockrestart_do(trans,
1991                         PTR_ERR_OR_ZERO((d = dirent_get_by_pos(trans, &dirent_iter,
1992                                           SPOS(inode->bi_dir, inode->bi_dir_offset,
1993                                                parent_snapshot))).k));
1994                 if (ret && !bch2_err_matches(ret, ENOENT))
1995                         break;
1996
1997                 if (!ret && !dirent_points_to_inode(d, inode)) {
1998                         bch2_trans_iter_exit(trans, &dirent_iter);
1999                         ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
2000                 }
2001
2002                 if (bch2_err_matches(ret, ENOENT)) {
2003                         if (fsck_err(c,  "unreachable inode %llu:%u, type %s nlink %u backptr %llu:%llu",
2004                                      inode->bi_inum, snapshot,
2005                                      bch2_d_type_str(inode_d_type(inode)),
2006                                      inode->bi_nlink,
2007                                      inode->bi_dir,
2008                                      inode->bi_dir_offset))
2009                                 ret = reattach_inode(trans, inode, snapshot);
2010                         break;
2011                 }
2012
2013                 bch2_trans_iter_exit(trans, &dirent_iter);
2014
2015                 if (!S_ISDIR(inode->bi_mode))
2016                         break;
2017
2018                 ret = path_down(c, p, inode->bi_inum, snapshot);
2019                 if (ret) {
2020                         bch_err(c, "memory allocation failure");
2021                         return ret;
2022                 }
2023
2024                 snapshot = parent_snapshot;
2025
2026                 ret = lookup_inode(trans, inode->bi_dir, inode, &snapshot);
2027                 if (ret) {
2028                         /* Should have been caught in dirents pass */
2029                         bch_err(c, "error looking up parent directory: %i", ret);
2030                         break;
2031                 }
2032
2033                 if (path_is_dup(p, inode->bi_inum, snapshot)) {
2034                         struct pathbuf_entry *i;
2035
2036                         /* XXX print path */
2037                         bch_err(c, "directory structure loop");
2038
2039                         darray_for_each(*p, i)
2040                                 pr_err("%llu:%u", i->inum, i->snapshot);
2041                         pr_err("%llu:%u", inode->bi_inum, snapshot);
2042
2043                         if (!fsck_err(c, "directory structure loop"))
2044                                 return 0;
2045
2046                         ret = commit_do(trans, NULL, NULL,
2047                                               BTREE_INSERT_NOFAIL|
2048                                               BTREE_INSERT_LAZY_RW,
2049                                         remove_backpointer(trans, inode));
2050                         if (ret) {
2051                                 bch_err(c, "error removing dirent: %i", ret);
2052                                 break;
2053                         }
2054
2055                         ret = reattach_inode(trans, inode, snapshot);
2056                 }
2057         }
2058 fsck_err:
2059         bch_err_fn(c, ret);
2060         return ret;
2061 }
2062
2063 /*
2064  * Check for unreachable inodes, as well as loops in the directory structure:
2065  * After bch2_check_dirents(), if an inode backpointer doesn't exist that means it's
2066  * unreachable:
2067  */
2068 int bch2_check_directory_structure(struct bch_fs *c)
2069 {
2070         struct btree_trans *trans = bch2_trans_get(c);
2071         struct btree_iter iter;
2072         struct bkey_s_c k;
2073         struct bch_inode_unpacked u;
2074         pathbuf path = { 0, };
2075         int ret;
2076
2077         for_each_btree_key(trans, iter, BTREE_ID_inodes, POS_MIN,
2078                            BTREE_ITER_INTENT|
2079                            BTREE_ITER_PREFETCH|
2080                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2081                 if (!bkey_is_inode(k.k))
2082                         continue;
2083
2084                 ret = bch2_inode_unpack(k, &u);
2085                 if (ret) {
2086                         /* Should have been caught earlier in fsck: */
2087                         bch_err(c, "error unpacking inode %llu: %i", k.k->p.offset, ret);
2088                         break;
2089                 }
2090
2091                 if (u.bi_flags & BCH_INODE_UNLINKED)
2092                         continue;
2093
2094                 ret = check_path(trans, &path, &u, iter.pos.snapshot);
2095                 if (ret)
2096                         break;
2097         }
2098         bch2_trans_iter_exit(trans, &iter);
2099         bch2_trans_put(trans);
2100         darray_exit(&path);
2101         bch_err_fn(c, ret);
2102         return ret;
2103 }
2104
2105 struct nlink_table {
2106         size_t          nr;
2107         size_t          size;
2108
2109         struct nlink {
2110                 u64     inum;
2111                 u32     snapshot;
2112                 u32     count;
2113         }               *d;
2114 };
2115
2116 static int add_nlink(struct bch_fs *c, struct nlink_table *t,
2117                      u64 inum, u32 snapshot)
2118 {
2119         if (t->nr == t->size) {
2120                 size_t new_size = max_t(size_t, 128UL, t->size * 2);
2121                 void *d = kvmalloc_array(new_size, sizeof(t->d[0]), GFP_KERNEL);
2122
2123                 if (!d) {
2124                         bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
2125                                 new_size);
2126                         return -BCH_ERR_ENOMEM_fsck_add_nlink;
2127                 }
2128
2129                 if (t->d)
2130                         memcpy(d, t->d, t->size * sizeof(t->d[0]));
2131                 kvfree(t->d);
2132
2133                 t->d = d;
2134                 t->size = new_size;
2135         }
2136
2137
2138         t->d[t->nr++] = (struct nlink) {
2139                 .inum           = inum,
2140                 .snapshot       = snapshot,
2141         };
2142
2143         return 0;
2144 }
2145
2146 static int nlink_cmp(const void *_l, const void *_r)
2147 {
2148         const struct nlink *l = _l;
2149         const struct nlink *r = _r;
2150
2151         return cmp_int(l->inum, r->inum) ?: cmp_int(l->snapshot, r->snapshot);
2152 }
2153
2154 static void inc_link(struct bch_fs *c, struct snapshots_seen *s,
2155                      struct nlink_table *links,
2156                      u64 range_start, u64 range_end, u64 inum, u32 snapshot)
2157 {
2158         struct nlink *link, key = {
2159                 .inum = inum, .snapshot = U32_MAX,
2160         };
2161
2162         if (inum < range_start || inum >= range_end)
2163                 return;
2164
2165         link = __inline_bsearch(&key, links->d, links->nr,
2166                                 sizeof(links->d[0]), nlink_cmp);
2167         if (!link)
2168                 return;
2169
2170         while (link > links->d && link[0].inum == link[-1].inum)
2171                 --link;
2172
2173         for (; link < links->d + links->nr && link->inum == inum; link++)
2174                 if (ref_visible(c, s, snapshot, link->snapshot)) {
2175                         link->count++;
2176                         if (link->snapshot >= snapshot)
2177                                 break;
2178                 }
2179 }
2180
2181 noinline_for_stack
2182 static int check_nlinks_find_hardlinks(struct bch_fs *c,
2183                                        struct nlink_table *t,
2184                                        u64 start, u64 *end)
2185 {
2186         struct btree_trans *trans = bch2_trans_get(c);
2187         struct btree_iter iter;
2188         struct bkey_s_c k;
2189         struct bch_inode_unpacked u;
2190         int ret = 0;
2191
2192         for_each_btree_key(trans, iter, BTREE_ID_inodes,
2193                            POS(0, start),
2194                            BTREE_ITER_INTENT|
2195                            BTREE_ITER_PREFETCH|
2196                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2197                 if (!bkey_is_inode(k.k))
2198                         continue;
2199
2200                 /* Should never fail, checked by bch2_inode_invalid: */
2201                 BUG_ON(bch2_inode_unpack(k, &u));
2202
2203                 /*
2204                  * Backpointer and directory structure checks are sufficient for
2205                  * directories, since they can't have hardlinks:
2206                  */
2207                 if (S_ISDIR(u.bi_mode))
2208                         continue;
2209
2210                 if (!u.bi_nlink)
2211                         continue;
2212
2213                 ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot);
2214                 if (ret) {
2215                         *end = k.k->p.offset;
2216                         ret = 0;
2217                         break;
2218                 }
2219
2220         }
2221         bch2_trans_iter_exit(trans, &iter);
2222         bch2_trans_put(trans);
2223
2224         if (ret)
2225                 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
2226
2227         return ret;
2228 }
2229
2230 noinline_for_stack
2231 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
2232                                      u64 range_start, u64 range_end)
2233 {
2234         struct btree_trans *trans = bch2_trans_get(c);
2235         struct snapshots_seen s;
2236         struct btree_iter iter;
2237         struct bkey_s_c k;
2238         struct bkey_s_c_dirent d;
2239         int ret;
2240
2241         snapshots_seen_init(&s);
2242
2243         for_each_btree_key(trans, iter, BTREE_ID_dirents, POS_MIN,
2244                            BTREE_ITER_INTENT|
2245                            BTREE_ITER_PREFETCH|
2246                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2247                 ret = snapshots_seen_update(c, &s, iter.btree_id, k.k->p);
2248                 if (ret)
2249                         break;
2250
2251                 switch (k.k->type) {
2252                 case KEY_TYPE_dirent:
2253                         d = bkey_s_c_to_dirent(k);
2254
2255                         if (d.v->d_type != DT_DIR &&
2256                             d.v->d_type != DT_SUBVOL)
2257                                 inc_link(c, &s, links, range_start, range_end,
2258                                          le64_to_cpu(d.v->d_inum),
2259                                          bch2_snapshot_equiv(c, d.k->p.snapshot));
2260                         break;
2261                 }
2262         }
2263         bch2_trans_iter_exit(trans, &iter);
2264
2265         if (ret)
2266                 bch_err(c, "error in fsck: btree error %i while walking dirents", ret);
2267
2268         bch2_trans_put(trans);
2269         snapshots_seen_exit(&s);
2270         return ret;
2271 }
2272
2273 static int check_nlinks_update_inode(struct btree_trans *trans, struct btree_iter *iter,
2274                                      struct bkey_s_c k,
2275                                      struct nlink_table *links,
2276                                      size_t *idx, u64 range_end)
2277 {
2278         struct bch_fs *c = trans->c;
2279         struct bch_inode_unpacked u;
2280         struct nlink *link = &links->d[*idx];
2281         int ret = 0;
2282
2283         if (k.k->p.offset >= range_end)
2284                 return 1;
2285
2286         if (!bkey_is_inode(k.k))
2287                 return 0;
2288
2289         BUG_ON(bch2_inode_unpack(k, &u));
2290
2291         if (S_ISDIR(u.bi_mode))
2292                 return 0;
2293
2294         if (!u.bi_nlink)
2295                 return 0;
2296
2297         while ((cmp_int(link->inum, k.k->p.offset) ?:
2298                 cmp_int(link->snapshot, k.k->p.snapshot)) < 0) {
2299                 BUG_ON(*idx == links->nr);
2300                 link = &links->d[++*idx];
2301         }
2302
2303         if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count, c,
2304                         "inode %llu type %s has wrong i_nlink (%u, should be %u)",
2305                         u.bi_inum, bch2_d_types[mode_to_type(u.bi_mode)],
2306                         bch2_inode_nlink_get(&u), link->count)) {
2307                 bch2_inode_nlink_set(&u, link->count);
2308                 ret = __write_inode(trans, &u, k.k->p.snapshot);
2309         }
2310 fsck_err:
2311         return ret;
2312 }
2313
2314 noinline_for_stack
2315 static int check_nlinks_update_hardlinks(struct bch_fs *c,
2316                                struct nlink_table *links,
2317                                u64 range_start, u64 range_end)
2318 {
2319         struct btree_iter iter;
2320         struct bkey_s_c k;
2321         size_t idx = 0;
2322         int ret = 0;
2323
2324         ret = bch2_trans_run(c,
2325                 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
2326                                 POS(0, range_start),
2327                                 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
2328                                 NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
2329                         check_nlinks_update_inode(trans, &iter, k, links, &idx, range_end)));
2330         if (ret < 0) {
2331                 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
2332                 return ret;
2333         }
2334
2335         return 0;
2336 }
2337
2338 int bch2_check_nlinks(struct bch_fs *c)
2339 {
2340         struct nlink_table links = { 0 };
2341         u64 this_iter_range_start, next_iter_range_start = 0;
2342         int ret = 0;
2343
2344         do {
2345                 this_iter_range_start = next_iter_range_start;
2346                 next_iter_range_start = U64_MAX;
2347
2348                 ret = check_nlinks_find_hardlinks(c, &links,
2349                                                   this_iter_range_start,
2350                                                   &next_iter_range_start);
2351
2352                 ret = check_nlinks_walk_dirents(c, &links,
2353                                           this_iter_range_start,
2354                                           next_iter_range_start);
2355                 if (ret)
2356                         break;
2357
2358                 ret = check_nlinks_update_hardlinks(c, &links,
2359                                          this_iter_range_start,
2360                                          next_iter_range_start);
2361                 if (ret)
2362                         break;
2363
2364                 links.nr = 0;
2365         } while (next_iter_range_start != U64_MAX);
2366
2367         kvfree(links.d);
2368         bch_err_fn(c, ret);
2369         return ret;
2370 }
2371
2372 static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter,
2373                              struct bkey_s_c k)
2374 {
2375         struct bkey_s_c_reflink_p p;
2376         struct bkey_i_reflink_p *u;
2377         int ret;
2378
2379         if (k.k->type != KEY_TYPE_reflink_p)
2380                 return 0;
2381
2382         p = bkey_s_c_to_reflink_p(k);
2383
2384         if (!p.v->front_pad && !p.v->back_pad)
2385                 return 0;
2386
2387         u = bch2_trans_kmalloc(trans, sizeof(*u));
2388         ret = PTR_ERR_OR_ZERO(u);
2389         if (ret)
2390                 return ret;
2391
2392         bkey_reassemble(&u->k_i, k);
2393         u->v.front_pad  = 0;
2394         u->v.back_pad   = 0;
2395
2396         return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_NORUN);
2397 }
2398
2399 int bch2_fix_reflink_p(struct bch_fs *c)
2400 {
2401         struct btree_iter iter;
2402         struct bkey_s_c k;
2403         int ret;
2404
2405         if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
2406                 return 0;
2407
2408         ret = bch2_trans_run(c,
2409                 for_each_btree_key_commit(trans, iter,
2410                                 BTREE_ID_extents, POS_MIN,
2411                                 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|
2412                                 BTREE_ITER_ALL_SNAPSHOTS, k,
2413                                 NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
2414                         fix_reflink_p_key(trans, &iter, k)));
2415         bch_err_fn(c, ret);
2416         return ret;
2417 }