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

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "bkey_buf.h"
   5 #include "btree_cache.h"
   6 #include "btree_update.h"
   7 #include "buckets.h"
   8 #include "darray.h"
   9 #include "dirent.h"
  10 #include "error.h"
  11 #include "fs-common.h"
  12 #include "fsck.h"
  13 #include "inode.h"
  14 #include "keylist.h"
  15 #include "recovery.h"
  16 #include "snapshot.h"
  17 #include "super.h"
  18 #include "xattr.h"
  19
  20 #include <linux/bsearch.h>
  21 #include <linux/dcache.h> /* struct qstr */
  22
  23 /*
  24  * XXX: this is handling transaction restarts without returning
  25  * -BCH_ERR_transaction_restart_nested, this is not how we do things anymore:
  26  */
  27 static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum,
  28                                     u32 snapshot)
  29 {
  30         u64 sectors = 0;
  31
  32         int ret = for_each_btree_key_upto(trans, iter, BTREE_ID_extents,
  33                                 SPOS(inum, 0, snapshot),
  34                                 POS(inum, U64_MAX),
  35                                 0, k, ({
  36                 if (bkey_extent_is_allocation(k.k))
  37                         sectors += k.k->size;
  38                 0;
  39         }));
  40
  41         return ret ?: sectors;
  42 }
  43
  44 static s64 bch2_count_subdirs(struct btree_trans *trans, u64 inum,
  45                                     u32 snapshot)
  46 {
  47         u64 subdirs = 0;
  48
  49         int ret = for_each_btree_key_upto(trans, iter, BTREE_ID_dirents,
  50                                     SPOS(inum, 0, snapshot),
  51                                     POS(inum, U64_MAX),
  52                                     0, k, ({
  53                 if (k.k->type == KEY_TYPE_dirent &&
  54                     bkey_s_c_to_dirent(k).v->d_type == DT_DIR)
  55                         subdirs++;
  56                 0;
  57         }));
  58
  59         return ret ?: subdirs;
  60 }
  61
  62 static int subvol_lookup(struct btree_trans *trans, u32 subvol,
  63                          u32 *snapshot, u64 *inum)
  64 {
  65         struct bch_subvolume s;
  66         int ret;
  67
  68         ret = bch2_subvolume_get(trans, subvol, false, 0, &s);
  69
  70         *snapshot = le32_to_cpu(s.snapshot);
  71         *inum = le64_to_cpu(s.inode);
  72         return ret;
  73 }
  74
  75 static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr,
  76                               struct bch_inode_unpacked *inode)
  77 {
  78         struct btree_iter iter;
  79         struct bkey_s_c k;
  80         int ret;
  81
  82         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
  83                              POS(0, inode_nr),
  84                              BTREE_ITER_ALL_SNAPSHOTS);
  85         k = bch2_btree_iter_peek(&iter);
  86         ret = bkey_err(k);
  87         if (ret)
  88                 goto err;
  89
  90         if (!k.k || !bkey_eq(k.k->p, POS(0, inode_nr))) {
  91                 ret = -BCH_ERR_ENOENT_inode;
  92                 goto err;
  93         }
  94
  95         ret = bch2_inode_unpack(k, inode);
  96 err:
  97         bch_err_msg(trans->c, ret, "fetching inode %llu", inode_nr);
  98         bch2_trans_iter_exit(trans, &iter);
  99         return ret;
 100 }
 101
 102 static int lookup_inode(struct btree_trans *trans, u64 inode_nr,
 103                           struct bch_inode_unpacked *inode,
 104                           u32 *snapshot)
 105 {
 106         struct btree_iter iter;
 107         struct bkey_s_c k;
 108         int ret;
 109
 110         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
 111                                SPOS(0, inode_nr, *snapshot), 0);
 112         ret = bkey_err(k);
 113         if (ret)
 114                 goto err;
 115
 116         ret = bkey_is_inode(k.k)
 117                 ? bch2_inode_unpack(k, inode)
 118                 : -BCH_ERR_ENOENT_inode;
 119         if (!ret)
 120                 *snapshot = iter.pos.snapshot;
 121 err:
 122         bch_err_msg(trans->c, ret, "fetching inode %llu:%u", inode_nr, *snapshot);
 123         bch2_trans_iter_exit(trans, &iter);
 124         return ret;
 125 }
 126
 127 static int __lookup_dirent(struct btree_trans *trans,
 128                            struct bch_hash_info hash_info,
 129                            subvol_inum dir, struct qstr *name,
 130                            u64 *target, unsigned *type)
 131 {
 132         struct btree_iter iter;
 133         struct bkey_s_c_dirent d;
 134         int ret;
 135
 136         ret = bch2_hash_lookup(trans, &iter, bch2_dirent_hash_desc,
 137                                &hash_info, dir, name, 0);
 138         if (ret)
 139                 return ret;
 140
 141         d = bkey_s_c_to_dirent(bch2_btree_iter_peek_slot(&iter));
 142         *target = le64_to_cpu(d.v->d_inum);
 143         *type = d.v->d_type;
 144         bch2_trans_iter_exit(trans, &iter);
 145         return 0;
 146 }
 147
 148 static int __write_inode(struct btree_trans *trans,
 149                          struct bch_inode_unpacked *inode,
 150                          u32 snapshot)
 151 {
 152         struct bkey_inode_buf *inode_p =
 153                 bch2_trans_kmalloc(trans, sizeof(*inode_p));
 154
 155         if (IS_ERR(inode_p))
 156                 return PTR_ERR(inode_p);
 157
 158         bch2_inode_pack(inode_p, inode);
 159         inode_p->inode.k.p.snapshot = snapshot;
 160
 161         return bch2_btree_insert_nonextent(trans, BTREE_ID_inodes,
 162                                 &inode_p->inode.k_i,
 163                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 164 }
 165
 166 static int fsck_write_inode(struct btree_trans *trans,
 167                             struct bch_inode_unpacked *inode,
 168                             u32 snapshot)
 169 {
 170         int ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
 171                             __write_inode(trans, inode, snapshot));
 172         bch_err_fn(trans->c, ret);
 173         return ret;
 174 }
 175
 176 static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
 177 {
 178         struct bch_fs *c = trans->c;
 179         struct btree_iter iter;
 180         struct bch_inode_unpacked dir_inode;
 181         struct bch_hash_info dir_hash_info;
 182         int ret;
 183
 184         ret = lookup_first_inode(trans, pos.inode, &dir_inode);
 185         if (ret)
 186                 goto err;
 187
 188         dir_hash_info = bch2_hash_info_init(c, &dir_inode);
 189
 190         bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_INTENT);
 191
 192         ret = bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
 193                                   &dir_hash_info, &iter,
 194                                   BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 195         bch2_trans_iter_exit(trans, &iter);
 196 err:
 197         bch_err_fn(c, ret);
 198         return ret;
 199 }
 200
 201 /* Get lost+found, create if it doesn't exist: */
 202 static int lookup_lostfound(struct btree_trans *trans, u32 snapshot,
 203                             struct bch_inode_unpacked *lostfound)
 204 {
 205         struct bch_fs *c = trans->c;
 206         struct qstr lostfound_str = QSTR("lost+found");
 207         u64 inum = 0;
 208         unsigned d_type = 0;
 209         int ret;
 210
 211         struct bch_snapshot_tree st;
 212         ret = bch2_snapshot_tree_lookup(trans,
 213                         bch2_snapshot_tree(c, snapshot), &st);
 214         if (ret)
 215                 return ret;
 216
 217         subvol_inum root_inum = { .subvol = le32_to_cpu(st.master_subvol) };
 218         u32 subvol_snapshot;
 219
 220         ret = subvol_lookup(trans, le32_to_cpu(st.master_subvol),
 221                             &subvol_snapshot, &root_inum.inum);
 222         bch_err_msg(c, ret, "looking up root subvol");
 223         if (ret)
 224                 return ret;
 225
 226         struct bch_inode_unpacked root_inode;
 227         struct bch_hash_info root_hash_info;
 228         ret = lookup_inode(trans, root_inum.inum, &root_inode, &snapshot);
 229         bch_err_msg(c, ret, "looking up root inode");
 230         if (ret)
 231                 return ret;
 232
 233         root_hash_info = bch2_hash_info_init(c, &root_inode);
 234
 235         ret = __lookup_dirent(trans, root_hash_info, root_inum,
 236                               &lostfound_str, &inum, &d_type);
 237         if (bch2_err_matches(ret, ENOENT))
 238                 goto create_lostfound;
 239
 240         bch_err_fn(c, ret);
 241         if (ret)
 242                 return ret;
 243
 244         if (d_type != DT_DIR) {
 245                 bch_err(c, "error looking up lost+found: not a directory");
 246                 return -BCH_ERR_ENOENT_not_directory;
 247         }
 248
 249         /*
 250          * The bch2_check_dirents pass has already run, dangling dirents
 251          * shouldn't exist here:
 252          */
 253         return lookup_inode(trans, inum, lostfound, &snapshot);
 254
 255 create_lostfound:
 256         /*
 257          * XXX: we could have a nicer log message here  if we had a nice way to
 258          * walk backpointers to print a path
 259          */
 260         bch_notice(c, "creating lost+found in snapshot %u", le32_to_cpu(st.root_snapshot));
 261
 262         u64 now = bch2_current_time(c);
 263         struct btree_iter lostfound_iter = { NULL };
 264         u64 cpu = raw_smp_processor_id();
 265
 266         bch2_inode_init_early(c, lostfound);
 267         bch2_inode_init_late(lostfound, now, 0, 0, S_IFDIR|0700, 0, &root_inode);
 268         lostfound->bi_dir = root_inode.bi_inum;
 269
 270         root_inode.bi_nlink++;
 271
 272         ret = bch2_inode_create(trans, &lostfound_iter, lostfound, snapshot, cpu);
 273         if (ret)
 274                 goto err;
 275
 276         bch2_btree_iter_set_snapshot(&lostfound_iter, snapshot);
 277         ret = bch2_btree_iter_traverse(&lostfound_iter);
 278         if (ret)
 279                 goto err;
 280
 281         ret =   bch2_dirent_create_snapshot(trans,
 282                                 root_inode.bi_inum, snapshot, &root_hash_info,
 283                                 mode_to_type(lostfound->bi_mode),
 284                                 &lostfound_str,
 285                                 lostfound->bi_inum,
 286                                 &lostfound->bi_dir_offset,
 287                                 BCH_HASH_SET_MUST_CREATE) ?:
 288                 bch2_inode_write_flags(trans, &lostfound_iter, lostfound,
 289                                        BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 290 err:
 291         bch_err_msg(c, ret, "creating lost+found");
 292         bch2_trans_iter_exit(trans, &lostfound_iter);
 293         return ret;
 294 }
 295
 296 static int reattach_inode(struct btree_trans *trans,
 297                           struct bch_inode_unpacked *inode,
 298                           u32 inode_snapshot)
 299 {
 300         struct bch_hash_info dir_hash;
 301         struct bch_inode_unpacked lostfound;
 302         char name_buf[20];
 303         struct qstr name;
 304         u64 dir_offset = 0;
 305         int ret;
 306
 307         ret = lookup_lostfound(trans, inode_snapshot, &lostfound);
 308         if (ret)
 309                 return ret;
 310
 311         if (S_ISDIR(inode->bi_mode)) {
 312                 lostfound.bi_nlink++;
 313
 314                 ret = __write_inode(trans, &lostfound, U32_MAX);
 315                 if (ret)
 316                         return ret;
 317         }
 318
 319         dir_hash = bch2_hash_info_init(trans->c, &lostfound);
 320
 321         snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
 322         name = (struct qstr) QSTR(name_buf);
 323
 324         ret = bch2_dirent_create_snapshot(trans,
 325                                 lostfound.bi_inum, inode_snapshot,
 326                                 &dir_hash,
 327                                 inode_d_type(inode),
 328                                 &name, inode->bi_inum, &dir_offset,
 329                                 BCH_HASH_SET_MUST_CREATE);
 330         if (ret)
 331                 return ret;
 332
 333         inode->bi_dir           = lostfound.bi_inum;
 334         inode->bi_dir_offset    = dir_offset;
 335
 336         return __write_inode(trans, inode, inode_snapshot);
 337 }
 338
 339 static int remove_backpointer(struct btree_trans *trans,
 340                               struct bch_inode_unpacked *inode)
 341 {
 342         struct btree_iter iter;
 343         struct bkey_s_c_dirent d;
 344         int ret;
 345
 346         d = bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_dirents,
 347                                      POS(inode->bi_dir, inode->bi_dir_offset), 0,
 348                                      dirent);
 349         ret =   bkey_err(d) ?:
 350                 __remove_dirent(trans, d.k->p);
 351         bch2_trans_iter_exit(trans, &iter);
 352         return ret;
 353 }
 354
 355 struct snapshots_seen_entry {
 356         u32                             id;
 357         u32                             equiv;
 358 };
 359
 360 struct snapshots_seen {
 361         struct bpos                     pos;
 362         DARRAY(struct snapshots_seen_entry) ids;
 363 };
 364
 365 static inline void snapshots_seen_exit(struct snapshots_seen *s)
 366 {
 367         darray_exit(&s->ids);
 368 }
 369
 370 static inline void snapshots_seen_init(struct snapshots_seen *s)
 371 {
 372         memset(s, 0, sizeof(*s));
 373 }
 374
 375 static int snapshots_seen_add_inorder(struct bch_fs *c, struct snapshots_seen *s, u32 id)
 376 {
 377         struct snapshots_seen_entry *i, n = {
 378                 .id     = id,
 379                 .equiv  = bch2_snapshot_equiv(c, id),
 380         };
 381         int ret = 0;
 382
 383         __darray_for_each(s->ids, i) {
 384                 if (i->id == id)
 385                         return 0;
 386                 if (i->id > id)
 387                         break;
 388         }
 389
 390         ret = darray_insert_item(&s->ids, i - s->ids.data, n);
 391         if (ret)
 392                 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
 393                         s->ids.size);
 394         return ret;
 395 }
 396
 397 static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s,
 398                                  enum btree_id btree_id, struct bpos pos)
 399 {
 400         struct snapshots_seen_entry n = {
 401                 .id     = pos.snapshot,
 402                 .equiv  = bch2_snapshot_equiv(c, pos.snapshot),
 403         };
 404         int ret = 0;
 405
 406         if (!bkey_eq(s->pos, pos))
 407                 s->ids.nr = 0;
 408
 409         s->pos = pos;
 410         s->pos.snapshot = n.equiv;
 411
 412         darray_for_each(s->ids, i) {
 413                 if (i->id == n.id)
 414                         return 0;
 415
 416                 /*
 417                  * We currently don't rigorously track for snapshot cleanup
 418                  * needing to be run, so it shouldn't be a fsck error yet:
 419                  */
 420                 if (i->equiv == n.equiv) {
 421                         bch_err(c, "snapshot deletion did not finish:\n"
 422                                 "  duplicate keys in btree %s at %llu:%llu snapshots %u, %u (equiv %u)\n",
 423                                 bch2_btree_id_str(btree_id),
 424                                 pos.inode, pos.offset,
 425                                 i->id, n.id, n.equiv);
 426                         set_bit(BCH_FS_need_delete_dead_snapshots, &c->flags);
 427                         return bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_delete_dead_snapshots);
 428                 }
 429         }
 430
 431         ret = darray_push(&s->ids, n);
 432         if (ret)
 433                 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
 434                         s->ids.size);
 435         return ret;
 436 }
 437
 438 /**
 439  * key_visible_in_snapshot - returns true if @id is a descendent of @ancestor,
 440  * and @ancestor hasn't been overwritten in @seen
 441  *
 442  * @c:          filesystem handle
 443  * @seen:       list of snapshot ids already seen at current position
 444  * @id:         descendent snapshot id
 445  * @ancestor:   ancestor snapshot id
 446  *
 447  * Returns:     whether key in @ancestor snapshot is visible in @id snapshot
 448  */
 449 static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *seen,
 450                                     u32 id, u32 ancestor)
 451 {
 452         ssize_t i;
 453
 454         EBUG_ON(id > ancestor);
 455         EBUG_ON(!bch2_snapshot_is_equiv(c, id));
 456         EBUG_ON(!bch2_snapshot_is_equiv(c, ancestor));
 457
 458         /* @ancestor should be the snapshot most recently added to @seen */
 459         EBUG_ON(ancestor != seen->pos.snapshot);
 460         EBUG_ON(ancestor != seen->ids.data[seen->ids.nr - 1].equiv);
 461
 462         if (id == ancestor)
 463                 return true;
 464
 465         if (!bch2_snapshot_is_ancestor(c, id, ancestor))
 466                 return false;
 467
 468         /*
 469          * We know that @id is a descendant of @ancestor, we're checking if
 470          * we've seen a key that overwrote @ancestor - i.e. also a descendent of
 471          * @ascestor and with @id as a descendent.
 472          *
 473          * But we already know that we're scanning IDs between @id and @ancestor
 474          * numerically, since snapshot ID lists are kept sorted, so if we find
 475          * an id that's an ancestor of @id we're done:
 476          */
 477
 478         for (i = seen->ids.nr - 2;
 479              i >= 0 && seen->ids.data[i].equiv >= id;
 480              --i)
 481                 if (bch2_snapshot_is_ancestor(c, id, seen->ids.data[i].equiv))
 482                         return false;
 483
 484         return true;
 485 }
 486
 487 /**
 488  * ref_visible - given a key with snapshot id @src that points to a key with
 489  * snapshot id @dst, test whether there is some snapshot in which @dst is
 490  * visible.
 491  *
 492  * @c:          filesystem handle
 493  * @s:          list of snapshot IDs already seen at @src
 494  * @src:        snapshot ID of src key
 495  * @dst:        snapshot ID of dst key
 496  * Returns:     true if there is some snapshot in which @dst is visible
 497  *
 498  * Assumes we're visiting @src keys in natural key order
 499  */
 500 static bool ref_visible(struct bch_fs *c, struct snapshots_seen *s,
 501                         u32 src, u32 dst)
 502 {
 503         return dst <= src
 504                 ? key_visible_in_snapshot(c, s, dst, src)
 505                 : bch2_snapshot_is_ancestor(c, src, dst);
 506 }
 507
 508 static int ref_visible2(struct bch_fs *c,
 509                         u32 src, struct snapshots_seen *src_seen,
 510                         u32 dst, struct snapshots_seen *dst_seen)
 511 {
 512         src = bch2_snapshot_equiv(c, src);
 513         dst = bch2_snapshot_equiv(c, dst);
 514
 515         if (dst > src) {
 516                 swap(dst, src);
 517                 swap(dst_seen, src_seen);
 518         }
 519         return key_visible_in_snapshot(c, src_seen, dst, src);
 520 }
 521
 522 #define for_each_visible_inode(_c, _s, _w, _snapshot, _i)                               \
 523         for (_i = (_w)->inodes.data; _i < (_w)->inodes.data + (_w)->inodes.nr &&        \
 524              (_i)->snapshot <= (_snapshot); _i++)                                       \
 525                 if (key_visible_in_snapshot(_c, _s, _i->snapshot, _snapshot))
 526
 527 struct inode_walker_entry {
 528         struct bch_inode_unpacked inode;
 529         u32                     snapshot;
 530         bool                    seen_this_pos;
 531         u64                     count;
 532 };
 533
 534 struct inode_walker {
 535         bool                            first_this_inode;
 536         bool                            recalculate_sums;
 537         struct bpos                     last_pos;
 538
 539         DARRAY(struct inode_walker_entry) inodes;
 540 };
 541
 542 static void inode_walker_exit(struct inode_walker *w)
 543 {
 544         darray_exit(&w->inodes);
 545 }
 546
 547 static struct inode_walker inode_walker_init(void)
 548 {
 549         return (struct inode_walker) { 0, };
 550 }
 551
 552 static int add_inode(struct bch_fs *c, struct inode_walker *w,
 553                      struct bkey_s_c inode)
 554 {
 555         struct bch_inode_unpacked u;
 556
 557         BUG_ON(bch2_inode_unpack(inode, &u));
 558
 559         return darray_push(&w->inodes, ((struct inode_walker_entry) {
 560                 .inode          = u,
 561                 .snapshot       = bch2_snapshot_equiv(c, inode.k->p.snapshot),
 562         }));
 563 }
 564
 565 static int get_inodes_all_snapshots(struct btree_trans *trans,
 566                                     struct inode_walker *w, u64 inum)
 567 {
 568         struct bch_fs *c = trans->c;
 569         struct btree_iter iter;
 570         struct bkey_s_c k;
 571         int ret;
 572
 573         w->recalculate_sums = false;
 574         w->inodes.nr = 0;
 575
 576         for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum),
 577                                      BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
 578                 if (k.k->p.offset != inum)
 579                         break;
 580
 581                 if (bkey_is_inode(k.k))
 582                         add_inode(c, w, k);
 583         }
 584         bch2_trans_iter_exit(trans, &iter);
 585
 586         if (ret)
 587                 return ret;
 588
 589         w->first_this_inode = true;
 590         return 0;
 591 }
 592
 593 static struct inode_walker_entry *
 594 lookup_inode_for_snapshot(struct bch_fs *c, struct inode_walker *w,
 595                           u32 snapshot, bool is_whiteout)
 596 {
 597         struct inode_walker_entry *i;
 598
 599         snapshot = bch2_snapshot_equiv(c, snapshot);
 600
 601         __darray_for_each(w->inodes, i)
 602                 if (bch2_snapshot_is_ancestor(c, snapshot, i->snapshot))
 603                         goto found;
 604
 605         return NULL;
 606 found:
 607         BUG_ON(snapshot > i->snapshot);
 608
 609         if (snapshot != i->snapshot && !is_whiteout) {
 610                 struct inode_walker_entry new = *i;
 611                 size_t pos;
 612                 int ret;
 613
 614                 new.snapshot = snapshot;
 615                 new.count = 0;
 616
 617                 bch_info(c, "have key for inode %llu:%u but have inode in ancestor snapshot %u",
 618                          w->last_pos.inode, snapshot, i->snapshot);
 619
 620                 while (i > w->inodes.data && i[-1].snapshot > snapshot)
 621                         --i;
 622
 623                 pos = i - w->inodes.data;
 624                 ret = darray_insert_item(&w->inodes, pos, new);
 625                 if (ret)
 626                         return ERR_PTR(ret);
 627
 628                 i = w->inodes.data + pos;
 629         }
 630
 631         return i;
 632 }
 633
 634 static struct inode_walker_entry *walk_inode(struct btree_trans *trans,
 635                                              struct inode_walker *w, struct bpos pos,
 636                                              bool is_whiteout)
 637 {
 638         if (w->last_pos.inode != pos.inode) {
 639                 int ret = get_inodes_all_snapshots(trans, w, pos.inode);
 640                 if (ret)
 641                         return ERR_PTR(ret);
 642         } else if (bkey_cmp(w->last_pos, pos)) {
 643                 darray_for_each(w->inodes, i)
 644                         i->seen_this_pos = false;
 645         }
 646
 647         w->last_pos = pos;
 648
 649         return lookup_inode_for_snapshot(trans->c, w, pos.snapshot, is_whiteout);
 650 }
 651
 652 static int __get_visible_inodes(struct btree_trans *trans,
 653                                 struct inode_walker *w,
 654                                 struct snapshots_seen *s,
 655                                 u64 inum)
 656 {
 657         struct bch_fs *c = trans->c;
 658         struct btree_iter iter;
 659         struct bkey_s_c k;
 660         int ret;
 661
 662         w->inodes.nr = 0;
 663
 664         for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum),
 665                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
 666                 u32 equiv = bch2_snapshot_equiv(c, k.k->p.snapshot);
 667
 668                 if (k.k->p.offset != inum)
 669                         break;
 670
 671                 if (!ref_visible(c, s, s->pos.snapshot, equiv))
 672                         continue;
 673
 674                 if (bkey_is_inode(k.k))
 675                         add_inode(c, w, k);
 676
 677                 if (equiv >= s->pos.snapshot)
 678                         break;
 679         }
 680         bch2_trans_iter_exit(trans, &iter);
 681
 682         return ret;
 683 }
 684
 685 static int check_key_has_snapshot(struct btree_trans *trans,
 686                                   struct btree_iter *iter,
 687                                   struct bkey_s_c k)
 688 {
 689         struct bch_fs *c = trans->c;
 690         struct printbuf buf = PRINTBUF;
 691         int ret = 0;
 692
 693         if (mustfix_fsck_err_on(!bch2_snapshot_equiv(c, k.k->p.snapshot), c,
 694                                 bkey_in_missing_snapshot,
 695                                 "key in missing snapshot: %s",
 696                                 (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
 697                 ret = bch2_btree_delete_at(trans, iter,
 698                                             BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?: 1;
 699 fsck_err:
 700         printbuf_exit(&buf);
 701         return ret;
 702 }
 703
 704 static int hash_redo_key(struct btree_trans *trans,
 705                          const struct bch_hash_desc desc,
 706                          struct bch_hash_info *hash_info,
 707                          struct btree_iter *k_iter, struct bkey_s_c k)
 708 {
 709         struct bkey_i *delete;
 710         struct bkey_i *tmp;
 711
 712         delete = bch2_trans_kmalloc(trans, sizeof(*delete));
 713         if (IS_ERR(delete))
 714                 return PTR_ERR(delete);
 715
 716         tmp = bch2_bkey_make_mut_noupdate(trans, k);
 717         if (IS_ERR(tmp))
 718                 return PTR_ERR(tmp);
 719
 720         bkey_init(&delete->k);
 721         delete->k.p = k_iter->pos;
 722         return  bch2_btree_iter_traverse(k_iter) ?:
 723                 bch2_trans_update(trans, k_iter, delete, 0) ?:
 724                 bch2_hash_set_snapshot(trans, desc, hash_info,
 725                                        (subvol_inum) { 0, k.k->p.inode },
 726                                        k.k->p.snapshot, tmp,
 727                                        BCH_HASH_SET_MUST_CREATE,
 728                                        BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
 729                 bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
 730 }
 731
 732 static int hash_check_key(struct btree_trans *trans,
 733                           const struct bch_hash_desc desc,
 734                           struct bch_hash_info *hash_info,
 735                           struct btree_iter *k_iter, struct bkey_s_c hash_k)
 736 {
 737         struct bch_fs *c = trans->c;
 738         struct btree_iter iter = { NULL };
 739         struct printbuf buf = PRINTBUF;
 740         struct bkey_s_c k;
 741         u64 hash;
 742         int ret = 0;
 743
 744         if (hash_k.k->type != desc.key_type)
 745                 return 0;
 746
 747         hash = desc.hash_bkey(hash_info, hash_k);
 748
 749         if (likely(hash == hash_k.k->p.offset))
 750                 return 0;
 751
 752         if (hash_k.k->p.offset < hash)
 753                 goto bad_hash;
 754
 755         for_each_btree_key_norestart(trans, iter, desc.btree_id,
 756                                      SPOS(hash_k.k->p.inode, hash, hash_k.k->p.snapshot),
 757                                      BTREE_ITER_SLOTS, k, ret) {
 758                 if (bkey_eq(k.k->p, hash_k.k->p))
 759                         break;
 760
 761                 if (fsck_err_on(k.k->type == desc.key_type &&
 762                                 !desc.cmp_bkey(k, hash_k), c,
 763                                 hash_table_key_duplicate,
 764                                 "duplicate hash table keys:\n%s",
 765                                 (printbuf_reset(&buf),
 766                                  bch2_bkey_val_to_text(&buf, c, hash_k),
 767                                  buf.buf))) {
 768                         ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0) ?: 1;
 769                         break;
 770                 }
 771
 772                 if (bkey_deleted(k.k)) {
 773                         bch2_trans_iter_exit(trans, &iter);
 774                         goto bad_hash;
 775                 }
 776         }
 777 out:
 778         bch2_trans_iter_exit(trans, &iter);
 779         printbuf_exit(&buf);
 780         return ret;
 781 bad_hash:
 782         if (fsck_err(c, hash_table_key_wrong_offset,
 783                      "hash table key at wrong offset: btree %s inode %llu offset %llu, hashed to %llu\n%s",
 784                      bch2_btree_id_str(desc.btree_id), hash_k.k->p.inode, hash_k.k->p.offset, hash,
 785                      (printbuf_reset(&buf),
 786                       bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf))) {
 787                 ret = hash_redo_key(trans, desc, hash_info, k_iter, hash_k);
 788                 bch_err_fn(c, ret);
 789                 if (ret)
 790                         return ret;
 791                 ret = -BCH_ERR_transaction_restart_nested;
 792         }
 793 fsck_err:
 794         goto out;
 795 }
 796
 797 static int check_inode_deleted_list(struct btree_trans *trans, struct bpos p)
 798 {
 799         struct btree_iter iter;
 800         struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_deleted_inodes, p, 0);
 801         int ret = bkey_err(k);
 802         if (ret)
 803                 return ret;
 804
 805         bch2_trans_iter_exit(trans, &iter);
 806         return k.k->type == KEY_TYPE_set;
 807 }
 808
 809 static int check_inode(struct btree_trans *trans,
 810                        struct btree_iter *iter,
 811                        struct bkey_s_c k,
 812                        struct bch_inode_unpacked *prev,
 813                        struct snapshots_seen *s,
 814                        bool full)
 815 {
 816         struct bch_fs *c = trans->c;
 817         struct bch_inode_unpacked u;
 818         bool do_update = false;
 819         int ret;
 820
 821         ret = check_key_has_snapshot(trans, iter, k);
 822         if (ret < 0)
 823                 goto err;
 824         if (ret)
 825                 return 0;
 826
 827         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
 828         if (ret)
 829                 goto err;
 830
 831         if (!bkey_is_inode(k.k))
 832                 return 0;
 833
 834         BUG_ON(bch2_inode_unpack(k, &u));
 835
 836         if (!full &&
 837             !(u.bi_flags & (BCH_INODE_i_size_dirty|
 838                             BCH_INODE_i_sectors_dirty|
 839                             BCH_INODE_unlinked)))
 840                 return 0;
 841
 842         if (prev->bi_inum != u.bi_inum)
 843                 *prev = u;
 844
 845         if (fsck_err_on(prev->bi_hash_seed      != u.bi_hash_seed ||
 846                         inode_d_type(prev)      != inode_d_type(&u),
 847                         c, inode_snapshot_mismatch,
 848                         "inodes in different snapshots don't match")) {
 849                 bch_err(c, "repair not implemented yet");
 850                 return -BCH_ERR_fsck_repair_unimplemented;
 851         }
 852
 853         if ((u.bi_flags & (BCH_INODE_i_size_dirty|BCH_INODE_unlinked)) &&
 854             bch2_key_has_snapshot_overwrites(trans, BTREE_ID_inodes, k.k->p)) {
 855                 struct bpos new_min_pos;
 856
 857                 ret = bch2_propagate_key_to_snapshot_leaves(trans, iter->btree_id, k, &new_min_pos);
 858                 if (ret)
 859                         goto err;
 860
 861                 u.bi_flags &= ~BCH_INODE_i_size_dirty|BCH_INODE_unlinked;
 862
 863                 ret = __write_inode(trans, &u, iter->pos.snapshot);
 864                 bch_err_msg(c, ret, "in fsck updating inode");
 865                 if (ret)
 866                         return ret;
 867
 868                 if (!bpos_eq(new_min_pos, POS_MIN))
 869                         bch2_btree_iter_set_pos(iter, bpos_predecessor(new_min_pos));
 870                 return 0;
 871         }
 872
 873         if (u.bi_flags & BCH_INODE_unlinked) {
 874                 ret = check_inode_deleted_list(trans, k.k->p);
 875                 if (ret < 0)
 876                         return ret;
 877
 878                 fsck_err_on(ret, c, unlinked_inode_not_on_deleted_list,
 879                             "inode %llu:%u unlinked, but not on deleted list",
 880                             u.bi_inum, k.k->p.snapshot);
 881                 ret = 0;
 882         }
 883
 884         if (u.bi_flags & BCH_INODE_unlinked &&
 885             (!c->sb.clean ||
 886              fsck_err(c, inode_unlinked_but_clean,
 887                       "filesystem marked clean, but inode %llu unlinked",
 888                       u.bi_inum))) {
 889                 ret = bch2_inode_rm_snapshot(trans, u.bi_inum, iter->pos.snapshot);
 890                 bch_err_msg(c, ret, "in fsck deleting inode");
 891                 return ret;
 892         }
 893
 894         if (u.bi_flags & BCH_INODE_i_size_dirty &&
 895             (!c->sb.clean ||
 896              fsck_err(c, inode_i_size_dirty_but_clean,
 897                       "filesystem marked clean, but inode %llu has i_size dirty",
 898                       u.bi_inum))) {
 899                 bch_verbose(c, "truncating inode %llu", u.bi_inum);
 900
 901                 /*
 902                  * XXX: need to truncate partial blocks too here - or ideally
 903                  * just switch units to bytes and that issue goes away
 904                  */
 905                 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
 906                                 SPOS(u.bi_inum, round_up(u.bi_size, block_bytes(c)) >> 9,
 907                                      iter->pos.snapshot),
 908                                 POS(u.bi_inum, U64_MAX),
 909                                 0, NULL);
 910                 bch_err_msg(c, ret, "in fsck truncating inode");
 911                 if (ret)
 912                         return ret;
 913
 914                 /*
 915                  * We truncated without our normal sector accounting hook, just
 916                  * make sure we recalculate it:
 917                  */
 918                 u.bi_flags |= BCH_INODE_i_sectors_dirty;
 919
 920                 u.bi_flags &= ~BCH_INODE_i_size_dirty;
 921                 do_update = true;
 922         }
 923
 924         if (u.bi_flags & BCH_INODE_i_sectors_dirty &&
 925             (!c->sb.clean ||
 926              fsck_err(c, inode_i_sectors_dirty_but_clean,
 927                       "filesystem marked clean, but inode %llu has i_sectors dirty",
 928                       u.bi_inum))) {
 929                 s64 sectors;
 930
 931                 bch_verbose(c, "recounting sectors for inode %llu",
 932                             u.bi_inum);
 933
 934                 sectors = bch2_count_inode_sectors(trans, u.bi_inum, iter->pos.snapshot);
 935                 if (sectors < 0) {
 936                         bch_err_msg(c, sectors, "in fsck recounting inode sectors");
 937                         return sectors;
 938                 }
 939
 940                 u.bi_sectors = sectors;
 941                 u.bi_flags &= ~BCH_INODE_i_sectors_dirty;
 942                 do_update = true;
 943         }
 944
 945         if (u.bi_flags & BCH_INODE_backptr_untrusted) {
 946                 u.bi_dir = 0;
 947                 u.bi_dir_offset = 0;
 948                 u.bi_flags &= ~BCH_INODE_backptr_untrusted;
 949                 do_update = true;
 950         }
 951
 952         if (do_update) {
 953                 ret = __write_inode(trans, &u, iter->pos.snapshot);
 954                 bch_err_msg(c, ret, "in fsck updating inode");
 955                 if (ret)
 956                         return ret;
 957         }
 958 err:
 959 fsck_err:
 960         bch_err_fn(c, ret);
 961         return ret;
 962 }
 963
 964 int bch2_check_inodes(struct bch_fs *c)
 965 {
 966         bool full = c->opts.fsck;
 967         struct bch_inode_unpacked prev = { 0 };
 968         struct snapshots_seen s;
 969
 970         snapshots_seen_init(&s);
 971
 972         int ret = bch2_trans_run(c,
 973                 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
 974                                 POS_MIN,
 975                                 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
 976                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
 977                         check_inode(trans, &iter, k, &prev, &s, full)));
 978
 979         snapshots_seen_exit(&s);
 980         bch_err_fn(c, ret);
 981         return ret;
 982 }
 983
 984 static struct bkey_s_c_dirent dirent_get_by_pos(struct btree_trans *trans,
 985                                                 struct btree_iter *iter,
 986                                                 struct bpos pos)
 987 {
 988         return bch2_bkey_get_iter_typed(trans, iter, BTREE_ID_dirents, pos, 0, dirent);
 989 }
 990
 991 static bool inode_points_to_dirent(struct bch_inode_unpacked *inode,
 992                                    struct bkey_s_c_dirent d)
 993 {
 994         return  inode->bi_dir           == d.k->p.inode &&
 995                 inode->bi_dir_offset    == d.k->p.offset;
 996 }
 997
 998 static bool dirent_points_to_inode(struct bkey_s_c_dirent d,
 999                                    struct bch_inode_unpacked *inode)
1000 {
1001         return d.v->d_type == DT_SUBVOL
1002                 ? le32_to_cpu(d.v->d_child_subvol)      == inode->bi_subvol
1003                 : le64_to_cpu(d.v->d_inum)              == inode->bi_inum;
1004 }
1005
1006 static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
1007 {
1008         struct bch_fs *c = trans->c;
1009         u32 restart_count = trans->restart_count;
1010         int ret = 0;
1011         s64 count2;
1012
1013         darray_for_each(w->inodes, i) {
1014                 if (i->inode.bi_sectors == i->count)
1015                         continue;
1016
1017                 count2 = bch2_count_inode_sectors(trans, w->last_pos.inode, i->snapshot);
1018
1019                 if (w->recalculate_sums)
1020                         i->count = count2;
1021
1022                 if (i->count != count2) {
1023                         bch_err(c, "fsck counted i_sectors wrong for inode %llu:%u: got %llu should be %llu",
1024                                 w->last_pos.inode, i->snapshot, i->count, count2);
1025                         return -BCH_ERR_internal_fsck_err;
1026                 }
1027
1028                 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_sectors_dirty),
1029                                 c, inode_i_sectors_wrong,
1030                                 "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
1031                                 w->last_pos.inode, i->snapshot,
1032                                 i->inode.bi_sectors, i->count)) {
1033                         i->inode.bi_sectors = i->count;
1034                         ret = fsck_write_inode(trans, &i->inode, i->snapshot);
1035                         if (ret)
1036                                 break;
1037                 }
1038         }
1039 fsck_err:
1040         bch_err_fn(c, ret);
1041         return ret ?: trans_was_restarted(trans, restart_count);
1042 }
1043
1044 struct extent_end {
1045         u32                     snapshot;
1046         u64                     offset;
1047         struct snapshots_seen   seen;
1048 };
1049
1050 struct extent_ends {
1051         struct bpos                     last_pos;
1052         DARRAY(struct extent_end)       e;
1053 };
1054
1055 static void extent_ends_reset(struct extent_ends *extent_ends)
1056 {
1057         darray_for_each(extent_ends->e, i)
1058                 snapshots_seen_exit(&i->seen);
1059         extent_ends->e.nr = 0;
1060 }
1061
1062 static void extent_ends_exit(struct extent_ends *extent_ends)
1063 {
1064         extent_ends_reset(extent_ends);
1065         darray_exit(&extent_ends->e);
1066 }
1067
1068 static void extent_ends_init(struct extent_ends *extent_ends)
1069 {
1070         memset(extent_ends, 0, sizeof(*extent_ends));
1071 }
1072
1073 static int extent_ends_at(struct bch_fs *c,
1074                           struct extent_ends *extent_ends,
1075                           struct snapshots_seen *seen,
1076                           struct bkey_s_c k)
1077 {
1078         struct extent_end *i, n = (struct extent_end) {
1079                 .offset         = k.k->p.offset,
1080                 .snapshot       = k.k->p.snapshot,
1081                 .seen           = *seen,
1082         };
1083
1084         n.seen.ids.data = kmemdup(seen->ids.data,
1085                               sizeof(seen->ids.data[0]) * seen->ids.size,
1086                               GFP_KERNEL);
1087         if (!n.seen.ids.data)
1088                 return -BCH_ERR_ENOMEM_fsck_extent_ends_at;
1089
1090         __darray_for_each(extent_ends->e, i) {
1091                 if (i->snapshot == k.k->p.snapshot) {
1092                         snapshots_seen_exit(&i->seen);
1093                         *i = n;
1094                         return 0;
1095                 }
1096
1097                 if (i->snapshot >= k.k->p.snapshot)
1098                         break;
1099         }
1100
1101         return darray_insert_item(&extent_ends->e, i - extent_ends->e.data, n);
1102 }
1103
1104 static int overlapping_extents_found(struct btree_trans *trans,
1105                                      enum btree_id btree,
1106                                      struct bpos pos1, struct snapshots_seen *pos1_seen,
1107                                      struct bkey pos2,
1108                                      bool *fixed,
1109                                      struct extent_end *extent_end)
1110 {
1111         struct bch_fs *c = trans->c;
1112         struct printbuf buf = PRINTBUF;
1113         struct btree_iter iter1, iter2 = { NULL };
1114         struct bkey_s_c k1, k2;
1115         int ret;
1116
1117         BUG_ON(bkey_le(pos1, bkey_start_pos(&pos2)));
1118
1119         bch2_trans_iter_init(trans, &iter1, btree, pos1,
1120                              BTREE_ITER_ALL_SNAPSHOTS|
1121                              BTREE_ITER_NOT_EXTENTS);
1122         k1 = bch2_btree_iter_peek_upto(&iter1, POS(pos1.inode, U64_MAX));
1123         ret = bkey_err(k1);
1124         if (ret)
1125                 goto err;
1126
1127         prt_str(&buf, "\n  ");
1128         bch2_bkey_val_to_text(&buf, c, k1);
1129
1130         if (!bpos_eq(pos1, k1.k->p)) {
1131                 prt_str(&buf, "\n  wanted\n  ");
1132                 bch2_bpos_to_text(&buf, pos1);
1133                 prt_str(&buf, "\n  ");
1134                 bch2_bkey_to_text(&buf, &pos2);
1135
1136                 bch_err(c, "%s: error finding first overlapping extent when repairing, got%s",
1137                         __func__, buf.buf);
1138                 ret = -BCH_ERR_internal_fsck_err;
1139                 goto err;
1140         }
1141
1142         bch2_trans_copy_iter(&iter2, &iter1);
1143
1144         while (1) {
1145                 bch2_btree_iter_advance(&iter2);
1146
1147                 k2 = bch2_btree_iter_peek_upto(&iter2, POS(pos1.inode, U64_MAX));
1148                 ret = bkey_err(k2);
1149                 if (ret)
1150                         goto err;
1151
1152                 if (bpos_ge(k2.k->p, pos2.p))
1153                         break;
1154         }
1155
1156         prt_str(&buf, "\n  ");
1157         bch2_bkey_val_to_text(&buf, c, k2);
1158
1159         if (bpos_gt(k2.k->p, pos2.p) ||
1160             pos2.size != k2.k->size) {
1161                 bch_err(c, "%s: error finding seconding overlapping extent when repairing%s",
1162                         __func__, buf.buf);
1163                 ret = -BCH_ERR_internal_fsck_err;
1164                 goto err;
1165         }
1166
1167         prt_printf(&buf, "\n  overwriting %s extent",
1168                    pos1.snapshot >= pos2.p.snapshot ? "first" : "second");
1169
1170         if (fsck_err(c, extent_overlapping,
1171                      "overlapping extents%s", buf.buf)) {
1172                 struct btree_iter *old_iter = &iter1;
1173                 struct disk_reservation res = { 0 };
1174
1175                 if (pos1.snapshot < pos2.p.snapshot) {
1176                         old_iter = &iter2;
1177                         swap(k1, k2);
1178                 }
1179
1180                 trans->extra_disk_res += bch2_bkey_sectors_compressed(k2);
1181
1182                 ret =   bch2_trans_update_extent_overwrite(trans, old_iter,
1183                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE,
1184                                 k1, k2) ?:
1185                         bch2_trans_commit(trans, &res, NULL, BCH_TRANS_COMMIT_no_enospc);
1186                 bch2_disk_reservation_put(c, &res);
1187
1188                 if (ret)
1189                         goto err;
1190
1191                 *fixed = true;
1192
1193                 if (pos1.snapshot == pos2.p.snapshot) {
1194                         /*
1195                          * We overwrote the first extent, and did the overwrite
1196                          * in the same snapshot:
1197                          */
1198                         extent_end->offset = bkey_start_offset(&pos2);
1199                 } else if (pos1.snapshot > pos2.p.snapshot) {
1200                         /*
1201                          * We overwrote the first extent in pos2's snapshot:
1202                          */
1203                         ret = snapshots_seen_add_inorder(c, pos1_seen, pos2.p.snapshot);
1204                 } else {
1205                         /*
1206                          * We overwrote the second extent - restart
1207                          * check_extent() from the top:
1208                          */
1209                         ret = -BCH_ERR_transaction_restart_nested;
1210                 }
1211         }
1212 fsck_err:
1213 err:
1214         bch2_trans_iter_exit(trans, &iter2);
1215         bch2_trans_iter_exit(trans, &iter1);
1216         printbuf_exit(&buf);
1217         return ret;
1218 }
1219
1220 static int check_overlapping_extents(struct btree_trans *trans,
1221                               struct snapshots_seen *seen,
1222                               struct extent_ends *extent_ends,
1223                               struct bkey_s_c k,
1224                               u32 equiv,
1225                               struct btree_iter *iter,
1226                               bool *fixed)
1227 {
1228         struct bch_fs *c = trans->c;
1229         int ret = 0;
1230
1231         /* transaction restart, running again */
1232         if (bpos_eq(extent_ends->last_pos, k.k->p))
1233                 return 0;
1234
1235         if (extent_ends->last_pos.inode != k.k->p.inode)
1236                 extent_ends_reset(extent_ends);
1237
1238         darray_for_each(extent_ends->e, i) {
1239                 if (i->offset <= bkey_start_offset(k.k))
1240                         continue;
1241
1242                 if (!ref_visible2(c,
1243                                   k.k->p.snapshot, seen,
1244                                   i->snapshot, &i->seen))
1245                         continue;
1246
1247                 ret = overlapping_extents_found(trans, iter->btree_id,
1248                                                 SPOS(iter->pos.inode,
1249                                                      i->offset,
1250                                                      i->snapshot),
1251                                                 &i->seen,
1252                                                 *k.k, fixed, i);
1253                 if (ret)
1254                         goto err;
1255         }
1256
1257         ret = extent_ends_at(c, extent_ends, seen, k);
1258         if (ret)
1259                 goto err;
1260
1261         extent_ends->last_pos = k.k->p;
1262 err:
1263         return ret;
1264 }
1265
1266 static int check_extent_overbig(struct btree_trans *trans, struct btree_iter *iter,
1267                                 struct bkey_s_c k)
1268 {
1269         struct bch_fs *c = trans->c;
1270         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1271         struct bch_extent_crc_unpacked crc;
1272         const union bch_extent_entry *i;
1273         unsigned encoded_extent_max_sectors = c->opts.encoded_extent_max >> 9;
1274
1275         bkey_for_each_crc(k.k, ptrs, crc, i)
1276                 if (crc_is_encoded(crc) &&
1277                     crc.uncompressed_size > encoded_extent_max_sectors) {
1278                         struct printbuf buf = PRINTBUF;
1279
1280                         bch2_bkey_val_to_text(&buf, c, k);
1281                         bch_err(c, "overbig encoded extent, please report this:\n  %s", buf.buf);
1282                         printbuf_exit(&buf);
1283                 }
1284
1285         return 0;
1286 }
1287
1288 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
1289                         struct bkey_s_c k,
1290                         struct inode_walker *inode,
1291                         struct snapshots_seen *s,
1292                         struct extent_ends *extent_ends)
1293 {
1294         struct bch_fs *c = trans->c;
1295         struct inode_walker_entry *i;
1296         struct printbuf buf = PRINTBUF;
1297         struct bpos equiv = k.k->p;
1298         int ret = 0;
1299
1300         equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1301
1302         ret = check_key_has_snapshot(trans, iter, k);
1303         if (ret) {
1304                 ret = ret < 0 ? ret : 0;
1305                 goto out;
1306         }
1307
1308         if (inode->last_pos.inode != k.k->p.inode) {
1309                 ret = check_i_sectors(trans, inode);
1310                 if (ret)
1311                         goto err;
1312         }
1313
1314         i = walk_inode(trans, inode, equiv, k.k->type == KEY_TYPE_whiteout);
1315         ret = PTR_ERR_OR_ZERO(i);
1316         if (ret)
1317                 goto err;
1318
1319         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1320         if (ret)
1321                 goto err;
1322
1323         if (k.k->type != KEY_TYPE_whiteout) {
1324                 if (fsck_err_on(!i, c, extent_in_missing_inode,
1325                                 "extent in missing inode:\n  %s",
1326                                 (printbuf_reset(&buf),
1327                                  bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1328                         goto delete;
1329
1330                 if (fsck_err_on(i &&
1331                                 !S_ISREG(i->inode.bi_mode) &&
1332                                 !S_ISLNK(i->inode.bi_mode),
1333                                 c, extent_in_non_reg_inode,
1334                                 "extent in non regular inode mode %o:\n  %s",
1335                                 i->inode.bi_mode,
1336                                 (printbuf_reset(&buf),
1337                                  bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1338                         goto delete;
1339
1340                 ret = check_overlapping_extents(trans, s, extent_ends, k,
1341                                                 equiv.snapshot, iter,
1342                                                 &inode->recalculate_sums);
1343                 if (ret)
1344                         goto err;
1345         }
1346
1347         /*
1348          * Check inodes in reverse order, from oldest snapshots to newest,
1349          * starting from the inode that matches this extent's snapshot. If we
1350          * didn't have one, iterate over all inodes:
1351          */
1352         if (!i)
1353                 i = inode->inodes.data + inode->inodes.nr - 1;
1354
1355         for (;
1356              inode->inodes.data && i >= inode->inodes.data;
1357              --i) {
1358                 if (i->snapshot > equiv.snapshot ||
1359                     !key_visible_in_snapshot(c, s, i->snapshot, equiv.snapshot))
1360                         continue;
1361
1362                 if (k.k->type != KEY_TYPE_whiteout) {
1363                         if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_size_dirty) &&
1364                                         k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9 &&
1365                                         !bkey_extent_is_reservation(k),
1366                                         c, extent_past_end_of_inode,
1367                                         "extent type past end of inode %llu:%u, i_size %llu\n  %s",
1368                                         i->inode.bi_inum, i->snapshot, i->inode.bi_size,
1369                                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1370                                 struct btree_iter iter2;
1371
1372                                 bch2_trans_copy_iter(&iter2, iter);
1373                                 bch2_btree_iter_set_snapshot(&iter2, i->snapshot);
1374                                 ret =   bch2_btree_iter_traverse(&iter2) ?:
1375                                         bch2_btree_delete_at(trans, &iter2,
1376                                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1377                                 bch2_trans_iter_exit(trans, &iter2);
1378                                 if (ret)
1379                                         goto err;
1380
1381                                 iter->k.type = KEY_TYPE_whiteout;
1382                         }
1383
1384                         if (bkey_extent_is_allocation(k.k))
1385                                 i->count += k.k->size;
1386                 }
1387
1388                 i->seen_this_pos = true;
1389         }
1390 out:
1391 err:
1392 fsck_err:
1393         printbuf_exit(&buf);
1394         bch_err_fn(c, ret);
1395         return ret;
1396 delete:
1397         ret = bch2_btree_delete_at(trans, iter, BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1398         goto out;
1399 }
1400
1401 /*
1402  * Walk extents: verify that extents have a corresponding S_ISREG inode, and
1403  * that i_size an i_sectors are consistent
1404  */
1405 int bch2_check_extents(struct bch_fs *c)
1406 {
1407         struct inode_walker w = inode_walker_init();
1408         struct snapshots_seen s;
1409         struct extent_ends extent_ends;
1410         struct disk_reservation res = { 0 };
1411
1412         snapshots_seen_init(&s);
1413         extent_ends_init(&extent_ends);
1414
1415         int ret = bch2_trans_run(c,
1416                 for_each_btree_key_commit(trans, iter, BTREE_ID_extents,
1417                                 POS(BCACHEFS_ROOT_INO, 0),
1418                                 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1419                                 &res, NULL,
1420                                 BCH_TRANS_COMMIT_no_enospc, ({
1421                         bch2_disk_reservation_put(c, &res);
1422                         check_extent(trans, &iter, k, &w, &s, &extent_ends) ?:
1423                         check_extent_overbig(trans, &iter, k);
1424                 })) ?:
1425                 check_i_sectors(trans, &w));
1426
1427         bch2_disk_reservation_put(c, &res);
1428         extent_ends_exit(&extent_ends);
1429         inode_walker_exit(&w);
1430         snapshots_seen_exit(&s);
1431
1432         bch_err_fn(c, ret);
1433         return ret;
1434 }
1435
1436 int bch2_check_indirect_extents(struct bch_fs *c)
1437 {
1438         struct disk_reservation res = { 0 };
1439
1440         int ret = bch2_trans_run(c,
1441                 for_each_btree_key_commit(trans, iter, BTREE_ID_reflink,
1442                                 POS_MIN,
1443                                 BTREE_ITER_PREFETCH, k,
1444                                 &res, NULL,
1445                                 BCH_TRANS_COMMIT_no_enospc, ({
1446                         bch2_disk_reservation_put(c, &res);
1447                         check_extent_overbig(trans, &iter, k);
1448                 })));
1449
1450         bch2_disk_reservation_put(c, &res);
1451         bch_err_fn(c, ret);
1452         return ret;
1453 }
1454
1455 static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
1456 {
1457         struct bch_fs *c = trans->c;
1458         u32 restart_count = trans->restart_count;
1459         int ret = 0;
1460         s64 count2;
1461
1462         darray_for_each(w->inodes, i) {
1463                 if (i->inode.bi_nlink == i->count)
1464                         continue;
1465
1466                 count2 = bch2_count_subdirs(trans, w->last_pos.inode, i->snapshot);
1467                 if (count2 < 0)
1468                         return count2;
1469
1470                 if (i->count != count2) {
1471                         bch_err(c, "fsck counted subdirectories wrong: got %llu should be %llu",
1472                                 i->count, count2);
1473                         i->count = count2;
1474                         if (i->inode.bi_nlink == i->count)
1475                                 continue;
1476                 }
1477
1478                 if (fsck_err_on(i->inode.bi_nlink != i->count,
1479                                 c, inode_dir_wrong_nlink,
1480                                 "directory %llu:%u with wrong i_nlink: got %u, should be %llu",
1481                                 w->last_pos.inode, i->snapshot, i->inode.bi_nlink, i->count)) {
1482                         i->inode.bi_nlink = i->count;
1483                         ret = fsck_write_inode(trans, &i->inode, i->snapshot);
1484                         if (ret)
1485                                 break;
1486                 }
1487         }
1488 fsck_err:
1489         bch_err_fn(c, ret);
1490         return ret ?: trans_was_restarted(trans, restart_count);
1491 }
1492
1493 static int check_dirent_target(struct btree_trans *trans,
1494                                struct btree_iter *iter,
1495                                struct bkey_s_c_dirent d,
1496                                struct bch_inode_unpacked *target,
1497                                u32 target_snapshot)
1498 {
1499         struct bch_fs *c = trans->c;
1500         struct bkey_i_dirent *n;
1501         struct printbuf buf = PRINTBUF;
1502         struct btree_iter bp_iter = { NULL };
1503         int ret = 0;
1504
1505         if (!target->bi_dir &&
1506             !target->bi_dir_offset) {
1507                 target->bi_dir          = d.k->p.inode;
1508                 target->bi_dir_offset   = d.k->p.offset;
1509
1510                 ret = __write_inode(trans, target, target_snapshot);
1511                 if (ret)
1512                         goto err;
1513         }
1514
1515         if (!inode_points_to_dirent(target, d)) {
1516                 struct bkey_s_c_dirent bp_dirent = dirent_get_by_pos(trans, &bp_iter,
1517                                       SPOS(target->bi_dir, target->bi_dir_offset, target_snapshot));
1518                 ret = bkey_err(bp_dirent);
1519                 if (ret && !bch2_err_matches(ret, ENOENT))
1520                         goto err;
1521
1522                 bool backpointer_exists = !ret;
1523                 ret = 0;
1524
1525                 bch2_bkey_val_to_text(&buf, c, d.s_c);
1526                 prt_newline(&buf);
1527                 if (backpointer_exists)
1528                         bch2_bkey_val_to_text(&buf, c, bp_dirent.s_c);
1529
1530                 if (fsck_err_on(S_ISDIR(target->bi_mode) && backpointer_exists,
1531                                 c, inode_dir_multiple_links,
1532                                 "directory %llu:%u with multiple links\n%s",
1533                                 target->bi_inum, target_snapshot, buf.buf)) {
1534                         ret = __remove_dirent(trans, d.k->p);
1535                         goto out;
1536                 }
1537
1538                 /*
1539                  * hardlinked file with nlink 0:
1540                  * We're just adjusting nlink here so check_nlinks() will pick
1541                  * it up, it ignores inodes with nlink 0
1542                  */
1543                 if (fsck_err_on(backpointer_exists && !target->bi_nlink,
1544                                 c, inode_multiple_links_but_nlink_0,
1545                                 "inode %llu:%u type %s has multiple links but i_nlink 0\n%s",
1546                                 target->bi_inum, target_snapshot, bch2_d_types[d.v->d_type], buf.buf)) {
1547                         target->bi_nlink++;
1548                         target->bi_flags &= ~BCH_INODE_unlinked;
1549
1550                         ret = __write_inode(trans, target, target_snapshot);
1551                         if (ret)
1552                                 goto err;
1553                 }
1554
1555                 if (fsck_err_on(!backpointer_exists,
1556                                 c, inode_wrong_backpointer,
1557                                 "inode %llu:%u has wrong backpointer:\n"
1558                                 "got       %llu:%llu\n"
1559                                 "should be %llu:%llu",
1560                                 target->bi_inum, target_snapshot,
1561                                 target->bi_dir,
1562                                 target->bi_dir_offset,
1563                                 d.k->p.inode,
1564                                 d.k->p.offset)) {
1565                         target->bi_dir          = d.k->p.inode;
1566                         target->bi_dir_offset   = d.k->p.offset;
1567
1568                         ret = __write_inode(trans, target, target_snapshot);
1569                         if (ret)
1570                                 goto err;
1571                 }
1572         }
1573
1574         if (fsck_err_on(d.v->d_type != inode_d_type(target),
1575                         c, dirent_d_type_wrong,
1576                         "incorrect d_type: got %s, should be %s:\n%s",
1577                         bch2_d_type_str(d.v->d_type),
1578                         bch2_d_type_str(inode_d_type(target)),
1579                         (printbuf_reset(&buf),
1580                          bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
1581                 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1582                 ret = PTR_ERR_OR_ZERO(n);
1583                 if (ret)
1584                         goto err;
1585
1586                 bkey_reassemble(&n->k_i, d.s_c);
1587                 n->v.d_type = inode_d_type(target);
1588
1589                 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1590                 if (ret)
1591                         goto err;
1592
1593                 d = dirent_i_to_s_c(n);
1594         }
1595
1596         if (fsck_err_on(d.v->d_type == DT_SUBVOL &&
1597                         target->bi_parent_subvol != le32_to_cpu(d.v->d_parent_subvol),
1598                         c, dirent_d_parent_subvol_wrong,
1599                         "dirent has wrong d_parent_subvol field: got %u, should be %u",
1600                         le32_to_cpu(d.v->d_parent_subvol),
1601                         target->bi_parent_subvol)) {
1602                 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1603                 ret = PTR_ERR_OR_ZERO(n);
1604                 if (ret)
1605                         goto err;
1606
1607                 bkey_reassemble(&n->k_i, d.s_c);
1608                 n->v.d_parent_subvol = cpu_to_le32(target->bi_parent_subvol);
1609
1610                 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1611                 if (ret)
1612                         goto err;
1613
1614                 d = dirent_i_to_s_c(n);
1615         }
1616 out:
1617 err:
1618 fsck_err:
1619         bch2_trans_iter_exit(trans, &bp_iter);
1620         printbuf_exit(&buf);
1621         bch_err_fn(c, ret);
1622         return ret;
1623 }
1624
1625 static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
1626                         struct bkey_s_c k,
1627                         struct bch_hash_info *hash_info,
1628                         struct inode_walker *dir,
1629                         struct inode_walker *target,
1630                         struct snapshots_seen *s)
1631 {
1632         struct bch_fs *c = trans->c;
1633         struct bkey_s_c_dirent d;
1634         struct inode_walker_entry *i;
1635         struct printbuf buf = PRINTBUF;
1636         struct bpos equiv;
1637         int ret = 0;
1638
1639         ret = check_key_has_snapshot(trans, iter, k);
1640         if (ret) {
1641                 ret = ret < 0 ? ret : 0;
1642                 goto out;
1643         }
1644
1645         equiv = k.k->p;
1646         equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1647
1648         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1649         if (ret)
1650                 goto err;
1651
1652         if (k.k->type == KEY_TYPE_whiteout)
1653                 goto out;
1654
1655         if (dir->last_pos.inode != k.k->p.inode) {
1656                 ret = check_subdir_count(trans, dir);
1657                 if (ret)
1658                         goto err;
1659         }
1660
1661         BUG_ON(!btree_iter_path(trans, iter)->should_be_locked);
1662
1663         i = walk_inode(trans, dir, equiv, k.k->type == KEY_TYPE_whiteout);
1664         ret = PTR_ERR_OR_ZERO(i);
1665         if (ret < 0)
1666                 goto err;
1667
1668         if (dir->first_this_inode && dir->inodes.nr)
1669                 *hash_info = bch2_hash_info_init(c, &dir->inodes.data[0].inode);
1670         dir->first_this_inode = false;
1671
1672         if (fsck_err_on(!i, c, dirent_in_missing_dir_inode,
1673                         "dirent in nonexisting directory:\n%s",
1674                         (printbuf_reset(&buf),
1675                          bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1676                 ret = bch2_btree_delete_at(trans, iter,
1677                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1678                 goto out;
1679         }
1680
1681         if (!i)
1682                 goto out;
1683
1684         if (fsck_err_on(!S_ISDIR(i->inode.bi_mode),
1685                         c, dirent_in_non_dir_inode,
1686                         "dirent in non directory inode type %s:\n%s",
1687                         bch2_d_type_str(inode_d_type(&i->inode)),
1688                         (printbuf_reset(&buf),
1689                          bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1690                 ret = bch2_btree_delete_at(trans, iter, 0);
1691                 goto out;
1692         }
1693
1694         ret = hash_check_key(trans, bch2_dirent_hash_desc, hash_info, iter, k);
1695         if (ret < 0)
1696                 goto err;
1697         if (ret) {
1698                 /* dirent has been deleted */
1699                 ret = 0;
1700                 goto out;
1701         }
1702
1703         if (k.k->type != KEY_TYPE_dirent)
1704                 goto out;
1705
1706         d = bkey_s_c_to_dirent(k);
1707
1708         if (d.v->d_type == DT_SUBVOL) {
1709                 struct bch_inode_unpacked subvol_root;
1710                 u32 target_subvol = le32_to_cpu(d.v->d_child_subvol);
1711                 u32 target_snapshot;
1712                 u64 target_inum;
1713
1714                 ret = subvol_lookup(trans, target_subvol,
1715                                       &target_snapshot, &target_inum);
1716                 if (ret && !bch2_err_matches(ret, ENOENT))
1717                         goto err;
1718
1719                 if (fsck_err_on(ret, c, dirent_to_missing_subvol,
1720                                 "dirent points to missing subvolume %u",
1721                                 le32_to_cpu(d.v->d_child_subvol))) {
1722                         ret = __remove_dirent(trans, d.k->p);
1723                         goto err;
1724                 }
1725
1726                 ret = lookup_inode(trans, target_inum,
1727                                    &subvol_root, &target_snapshot);
1728                 if (ret && !bch2_err_matches(ret, ENOENT))
1729                         goto err;
1730
1731                 if (fsck_err_on(ret, c, subvol_to_missing_root,
1732                                 "subvolume %u points to missing subvolume root %llu",
1733                                 target_subvol,
1734                                 target_inum)) {
1735                         bch_err(c, "repair not implemented yet");
1736                         ret = -EINVAL;
1737                         goto err;
1738                 }
1739
1740                 if (fsck_err_on(subvol_root.bi_subvol != target_subvol,
1741                                 c, subvol_root_wrong_bi_subvol,
1742                                 "subvol root %llu has wrong bi_subvol field: got %u, should be %u",
1743                                 target_inum,
1744                                 subvol_root.bi_subvol, target_subvol)) {
1745                         subvol_root.bi_subvol = target_subvol;
1746                         ret = __write_inode(trans, &subvol_root, target_snapshot);
1747                         if (ret)
1748                                 goto err;
1749                 }
1750
1751                 ret = check_dirent_target(trans, iter, d, &subvol_root,
1752                                           target_snapshot);
1753                 if (ret)
1754                         goto err;
1755         } else {
1756                 ret = __get_visible_inodes(trans, target, s, le64_to_cpu(d.v->d_inum));
1757                 if (ret)
1758                         goto err;
1759
1760                 if (fsck_err_on(!target->inodes.nr,
1761                                 c, dirent_to_missing_inode,
1762                                 "dirent points to missing inode: (equiv %u)\n%s",
1763                                 equiv.snapshot,
1764                                 (printbuf_reset(&buf),
1765                                  bch2_bkey_val_to_text(&buf, c, k),
1766                                  buf.buf))) {
1767                         ret = __remove_dirent(trans, d.k->p);
1768                         if (ret)
1769                                 goto err;
1770                 }
1771
1772                 darray_for_each(target->inodes, i) {
1773                         ret = check_dirent_target(trans, iter, d,
1774                                                   &i->inode, i->snapshot);
1775                         if (ret)
1776                                 goto err;
1777                 }
1778         }
1779
1780         if (d.v->d_type == DT_DIR)
1781                 for_each_visible_inode(c, s, dir, equiv.snapshot, i)
1782                         i->count++;
1783
1784 out:
1785 err:
1786 fsck_err:
1787         printbuf_exit(&buf);
1788         bch_err_fn(c, ret);
1789         return ret;
1790 }
1791
1792 /*
1793  * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
1794  * validate d_type
1795  */
1796 int bch2_check_dirents(struct bch_fs *c)
1797 {
1798         struct inode_walker dir = inode_walker_init();
1799         struct inode_walker target = inode_walker_init();
1800         struct snapshots_seen s;
1801         struct bch_hash_info hash_info;
1802
1803         snapshots_seen_init(&s);
1804
1805         int ret = bch2_trans_run(c,
1806                 for_each_btree_key_commit(trans, iter, BTREE_ID_dirents,
1807                                 POS(BCACHEFS_ROOT_INO, 0),
1808                                 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
1809                                 k,
1810                                 NULL, NULL,
1811                                 BCH_TRANS_COMMIT_no_enospc,
1812                         check_dirent(trans, &iter, k, &hash_info, &dir, &target, &s)));
1813
1814         snapshots_seen_exit(&s);
1815         inode_walker_exit(&dir);
1816         inode_walker_exit(&target);
1817         bch_err_fn(c, ret);
1818         return ret;
1819 }
1820
1821 static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
1822                        struct bkey_s_c k,
1823                        struct bch_hash_info *hash_info,
1824                        struct inode_walker *inode)
1825 {
1826         struct bch_fs *c = trans->c;
1827         struct inode_walker_entry *i;
1828         int ret;
1829
1830         ret = check_key_has_snapshot(trans, iter, k);
1831         if (ret)
1832                 return ret;
1833
1834         i = walk_inode(trans, inode, k.k->p, k.k->type == KEY_TYPE_whiteout);
1835         ret = PTR_ERR_OR_ZERO(i);
1836         if (ret)
1837                 return ret;
1838
1839         if (inode->first_this_inode && inode->inodes.nr)
1840                 *hash_info = bch2_hash_info_init(c, &inode->inodes.data[0].inode);
1841         inode->first_this_inode = false;
1842
1843         if (fsck_err_on(!i, c, xattr_in_missing_inode,
1844                         "xattr for missing inode %llu",
1845                         k.k->p.inode))
1846                 return bch2_btree_delete_at(trans, iter, 0);
1847
1848         if (!i)
1849                 return 0;
1850
1851         ret = hash_check_key(trans, bch2_xattr_hash_desc, hash_info, iter, k);
1852 fsck_err:
1853         bch_err_fn(c, ret);
1854         return ret;
1855 }
1856
1857 /*
1858  * Walk xattrs: verify that they all have a corresponding inode
1859  */
1860 int bch2_check_xattrs(struct bch_fs *c)
1861 {
1862         struct inode_walker inode = inode_walker_init();
1863         struct bch_hash_info hash_info;
1864         int ret = 0;
1865
1866         ret = bch2_trans_run(c,
1867                 for_each_btree_key_commit(trans, iter, BTREE_ID_xattrs,
1868                         POS(BCACHEFS_ROOT_INO, 0),
1869                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
1870                         k,
1871                         NULL, NULL,
1872                         BCH_TRANS_COMMIT_no_enospc,
1873                 check_xattr(trans, &iter, k, &hash_info, &inode)));
1874         bch_err_fn(c, ret);
1875         return ret;
1876 }
1877
1878 static int check_root_trans(struct btree_trans *trans)
1879 {
1880         struct bch_fs *c = trans->c;
1881         struct bch_inode_unpacked root_inode;
1882         u32 snapshot;
1883         u64 inum;
1884         int ret;
1885
1886         ret = subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
1887         if (ret && !bch2_err_matches(ret, ENOENT))
1888                 return ret;
1889
1890         if (mustfix_fsck_err_on(ret, c, root_subvol_missing,
1891                                 "root subvol missing")) {
1892                 struct bkey_i_subvolume root_subvol;
1893
1894                 snapshot        = U32_MAX;
1895                 inum            = BCACHEFS_ROOT_INO;
1896
1897                 bkey_subvolume_init(&root_subvol.k_i);
1898                 root_subvol.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1899                 root_subvol.v.flags     = 0;
1900                 root_subvol.v.snapshot  = cpu_to_le32(snapshot);
1901                 root_subvol.v.inode     = cpu_to_le64(inum);
1902                 ret = bch2_btree_insert_trans(trans, BTREE_ID_subvolumes, &root_subvol.k_i, 0);
1903                 bch_err_msg(c, ret, "writing root subvol");
1904                 if (ret)
1905                         goto err;
1906         }
1907
1908         ret = lookup_inode(trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
1909         if (ret && !bch2_err_matches(ret, ENOENT))
1910                 return ret;
1911
1912         if (mustfix_fsck_err_on(ret, c, root_dir_missing,
1913                                 "root directory missing") ||
1914             mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode),
1915                                 c, root_inode_not_dir,
1916                                 "root inode not a directory")) {
1917                 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755,
1918                                 0, NULL);
1919                 root_inode.bi_inum = inum;
1920
1921                 ret = __write_inode(trans, &root_inode, snapshot);
1922                 bch_err_msg(c, ret, "writing root inode");
1923         }
1924 err:
1925 fsck_err:
1926         return ret;
1927 }
1928
1929 /* Get root directory, create if it doesn't exist: */
1930 int bch2_check_root(struct bch_fs *c)
1931 {
1932         int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1933                 check_root_trans(trans));
1934         bch_err_fn(c, ret);
1935         return ret;
1936 }
1937
1938 struct pathbuf_entry {
1939         u64     inum;
1940         u32     snapshot;
1941 };
1942
1943 typedef DARRAY(struct pathbuf_entry) pathbuf;
1944
1945 static bool path_is_dup(pathbuf *p, u64 inum, u32 snapshot)
1946 {
1947         darray_for_each(*p, i)
1948                 if (i->inum     == inum &&
1949                     i->snapshot == snapshot)
1950                         return true;
1951         return false;
1952 }
1953
1954 static int path_down(struct bch_fs *c, pathbuf *p,
1955                      u64 inum, u32 snapshot)
1956 {
1957         int ret = darray_push(p, ((struct pathbuf_entry) {
1958                 .inum           = inum,
1959                 .snapshot       = snapshot,
1960         }));
1961
1962         if (ret)
1963                 bch_err(c, "fsck: error allocating memory for pathbuf, size %zu",
1964                         p->size);
1965         return ret;
1966 }
1967
1968 /*
1969  * Check that a given inode is reachable from the root:
1970  *
1971  * XXX: we should also be verifying that inodes are in the right subvolumes
1972  */
1973 static int check_path(struct btree_trans *trans,
1974                       pathbuf *p,
1975                       struct bch_inode_unpacked *inode,
1976                       u32 snapshot)
1977 {
1978         struct bch_fs *c = trans->c;
1979         int ret = 0;
1980
1981         snapshot = bch2_snapshot_equiv(c, snapshot);
1982         p->nr = 0;
1983
1984         while (!(inode->bi_inum == BCACHEFS_ROOT_INO &&
1985                  inode->bi_subvol == BCACHEFS_ROOT_SUBVOL)) {
1986                 struct btree_iter dirent_iter;
1987                 struct bkey_s_c_dirent d;
1988                 u32 parent_snapshot = snapshot;
1989
1990                 if (inode->bi_subvol) {
1991                         u64 inum;
1992
1993                         ret = subvol_lookup(trans, inode->bi_parent_subvol,
1994                                             &parent_snapshot, &inum);
1995                         if (ret)
1996                                 break;
1997                 }
1998
1999                 d = dirent_get_by_pos(trans, &dirent_iter,
2000                                       SPOS(inode->bi_dir, inode->bi_dir_offset,
2001                                            parent_snapshot));
2002                 ret = bkey_err(d.s_c);
2003                 if (ret && !bch2_err_matches(ret, ENOENT))
2004                         break;
2005
2006                 if (!ret && !dirent_points_to_inode(d, inode)) {
2007                         bch2_trans_iter_exit(trans, &dirent_iter);
2008                         ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
2009                 }
2010
2011                 if (bch2_err_matches(ret, ENOENT)) {
2012                         if (fsck_err(c,  inode_unreachable,
2013                                      "unreachable inode %llu:%u, type %s nlink %u backptr %llu:%llu",
2014                                      inode->bi_inum, snapshot,
2015                                      bch2_d_type_str(inode_d_type(inode)),
2016                                      inode->bi_nlink,
2017                                      inode->bi_dir,
2018                                      inode->bi_dir_offset))
2019                                 ret = reattach_inode(trans, inode, snapshot);
2020                         break;
2021                 }
2022
2023                 bch2_trans_iter_exit(trans, &dirent_iter);
2024
2025                 if (!S_ISDIR(inode->bi_mode))
2026                         break;
2027
2028                 ret = path_down(c, p, inode->bi_inum, snapshot);
2029                 if (ret) {
2030                         bch_err(c, "memory allocation failure");
2031                         return ret;
2032                 }
2033
2034                 snapshot = parent_snapshot;
2035
2036                 ret = lookup_inode(trans, inode->bi_dir, inode, &snapshot);
2037                 if (ret) {
2038                         /* Should have been caught in dirents pass */
2039                         if (!bch2_err_matches(ret, BCH_ERR_transaction_restart))
2040                                 bch_err(c, "error looking up parent directory: %i", ret);
2041                         break;
2042                 }
2043
2044                 if (path_is_dup(p, inode->bi_inum, snapshot)) {
2045                         /* XXX print path */
2046                         bch_err(c, "directory structure loop");
2047
2048                         darray_for_each(*p, i)
2049                                 pr_err("%llu:%u", i->inum, i->snapshot);
2050                         pr_err("%llu:%u", inode->bi_inum, snapshot);
2051
2052                         if (!fsck_err(c, dir_loop, "directory structure loop"))
2053                                 return 0;
2054
2055                         ret = remove_backpointer(trans, inode);
2056                         if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
2057                                 bch_err_msg(c, ret, "removing dirent");
2058                         if (ret)
2059                                 break;
2060
2061                         ret = reattach_inode(trans, inode, snapshot);
2062                         if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
2063                                 bch_err_msg(c, ret, "reattaching inode %llu", inode->bi_inum);
2064                         break;
2065                 }
2066         }
2067 fsck_err:
2068         bch_err_fn(c, ret);
2069         return ret;
2070 }
2071
2072 /*
2073  * Check for unreachable inodes, as well as loops in the directory structure:
2074  * After bch2_check_dirents(), if an inode backpointer doesn't exist that means it's
2075  * unreachable:
2076  */
2077 int bch2_check_directory_structure(struct bch_fs *c)
2078 {
2079         struct bch_inode_unpacked u;
2080         pathbuf path = { 0, };
2081         int ret;
2082
2083         ret = bch2_trans_run(c,
2084                 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes, POS_MIN,
2085                                           BTREE_ITER_INTENT|
2086                                           BTREE_ITER_PREFETCH|
2087                                           BTREE_ITER_ALL_SNAPSHOTS, k,
2088                                           NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
2089                         if (!bkey_is_inode(k.k))
2090                                 continue;
2091
2092                         BUG_ON(bch2_inode_unpack(k, &u));
2093
2094                         if (u.bi_flags & BCH_INODE_unlinked)
2095                                 continue;
2096
2097                         check_path(trans, &path, &u, iter.pos.snapshot);
2098                 })));
2099         darray_exit(&path);
2100
2101         bch_err_fn(c, ret);
2102         return ret;
2103 }
2104
2105 struct nlink_table {
2106         size_t          nr;
2107         size_t          size;
2108
2109         struct nlink {
2110                 u64     inum;
2111                 u32     snapshot;
2112                 u32     count;
2113         }               *d;
2114 };
2115
2116 static int add_nlink(struct bch_fs *c, struct nlink_table *t,
2117                      u64 inum, u32 snapshot)
2118 {
2119         if (t->nr == t->size) {
2120                 size_t new_size = max_t(size_t, 128UL, t->size * 2);
2121                 void *d = kvmalloc_array(new_size, sizeof(t->d[0]), GFP_KERNEL);
2122
2123                 if (!d) {
2124                         bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
2125                                 new_size);
2126                         return -BCH_ERR_ENOMEM_fsck_add_nlink;
2127                 }
2128
2129                 if (t->d)
2130                         memcpy(d, t->d, t->size * sizeof(t->d[0]));
2131                 kvfree(t->d);
2132
2133                 t->d = d;
2134                 t->size = new_size;
2135         }
2136
2137
2138         t->d[t->nr++] = (struct nlink) {
2139                 .inum           = inum,
2140                 .snapshot       = snapshot,
2141         };
2142
2143         return 0;
2144 }
2145
2146 static int nlink_cmp(const void *_l, const void *_r)
2147 {
2148         const struct nlink *l = _l;
2149         const struct nlink *r = _r;
2150
2151         return cmp_int(l->inum, r->inum);
2152 }
2153
2154 static void inc_link(struct bch_fs *c, struct snapshots_seen *s,
2155                      struct nlink_table *links,
2156                      u64 range_start, u64 range_end, u64 inum, u32 snapshot)
2157 {
2158         struct nlink *link, key = {
2159                 .inum = inum, .snapshot = U32_MAX,
2160         };
2161
2162         if (inum < range_start || inum >= range_end)
2163                 return;
2164
2165         link = __inline_bsearch(&key, links->d, links->nr,
2166                                 sizeof(links->d[0]), nlink_cmp);
2167         if (!link)
2168                 return;
2169
2170         while (link > links->d && link[0].inum == link[-1].inum)
2171                 --link;
2172
2173         for (; link < links->d + links->nr && link->inum == inum; link++)
2174                 if (ref_visible(c, s, snapshot, link->snapshot)) {
2175                         link->count++;
2176                         if (link->snapshot >= snapshot)
2177                                 break;
2178                 }
2179 }
2180
2181 noinline_for_stack
2182 static int check_nlinks_find_hardlinks(struct bch_fs *c,
2183                                        struct nlink_table *t,
2184                                        u64 start, u64 *end)
2185 {
2186         int ret = bch2_trans_run(c,
2187                 for_each_btree_key(trans, iter, BTREE_ID_inodes,
2188                                    POS(0, start),
2189                                    BTREE_ITER_INTENT|
2190                                    BTREE_ITER_PREFETCH|
2191                                    BTREE_ITER_ALL_SNAPSHOTS, k, ({
2192                         if (!bkey_is_inode(k.k))
2193                                 continue;
2194
2195                         /* Should never fail, checked by bch2_inode_invalid: */
2196                         struct bch_inode_unpacked u;
2197                         BUG_ON(bch2_inode_unpack(k, &u));
2198
2199                         /*
2200                          * Backpointer and directory structure checks are sufficient for
2201                          * directories, since they can't have hardlinks:
2202                          */
2203                         if (S_ISDIR(u.bi_mode))
2204                                 continue;
2205
2206                         if (!u.bi_nlink)
2207                                 continue;
2208
2209                         ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot);
2210                         if (ret) {
2211                                 *end = k.k->p.offset;
2212                                 ret = 0;
2213                                 break;
2214                         }
2215                         0;
2216                 })));
2217
2218         bch_err_fn(c, ret);
2219         return ret;
2220 }
2221
2222 noinline_for_stack
2223 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
2224                                      u64 range_start, u64 range_end)
2225 {
2226         struct snapshots_seen s;
2227
2228         snapshots_seen_init(&s);
2229
2230         int ret = bch2_trans_run(c,
2231                 for_each_btree_key(trans, iter, BTREE_ID_dirents, POS_MIN,
2232                                    BTREE_ITER_INTENT|
2233                                    BTREE_ITER_PREFETCH|
2234                                    BTREE_ITER_ALL_SNAPSHOTS, k, ({
2235                         ret = snapshots_seen_update(c, &s, iter.btree_id, k.k->p);
2236                         if (ret)
2237                                 break;
2238
2239                         if (k.k->type == KEY_TYPE_dirent) {
2240                                 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
2241
2242                                 if (d.v->d_type != DT_DIR &&
2243                                     d.v->d_type != DT_SUBVOL)
2244                                         inc_link(c, &s, links, range_start, range_end,
2245                                                  le64_to_cpu(d.v->d_inum),
2246                                                  bch2_snapshot_equiv(c, d.k->p.snapshot));
2247                         }
2248                         0;
2249                 })));
2250
2251         snapshots_seen_exit(&s);
2252
2253         bch_err_fn(c, ret);
2254         return ret;
2255 }
2256
2257 static int check_nlinks_update_inode(struct btree_trans *trans, struct btree_iter *iter,
2258                                      struct bkey_s_c k,
2259                                      struct nlink_table *links,
2260                                      size_t *idx, u64 range_end)
2261 {
2262         struct bch_fs *c = trans->c;
2263         struct bch_inode_unpacked u;
2264         struct nlink *link = &links->d[*idx];
2265         int ret = 0;
2266
2267         if (k.k->p.offset >= range_end)
2268                 return 1;
2269
2270         if (!bkey_is_inode(k.k))
2271                 return 0;
2272
2273         BUG_ON(bch2_inode_unpack(k, &u));
2274
2275         if (S_ISDIR(u.bi_mode))
2276                 return 0;
2277
2278         if (!u.bi_nlink)
2279                 return 0;
2280
2281         while ((cmp_int(link->inum, k.k->p.offset) ?:
2282                 cmp_int(link->snapshot, k.k->p.snapshot)) < 0) {
2283                 BUG_ON(*idx == links->nr);
2284                 link = &links->d[++*idx];
2285         }
2286
2287         if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count,
2288                         c, inode_wrong_nlink,
2289                         "inode %llu type %s has wrong i_nlink (%u, should be %u)",
2290                         u.bi_inum, bch2_d_types[mode_to_type(u.bi_mode)],
2291                         bch2_inode_nlink_get(&u), link->count)) {
2292                 bch2_inode_nlink_set(&u, link->count);
2293                 ret = __write_inode(trans, &u, k.k->p.snapshot);
2294         }
2295 fsck_err:
2296         return ret;
2297 }
2298
2299 noinline_for_stack
2300 static int check_nlinks_update_hardlinks(struct bch_fs *c,
2301                                struct nlink_table *links,
2302                                u64 range_start, u64 range_end)
2303 {
2304         size_t idx = 0;
2305
2306         int ret = bch2_trans_run(c,
2307                 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
2308                                 POS(0, range_start),
2309                                 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
2310                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2311                         check_nlinks_update_inode(trans, &iter, k, links, &idx, range_end)));
2312         if (ret < 0) {
2313                 bch_err(c, "error in fsck walking inodes: %s", bch2_err_str(ret));
2314                 return ret;
2315         }
2316
2317         return 0;
2318 }
2319
2320 int bch2_check_nlinks(struct bch_fs *c)
2321 {
2322         struct nlink_table links = { 0 };
2323         u64 this_iter_range_start, next_iter_range_start = 0;
2324         int ret = 0;
2325
2326         do {
2327                 this_iter_range_start = next_iter_range_start;
2328                 next_iter_range_start = U64_MAX;
2329
2330                 ret = check_nlinks_find_hardlinks(c, &links,
2331                                                   this_iter_range_start,
2332                                                   &next_iter_range_start);
2333
2334                 ret = check_nlinks_walk_dirents(c, &links,
2335                                           this_iter_range_start,
2336                                           next_iter_range_start);
2337                 if (ret)
2338                         break;
2339
2340                 ret = check_nlinks_update_hardlinks(c, &links,
2341                                          this_iter_range_start,
2342                                          next_iter_range_start);
2343                 if (ret)
2344                         break;
2345
2346                 links.nr = 0;
2347         } while (next_iter_range_start != U64_MAX);
2348
2349         kvfree(links.d);
2350         bch_err_fn(c, ret);
2351         return ret;
2352 }
2353
2354 static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter,
2355                              struct bkey_s_c k)
2356 {
2357         struct bkey_s_c_reflink_p p;
2358         struct bkey_i_reflink_p *u;
2359
2360         if (k.k->type != KEY_TYPE_reflink_p)
2361                 return 0;
2362
2363         p = bkey_s_c_to_reflink_p(k);
2364
2365         if (!p.v->front_pad && !p.v->back_pad)
2366                 return 0;
2367
2368         u = bch2_trans_kmalloc(trans, sizeof(*u));
2369         int ret = PTR_ERR_OR_ZERO(u);
2370         if (ret)
2371                 return ret;
2372
2373         bkey_reassemble(&u->k_i, k);
2374         u->v.front_pad  = 0;
2375         u->v.back_pad   = 0;
2376
2377         return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_NORUN);
2378 }
2379
2380 int bch2_fix_reflink_p(struct bch_fs *c)
2381 {
2382         if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
2383                 return 0;
2384
2385         int ret = bch2_trans_run(c,
2386                 for_each_btree_key_commit(trans, iter,
2387                                 BTREE_ID_extents, POS_MIN,
2388                                 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|
2389                                 BTREE_ITER_ALL_SNAPSHOTS, k,
2390                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2391                         fix_reflink_p_key(trans, &iter, k)));
2392         bch_err_fn(c, ret);
2393         return ret;
2394 }