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 fonud", 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         if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
 130                 bch_err(trans->c, "error fetching inode %llu: %s",
 131                         inode_nr, bch2_err_str(ret));
 132         bch2_trans_iter_exit(trans, &iter);
 133         return ret;
 134 }
 135
 136 static int __lookup_inode(struct btree_trans *trans, u64 inode_nr,
 137                           struct bch_inode_unpacked *inode,
 138                           u32 *snapshot)
 139 {
 140         struct btree_iter iter;
 141         struct bkey_s_c k;
 142         int ret;
 143
 144         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
 145                                SPOS(0, inode_nr, *snapshot), 0);
 146         ret = bkey_err(k);
 147         if (ret)
 148                 goto err;
 149
 150         ret = bkey_is_inode(k.k)
 151                 ? bch2_inode_unpack(k, inode)
 152                 : -BCH_ERR_ENOENT_inode;
 153         if (!ret)
 154                 *snapshot = iter.pos.snapshot;
 155 err:
 156         if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
 157                 bch_err(trans->c, "error fetching inode %llu:%u: %s",
 158                         inode_nr, *snapshot, bch2_err_str(ret));
 159         bch2_trans_iter_exit(trans, &iter);
 160         return ret;
 161 }
 162
 163 static int lookup_inode(struct btree_trans *trans, u64 inode_nr,
 164                         struct bch_inode_unpacked *inode,
 165                         u32 *snapshot)
 166 {
 167         return lockrestart_do(trans, __lookup_inode(trans, inode_nr, inode, snapshot));
 168 }
 169
 170 static int __lookup_dirent(struct btree_trans *trans,
 171                            struct bch_hash_info hash_info,
 172                            subvol_inum dir, struct qstr *name,
 173                            u64 *target, unsigned *type)
 174 {
 175         struct btree_iter iter;
 176         struct bkey_s_c_dirent d;
 177         int ret;
 178
 179         ret = bch2_hash_lookup(trans, &iter, bch2_dirent_hash_desc,
 180                                &hash_info, dir, name, 0);
 181         if (ret)
 182                 return ret;
 183
 184         d = bkey_s_c_to_dirent(bch2_btree_iter_peek_slot(&iter));
 185         *target = le64_to_cpu(d.v->d_inum);
 186         *type = d.v->d_type;
 187         bch2_trans_iter_exit(trans, &iter);
 188         return 0;
 189 }
 190
 191 static int __write_inode(struct btree_trans *trans,
 192                          struct bch_inode_unpacked *inode,
 193                          u32 snapshot)
 194 {
 195         struct bkey_inode_buf *inode_p =
 196                 bch2_trans_kmalloc(trans, sizeof(*inode_p));
 197
 198         if (IS_ERR(inode_p))
 199                 return PTR_ERR(inode_p);
 200
 201         bch2_inode_pack(inode_p, inode);
 202         inode_p->inode.k.p.snapshot = snapshot;
 203
 204         return bch2_btree_insert_nonextent(trans, BTREE_ID_inodes,
 205                                 &inode_p->inode.k_i,
 206                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 207 }
 208
 209 static int write_inode(struct btree_trans *trans,
 210                        struct bch_inode_unpacked *inode,
 211                        u32 snapshot)
 212 {
 213         int ret = commit_do(trans, NULL, NULL,
 214                                   BTREE_INSERT_NOFAIL|
 215                                   BTREE_INSERT_LAZY_RW,
 216                                   __write_inode(trans, inode, snapshot));
 217         if (ret)
 218                 bch_err(trans->c, "error in fsck: error updating inode: %s",
 219                         bch2_err_str(ret));
 220         return ret;
 221 }
 222
 223 static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
 224 {
 225         struct bch_fs *c = trans->c;
 226         struct btree_iter iter;
 227         struct bch_inode_unpacked dir_inode;
 228         struct bch_hash_info dir_hash_info;
 229         int ret;
 230
 231         ret = lookup_first_inode(trans, pos.inode, &dir_inode);
 232         if (ret)
 233                 goto err;
 234
 235         dir_hash_info = bch2_hash_info_init(c, &dir_inode);
 236
 237         bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_INTENT);
 238
 239         ret = bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
 240                                   &dir_hash_info, &iter,
 241                                   BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 242         bch2_trans_iter_exit(trans, &iter);
 243 err:
 244         if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
 245                 bch_err_fn(c, ret);
 246         return ret;
 247 }
 248
 249 /* Get lost+found, create if it doesn't exist: */
 250 static int lookup_lostfound(struct btree_trans *trans, u32 subvol,
 251                             struct bch_inode_unpacked *lostfound)
 252 {
 253         struct bch_fs *c = trans->c;
 254         struct bch_inode_unpacked root;
 255         struct bch_hash_info root_hash_info;
 256         struct qstr lostfound_str = QSTR("lost+found");
 257         subvol_inum root_inum = { .subvol = subvol };
 258         u64 inum = 0;
 259         unsigned d_type = 0;
 260         u32 snapshot;
 261         int ret;
 262
 263         ret = __subvol_lookup(trans, subvol, &snapshot, &root_inum.inum);
 264         if (ret)
 265                 return ret;
 266
 267         ret = __lookup_inode(trans, root_inum.inum, &root, &snapshot);
 268         if (ret)
 269                 return ret;
 270
 271         root_hash_info = bch2_hash_info_init(c, &root);
 272
 273         ret = __lookup_dirent(trans, root_hash_info, root_inum,
 274                             &lostfound_str, &inum, &d_type);
 275         if (bch2_err_matches(ret, ENOENT)) {
 276                 bch_notice(c, "creating lost+found");
 277                 goto create_lostfound;
 278         }
 279
 280         if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
 281                 bch_err(c, "error looking up lost+found: %s", bch2_err_str(ret));
 282         if (ret)
 283                 return ret;
 284
 285         if (d_type != DT_DIR) {
 286                 bch_err(c, "error looking up lost+found: not a directory");
 287                 return ret;
 288         }
 289
 290         /*
 291          * The bch2_check_dirents pass has already run, dangling dirents
 292          * shouldn't exist here:
 293          */
 294         return __lookup_inode(trans, inum, lostfound, &snapshot);
 295
 296 create_lostfound:
 297         bch2_inode_init_early(c, lostfound);
 298
 299         ret = bch2_create_trans(trans, root_inum, &root,
 300                                 lostfound, &lostfound_str,
 301                                 0, 0, S_IFDIR|0700, 0, NULL, NULL,
 302                                 (subvol_inum) { }, 0);
 303         if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
 304                 bch_err(c, "error creating lost+found: %s", bch2_err_str(ret));
 305         return ret;
 306 }
 307
 308 static int __reattach_inode(struct btree_trans *trans,
 309                           struct bch_inode_unpacked *inode,
 310                           u32 inode_snapshot)
 311 {
 312         struct bch_hash_info dir_hash;
 313         struct bch_inode_unpacked lostfound;
 314         char name_buf[20];
 315         struct qstr name;
 316         u64 dir_offset = 0;
 317         u32 subvol;
 318         int ret;
 319
 320         ret = __snapshot_lookup_subvol(trans, inode_snapshot, &subvol);
 321         if (ret)
 322                 return ret;
 323
 324         ret = lookup_lostfound(trans, subvol, &lostfound);
 325         if (ret)
 326                 return ret;
 327
 328         if (S_ISDIR(inode->bi_mode)) {
 329                 lostfound.bi_nlink++;
 330
 331                 ret = __write_inode(trans, &lostfound, U32_MAX);
 332                 if (ret)
 333                         return ret;
 334         }
 335
 336         dir_hash = bch2_hash_info_init(trans->c, &lostfound);
 337
 338         snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
 339         name = (struct qstr) QSTR(name_buf);
 340
 341         ret = bch2_dirent_create(trans,
 342                                  (subvol_inum) {
 343                                         .subvol = subvol,
 344                                         .inum = lostfound.bi_inum,
 345                                  },
 346                                  &dir_hash,
 347                                  inode_d_type(inode),
 348                                  &name, inode->bi_inum, &dir_offset,
 349                                  BCH_HASH_SET_MUST_CREATE);
 350         if (ret)
 351                 return ret;
 352
 353         inode->bi_dir           = lostfound.bi_inum;
 354         inode->bi_dir_offset    = dir_offset;
 355
 356         return __write_inode(trans, inode, inode_snapshot);
 357 }
 358
 359 static int reattach_inode(struct btree_trans *trans,
 360                           struct bch_inode_unpacked *inode,
 361                           u32 inode_snapshot)
 362 {
 363         int ret = commit_do(trans, NULL, NULL,
 364                                   BTREE_INSERT_LAZY_RW|
 365                                   BTREE_INSERT_NOFAIL,
 366                         __reattach_inode(trans, inode, inode_snapshot));
 367         if (ret) {
 368                 bch_err(trans->c, "error reattaching inode %llu: %s",
 369                         inode->bi_inum, bch2_err_str(ret));
 370                 return ret;
 371         }
 372
 373         return ret;
 374 }
 375
 376 static int remove_backpointer(struct btree_trans *trans,
 377                               struct bch_inode_unpacked *inode)
 378 {
 379         struct btree_iter iter;
 380         struct bkey_s_c_dirent d;
 381         int ret;
 382
 383         d = bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_dirents,
 384                                      POS(inode->bi_dir, inode->bi_dir_offset), 0,
 385                                      dirent);
 386         ret =   bkey_err(d) ?:
 387                 __remove_dirent(trans, d.k->p);
 388         bch2_trans_iter_exit(trans, &iter);
 389         return ret;
 390 }
 391
 392 struct snapshots_seen_entry {
 393         u32                             id;
 394         u32                             equiv;
 395 };
 396
 397 struct snapshots_seen {
 398         struct bpos                     pos;
 399         DARRAY(struct snapshots_seen_entry) ids;
 400 };
 401
 402 static inline void snapshots_seen_exit(struct snapshots_seen *s)
 403 {
 404         darray_exit(&s->ids);
 405 }
 406
 407 static inline void snapshots_seen_init(struct snapshots_seen *s)
 408 {
 409         memset(s, 0, sizeof(*s));
 410 }
 411
 412 static int snapshots_seen_add_inorder(struct bch_fs *c, struct snapshots_seen *s, u32 id)
 413 {
 414         struct snapshots_seen_entry *i, n = {
 415                 .id     = id,
 416                 .equiv  = bch2_snapshot_equiv(c, id),
 417         };
 418         int ret = 0;
 419
 420         darray_for_each(s->ids, i) {
 421                 if (i->id == id)
 422                         return 0;
 423                 if (i->id > id)
 424                         break;
 425         }
 426
 427         ret = darray_insert_item(&s->ids, i - s->ids.data, n);
 428         if (ret)
 429                 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
 430                         s->ids.size);
 431         return ret;
 432 }
 433
 434 static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s,
 435                                  enum btree_id btree_id, struct bpos pos)
 436 {
 437         struct snapshots_seen_entry *i, n = {
 438                 .id     = pos.snapshot,
 439                 .equiv  = bch2_snapshot_equiv(c, pos.snapshot),
 440         };
 441         int ret = 0;
 442
 443         if (!bkey_eq(s->pos, pos))
 444                 s->ids.nr = 0;
 445
 446         s->pos = pos;
 447         s->pos.snapshot = n.equiv;
 448
 449         darray_for_each(s->ids, i) {
 450                 if (i->id == n.id)
 451                         return 0;
 452
 453                 /*
 454                  * We currently don't rigorously track for snapshot cleanup
 455                  * needing to be run, so it shouldn't be a fsck error yet:
 456                  */
 457                 if (i->equiv == n.equiv) {
 458                         bch_err(c, "snapshot deletion did not finish:\n"
 459                                 "  duplicate keys in btree %s at %llu:%llu snapshots %u, %u (equiv %u)\n",
 460                                 bch2_btree_ids[btree_id],
 461                                 pos.inode, pos.offset,
 462                                 i->id, n.id, n.equiv);
 463                         return bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_delete_dead_snapshots);
 464                 }
 465         }
 466
 467         ret = darray_push(&s->ids, n);
 468         if (ret)
 469                 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
 470                         s->ids.size);
 471         return ret;
 472 }
 473
 474 /**
 475  * key_visible_in_snapshot - returns true if @id is a descendent of @ancestor,
 476  * and @ancestor hasn't been overwritten in @seen
 477  *
 478  * That is, returns whether key in @ancestor snapshot is visible in @id snapshot
 479  */
 480 static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *seen,
 481                                     u32 id, u32 ancestor)
 482 {
 483         ssize_t i;
 484
 485         EBUG_ON(id > ancestor);
 486         EBUG_ON(!bch2_snapshot_is_equiv(c, id));
 487         EBUG_ON(!bch2_snapshot_is_equiv(c, ancestor));
 488
 489         /* @ancestor should be the snapshot most recently added to @seen */
 490         EBUG_ON(ancestor != seen->pos.snapshot);
 491         EBUG_ON(ancestor != seen->ids.data[seen->ids.nr - 1].equiv);
 492
 493         if (id == ancestor)
 494                 return true;
 495
 496         if (!bch2_snapshot_is_ancestor(c, id, ancestor))
 497                 return false;
 498
 499         /*
 500          * We know that @id is a descendant of @ancestor, we're checking if
 501          * we've seen a key that overwrote @ancestor - i.e. also a descendent of
 502          * @ascestor and with @id as a descendent.
 503          *
 504          * But we already know that we're scanning IDs between @id and @ancestor
 505          * numerically, since snapshot ID lists are kept sorted, so if we find
 506          * an id that's an ancestor of @id we're done:
 507          */
 508
 509         for (i = seen->ids.nr - 2;
 510              i >= 0 && seen->ids.data[i].equiv >= id;
 511              --i)
 512                 if (bch2_snapshot_is_ancestor(c, id, seen->ids.data[i].equiv))
 513                         return false;
 514
 515         return true;
 516 }
 517
 518 /**
 519  * ref_visible - given a key with snapshot id @src that points to a key with
 520  * snapshot id @dst, test whether there is some snapshot in which @dst is
 521  * visible.
 522  *
 523  * This assumes we're visiting @src keys in natural key order.
 524  *
 525  * @s   - list of snapshot IDs already seen at @src
 526  * @src - snapshot ID of src key
 527  * @dst - snapshot ID of dst key
 528  */
 529 static int ref_visible(struct bch_fs *c, struct snapshots_seen *s,
 530                        u32 src, u32 dst)
 531 {
 532         return dst <= src
 533                 ? key_visible_in_snapshot(c, s, dst, src)
 534                 : bch2_snapshot_is_ancestor(c, src, dst);
 535 }
 536
 537 static int ref_visible2(struct bch_fs *c,
 538                         u32 src, struct snapshots_seen *src_seen,
 539                         u32 dst, struct snapshots_seen *dst_seen)
 540 {
 541         src = bch2_snapshot_equiv(c, src);
 542         dst = bch2_snapshot_equiv(c, dst);
 543
 544         if (dst > src) {
 545                 swap(dst, src);
 546                 swap(dst_seen, src_seen);
 547         }
 548         return key_visible_in_snapshot(c, src_seen, dst, src);
 549 }
 550
 551 #define for_each_visible_inode(_c, _s, _w, _snapshot, _i)                               \
 552         for (_i = (_w)->inodes.data; _i < (_w)->inodes.data + (_w)->inodes.nr &&        \
 553              (_i)->snapshot <= (_snapshot); _i++)                                       \
 554                 if (key_visible_in_snapshot(_c, _s, _i->snapshot, _snapshot))
 555
 556 struct inode_walker_entry {
 557         struct bch_inode_unpacked inode;
 558         u32                     snapshot;
 559         bool                    seen_this_pos;
 560         u64                     count;
 561 };
 562
 563 struct inode_walker {
 564         bool                            first_this_inode;
 565         bool                            recalculate_sums;
 566         struct bpos                     last_pos;
 567
 568         DARRAY(struct inode_walker_entry) inodes;
 569 };
 570
 571 static void inode_walker_exit(struct inode_walker *w)
 572 {
 573         darray_exit(&w->inodes);
 574 }
 575
 576 static struct inode_walker inode_walker_init(void)
 577 {
 578         return (struct inode_walker) { 0, };
 579 }
 580
 581 static int add_inode(struct bch_fs *c, struct inode_walker *w,
 582                      struct bkey_s_c inode)
 583 {
 584         struct bch_inode_unpacked u;
 585
 586         BUG_ON(bch2_inode_unpack(inode, &u));
 587
 588         return darray_push(&w->inodes, ((struct inode_walker_entry) {
 589                 .inode          = u,
 590                 .snapshot       = bch2_snapshot_equiv(c, inode.k->p.snapshot),
 591         }));
 592 }
 593
 594 static int get_inodes_all_snapshots(struct btree_trans *trans,
 595                                     struct inode_walker *w, u64 inum)
 596 {
 597         struct bch_fs *c = trans->c;
 598         struct btree_iter iter;
 599         struct bkey_s_c k;
 600         u32 restart_count = trans->restart_count;
 601         int ret;
 602
 603         w->recalculate_sums = false;
 604         w->inodes.nr = 0;
 605
 606         for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, inum),
 607                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
 608                 if (k.k->p.offset != inum)
 609                         break;
 610
 611                 if (bkey_is_inode(k.k))
 612                         add_inode(c, w, k);
 613         }
 614         bch2_trans_iter_exit(trans, &iter);
 615
 616         if (ret)
 617                 return ret;
 618
 619         w->first_this_inode = true;
 620
 621         if (trans_was_restarted(trans, restart_count))
 622                 return -BCH_ERR_transaction_restart_nested;
 623
 624         return 0;
 625 }
 626
 627 static struct inode_walker_entry *
 628 lookup_inode_for_snapshot(struct bch_fs *c, struct inode_walker *w,
 629                           u32 snapshot, bool is_whiteout)
 630 {
 631         struct inode_walker_entry *i;
 632
 633         snapshot = bch2_snapshot_equiv(c, snapshot);
 634
 635         darray_for_each(w->inodes, i)
 636                 if (bch2_snapshot_is_ancestor(c, snapshot, i->snapshot))
 637                         goto found;
 638
 639         return NULL;
 640 found:
 641         BUG_ON(snapshot > i->snapshot);
 642
 643         if (snapshot != i->snapshot && !is_whiteout) {
 644                 struct inode_walker_entry new = *i;
 645                 size_t pos;
 646                 int ret;
 647
 648                 new.snapshot = snapshot;
 649                 new.count = 0;
 650
 651                 bch_info(c, "have key for inode %llu:%u but have inode in ancestor snapshot %u",
 652                          w->last_pos.inode, snapshot, i->snapshot);
 653
 654                 while (i > w->inodes.data && i[-1].snapshot > snapshot)
 655                         --i;
 656
 657                 pos = i - w->inodes.data;
 658                 ret = darray_insert_item(&w->inodes, pos, new);
 659                 if (ret)
 660                         return ERR_PTR(ret);
 661
 662                 i = w->inodes.data + pos;
 663         }
 664
 665         return i;
 666 }
 667
 668 static struct inode_walker_entry *walk_inode(struct btree_trans *trans,
 669                                              struct inode_walker *w, struct bpos pos,
 670                                              bool is_whiteout)
 671 {
 672         if (w->last_pos.inode != pos.inode) {
 673                 int ret = get_inodes_all_snapshots(trans, w, pos.inode);
 674                 if (ret)
 675                         return ERR_PTR(ret);
 676         } else if (bkey_cmp(w->last_pos, pos)) {
 677                 struct inode_walker_entry *i;
 678
 679                 darray_for_each(w->inodes, i)
 680                         i->seen_this_pos = false;
 681
 682         }
 683
 684         w->last_pos = pos;
 685
 686         return lookup_inode_for_snapshot(trans->c, w, pos.snapshot, is_whiteout);
 687 }
 688
 689 static int __get_visible_inodes(struct btree_trans *trans,
 690                                 struct inode_walker *w,
 691                                 struct snapshots_seen *s,
 692                                 u64 inum)
 693 {
 694         struct bch_fs *c = trans->c;
 695         struct btree_iter iter;
 696         struct bkey_s_c k;
 697         int ret;
 698
 699         w->inodes.nr = 0;
 700
 701         for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum),
 702                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
 703                 u32 equiv = bch2_snapshot_equiv(c, k.k->p.snapshot);
 704
 705                 if (k.k->p.offset != inum)
 706                         break;
 707
 708                 if (!ref_visible(c, s, s->pos.snapshot, equiv))
 709                         continue;
 710
 711                 if (bkey_is_inode(k.k))
 712                         add_inode(c, w, k);
 713
 714                 if (equiv >= s->pos.snapshot)
 715                         break;
 716         }
 717         bch2_trans_iter_exit(trans, &iter);
 718
 719         return ret;
 720 }
 721
 722 static int check_key_has_snapshot(struct btree_trans *trans,
 723                                   struct btree_iter *iter,
 724                                   struct bkey_s_c k)
 725 {
 726         struct bch_fs *c = trans->c;
 727         struct printbuf buf = PRINTBUF;
 728         int ret = 0;
 729
 730         if (mustfix_fsck_err_on(!bch2_snapshot_equiv(c, k.k->p.snapshot), c,
 731                         "key in missing snapshot: %s",
 732                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
 733                 ret = bch2_btree_delete_at(trans, iter,
 734                                             BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?: 1;
 735 fsck_err:
 736         printbuf_exit(&buf);
 737         return ret;
 738 }
 739
 740 static int hash_redo_key(struct btree_trans *trans,
 741                          const struct bch_hash_desc desc,
 742                          struct bch_hash_info *hash_info,
 743                          struct btree_iter *k_iter, struct bkey_s_c k)
 744 {
 745         struct bkey_i *delete;
 746         struct bkey_i *tmp;
 747
 748         delete = bch2_trans_kmalloc(trans, sizeof(*delete));
 749         if (IS_ERR(delete))
 750                 return PTR_ERR(delete);
 751
 752         tmp = bch2_bkey_make_mut_noupdate(trans, k);
 753         if (IS_ERR(tmp))
 754                 return PTR_ERR(tmp);
 755
 756         bkey_init(&delete->k);
 757         delete->k.p = k_iter->pos;
 758         return  bch2_btree_iter_traverse(k_iter) ?:
 759                 bch2_trans_update(trans, k_iter, delete, 0) ?:
 760                 bch2_hash_set_snapshot(trans, desc, hash_info,
 761                                        (subvol_inum) { 0, k.k->p.inode },
 762                                        k.k->p.snapshot, tmp,
 763                                        BCH_HASH_SET_MUST_CREATE,
 764                                        BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
 765                 bch2_trans_commit(trans, NULL, NULL,
 766                                   BTREE_INSERT_NOFAIL|
 767                                   BTREE_INSERT_LAZY_RW);
 768 }
 769
 770 static int hash_check_key(struct btree_trans *trans,
 771                           const struct bch_hash_desc desc,
 772                           struct bch_hash_info *hash_info,
 773                           struct btree_iter *k_iter, struct bkey_s_c hash_k)
 774 {
 775         struct bch_fs *c = trans->c;
 776         struct btree_iter iter = { NULL };
 777         struct printbuf buf = PRINTBUF;
 778         struct bkey_s_c k;
 779         u64 hash;
 780         int ret = 0;
 781
 782         if (hash_k.k->type != desc.key_type)
 783                 return 0;
 784
 785         hash = desc.hash_bkey(hash_info, hash_k);
 786
 787         if (likely(hash == hash_k.k->p.offset))
 788                 return 0;
 789
 790         if (hash_k.k->p.offset < hash)
 791                 goto bad_hash;
 792
 793         for_each_btree_key_norestart(trans, iter, desc.btree_id,
 794                                      SPOS(hash_k.k->p.inode, hash, hash_k.k->p.snapshot),
 795                                      BTREE_ITER_SLOTS, k, ret) {
 796                 if (bkey_eq(k.k->p, hash_k.k->p))
 797                         break;
 798
 799                 if (fsck_err_on(k.k->type == desc.key_type &&
 800                                 !desc.cmp_bkey(k, hash_k), c,
 801                                 "duplicate hash table keys:\n%s",
 802                                 (printbuf_reset(&buf),
 803                                  bch2_bkey_val_to_text(&buf, c, hash_k),
 804                                  buf.buf))) {
 805                         ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0) ?: 1;
 806                         break;
 807                 }
 808
 809                 if (bkey_deleted(k.k)) {
 810                         bch2_trans_iter_exit(trans, &iter);
 811                         goto bad_hash;
 812                 }
 813         }
 814 out:
 815         bch2_trans_iter_exit(trans, &iter);
 816         printbuf_exit(&buf);
 817         return ret;
 818 bad_hash:
 819         if (fsck_err(c, "hash table key at wrong offset: btree %s inode %llu offset %llu, hashed to %llu\n%s",
 820                      bch2_btree_ids[desc.btree_id], hash_k.k->p.inode, hash_k.k->p.offset, hash,
 821                      (printbuf_reset(&buf),
 822                       bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf))) {
 823                 ret = hash_redo_key(trans, desc, hash_info, k_iter, hash_k);
 824                 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
 825                         bch_err(c, "hash_redo_key err %s", bch2_err_str(ret));
 826                 if (ret)
 827                         return ret;
 828                 ret = -BCH_ERR_transaction_restart_nested;
 829         }
 830 fsck_err:
 831         goto out;
 832 }
 833
 834 static int check_inode(struct btree_trans *trans,
 835                        struct btree_iter *iter,
 836                        struct bkey_s_c k,
 837                        struct bch_inode_unpacked *prev,
 838                        struct snapshots_seen *s,
 839                        bool full)
 840 {
 841         struct bch_fs *c = trans->c;
 842         struct bch_inode_unpacked u;
 843         bool do_update = false;
 844         int ret;
 845
 846         ret = check_key_has_snapshot(trans, iter, k);
 847         if (ret < 0)
 848                 goto err;
 849         if (ret)
 850                 return 0;
 851
 852         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
 853         if (ret)
 854                 goto err;
 855
 856         if (!bkey_is_inode(k.k))
 857                 return 0;
 858
 859         BUG_ON(bch2_inode_unpack(k, &u));
 860
 861         if (!full &&
 862             !(u.bi_flags & (BCH_INODE_I_SIZE_DIRTY|
 863                             BCH_INODE_I_SECTORS_DIRTY|
 864                             BCH_INODE_UNLINKED)))
 865                 return 0;
 866
 867         if (prev->bi_inum != u.bi_inum)
 868                 *prev = u;
 869
 870         if (fsck_err_on(prev->bi_hash_seed      != u.bi_hash_seed ||
 871                         inode_d_type(prev)      != inode_d_type(&u), c,
 872                         "inodes in different snapshots don't match")) {
 873                 bch_err(c, "repair not implemented yet");
 874                 return -EINVAL;
 875         }
 876
 877         if ((u.bi_flags & (BCH_INODE_I_SIZE_DIRTY|BCH_INODE_UNLINKED)) &&
 878             bch2_key_has_snapshot_overwrites(trans, BTREE_ID_inodes, k.k->p)) {
 879                 struct bpos new_min_pos;
 880
 881                 ret = bch2_propagate_key_to_snapshot_leaves(trans, iter->btree_id, k, &new_min_pos);
 882                 if (ret)
 883                         goto err;
 884
 885                 u.bi_flags &= ~BCH_INODE_I_SIZE_DIRTY|BCH_INODE_UNLINKED;
 886
 887                 ret = __write_inode(trans, &u, iter->pos.snapshot);
 888                 if (ret) {
 889                         bch_err_msg(c, ret, "in fsck: error updating inode");
 890                         return ret;
 891                 }
 892
 893                 if (!bpos_eq(new_min_pos, POS_MIN))
 894                         bch2_btree_iter_set_pos(iter, bpos_predecessor(new_min_pos));
 895                 return 0;
 896         }
 897
 898         if (u.bi_flags & BCH_INODE_UNLINKED &&
 899             (!c->sb.clean ||
 900              fsck_err(c, "filesystem marked clean, but inode %llu unlinked",
 901                       u.bi_inum))) {
 902                 bch2_trans_unlock(trans);
 903                 bch2_fs_lazy_rw(c);
 904
 905                 ret = bch2_inode_rm_snapshot(trans, u.bi_inum, iter->pos.snapshot);
 906                 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
 907                         bch_err(c, "error in fsck: error while deleting inode: %s",
 908                                 bch2_err_str(ret));
 909                 return ret;
 910         }
 911
 912         if (u.bi_flags & BCH_INODE_I_SIZE_DIRTY &&
 913             (!c->sb.clean ||
 914              fsck_err(c, "filesystem marked clean, but inode %llu has i_size dirty",
 915                       u.bi_inum))) {
 916                 bch_verbose(c, "truncating inode %llu", u.bi_inum);
 917
 918                 bch2_trans_unlock(trans);
 919                 bch2_fs_lazy_rw(c);
 920
 921                 /*
 922                  * XXX: need to truncate partial blocks too here - or ideally
 923                  * just switch units to bytes and that issue goes away
 924                  */
 925                 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
 926                                 SPOS(u.bi_inum, round_up(u.bi_size, block_bytes(c)) >> 9,
 927                                      iter->pos.snapshot),
 928                                 POS(u.bi_inum, U64_MAX),
 929                                 0, NULL);
 930                 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
 931                         bch_err(c, "error in fsck: error truncating inode: %s",
 932                                 bch2_err_str(ret));
 933                 if (ret)
 934                         return ret;
 935
 936                 /*
 937                  * We truncated without our normal sector accounting hook, just
 938                  * make sure we recalculate it:
 939                  */
 940                 u.bi_flags |= BCH_INODE_I_SECTORS_DIRTY;
 941
 942                 u.bi_flags &= ~BCH_INODE_I_SIZE_DIRTY;
 943                 do_update = true;
 944         }
 945
 946         if (u.bi_flags & BCH_INODE_I_SECTORS_DIRTY &&
 947             (!c->sb.clean ||
 948              fsck_err(c, "filesystem marked clean, but inode %llu has i_sectors dirty",
 949                       u.bi_inum))) {
 950                 s64 sectors;
 951
 952                 bch_verbose(c, "recounting sectors for inode %llu",
 953                             u.bi_inum);
 954
 955                 sectors = bch2_count_inode_sectors(trans, u.bi_inum, iter->pos.snapshot);
 956                 if (sectors < 0) {
 957                         bch_err(c, "error in fsck: error recounting inode sectors: %s",
 958                                 bch2_err_str(sectors));
 959                         return sectors;
 960                 }
 961
 962                 u.bi_sectors = sectors;
 963                 u.bi_flags &= ~BCH_INODE_I_SECTORS_DIRTY;
 964                 do_update = true;
 965         }
 966
 967         if (u.bi_flags & BCH_INODE_BACKPTR_UNTRUSTED) {
 968                 u.bi_dir = 0;
 969                 u.bi_dir_offset = 0;
 970                 u.bi_flags &= ~BCH_INODE_BACKPTR_UNTRUSTED;
 971                 do_update = true;
 972         }
 973
 974         if (do_update) {
 975                 ret = __write_inode(trans, &u, iter->pos.snapshot);
 976                 if (ret) {
 977                         bch_err_msg(c, ret, "in fsck: error updating inode");
 978                         return ret;
 979                 }
 980         }
 981 err:
 982 fsck_err:
 983         if (ret)
 984                 bch_err_fn(c, ret);
 985         return ret;
 986 }
 987
 988 noinline_for_stack
 989 int bch2_check_inodes(struct bch_fs *c)
 990 {
 991         bool full = c->opts.fsck;
 992         struct btree_trans trans;
 993         struct btree_iter iter;
 994         struct bch_inode_unpacked prev = { 0 };
 995         struct snapshots_seen s;
 996         struct bkey_s_c k;
 997         int ret;
 998
 999         snapshots_seen_init(&s);
1000         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1001
1002         ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_inodes,
1003                         POS_MIN,
1004                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1005                         NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
1006                 check_inode(&trans, &iter, k, &prev, &s, full));
1007
1008         bch2_trans_exit(&trans);
1009         snapshots_seen_exit(&s);
1010         if (ret)
1011                 bch_err_fn(c, ret);
1012         return ret;
1013 }
1014
1015 static struct bkey_s_c_dirent dirent_get_by_pos(struct btree_trans *trans,
1016                                                 struct btree_iter *iter,
1017                                                 struct bpos pos)
1018 {
1019         return bch2_bkey_get_iter_typed(trans, iter, BTREE_ID_dirents, pos, 0, dirent);
1020 }
1021
1022 static bool inode_points_to_dirent(struct bch_inode_unpacked *inode,
1023                                    struct bkey_s_c_dirent d)
1024 {
1025         return  inode->bi_dir           == d.k->p.inode &&
1026                 inode->bi_dir_offset    == d.k->p.offset;
1027 }
1028
1029 static bool dirent_points_to_inode(struct bkey_s_c_dirent d,
1030                                    struct bch_inode_unpacked *inode)
1031 {
1032         return d.v->d_type == DT_SUBVOL
1033                 ? le32_to_cpu(d.v->d_child_subvol)      == inode->bi_subvol
1034                 : le64_to_cpu(d.v->d_inum)              == inode->bi_inum;
1035 }
1036
1037 static int inode_backpointer_exists(struct btree_trans *trans,
1038                                     struct bch_inode_unpacked *inode,
1039                                     u32 snapshot)
1040 {
1041         struct btree_iter iter;
1042         struct bkey_s_c_dirent d;
1043         int ret;
1044
1045         d = dirent_get_by_pos(trans, &iter,
1046                         SPOS(inode->bi_dir, inode->bi_dir_offset, snapshot));
1047         ret = bkey_err(d);
1048         if (ret)
1049                 return bch2_err_matches(ret, ENOENT) ? 0 : ret;
1050
1051         ret = dirent_points_to_inode(d, inode);
1052         bch2_trans_iter_exit(trans, &iter);
1053         return ret;
1054 }
1055
1056 static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
1057 {
1058         struct bch_fs *c = trans->c;
1059         struct inode_walker_entry *i;
1060         u32 restart_count = trans->restart_count;
1061         int ret = 0;
1062         s64 count2;
1063
1064         darray_for_each(w->inodes, i) {
1065                 if (i->inode.bi_sectors == i->count)
1066                         continue;
1067
1068                 count2 = bch2_count_inode_sectors(trans, w->last_pos.inode, i->snapshot);
1069
1070                 if (w->recalculate_sums)
1071                         i->count = count2;
1072
1073                 if (i->count != count2) {
1074                         bch_err(c, "fsck counted i_sectors wrong for inode %llu:%u: got %llu should be %llu",
1075                                 w->last_pos.inode, i->snapshot, i->count, count2);
1076                         return -BCH_ERR_internal_fsck_err;
1077                 }
1078
1079                 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SECTORS_DIRTY), c,
1080                             "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
1081                             w->last_pos.inode, i->snapshot,
1082                             i->inode.bi_sectors, i->count)) {
1083                         i->inode.bi_sectors = i->count;
1084                         ret = write_inode(trans, &i->inode, i->snapshot);
1085                         if (ret)
1086                                 break;
1087                 }
1088         }
1089 fsck_err:
1090         if (ret)
1091                 bch_err_fn(c, ret);
1092         if (!ret && trans_was_restarted(trans, restart_count))
1093                 ret = -BCH_ERR_transaction_restart_nested;
1094         return ret;
1095 }
1096
1097 struct extent_end {
1098         u32                     snapshot;
1099         u64                     offset;
1100         struct snapshots_seen   seen;
1101 };
1102
1103 struct extent_ends {
1104         struct bpos                     last_pos;
1105         DARRAY(struct extent_end)       e;
1106 };
1107
1108 static void extent_ends_reset(struct extent_ends *extent_ends)
1109 {
1110         struct extent_end *i;
1111
1112         darray_for_each(extent_ends->e, i)
1113                 snapshots_seen_exit(&i->seen);
1114
1115         extent_ends->e.nr = 0;
1116 }
1117
1118 static void extent_ends_exit(struct extent_ends *extent_ends)
1119 {
1120         extent_ends_reset(extent_ends);
1121         darray_exit(&extent_ends->e);
1122 }
1123
1124 static void extent_ends_init(struct extent_ends *extent_ends)
1125 {
1126         memset(extent_ends, 0, sizeof(*extent_ends));
1127 }
1128
1129 static int extent_ends_at(struct bch_fs *c,
1130                           struct extent_ends *extent_ends,
1131                           struct snapshots_seen *seen,
1132                           struct bkey_s_c k)
1133 {
1134         struct extent_end *i, n = (struct extent_end) {
1135                 .offset         = k.k->p.offset,
1136                 .snapshot       = k.k->p.snapshot,
1137                 .seen           = *seen,
1138         };
1139
1140         n.seen.ids.data = kmemdup(seen->ids.data,
1141                               sizeof(seen->ids.data[0]) * seen->ids.size,
1142                               GFP_KERNEL);
1143         if (!n.seen.ids.data)
1144                 return -BCH_ERR_ENOMEM_fsck_extent_ends_at;
1145
1146         darray_for_each(extent_ends->e, i) {
1147                 if (i->snapshot == k.k->p.snapshot) {
1148                         snapshots_seen_exit(&i->seen);
1149                         *i = n;
1150                         return 0;
1151                 }
1152
1153                 if (i->snapshot >= k.k->p.snapshot)
1154                         break;
1155         }
1156
1157         return darray_insert_item(&extent_ends->e, i - extent_ends->e.data, n);
1158 }
1159
1160 static int overlapping_extents_found(struct btree_trans *trans,
1161                                      enum btree_id btree,
1162                                      struct bpos pos1, struct snapshots_seen *pos1_seen,
1163                                      struct bkey pos2,
1164                                      bool *fixed,
1165                                      struct extent_end *extent_end)
1166 {
1167         struct bch_fs *c = trans->c;
1168         struct printbuf buf = PRINTBUF;
1169         struct btree_iter iter1, iter2 = { NULL };
1170         struct bkey_s_c k1, k2;
1171         int ret;
1172
1173         BUG_ON(bkey_le(pos1, bkey_start_pos(&pos2)));
1174
1175         bch2_trans_iter_init(trans, &iter1, btree, pos1,
1176                              BTREE_ITER_ALL_SNAPSHOTS|
1177                              BTREE_ITER_NOT_EXTENTS);
1178         k1 = bch2_btree_iter_peek_upto(&iter1, POS(pos1.inode, U64_MAX));
1179         ret = bkey_err(k1);
1180         if (ret)
1181                 goto err;
1182
1183         prt_str(&buf, "\n  ");
1184         bch2_bkey_val_to_text(&buf, c, k1);
1185
1186         if (!bpos_eq(pos1, k1.k->p)) {
1187                 prt_str(&buf, "\n  wanted\n  ");
1188                 bch2_bpos_to_text(&buf, pos1);
1189                 prt_str(&buf, "\n  ");
1190                 bch2_bkey_to_text(&buf, &pos2);
1191
1192                 bch_err(c, "%s: error finding first overlapping extent when repairing, got%s",
1193                         __func__, buf.buf);
1194                 ret = -BCH_ERR_internal_fsck_err;
1195                 goto err;
1196         }
1197
1198         bch2_trans_copy_iter(&iter2, &iter1);
1199
1200         while (1) {
1201                 bch2_btree_iter_advance(&iter2);
1202
1203                 k2 = bch2_btree_iter_peek_upto(&iter2, POS(pos1.inode, U64_MAX));
1204                 ret = bkey_err(k2);
1205                 if (ret)
1206                         goto err;
1207
1208                 if (bpos_ge(k2.k->p, pos2.p))
1209                         break;
1210         }
1211
1212         prt_str(&buf, "\n  ");
1213         bch2_bkey_val_to_text(&buf, c, k2);
1214
1215         if (bpos_gt(k2.k->p, pos2.p) ||
1216             pos2.size != k2.k->size) {
1217                 bch_err(c, "%s: error finding seconding overlapping extent when repairing%s",
1218                         __func__, buf.buf);
1219                 ret = -BCH_ERR_internal_fsck_err;
1220                 goto err;
1221         }
1222
1223         prt_printf(&buf, "\n  overwriting %s extent",
1224                    pos1.snapshot >= pos2.p.snapshot ? "first" : "second");
1225
1226         if (fsck_err(c, "overlapping extents%s", buf.buf)) {
1227                 struct btree_iter *old_iter = &iter1;
1228                 struct disk_reservation res = { 0 };
1229
1230                 if (pos1.snapshot < pos2.p.snapshot) {
1231                         old_iter = &iter2;
1232                         swap(k1, k2);
1233                 }
1234
1235                 trans->extra_journal_res += bch2_bkey_sectors_compressed(k2);
1236
1237                 ret =   bch2_trans_update_extent_overwrite(trans, old_iter,
1238                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE,
1239                                 k1, k2) ?:
1240                         bch2_trans_commit(trans, &res, NULL,
1241                                 BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL);
1242                 bch2_disk_reservation_put(c, &res);
1243
1244                 if (ret)
1245                         goto err;
1246
1247                 *fixed = true;
1248
1249                 if (pos1.snapshot == pos2.p.snapshot) {
1250                         /*
1251                          * We overwrote the first extent, and did the overwrite
1252                          * in the same snapshot:
1253                          */
1254                         extent_end->offset = bkey_start_offset(&pos2);
1255                 } else if (pos1.snapshot > pos2.p.snapshot) {
1256                         /*
1257                          * We overwrote the first extent in pos2's snapshot:
1258                          */
1259                         ret = snapshots_seen_add_inorder(c, pos1_seen, pos2.p.snapshot);
1260                 } else {
1261                         /*
1262                          * We overwrote the second extent - restart
1263                          * check_extent() from the top:
1264                          */
1265                         ret = -BCH_ERR_transaction_restart_nested;
1266                 }
1267         }
1268 fsck_err:
1269 err:
1270         bch2_trans_iter_exit(trans, &iter2);
1271         bch2_trans_iter_exit(trans, &iter1);
1272         printbuf_exit(&buf);
1273         return ret;
1274 }
1275
1276 static int check_overlapping_extents(struct btree_trans *trans,
1277                               struct snapshots_seen *seen,
1278                               struct extent_ends *extent_ends,
1279                               struct bkey_s_c k,
1280                               u32 equiv,
1281                               struct btree_iter *iter,
1282                               bool *fixed)
1283 {
1284         struct bch_fs *c = trans->c;
1285         struct extent_end *i;
1286         int ret = 0;
1287
1288         /* transaction restart, running again */
1289         if (bpos_eq(extent_ends->last_pos, k.k->p))
1290                 return 0;
1291
1292         if (extent_ends->last_pos.inode != k.k->p.inode)
1293                 extent_ends_reset(extent_ends);
1294
1295         darray_for_each(extent_ends->e, i) {
1296                 if (i->offset <= bkey_start_offset(k.k))
1297                         continue;
1298
1299                 if (!ref_visible2(c,
1300                                   k.k->p.snapshot, seen,
1301                                   i->snapshot, &i->seen))
1302                         continue;
1303
1304                 ret = overlapping_extents_found(trans, iter->btree_id,
1305                                                 SPOS(iter->pos.inode,
1306                                                      i->offset,
1307                                                      i->snapshot),
1308                                                 &i->seen,
1309                                                 *k.k, fixed, i);
1310                 if (ret)
1311                         goto err;
1312         }
1313
1314         ret = extent_ends_at(c, extent_ends, seen, k);
1315         if (ret)
1316                 goto err;
1317
1318         extent_ends->last_pos = k.k->p;
1319 err:
1320         return ret;
1321 }
1322
1323 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
1324                         struct bkey_s_c k,
1325                         struct inode_walker *inode,
1326                         struct snapshots_seen *s,
1327                         struct extent_ends *extent_ends)
1328 {
1329         struct bch_fs *c = trans->c;
1330         struct inode_walker_entry *i;
1331         struct printbuf buf = PRINTBUF;
1332         struct bpos equiv = k.k->p;
1333         int ret = 0;
1334
1335         equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1336
1337         ret = check_key_has_snapshot(trans, iter, k);
1338         if (ret) {
1339                 ret = ret < 0 ? ret : 0;
1340                 goto out;
1341         }
1342
1343         if (inode->last_pos.inode != k.k->p.inode) {
1344                 ret = check_i_sectors(trans, inode);
1345                 if (ret)
1346                         goto err;
1347         }
1348
1349         i = walk_inode(trans, inode, equiv, k.k->type == KEY_TYPE_whiteout);
1350         ret = PTR_ERR_OR_ZERO(i);
1351         if (ret)
1352                 goto err;
1353
1354         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1355         if (ret)
1356                 goto err;
1357
1358         if (k.k->type != KEY_TYPE_whiteout) {
1359                 if (fsck_err_on(!i, c,
1360                                 "extent in missing inode:\n  %s",
1361                                 (printbuf_reset(&buf),
1362                                  bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1363                         goto delete;
1364
1365                 if (fsck_err_on(i &&
1366                                 !S_ISREG(i->inode.bi_mode) &&
1367                                 !S_ISLNK(i->inode.bi_mode), c,
1368                                 "extent in non regular inode mode %o:\n  %s",
1369                                 i->inode.bi_mode,
1370                                 (printbuf_reset(&buf),
1371                                  bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1372                         goto delete;
1373
1374                 ret = check_overlapping_extents(trans, s, extent_ends, k,
1375                                                 equiv.snapshot, iter,
1376                                                 &inode->recalculate_sums);
1377                 if (ret)
1378                         goto err;
1379         }
1380
1381         /*
1382          * Check inodes in reverse order, from oldest snapshots to newest,
1383          * starting from the inode that matches this extent's snapshot. If we
1384          * didn't have one, iterate over all inodes:
1385          */
1386         if (!i)
1387                 i = inode->inodes.data + inode->inodes.nr - 1;
1388
1389         for (;
1390              inode->inodes.data && i >= inode->inodes.data;
1391              --i) {
1392                 if (i->snapshot > equiv.snapshot ||
1393                     !key_visible_in_snapshot(c, s, i->snapshot, equiv.snapshot))
1394                         continue;
1395
1396                 if (k.k->type != KEY_TYPE_whiteout) {
1397                         if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SIZE_DIRTY) &&
1398                                         k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9 &&
1399                                         !bkey_extent_is_reservation(k), c,
1400                                         "extent type past end of inode %llu:%u, i_size %llu\n  %s",
1401                                         i->inode.bi_inum, i->snapshot, i->inode.bi_size,
1402                                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1403                                 struct btree_iter iter2;
1404
1405                                 bch2_trans_copy_iter(&iter2, iter);
1406                                 bch2_btree_iter_set_snapshot(&iter2, i->snapshot);
1407                                 ret =   bch2_btree_iter_traverse(&iter2) ?:
1408                                         bch2_btree_delete_at(trans, &iter2,
1409                                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1410                                 bch2_trans_iter_exit(trans, &iter2);
1411                                 if (ret)
1412                                         goto err;
1413
1414                                 iter->k.type = KEY_TYPE_whiteout;
1415                         }
1416
1417                         if (bkey_extent_is_allocation(k.k))
1418                                 i->count += k.k->size;
1419                 }
1420
1421                 i->seen_this_pos = true;
1422         }
1423 out:
1424 err:
1425 fsck_err:
1426         printbuf_exit(&buf);
1427
1428         if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
1429                 bch_err_fn(c, ret);
1430         return ret;
1431 delete:
1432         ret = bch2_btree_delete_at(trans, iter, BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1433         goto out;
1434 }
1435
1436 /*
1437  * Walk extents: verify that extents have a corresponding S_ISREG inode, and
1438  * that i_size an i_sectors are consistent
1439  */
1440 int bch2_check_extents(struct bch_fs *c)
1441 {
1442         struct inode_walker w = inode_walker_init();
1443         struct snapshots_seen s;
1444         struct btree_trans trans;
1445         struct btree_iter iter;
1446         struct bkey_s_c k;
1447         struct extent_ends extent_ends;
1448         struct disk_reservation res = { 0 };
1449         int ret = 0;
1450
1451         snapshots_seen_init(&s);
1452         extent_ends_init(&extent_ends);
1453         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 4096);
1454
1455         ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_extents,
1456                         POS(BCACHEFS_ROOT_INO, 0),
1457                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1458                         &res, NULL,
1459                         BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL, ({
1460                 bch2_disk_reservation_put(c, &res);
1461                 check_extent(&trans, &iter, k, &w, &s, &extent_ends);
1462         })) ?:
1463         check_i_sectors(&trans, &w);
1464
1465         bch2_disk_reservation_put(c, &res);
1466         extent_ends_exit(&extent_ends);
1467         inode_walker_exit(&w);
1468         bch2_trans_exit(&trans);
1469         snapshots_seen_exit(&s);
1470
1471         if (ret)
1472                 bch_err_fn(c, ret);
1473         return ret;
1474 }
1475
1476 static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
1477 {
1478         struct bch_fs *c = trans->c;
1479         struct inode_walker_entry *i;
1480         u32 restart_count = trans->restart_count;
1481         int ret = 0;
1482         s64 count2;
1483
1484         darray_for_each(w->inodes, i) {
1485                 if (i->inode.bi_nlink == i->count)
1486                         continue;
1487
1488                 count2 = bch2_count_subdirs(trans, w->last_pos.inode, i->snapshot);
1489                 if (count2 < 0)
1490                         return count2;
1491
1492                 if (i->count != count2) {
1493                         bch_err(c, "fsck counted subdirectories wrong: got %llu should be %llu",
1494                                 i->count, count2);
1495                         i->count = count2;
1496                         if (i->inode.bi_nlink == i->count)
1497                                 continue;
1498                 }
1499
1500                 if (fsck_err_on(i->inode.bi_nlink != i->count, c,
1501                                 "directory %llu:%u with wrong i_nlink: got %u, should be %llu",
1502                                 w->last_pos.inode, i->snapshot, i->inode.bi_nlink, i->count)) {
1503                         i->inode.bi_nlink = i->count;
1504                         ret = write_inode(trans, &i->inode, i->snapshot);
1505                         if (ret)
1506                                 break;
1507                 }
1508         }
1509 fsck_err:
1510         if (ret)
1511                 bch_err_fn(c, ret);
1512         if (!ret && trans_was_restarted(trans, restart_count))
1513                 ret = -BCH_ERR_transaction_restart_nested;
1514         return ret;
1515 }
1516
1517 static int check_dirent_target(struct btree_trans *trans,
1518                                struct btree_iter *iter,
1519                                struct bkey_s_c_dirent d,
1520                                struct bch_inode_unpacked *target,
1521                                u32 target_snapshot)
1522 {
1523         struct bch_fs *c = trans->c;
1524         struct bkey_i_dirent *n;
1525         bool backpointer_exists = true;
1526         struct printbuf buf = PRINTBUF;
1527         int ret = 0;
1528
1529         if (!target->bi_dir &&
1530             !target->bi_dir_offset) {
1531                 target->bi_dir          = d.k->p.inode;
1532                 target->bi_dir_offset   = d.k->p.offset;
1533
1534                 ret = __write_inode(trans, target, target_snapshot);
1535                 if (ret)
1536                         goto err;
1537         }
1538
1539         if (!inode_points_to_dirent(target, d)) {
1540                 ret = inode_backpointer_exists(trans, target, d.k->p.snapshot);
1541                 if (ret < 0)
1542                         goto err;
1543
1544                 backpointer_exists = ret;
1545                 ret = 0;
1546
1547                 if (fsck_err_on(S_ISDIR(target->bi_mode) &&
1548                                 backpointer_exists, c,
1549                                 "directory %llu with multiple links",
1550                                 target->bi_inum)) {
1551                         ret = __remove_dirent(trans, d.k->p);
1552                         goto out;
1553                 }
1554
1555                 if (fsck_err_on(backpointer_exists &&
1556                                 !target->bi_nlink, c,
1557                                 "inode %llu type %s has multiple links but i_nlink 0",
1558                                 target->bi_inum, bch2_d_types[d.v->d_type])) {
1559                         target->bi_nlink++;
1560                         target->bi_flags &= ~BCH_INODE_UNLINKED;
1561
1562                         ret = __write_inode(trans, target, target_snapshot);
1563                         if (ret)
1564                                 goto err;
1565                 }
1566
1567                 if (fsck_err_on(!backpointer_exists, c,
1568                                 "inode %llu:%u has wrong backpointer:\n"
1569                                 "got       %llu:%llu\n"
1570                                 "should be %llu:%llu",
1571                                 target->bi_inum, target_snapshot,
1572                                 target->bi_dir,
1573                                 target->bi_dir_offset,
1574                                 d.k->p.inode,
1575                                 d.k->p.offset)) {
1576                         target->bi_dir          = d.k->p.inode;
1577                         target->bi_dir_offset   = d.k->p.offset;
1578
1579                         ret = __write_inode(trans, target, target_snapshot);
1580                         if (ret)
1581                                 goto err;
1582                 }
1583         }
1584
1585         if (fsck_err_on(d.v->d_type != inode_d_type(target), c,
1586                         "incorrect d_type: got %s, should be %s:\n%s",
1587                         bch2_d_type_str(d.v->d_type),
1588                         bch2_d_type_str(inode_d_type(target)),
1589                         (printbuf_reset(&buf),
1590                          bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
1591                 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1592                 ret = PTR_ERR_OR_ZERO(n);
1593                 if (ret)
1594                         goto err;
1595
1596                 bkey_reassemble(&n->k_i, d.s_c);
1597                 n->v.d_type = inode_d_type(target);
1598
1599                 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1600                 if (ret)
1601                         goto err;
1602
1603                 d = dirent_i_to_s_c(n);
1604         }
1605
1606         if (d.v->d_type == DT_SUBVOL &&
1607             target->bi_parent_subvol != le32_to_cpu(d.v->d_parent_subvol) &&
1608             (c->sb.version < bcachefs_metadata_version_subvol_dirent ||
1609              fsck_err(c, "dirent has wrong d_parent_subvol field: got %u, should be %u",
1610                       le32_to_cpu(d.v->d_parent_subvol),
1611                       target->bi_parent_subvol))) {
1612                 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1613                 ret = PTR_ERR_OR_ZERO(n);
1614                 if (ret)
1615                         goto err;
1616
1617                 bkey_reassemble(&n->k_i, d.s_c);
1618                 n->v.d_parent_subvol = cpu_to_le32(target->bi_parent_subvol);
1619
1620                 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1621                 if (ret)
1622                         goto err;
1623
1624                 d = dirent_i_to_s_c(n);
1625         }
1626 out:
1627 err:
1628 fsck_err:
1629         printbuf_exit(&buf);
1630
1631         if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
1632                 bch_err_fn(c, ret);
1633         return ret;
1634 }
1635
1636 static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
1637                         struct bkey_s_c k,
1638                         struct bch_hash_info *hash_info,
1639                         struct inode_walker *dir,
1640                         struct inode_walker *target,
1641                         struct snapshots_seen *s)
1642 {
1643         struct bch_fs *c = trans->c;
1644         struct bkey_s_c_dirent d;
1645         struct inode_walker_entry *i;
1646         struct printbuf buf = PRINTBUF;
1647         struct bpos equiv;
1648         int ret = 0;
1649
1650         ret = check_key_has_snapshot(trans, iter, k);
1651         if (ret) {
1652                 ret = ret < 0 ? ret : 0;
1653                 goto out;
1654         }
1655
1656         equiv = k.k->p;
1657         equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1658
1659         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1660         if (ret)
1661                 goto err;
1662
1663         if (k.k->type == KEY_TYPE_whiteout)
1664                 goto out;
1665
1666         if (dir->last_pos.inode != k.k->p.inode) {
1667                 ret = check_subdir_count(trans, dir);
1668                 if (ret)
1669                         goto err;
1670         }
1671
1672         BUG_ON(!iter->path->should_be_locked);
1673
1674         i = walk_inode(trans, dir, equiv, k.k->type == KEY_TYPE_whiteout);
1675         ret = PTR_ERR_OR_ZERO(i);
1676         if (ret < 0)
1677                 goto err;
1678
1679         if (dir->first_this_inode && dir->inodes.nr)
1680                 *hash_info = bch2_hash_info_init(c, &dir->inodes.data[0].inode);
1681         dir->first_this_inode = false;
1682
1683         if (fsck_err_on(!i, c,
1684                         "dirent in nonexisting directory:\n%s",
1685                         (printbuf_reset(&buf),
1686                          bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1687                 ret = bch2_btree_delete_at(trans, iter,
1688                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1689                 goto out;
1690         }
1691
1692         if (!i)
1693                 goto out;
1694
1695         if (fsck_err_on(!S_ISDIR(i->inode.bi_mode), c,
1696                         "dirent in non directory inode type %s:\n%s",
1697                         bch2_d_type_str(inode_d_type(&i->inode)),
1698                         (printbuf_reset(&buf),
1699                          bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1700                 ret = bch2_btree_delete_at(trans, iter, 0);
1701                 goto out;
1702         }
1703
1704         ret = hash_check_key(trans, bch2_dirent_hash_desc, hash_info, iter, k);
1705         if (ret < 0)
1706                 goto err;
1707         if (ret) {
1708                 /* dirent has been deleted */
1709                 ret = 0;
1710                 goto out;
1711         }
1712
1713         if (k.k->type != KEY_TYPE_dirent)
1714                 goto out;
1715
1716         d = bkey_s_c_to_dirent(k);
1717
1718         if (d.v->d_type == DT_SUBVOL) {
1719                 struct bch_inode_unpacked subvol_root;
1720                 u32 target_subvol = le32_to_cpu(d.v->d_child_subvol);
1721                 u32 target_snapshot;
1722                 u64 target_inum;
1723
1724                 ret = __subvol_lookup(trans, target_subvol,
1725                                       &target_snapshot, &target_inum);
1726                 if (ret && !bch2_err_matches(ret, ENOENT))
1727                         goto err;
1728
1729                 if (fsck_err_on(ret, c,
1730                                 "dirent points to missing subvolume %u",
1731                                 le32_to_cpu(d.v->d_child_subvol))) {
1732                         ret = __remove_dirent(trans, d.k->p);
1733                         goto err;
1734                 }
1735
1736                 ret = __lookup_inode(trans, target_inum,
1737                                    &subvol_root, &target_snapshot);
1738                 if (ret && !bch2_err_matches(ret, ENOENT))
1739                         goto err;
1740
1741                 if (fsck_err_on(ret, c,
1742                                 "subvolume %u points to missing subvolume root %llu",
1743                                 target_subvol,
1744                                 target_inum)) {
1745                         bch_err(c, "repair not implemented yet");
1746                         ret = -EINVAL;
1747                         goto err;
1748                 }
1749
1750                 if (fsck_err_on(subvol_root.bi_subvol != target_subvol, c,
1751                                 "subvol root %llu has wrong bi_subvol field: got %u, should be %u",
1752                                 target_inum,
1753                                 subvol_root.bi_subvol, target_subvol)) {
1754                         subvol_root.bi_subvol = target_subvol;
1755                         ret = __write_inode(trans, &subvol_root, target_snapshot);
1756                         if (ret)
1757                                 goto err;
1758                 }
1759
1760                 ret = check_dirent_target(trans, iter, d, &subvol_root,
1761                                           target_snapshot);
1762                 if (ret)
1763                         goto err;
1764         } else {
1765                 ret = __get_visible_inodes(trans, target, s, le64_to_cpu(d.v->d_inum));
1766                 if (ret)
1767                         goto err;
1768
1769                 if (fsck_err_on(!target->inodes.nr, c,
1770                                 "dirent points to missing inode: (equiv %u)\n%s",
1771                                 equiv.snapshot,
1772                                 (printbuf_reset(&buf),
1773                                  bch2_bkey_val_to_text(&buf, c, k),
1774                                  buf.buf))) {
1775                         ret = __remove_dirent(trans, d.k->p);
1776                         if (ret)
1777                                 goto err;
1778                 }
1779
1780                 darray_for_each(target->inodes, i) {
1781                         ret = check_dirent_target(trans, iter, d,
1782                                                   &i->inode, i->snapshot);
1783                         if (ret)
1784                                 goto err;
1785                 }
1786         }
1787
1788         if (d.v->d_type == DT_DIR)
1789                 for_each_visible_inode(c, s, dir, equiv.snapshot, i)
1790                         i->count++;
1791
1792 out:
1793 err:
1794 fsck_err:
1795         printbuf_exit(&buf);
1796
1797         if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
1798                 bch_err_fn(c, ret);
1799         return ret;
1800 }
1801
1802 /*
1803  * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
1804  * validate d_type
1805  */
1806 int bch2_check_dirents(struct bch_fs *c)
1807 {
1808         struct inode_walker dir = inode_walker_init();
1809         struct inode_walker target = inode_walker_init();
1810         struct snapshots_seen s;
1811         struct bch_hash_info hash_info;
1812         struct btree_trans trans;
1813         struct btree_iter iter;
1814         struct bkey_s_c k;
1815         int ret = 0;
1816
1817         snapshots_seen_init(&s);
1818         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1819
1820         ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_dirents,
1821                         POS(BCACHEFS_ROOT_INO, 0),
1822                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
1823                         k,
1824                         NULL, NULL,
1825                         BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
1826                 check_dirent(&trans, &iter, k, &hash_info, &dir, &target, &s));
1827
1828         bch2_trans_exit(&trans);
1829         snapshots_seen_exit(&s);
1830         inode_walker_exit(&dir);
1831         inode_walker_exit(&target);
1832
1833         if (ret)
1834                 bch_err_fn(c, ret);
1835         return ret;
1836 }
1837
1838 static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
1839                        struct bkey_s_c k,
1840                        struct bch_hash_info *hash_info,
1841                        struct inode_walker *inode)
1842 {
1843         struct bch_fs *c = trans->c;
1844         struct inode_walker_entry *i;
1845         int ret;
1846
1847         ret = check_key_has_snapshot(trans, iter, k);
1848         if (ret)
1849                 return ret;
1850
1851         i = walk_inode(trans, inode, k.k->p, k.k->type == KEY_TYPE_whiteout);
1852         ret = PTR_ERR_OR_ZERO(i);
1853         if (ret)
1854                 return ret;
1855
1856         if (inode->first_this_inode && inode->inodes.nr)
1857                 *hash_info = bch2_hash_info_init(c, &inode->inodes.data[0].inode);
1858         inode->first_this_inode = false;
1859
1860         if (fsck_err_on(!i, c,
1861                         "xattr for missing inode %llu",
1862                         k.k->p.inode))
1863                 return bch2_btree_delete_at(trans, iter, 0);
1864
1865         if (!i)
1866                 return 0;
1867
1868         ret = hash_check_key(trans, bch2_xattr_hash_desc, hash_info, iter, k);
1869 fsck_err:
1870         if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
1871                 bch_err_fn(c, ret);
1872         return ret;
1873 }
1874
1875 /*
1876  * Walk xattrs: verify that they all have a corresponding inode
1877  */
1878 int bch2_check_xattrs(struct bch_fs *c)
1879 {
1880         struct inode_walker inode = inode_walker_init();
1881         struct bch_hash_info hash_info;
1882         struct btree_trans trans;
1883         struct btree_iter iter;
1884         struct bkey_s_c k;
1885         int ret = 0;
1886
1887         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1888
1889         ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_xattrs,
1890                         POS(BCACHEFS_ROOT_INO, 0),
1891                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
1892                         k,
1893                         NULL, NULL,
1894                         BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
1895                 check_xattr(&trans, &iter, k, &hash_info, &inode));
1896
1897         bch2_trans_exit(&trans);
1898
1899         if (ret)
1900                 bch_err_fn(c, ret);
1901         return ret;
1902 }
1903
1904 static int check_root_trans(struct btree_trans *trans)
1905 {
1906         struct bch_fs *c = trans->c;
1907         struct bch_inode_unpacked root_inode;
1908         u32 snapshot;
1909         u64 inum;
1910         int ret;
1911
1912         ret = __subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
1913         if (ret && !bch2_err_matches(ret, ENOENT))
1914                 return ret;
1915
1916         if (mustfix_fsck_err_on(ret, c, "root subvol missing")) {
1917                 struct bkey_i_subvolume root_subvol;
1918
1919                 snapshot        = U32_MAX;
1920                 inum            = BCACHEFS_ROOT_INO;
1921
1922                 bkey_subvolume_init(&root_subvol.k_i);
1923                 root_subvol.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1924                 root_subvol.v.flags     = 0;
1925                 root_subvol.v.snapshot  = cpu_to_le32(snapshot);
1926                 root_subvol.v.inode     = cpu_to_le64(inum);
1927                 ret = commit_do(trans, NULL, NULL,
1928                                       BTREE_INSERT_NOFAIL|
1929                                       BTREE_INSERT_LAZY_RW,
1930                         __bch2_btree_insert(trans, BTREE_ID_subvolumes,
1931                                             &root_subvol.k_i, 0));
1932                 if (ret) {
1933                         bch_err(c, "error writing root subvol: %s", bch2_err_str(ret));
1934                         goto err;
1935                 }
1936
1937         }
1938
1939         ret = __lookup_inode(trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
1940         if (ret && !bch2_err_matches(ret, ENOENT))
1941                 return ret;
1942
1943         if (mustfix_fsck_err_on(ret, c, "root directory missing") ||
1944             mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode), c,
1945                                 "root inode not a directory")) {
1946                 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755,
1947                                 0, NULL);
1948                 root_inode.bi_inum = inum;
1949
1950                 ret = __write_inode(trans, &root_inode, snapshot);
1951                 if (ret)
1952                         bch_err(c, "error writing root inode: %s", bch2_err_str(ret));
1953         }
1954 err:
1955 fsck_err:
1956         return ret;
1957 }
1958
1959 /* Get root directory, create if it doesn't exist: */
1960 int bch2_check_root(struct bch_fs *c)
1961 {
1962         int ret;
1963
1964         ret = bch2_trans_do(c, NULL, NULL,
1965                              BTREE_INSERT_NOFAIL|
1966                              BTREE_INSERT_LAZY_RW,
1967                 check_root_trans(&trans));
1968
1969         if (ret)
1970                 bch_err_fn(c, ret);
1971         return ret;
1972 }
1973
1974 struct pathbuf_entry {
1975         u64     inum;
1976         u32     snapshot;
1977 };
1978
1979 typedef DARRAY(struct pathbuf_entry) pathbuf;
1980
1981 static bool path_is_dup(pathbuf *p, u64 inum, u32 snapshot)
1982 {
1983         struct pathbuf_entry *i;
1984
1985         darray_for_each(*p, i)
1986                 if (i->inum     == inum &&
1987                     i->snapshot == snapshot)
1988                         return true;
1989
1990         return false;
1991 }
1992
1993 static int path_down(struct bch_fs *c, pathbuf *p,
1994                      u64 inum, u32 snapshot)
1995 {
1996         int ret = darray_push(p, ((struct pathbuf_entry) {
1997                 .inum           = inum,
1998                 .snapshot       = snapshot,
1999         }));
2000
2001         if (ret)
2002                 bch_err(c, "fsck: error allocating memory for pathbuf, size %zu",
2003                         p->size);
2004         return ret;
2005 }
2006
2007 /*
2008  * Check that a given inode is reachable from the root:
2009  *
2010  * XXX: we should also be verifying that inodes are in the right subvolumes
2011  */
2012 static int check_path(struct btree_trans *trans,
2013                       pathbuf *p,
2014                       struct bch_inode_unpacked *inode,
2015                       u32 snapshot)
2016 {
2017         struct bch_fs *c = trans->c;
2018         int ret = 0;
2019
2020         snapshot = bch2_snapshot_equiv(c, snapshot);
2021         p->nr = 0;
2022
2023         while (!(inode->bi_inum == BCACHEFS_ROOT_INO &&
2024                  inode->bi_subvol == BCACHEFS_ROOT_SUBVOL)) {
2025                 struct btree_iter dirent_iter;
2026                 struct bkey_s_c_dirent d;
2027                 u32 parent_snapshot = snapshot;
2028
2029                 if (inode->bi_subvol) {
2030                         u64 inum;
2031
2032                         ret = subvol_lookup(trans, inode->bi_parent_subvol,
2033                                             &parent_snapshot, &inum);
2034                         if (ret)
2035                                 break;
2036                 }
2037
2038                 ret = lockrestart_do(trans,
2039                         PTR_ERR_OR_ZERO((d = dirent_get_by_pos(trans, &dirent_iter,
2040                                           SPOS(inode->bi_dir, inode->bi_dir_offset,
2041                                                parent_snapshot))).k));
2042                 if (ret && !bch2_err_matches(ret, ENOENT))
2043                         break;
2044
2045                 if (!ret && !dirent_points_to_inode(d, inode)) {
2046                         bch2_trans_iter_exit(trans, &dirent_iter);
2047                         ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
2048                 }
2049
2050                 if (bch2_err_matches(ret, ENOENT)) {
2051                         if (fsck_err(c,  "unreachable inode %llu:%u, type %s nlink %u backptr %llu:%llu",
2052                                      inode->bi_inum, snapshot,
2053                                      bch2_d_type_str(inode_d_type(inode)),
2054                                      inode->bi_nlink,
2055                                      inode->bi_dir,
2056                                      inode->bi_dir_offset))
2057                                 ret = reattach_inode(trans, inode, snapshot);
2058                         break;
2059                 }
2060
2061                 bch2_trans_iter_exit(trans, &dirent_iter);
2062
2063                 if (!S_ISDIR(inode->bi_mode))
2064                         break;
2065
2066                 ret = path_down(c, p, inode->bi_inum, snapshot);
2067                 if (ret) {
2068                         bch_err(c, "memory allocation failure");
2069                         return ret;
2070                 }
2071
2072                 snapshot = parent_snapshot;
2073
2074                 ret = lookup_inode(trans, inode->bi_dir, inode, &snapshot);
2075                 if (ret) {
2076                         /* Should have been caught in dirents pass */
2077                         bch_err(c, "error looking up parent directory: %i", ret);
2078                         break;
2079                 }
2080
2081                 if (path_is_dup(p, inode->bi_inum, snapshot)) {
2082                         struct pathbuf_entry *i;
2083
2084                         /* XXX print path */
2085                         bch_err(c, "directory structure loop");
2086
2087                         darray_for_each(*p, i)
2088                                 pr_err("%llu:%u", i->inum, i->snapshot);
2089                         pr_err("%llu:%u", inode->bi_inum, snapshot);
2090
2091                         if (!fsck_err(c, "directory structure loop"))
2092                                 return 0;
2093
2094                         ret = commit_do(trans, NULL, NULL,
2095                                               BTREE_INSERT_NOFAIL|
2096                                               BTREE_INSERT_LAZY_RW,
2097                                         remove_backpointer(trans, inode));
2098                         if (ret) {
2099                                 bch_err(c, "error removing dirent: %i", ret);
2100                                 break;
2101                         }
2102
2103                         ret = reattach_inode(trans, inode, snapshot);
2104                 }
2105         }
2106 fsck_err:
2107         if (ret)
2108                 bch_err_fn(c, ret);
2109         return ret;
2110 }
2111
2112 /*
2113  * Check for unreachable inodes, as well as loops in the directory structure:
2114  * After bch2_check_dirents(), if an inode backpointer doesn't exist that means it's
2115  * unreachable:
2116  */
2117 int bch2_check_directory_structure(struct bch_fs *c)
2118 {
2119         struct btree_trans trans;
2120         struct btree_iter iter;
2121         struct bkey_s_c k;
2122         struct bch_inode_unpacked u;
2123         pathbuf path = { 0, };
2124         int ret;
2125
2126         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2127
2128         for_each_btree_key(&trans, iter, BTREE_ID_inodes, POS_MIN,
2129                            BTREE_ITER_INTENT|
2130                            BTREE_ITER_PREFETCH|
2131                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2132                 if (!bkey_is_inode(k.k))
2133                         continue;
2134
2135                 ret = bch2_inode_unpack(k, &u);
2136                 if (ret) {
2137                         /* Should have been caught earlier in fsck: */
2138                         bch_err(c, "error unpacking inode %llu: %i", k.k->p.offset, ret);
2139                         break;
2140                 }
2141
2142                 if (u.bi_flags & BCH_INODE_UNLINKED)
2143                         continue;
2144
2145                 ret = check_path(&trans, &path, &u, iter.pos.snapshot);
2146                 if (ret)
2147                         break;
2148         }
2149         bch2_trans_iter_exit(&trans, &iter);
2150         bch2_trans_exit(&trans);
2151         darray_exit(&path);
2152
2153         if (ret)
2154                 bch_err_fn(c, ret);
2155         return ret;
2156 }
2157
2158 /* check_nlink pass: */
2159
2160 struct nlink_table {
2161         size_t          nr;
2162         size_t          size;
2163
2164         struct nlink {
2165                 u64     inum;
2166                 u32     snapshot;
2167                 u32     count;
2168         }               *d;
2169 };
2170
2171 static int add_nlink(struct bch_fs *c, struct nlink_table *t,
2172                      u64 inum, u32 snapshot)
2173 {
2174         if (t->nr == t->size) {
2175                 size_t new_size = max_t(size_t, 128UL, t->size * 2);
2176                 void *d = kvmalloc_array(new_size, sizeof(t->d[0]), GFP_KERNEL);
2177
2178                 if (!d) {
2179                         bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
2180                                 new_size);
2181                         return -BCH_ERR_ENOMEM_fsck_add_nlink;
2182                 }
2183
2184                 if (t->d)
2185                         memcpy(d, t->d, t->size * sizeof(t->d[0]));
2186                 kvfree(t->d);
2187
2188                 t->d = d;
2189                 t->size = new_size;
2190         }
2191
2192
2193         t->d[t->nr++] = (struct nlink) {
2194                 .inum           = inum,
2195                 .snapshot       = snapshot,
2196         };
2197
2198         return 0;
2199 }
2200
2201 static int nlink_cmp(const void *_l, const void *_r)
2202 {
2203         const struct nlink *l = _l;
2204         const struct nlink *r = _r;
2205
2206         return cmp_int(l->inum, r->inum) ?: cmp_int(l->snapshot, r->snapshot);
2207 }
2208
2209 static void inc_link(struct bch_fs *c, struct snapshots_seen *s,
2210                      struct nlink_table *links,
2211                      u64 range_start, u64 range_end, u64 inum, u32 snapshot)
2212 {
2213         struct nlink *link, key = {
2214                 .inum = inum, .snapshot = U32_MAX,
2215         };
2216
2217         if (inum < range_start || inum >= range_end)
2218                 return;
2219
2220         link = __inline_bsearch(&key, links->d, links->nr,
2221                                 sizeof(links->d[0]), nlink_cmp);
2222         if (!link)
2223                 return;
2224
2225         while (link > links->d && link[0].inum == link[-1].inum)
2226                 --link;
2227
2228         for (; link < links->d + links->nr && link->inum == inum; link++)
2229                 if (ref_visible(c, s, snapshot, link->snapshot)) {
2230                         link->count++;
2231                         if (link->snapshot >= snapshot)
2232                                 break;
2233                 }
2234 }
2235
2236 noinline_for_stack
2237 static int check_nlinks_find_hardlinks(struct bch_fs *c,
2238                                        struct nlink_table *t,
2239                                        u64 start, u64 *end)
2240 {
2241         struct btree_trans trans;
2242         struct btree_iter iter;
2243         struct bkey_s_c k;
2244         struct bch_inode_unpacked u;
2245         int ret = 0;
2246
2247         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2248
2249         for_each_btree_key(&trans, iter, BTREE_ID_inodes,
2250                            POS(0, start),
2251                            BTREE_ITER_INTENT|
2252                            BTREE_ITER_PREFETCH|
2253                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2254                 if (!bkey_is_inode(k.k))
2255                         continue;
2256
2257                 /* Should never fail, checked by bch2_inode_invalid: */
2258                 BUG_ON(bch2_inode_unpack(k, &u));
2259
2260                 /*
2261                  * Backpointer and directory structure checks are sufficient for
2262                  * directories, since they can't have hardlinks:
2263                  */
2264                 if (S_ISDIR(u.bi_mode))
2265                         continue;
2266
2267                 if (!u.bi_nlink)
2268                         continue;
2269
2270                 ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot);
2271                 if (ret) {
2272                         *end = k.k->p.offset;
2273                         ret = 0;
2274                         break;
2275                 }
2276
2277         }
2278         bch2_trans_iter_exit(&trans, &iter);
2279         bch2_trans_exit(&trans);
2280
2281         if (ret)
2282                 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
2283
2284         return ret;
2285 }
2286
2287 noinline_for_stack
2288 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
2289                                      u64 range_start, u64 range_end)
2290 {
2291         struct btree_trans trans;
2292         struct snapshots_seen s;
2293         struct btree_iter iter;
2294         struct bkey_s_c k;
2295         struct bkey_s_c_dirent d;
2296         int ret;
2297
2298         snapshots_seen_init(&s);
2299
2300         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2301
2302         for_each_btree_key(&trans, iter, BTREE_ID_dirents, POS_MIN,
2303                            BTREE_ITER_INTENT|
2304                            BTREE_ITER_PREFETCH|
2305                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2306                 ret = snapshots_seen_update(c, &s, iter.btree_id, k.k->p);
2307                 if (ret)
2308                         break;
2309
2310                 switch (k.k->type) {
2311                 case KEY_TYPE_dirent:
2312                         d = bkey_s_c_to_dirent(k);
2313
2314                         if (d.v->d_type != DT_DIR &&
2315                             d.v->d_type != DT_SUBVOL)
2316                                 inc_link(c, &s, links, range_start, range_end,
2317                                          le64_to_cpu(d.v->d_inum),
2318                                          bch2_snapshot_equiv(c, d.k->p.snapshot));
2319                         break;
2320                 }
2321         }
2322         bch2_trans_iter_exit(&trans, &iter);
2323
2324         if (ret)
2325                 bch_err(c, "error in fsck: btree error %i while walking dirents", ret);
2326
2327         bch2_trans_exit(&trans);
2328         snapshots_seen_exit(&s);
2329         return ret;
2330 }
2331
2332 static int check_nlinks_update_inode(struct btree_trans *trans, struct btree_iter *iter,
2333                                      struct bkey_s_c k,
2334                                      struct nlink_table *links,
2335                                      size_t *idx, u64 range_end)
2336 {
2337         struct bch_fs *c = trans->c;
2338         struct bch_inode_unpacked u;
2339         struct nlink *link = &links->d[*idx];
2340         int ret = 0;
2341
2342         if (k.k->p.offset >= range_end)
2343                 return 1;
2344
2345         if (!bkey_is_inode(k.k))
2346                 return 0;
2347
2348         BUG_ON(bch2_inode_unpack(k, &u));
2349
2350         if (S_ISDIR(u.bi_mode))
2351                 return 0;
2352
2353         if (!u.bi_nlink)
2354                 return 0;
2355
2356         while ((cmp_int(link->inum, k.k->p.offset) ?:
2357                 cmp_int(link->snapshot, k.k->p.snapshot)) < 0) {
2358                 BUG_ON(*idx == links->nr);
2359                 link = &links->d[++*idx];
2360         }
2361
2362         if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count, c,
2363                         "inode %llu type %s has wrong i_nlink (%u, should be %u)",
2364                         u.bi_inum, bch2_d_types[mode_to_type(u.bi_mode)],
2365                         bch2_inode_nlink_get(&u), link->count)) {
2366                 bch2_inode_nlink_set(&u, link->count);
2367                 ret = __write_inode(trans, &u, k.k->p.snapshot);
2368         }
2369 fsck_err:
2370         return ret;
2371 }
2372
2373 noinline_for_stack
2374 static int check_nlinks_update_hardlinks(struct bch_fs *c,
2375                                struct nlink_table *links,
2376                                u64 range_start, u64 range_end)
2377 {
2378         struct btree_trans trans;
2379         struct btree_iter iter;
2380         struct bkey_s_c k;
2381         size_t idx = 0;
2382         int ret = 0;
2383
2384         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2385
2386         ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_inodes,
2387                         POS(0, range_start),
2388                         BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
2389                         NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
2390                 check_nlinks_update_inode(&trans, &iter, k, links, &idx, range_end));
2391
2392         bch2_trans_exit(&trans);
2393
2394         if (ret < 0) {
2395                 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
2396                 return ret;
2397         }
2398
2399         return 0;
2400 }
2401
2402 int bch2_check_nlinks(struct bch_fs *c)
2403 {
2404         struct nlink_table links = { 0 };
2405         u64 this_iter_range_start, next_iter_range_start = 0;
2406         int ret = 0;
2407
2408         do {
2409                 this_iter_range_start = next_iter_range_start;
2410                 next_iter_range_start = U64_MAX;
2411
2412                 ret = check_nlinks_find_hardlinks(c, &links,
2413                                                   this_iter_range_start,
2414                                                   &next_iter_range_start);
2415
2416                 ret = check_nlinks_walk_dirents(c, &links,
2417                                           this_iter_range_start,
2418                                           next_iter_range_start);
2419                 if (ret)
2420                         break;
2421
2422                 ret = check_nlinks_update_hardlinks(c, &links,
2423                                          this_iter_range_start,
2424                                          next_iter_range_start);
2425                 if (ret)
2426                         break;
2427
2428                 links.nr = 0;
2429         } while (next_iter_range_start != U64_MAX);
2430
2431         kvfree(links.d);
2432
2433         if (ret)
2434                 bch_err_fn(c, ret);
2435         return ret;
2436 }
2437
2438 static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter,
2439                              struct bkey_s_c k)
2440 {
2441         struct bkey_s_c_reflink_p p;
2442         struct bkey_i_reflink_p *u;
2443         int ret;
2444
2445         if (k.k->type != KEY_TYPE_reflink_p)
2446                 return 0;
2447
2448         p = bkey_s_c_to_reflink_p(k);
2449
2450         if (!p.v->front_pad && !p.v->back_pad)
2451                 return 0;
2452
2453         u = bch2_trans_kmalloc(trans, sizeof(*u));
2454         ret = PTR_ERR_OR_ZERO(u);
2455         if (ret)
2456                 return ret;
2457
2458         bkey_reassemble(&u->k_i, k);
2459         u->v.front_pad  = 0;
2460         u->v.back_pad   = 0;
2461
2462         return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_NORUN);
2463 }
2464
2465 int bch2_fix_reflink_p(struct bch_fs *c)
2466 {
2467         struct btree_iter iter;
2468         struct bkey_s_c k;
2469         int ret;
2470
2471         if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
2472                 return 0;
2473
2474         ret = bch2_trans_run(c,
2475                 for_each_btree_key_commit(&trans, iter,
2476                                 BTREE_ID_extents, POS_MIN,
2477                                 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|
2478                                 BTREE_ITER_ALL_SNAPSHOTS, k,
2479                                 NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
2480                         fix_reflink_p_key(&trans, &iter, k)));
2481
2482         if (ret)
2483                 bch_err_fn(c, ret);
2484         return ret;
2485 }