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         bch2_trans_iter_exit(trans, &iter);
 123         return ret;
 124 }
 125
 126 static int __lookup_dirent(struct btree_trans *trans,
 127                            struct bch_hash_info hash_info,
 128                            subvol_inum dir, struct qstr *name,
 129                            u64 *target, unsigned *type)
 130 {
 131         struct btree_iter iter;
 132         struct bkey_s_c_dirent d;
 133         int ret;
 134
 135         ret = bch2_hash_lookup(trans, &iter, bch2_dirent_hash_desc,
 136                                &hash_info, dir, name, 0);
 137         if (ret)
 138                 return ret;
 139
 140         d = bkey_s_c_to_dirent(bch2_btree_iter_peek_slot(&iter));
 141         *target = le64_to_cpu(d.v->d_inum);
 142         *type = d.v->d_type;
 143         bch2_trans_iter_exit(trans, &iter);
 144         return 0;
 145 }
 146
 147 static int __write_inode(struct btree_trans *trans,
 148                          struct bch_inode_unpacked *inode,
 149                          u32 snapshot)
 150 {
 151         struct bkey_inode_buf *inode_p =
 152                 bch2_trans_kmalloc(trans, sizeof(*inode_p));
 153
 154         if (IS_ERR(inode_p))
 155                 return PTR_ERR(inode_p);
 156
 157         bch2_inode_pack(inode_p, inode);
 158         inode_p->inode.k.p.snapshot = snapshot;
 159
 160         return bch2_btree_insert_nonextent(trans, BTREE_ID_inodes,
 161                                 &inode_p->inode.k_i,
 162                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 163 }
 164
 165 static int fsck_write_inode(struct btree_trans *trans,
 166                             struct bch_inode_unpacked *inode,
 167                             u32 snapshot)
 168 {
 169         int ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
 170                             __write_inode(trans, inode, snapshot));
 171         bch_err_fn(trans->c, ret);
 172         return ret;
 173 }
 174
 175 static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
 176 {
 177         struct bch_fs *c = trans->c;
 178         struct btree_iter iter;
 179         struct bch_inode_unpacked dir_inode;
 180         struct bch_hash_info dir_hash_info;
 181         int ret;
 182
 183         ret = lookup_first_inode(trans, pos.inode, &dir_inode);
 184         if (ret)
 185                 goto err;
 186
 187         dir_hash_info = bch2_hash_info_init(c, &dir_inode);
 188
 189         bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_INTENT);
 190
 191         ret = bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
 192                                   &dir_hash_info, &iter,
 193                                   BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 194         bch2_trans_iter_exit(trans, &iter);
 195 err:
 196         bch_err_fn(c, ret);
 197         return ret;
 198 }
 199
 200 /* Get lost+found, create if it doesn't exist: */
 201 static int lookup_lostfound(struct btree_trans *trans, u32 snapshot,
 202                             struct bch_inode_unpacked *lostfound)
 203 {
 204         struct bch_fs *c = trans->c;
 205         struct qstr lostfound_str = QSTR("lost+found");
 206         u64 inum = 0;
 207         unsigned d_type = 0;
 208         int ret;
 209
 210         struct bch_snapshot_tree st;
 211         ret = bch2_snapshot_tree_lookup(trans,
 212                         bch2_snapshot_tree(c, snapshot), &st);
 213         if (ret)
 214                 return ret;
 215
 216         subvol_inum root_inum = { .subvol = le32_to_cpu(st.master_subvol) };
 217         u32 subvol_snapshot;
 218
 219         ret = subvol_lookup(trans, le32_to_cpu(st.master_subvol),
 220                             &subvol_snapshot, &root_inum.inum);
 221         bch_err_msg(c, ret, "looking up root subvol");
 222         if (ret)
 223                 return ret;
 224
 225         struct bch_inode_unpacked root_inode;
 226         struct bch_hash_info root_hash_info;
 227         ret = lookup_inode(trans, root_inum.inum, &root_inode, &snapshot);
 228         bch_err_msg(c, ret, "looking up root inode");
 229         if (ret)
 230                 return ret;
 231
 232         root_hash_info = bch2_hash_info_init(c, &root_inode);
 233
 234         ret = __lookup_dirent(trans, root_hash_info, root_inum,
 235                               &lostfound_str, &inum, &d_type);
 236         if (bch2_err_matches(ret, ENOENT))
 237                 goto create_lostfound;
 238
 239         bch_err_fn(c, ret);
 240         if (ret)
 241                 return ret;
 242
 243         if (d_type != DT_DIR) {
 244                 bch_err(c, "error looking up lost+found: not a directory");
 245                 return -BCH_ERR_ENOENT_not_directory;
 246         }
 247
 248         /*
 249          * The bch2_check_dirents pass has already run, dangling dirents
 250          * shouldn't exist here:
 251          */
 252         ret = lookup_inode(trans, inum, lostfound, &snapshot);
 253         bch_err_msg(c, ret, "looking up lost+found");
 254         return ret;
 255
 256 create_lostfound:
 257         /*
 258          * XXX: we could have a nicer log message here  if we had a nice way to
 259          * walk backpointers to print a path
 260          */
 261         bch_notice(c, "creating lost+found in snapshot %u", le32_to_cpu(st.root_snapshot));
 262
 263         u64 now = bch2_current_time(c);
 264         struct btree_iter lostfound_iter = { NULL };
 265         u64 cpu = raw_smp_processor_id();
 266
 267         bch2_inode_init_early(c, lostfound);
 268         bch2_inode_init_late(lostfound, now, 0, 0, S_IFDIR|0700, 0, &root_inode);
 269         lostfound->bi_dir = root_inode.bi_inum;
 270
 271         root_inode.bi_nlink++;
 272
 273         ret = bch2_inode_create(trans, &lostfound_iter, lostfound, snapshot, cpu);
 274         if (ret)
 275                 goto err;
 276
 277         bch2_btree_iter_set_snapshot(&lostfound_iter, snapshot);
 278         ret = bch2_btree_iter_traverse(&lostfound_iter);
 279         if (ret)
 280                 goto err;
 281
 282         ret =   bch2_dirent_create_snapshot(trans,
 283                                 root_inode.bi_inum, snapshot, &root_hash_info,
 284                                 mode_to_type(lostfound->bi_mode),
 285                                 &lostfound_str,
 286                                 lostfound->bi_inum,
 287                                 &lostfound->bi_dir_offset,
 288                                 BCH_HASH_SET_MUST_CREATE) ?:
 289                 bch2_inode_write_flags(trans, &lostfound_iter, lostfound,
 290                                        BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 291 err:
 292         bch_err_msg(c, ret, "creating lost+found");
 293         bch2_trans_iter_exit(trans, &lostfound_iter);
 294         return ret;
 295 }
 296
 297 static int reattach_inode(struct btree_trans *trans,
 298                           struct bch_inode_unpacked *inode,
 299                           u32 inode_snapshot)
 300 {
 301         struct bch_hash_info dir_hash;
 302         struct bch_inode_unpacked lostfound;
 303         char name_buf[20];
 304         struct qstr name;
 305         u64 dir_offset = 0;
 306         int ret;
 307
 308         ret = lookup_lostfound(trans, inode_snapshot, &lostfound);
 309         if (ret)
 310                 return ret;
 311
 312         if (S_ISDIR(inode->bi_mode)) {
 313                 lostfound.bi_nlink++;
 314
 315                 ret = __write_inode(trans, &lostfound, U32_MAX);
 316                 if (ret)
 317                         return ret;
 318         }
 319
 320         dir_hash = bch2_hash_info_init(trans->c, &lostfound);
 321
 322         snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
 323         name = (struct qstr) QSTR(name_buf);
 324
 325         ret = bch2_dirent_create_snapshot(trans,
 326                                 lostfound.bi_inum, inode_snapshot,
 327                                 &dir_hash,
 328                                 inode_d_type(inode),
 329                                 &name, inode->bi_inum, &dir_offset,
 330                                 BCH_HASH_SET_MUST_CREATE);
 331         if (ret)
 332                 return ret;
 333
 334         inode->bi_dir           = lostfound.bi_inum;
 335         inode->bi_dir_offset    = dir_offset;
 336
 337         return __write_inode(trans, inode, inode_snapshot);
 338 }
 339
 340 static int remove_backpointer(struct btree_trans *trans,
 341                               struct bch_inode_unpacked *inode)
 342 {
 343         struct btree_iter iter;
 344         struct bkey_s_c_dirent d;
 345         int ret;
 346
 347         d = bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_dirents,
 348                                      POS(inode->bi_dir, inode->bi_dir_offset), 0,
 349                                      dirent);
 350         ret =   bkey_err(d) ?:
 351                 __remove_dirent(trans, d.k->p);
 352         bch2_trans_iter_exit(trans, &iter);
 353         return ret;
 354 }
 355
 356 struct snapshots_seen_entry {
 357         u32                             id;
 358         u32                             equiv;
 359 };
 360
 361 struct snapshots_seen {
 362         struct bpos                     pos;
 363         DARRAY(struct snapshots_seen_entry) ids;
 364 };
 365
 366 static inline void snapshots_seen_exit(struct snapshots_seen *s)
 367 {
 368         darray_exit(&s->ids);
 369 }
 370
 371 static inline void snapshots_seen_init(struct snapshots_seen *s)
 372 {
 373         memset(s, 0, sizeof(*s));
 374 }
 375
 376 static int snapshots_seen_add_inorder(struct bch_fs *c, struct snapshots_seen *s, u32 id)
 377 {
 378         struct snapshots_seen_entry *i, n = {
 379                 .id     = id,
 380                 .equiv  = bch2_snapshot_equiv(c, id),
 381         };
 382         int ret = 0;
 383
 384         __darray_for_each(s->ids, i) {
 385                 if (i->id == id)
 386                         return 0;
 387                 if (i->id > id)
 388                         break;
 389         }
 390
 391         ret = darray_insert_item(&s->ids, i - s->ids.data, n);
 392         if (ret)
 393                 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
 394                         s->ids.size);
 395         return ret;
 396 }
 397
 398 static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s,
 399                                  enum btree_id btree_id, struct bpos pos)
 400 {
 401         struct snapshots_seen_entry n = {
 402                 .id     = pos.snapshot,
 403                 .equiv  = bch2_snapshot_equiv(c, pos.snapshot),
 404         };
 405         int ret = 0;
 406
 407         if (!bkey_eq(s->pos, pos))
 408                 s->ids.nr = 0;
 409
 410         s->pos = pos;
 411         s->pos.snapshot = n.equiv;
 412
 413         darray_for_each(s->ids, i) {
 414                 if (i->id == n.id)
 415                         return 0;
 416
 417                 /*
 418                  * We currently don't rigorously track for snapshot cleanup
 419                  * needing to be run, so it shouldn't be a fsck error yet:
 420                  */
 421                 if (i->equiv == n.equiv) {
 422                         bch_err(c, "snapshot deletion did not finish:\n"
 423                                 "  duplicate keys in btree %s at %llu:%llu snapshots %u, %u (equiv %u)\n",
 424                                 bch2_btree_id_str(btree_id),
 425                                 pos.inode, pos.offset,
 426                                 i->id, n.id, n.equiv);
 427                         set_bit(BCH_FS_need_delete_dead_snapshots, &c->flags);
 428                         return bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_delete_dead_snapshots);
 429                 }
 430         }
 431
 432         ret = darray_push(&s->ids, n);
 433         if (ret)
 434                 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
 435                         s->ids.size);
 436         return ret;
 437 }
 438
 439 /**
 440  * key_visible_in_snapshot - returns true if @id is a descendent of @ancestor,
 441  * and @ancestor hasn't been overwritten in @seen
 442  *
 443  * @c:          filesystem handle
 444  * @seen:       list of snapshot ids already seen at current position
 445  * @id:         descendent snapshot id
 446  * @ancestor:   ancestor snapshot id
 447  *
 448  * Returns:     whether key in @ancestor snapshot is visible in @id snapshot
 449  */
 450 static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *seen,
 451                                     u32 id, u32 ancestor)
 452 {
 453         ssize_t i;
 454
 455         EBUG_ON(id > ancestor);
 456         EBUG_ON(!bch2_snapshot_is_equiv(c, id));
 457         EBUG_ON(!bch2_snapshot_is_equiv(c, ancestor));
 458
 459         /* @ancestor should be the snapshot most recently added to @seen */
 460         EBUG_ON(ancestor != seen->pos.snapshot);
 461         EBUG_ON(ancestor != seen->ids.data[seen->ids.nr - 1].equiv);
 462
 463         if (id == ancestor)
 464                 return true;
 465
 466         if (!bch2_snapshot_is_ancestor(c, id, ancestor))
 467                 return false;
 468
 469         /*
 470          * We know that @id is a descendant of @ancestor, we're checking if
 471          * we've seen a key that overwrote @ancestor - i.e. also a descendent of
 472          * @ascestor and with @id as a descendent.
 473          *
 474          * But we already know that we're scanning IDs between @id and @ancestor
 475          * numerically, since snapshot ID lists are kept sorted, so if we find
 476          * an id that's an ancestor of @id we're done:
 477          */
 478
 479         for (i = seen->ids.nr - 2;
 480              i >= 0 && seen->ids.data[i].equiv >= id;
 481              --i)
 482                 if (bch2_snapshot_is_ancestor(c, id, seen->ids.data[i].equiv))
 483                         return false;
 484
 485         return true;
 486 }
 487
 488 /**
 489  * ref_visible - given a key with snapshot id @src that points to a key with
 490  * snapshot id @dst, test whether there is some snapshot in which @dst is
 491  * visible.
 492  *
 493  * @c:          filesystem handle
 494  * @s:          list of snapshot IDs already seen at @src
 495  * @src:        snapshot ID of src key
 496  * @dst:        snapshot ID of dst key
 497  * Returns:     true if there is some snapshot in which @dst is visible
 498  *
 499  * Assumes we're visiting @src keys in natural key order
 500  */
 501 static bool ref_visible(struct bch_fs *c, struct snapshots_seen *s,
 502                         u32 src, u32 dst)
 503 {
 504         return dst <= src
 505                 ? key_visible_in_snapshot(c, s, dst, src)
 506                 : bch2_snapshot_is_ancestor(c, src, dst);
 507 }
 508
 509 static int ref_visible2(struct bch_fs *c,
 510                         u32 src, struct snapshots_seen *src_seen,
 511                         u32 dst, struct snapshots_seen *dst_seen)
 512 {
 513         src = bch2_snapshot_equiv(c, src);
 514         dst = bch2_snapshot_equiv(c, dst);
 515
 516         if (dst > src) {
 517                 swap(dst, src);
 518                 swap(dst_seen, src_seen);
 519         }
 520         return key_visible_in_snapshot(c, src_seen, dst, src);
 521 }
 522
 523 #define for_each_visible_inode(_c, _s, _w, _snapshot, _i)                               \
 524         for (_i = (_w)->inodes.data; _i < (_w)->inodes.data + (_w)->inodes.nr &&        \
 525              (_i)->snapshot <= (_snapshot); _i++)                                       \
 526                 if (key_visible_in_snapshot(_c, _s, _i->snapshot, _snapshot))
 527
 528 struct inode_walker_entry {
 529         struct bch_inode_unpacked inode;
 530         u32                     snapshot;
 531         bool                    seen_this_pos;
 532         u64                     count;
 533 };
 534
 535 struct inode_walker {
 536         bool                            first_this_inode;
 537         bool                            recalculate_sums;
 538         struct bpos                     last_pos;
 539
 540         DARRAY(struct inode_walker_entry) inodes;
 541 };
 542
 543 static void inode_walker_exit(struct inode_walker *w)
 544 {
 545         darray_exit(&w->inodes);
 546 }
 547
 548 static struct inode_walker inode_walker_init(void)
 549 {
 550         return (struct inode_walker) { 0, };
 551 }
 552
 553 static int add_inode(struct bch_fs *c, struct inode_walker *w,
 554                      struct bkey_s_c inode)
 555 {
 556         struct bch_inode_unpacked u;
 557
 558         BUG_ON(bch2_inode_unpack(inode, &u));
 559
 560         return darray_push(&w->inodes, ((struct inode_walker_entry) {
 561                 .inode          = u,
 562                 .snapshot       = bch2_snapshot_equiv(c, inode.k->p.snapshot),
 563         }));
 564 }
 565
 566 static int get_inodes_all_snapshots(struct btree_trans *trans,
 567                                     struct inode_walker *w, u64 inum)
 568 {
 569         struct bch_fs *c = trans->c;
 570         struct btree_iter iter;
 571         struct bkey_s_c k;
 572         int ret;
 573
 574         w->recalculate_sums = false;
 575         w->inodes.nr = 0;
 576
 577         for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum),
 578                                      BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
 579                 if (k.k->p.offset != inum)
 580                         break;
 581
 582                 if (bkey_is_inode(k.k))
 583                         add_inode(c, w, k);
 584         }
 585         bch2_trans_iter_exit(trans, &iter);
 586
 587         if (ret)
 588                 return ret;
 589
 590         w->first_this_inode = true;
 591         return 0;
 592 }
 593
 594 static struct inode_walker_entry *
 595 lookup_inode_for_snapshot(struct bch_fs *c, struct inode_walker *w,
 596                           u32 snapshot, bool is_whiteout)
 597 {
 598         struct inode_walker_entry *i;
 599
 600         snapshot = bch2_snapshot_equiv(c, snapshot);
 601
 602         __darray_for_each(w->inodes, i)
 603                 if (bch2_snapshot_is_ancestor(c, snapshot, i->snapshot))
 604                         goto found;
 605
 606         return NULL;
 607 found:
 608         BUG_ON(snapshot > i->snapshot);
 609
 610         if (snapshot != i->snapshot && !is_whiteout) {
 611                 struct inode_walker_entry new = *i;
 612                 size_t pos;
 613                 int ret;
 614
 615                 new.snapshot = snapshot;
 616                 new.count = 0;
 617
 618                 bch_info(c, "have key for inode %llu:%u but have inode in ancestor snapshot %u",
 619                          w->last_pos.inode, snapshot, i->snapshot);
 620
 621                 while (i > w->inodes.data && i[-1].snapshot > snapshot)
 622                         --i;
 623
 624                 pos = i - w->inodes.data;
 625                 ret = darray_insert_item(&w->inodes, pos, new);
 626                 if (ret)
 627                         return ERR_PTR(ret);
 628
 629                 i = w->inodes.data + pos;
 630         }
 631
 632         return i;
 633 }
 634
 635 static struct inode_walker_entry *walk_inode(struct btree_trans *trans,
 636                                              struct inode_walker *w, struct bpos pos,
 637                                              bool is_whiteout)
 638 {
 639         if (w->last_pos.inode != pos.inode) {
 640                 int ret = get_inodes_all_snapshots(trans, w, pos.inode);
 641                 if (ret)
 642                         return ERR_PTR(ret);
 643         } else if (bkey_cmp(w->last_pos, pos)) {
 644                 darray_for_each(w->inodes, i)
 645                         i->seen_this_pos = false;
 646         }
 647
 648         w->last_pos = pos;
 649
 650         return lookup_inode_for_snapshot(trans->c, w, pos.snapshot, is_whiteout);
 651 }
 652
 653 static int __get_visible_inodes(struct btree_trans *trans,
 654                                 struct inode_walker *w,
 655                                 struct snapshots_seen *s,
 656                                 u64 inum)
 657 {
 658         struct bch_fs *c = trans->c;
 659         struct btree_iter iter;
 660         struct bkey_s_c k;
 661         int ret;
 662
 663         w->inodes.nr = 0;
 664
 665         for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum),
 666                            BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
 667                 u32 equiv = bch2_snapshot_equiv(c, k.k->p.snapshot);
 668
 669                 if (k.k->p.offset != inum)
 670                         break;
 671
 672                 if (!ref_visible(c, s, s->pos.snapshot, equiv))
 673                         continue;
 674
 675                 if (bkey_is_inode(k.k))
 676                         add_inode(c, w, k);
 677
 678                 if (equiv >= s->pos.snapshot)
 679                         break;
 680         }
 681         bch2_trans_iter_exit(trans, &iter);
 682
 683         return ret;
 684 }
 685
 686 static int check_key_has_snapshot(struct btree_trans *trans,
 687                                   struct btree_iter *iter,
 688                                   struct bkey_s_c k)
 689 {
 690         struct bch_fs *c = trans->c;
 691         struct printbuf buf = PRINTBUF;
 692         int ret = 0;
 693
 694         if (mustfix_fsck_err_on(!bch2_snapshot_equiv(c, k.k->p.snapshot), c,
 695                                 bkey_in_missing_snapshot,
 696                                 "key in missing snapshot: %s",
 697                                 (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
 698                 ret = bch2_btree_delete_at(trans, iter,
 699                                             BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?: 1;
 700 fsck_err:
 701         printbuf_exit(&buf);
 702         return ret;
 703 }
 704
 705 static int hash_redo_key(struct btree_trans *trans,
 706                          const struct bch_hash_desc desc,
 707                          struct bch_hash_info *hash_info,
 708                          struct btree_iter *k_iter, struct bkey_s_c k)
 709 {
 710         struct bkey_i *delete;
 711         struct bkey_i *tmp;
 712
 713         delete = bch2_trans_kmalloc(trans, sizeof(*delete));
 714         if (IS_ERR(delete))
 715                 return PTR_ERR(delete);
 716
 717         tmp = bch2_bkey_make_mut_noupdate(trans, k);
 718         if (IS_ERR(tmp))
 719                 return PTR_ERR(tmp);
 720
 721         bkey_init(&delete->k);
 722         delete->k.p = k_iter->pos;
 723         return  bch2_btree_iter_traverse(k_iter) ?:
 724                 bch2_trans_update(trans, k_iter, delete, 0) ?:
 725                 bch2_hash_set_snapshot(trans, desc, hash_info,
 726                                        (subvol_inum) { 0, k.k->p.inode },
 727                                        k.k->p.snapshot, tmp,
 728                                        BCH_HASH_SET_MUST_CREATE,
 729                                        BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
 730                 bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
 731 }
 732
 733 static int hash_check_key(struct btree_trans *trans,
 734                           const struct bch_hash_desc desc,
 735                           struct bch_hash_info *hash_info,
 736                           struct btree_iter *k_iter, struct bkey_s_c hash_k)
 737 {
 738         struct bch_fs *c = trans->c;
 739         struct btree_iter iter = { NULL };
 740         struct printbuf buf = PRINTBUF;
 741         struct bkey_s_c k;
 742         u64 hash;
 743         int ret = 0;
 744
 745         if (hash_k.k->type != desc.key_type)
 746                 return 0;
 747
 748         hash = desc.hash_bkey(hash_info, hash_k);
 749
 750         if (likely(hash == hash_k.k->p.offset))
 751                 return 0;
 752
 753         if (hash_k.k->p.offset < hash)
 754                 goto bad_hash;
 755
 756         for_each_btree_key_norestart(trans, iter, desc.btree_id,
 757                                      SPOS(hash_k.k->p.inode, hash, hash_k.k->p.snapshot),
 758                                      BTREE_ITER_SLOTS, k, ret) {
 759                 if (bkey_eq(k.k->p, hash_k.k->p))
 760                         break;
 761
 762                 if (fsck_err_on(k.k->type == desc.key_type &&
 763                                 !desc.cmp_bkey(k, hash_k), c,
 764                                 hash_table_key_duplicate,
 765                                 "duplicate hash table keys:\n%s",
 766                                 (printbuf_reset(&buf),
 767                                  bch2_bkey_val_to_text(&buf, c, hash_k),
 768                                  buf.buf))) {
 769                         ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0) ?: 1;
 770                         break;
 771                 }
 772
 773                 if (bkey_deleted(k.k)) {
 774                         bch2_trans_iter_exit(trans, &iter);
 775                         goto bad_hash;
 776                 }
 777         }
 778 out:
 779         bch2_trans_iter_exit(trans, &iter);
 780         printbuf_exit(&buf);
 781         return ret;
 782 bad_hash:
 783         if (fsck_err(c, hash_table_key_wrong_offset,
 784                      "hash table key at wrong offset: btree %s inode %llu offset %llu, hashed to %llu\n%s",
 785                      bch2_btree_id_str(desc.btree_id), hash_k.k->p.inode, hash_k.k->p.offset, hash,
 786                      (printbuf_reset(&buf),
 787                       bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf))) {
 788                 ret = hash_redo_key(trans, desc, hash_info, k_iter, hash_k);
 789                 bch_err_fn(c, ret);
 790                 if (ret)
 791                         return ret;
 792                 ret = -BCH_ERR_transaction_restart_nested;
 793         }
 794 fsck_err:
 795         goto out;
 796 }
 797
 798 static int check_inode_deleted_list(struct btree_trans *trans, struct bpos p)
 799 {
 800         struct btree_iter iter;
 801         struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_deleted_inodes, p, 0);
 802         int ret = bkey_err(k);
 803         if (ret)
 804                 return ret;
 805
 806         bch2_trans_iter_exit(trans, &iter);
 807         return k.k->type == KEY_TYPE_set;
 808 }
 809
 810 static int check_inode(struct btree_trans *trans,
 811                        struct btree_iter *iter,
 812                        struct bkey_s_c k,
 813                        struct bch_inode_unpacked *prev,
 814                        struct snapshots_seen *s,
 815                        bool full)
 816 {
 817         struct bch_fs *c = trans->c;
 818         struct bch_inode_unpacked u;
 819         bool do_update = false;
 820         int ret;
 821
 822         ret = check_key_has_snapshot(trans, iter, k);
 823         if (ret < 0)
 824                 goto err;
 825         if (ret)
 826                 return 0;
 827
 828         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
 829         if (ret)
 830                 goto err;
 831
 832         if (!bkey_is_inode(k.k))
 833                 return 0;
 834
 835         BUG_ON(bch2_inode_unpack(k, &u));
 836
 837         if (!full &&
 838             !(u.bi_flags & (BCH_INODE_i_size_dirty|
 839                             BCH_INODE_i_sectors_dirty|
 840                             BCH_INODE_unlinked)))
 841                 return 0;
 842
 843         if (prev->bi_inum != u.bi_inum)
 844                 *prev = u;
 845
 846         if (fsck_err_on(prev->bi_hash_seed      != u.bi_hash_seed ||
 847                         inode_d_type(prev)      != inode_d_type(&u),
 848                         c, inode_snapshot_mismatch,
 849                         "inodes in different snapshots don't match")) {
 850                 bch_err(c, "repair not implemented yet");
 851                 return -BCH_ERR_fsck_repair_unimplemented;
 852         }
 853
 854         if ((u.bi_flags & (BCH_INODE_i_size_dirty|BCH_INODE_unlinked)) &&
 855             bch2_key_has_snapshot_overwrites(trans, BTREE_ID_inodes, k.k->p)) {
 856                 struct bpos new_min_pos;
 857
 858                 ret = bch2_propagate_key_to_snapshot_leaves(trans, iter->btree_id, k, &new_min_pos);
 859                 if (ret)
 860                         goto err;
 861
 862                 u.bi_flags &= ~BCH_INODE_i_size_dirty|BCH_INODE_unlinked;
 863
 864                 ret = __write_inode(trans, &u, iter->pos.snapshot);
 865                 bch_err_msg(c, ret, "in fsck updating inode");
 866                 if (ret)
 867                         return ret;
 868
 869                 if (!bpos_eq(new_min_pos, POS_MIN))
 870                         bch2_btree_iter_set_pos(iter, bpos_predecessor(new_min_pos));
 871                 return 0;
 872         }
 873
 874         if (u.bi_flags & BCH_INODE_unlinked) {
 875                 ret = check_inode_deleted_list(trans, k.k->p);
 876                 if (ret < 0)
 877                         return ret;
 878
 879                 fsck_err_on(ret, c, unlinked_inode_not_on_deleted_list,
 880                             "inode %llu:%u unlinked, but not on deleted list",
 881                             u.bi_inum, k.k->p.snapshot);
 882                 ret = 0;
 883         }
 884
 885         if (u.bi_flags & BCH_INODE_unlinked &&
 886             (!c->sb.clean ||
 887              fsck_err(c, inode_unlinked_but_clean,
 888                       "filesystem marked clean, but inode %llu unlinked",
 889                       u.bi_inum))) {
 890                 ret = bch2_inode_rm_snapshot(trans, u.bi_inum, iter->pos.snapshot);
 891                 bch_err_msg(c, ret, "in fsck deleting inode");
 892                 return ret;
 893         }
 894
 895         if (u.bi_flags & BCH_INODE_i_size_dirty &&
 896             (!c->sb.clean ||
 897              fsck_err(c, inode_i_size_dirty_but_clean,
 898                       "filesystem marked clean, but inode %llu has i_size dirty",
 899                       u.bi_inum))) {
 900                 bch_verbose(c, "truncating inode %llu", u.bi_inum);
 901
 902                 /*
 903                  * XXX: need to truncate partial blocks too here - or ideally
 904                  * just switch units to bytes and that issue goes away
 905                  */
 906                 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
 907                                 SPOS(u.bi_inum, round_up(u.bi_size, block_bytes(c)) >> 9,
 908                                      iter->pos.snapshot),
 909                                 POS(u.bi_inum, U64_MAX),
 910                                 0, NULL);
 911                 bch_err_msg(c, ret, "in fsck truncating inode");
 912                 if (ret)
 913                         return ret;
 914
 915                 /*
 916                  * We truncated without our normal sector accounting hook, just
 917                  * make sure we recalculate it:
 918                  */
 919                 u.bi_flags |= BCH_INODE_i_sectors_dirty;
 920
 921                 u.bi_flags &= ~BCH_INODE_i_size_dirty;
 922                 do_update = true;
 923         }
 924
 925         if (u.bi_flags & BCH_INODE_i_sectors_dirty &&
 926             (!c->sb.clean ||
 927              fsck_err(c, inode_i_sectors_dirty_but_clean,
 928                       "filesystem marked clean, but inode %llu has i_sectors dirty",
 929                       u.bi_inum))) {
 930                 s64 sectors;
 931
 932                 bch_verbose(c, "recounting sectors for inode %llu",
 933                             u.bi_inum);
 934
 935                 sectors = bch2_count_inode_sectors(trans, u.bi_inum, iter->pos.snapshot);
 936                 if (sectors < 0) {
 937                         bch_err_msg(c, sectors, "in fsck recounting inode sectors");
 938                         return sectors;
 939                 }
 940
 941                 u.bi_sectors = sectors;
 942                 u.bi_flags &= ~BCH_INODE_i_sectors_dirty;
 943                 do_update = true;
 944         }
 945
 946         if (u.bi_flags & BCH_INODE_backptr_untrusted) {
 947                 u.bi_dir = 0;
 948                 u.bi_dir_offset = 0;
 949                 u.bi_flags &= ~BCH_INODE_backptr_untrusted;
 950                 do_update = true;
 951         }
 952
 953         if (do_update) {
 954                 ret = __write_inode(trans, &u, iter->pos.snapshot);
 955                 bch_err_msg(c, ret, "in fsck updating inode");
 956                 if (ret)
 957                         return ret;
 958         }
 959 err:
 960 fsck_err:
 961         bch_err_fn(c, ret);
 962         return ret;
 963 }
 964
 965 int bch2_check_inodes(struct bch_fs *c)
 966 {
 967         bool full = c->opts.fsck;
 968         struct bch_inode_unpacked prev = { 0 };
 969         struct snapshots_seen s;
 970
 971         snapshots_seen_init(&s);
 972
 973         int ret = bch2_trans_run(c,
 974                 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
 975                                 POS_MIN,
 976                                 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
 977                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
 978                         check_inode(trans, &iter, k, &prev, &s, full)));
 979
 980         snapshots_seen_exit(&s);
 981         bch_err_fn(c, ret);
 982         return ret;
 983 }
 984
 985 static struct bkey_s_c_dirent dirent_get_by_pos(struct btree_trans *trans,
 986                                                 struct btree_iter *iter,
 987                                                 struct bpos pos)
 988 {
 989         return bch2_bkey_get_iter_typed(trans, iter, BTREE_ID_dirents, pos, 0, dirent);
 990 }
 991
 992 static bool inode_points_to_dirent(struct bch_inode_unpacked *inode,
 993                                    struct bkey_s_c_dirent d)
 994 {
 995         return  inode->bi_dir           == d.k->p.inode &&
 996                 inode->bi_dir_offset    == d.k->p.offset;
 997 }
 998
 999 static bool dirent_points_to_inode(struct bkey_s_c_dirent d,
1000                                    struct bch_inode_unpacked *inode)
1001 {
1002         return d.v->d_type == DT_SUBVOL
1003                 ? le32_to_cpu(d.v->d_child_subvol)      == inode->bi_subvol
1004                 : le64_to_cpu(d.v->d_inum)              == inode->bi_inum;
1005 }
1006
1007 static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
1008 {
1009         struct bch_fs *c = trans->c;
1010         u32 restart_count = trans->restart_count;
1011         int ret = 0;
1012         s64 count2;
1013
1014         darray_for_each(w->inodes, i) {
1015                 if (i->inode.bi_sectors == i->count)
1016                         continue;
1017
1018                 count2 = bch2_count_inode_sectors(trans, w->last_pos.inode, i->snapshot);
1019
1020                 if (w->recalculate_sums)
1021                         i->count = count2;
1022
1023                 if (i->count != count2) {
1024                         bch_err(c, "fsck counted i_sectors wrong for inode %llu:%u: got %llu should be %llu",
1025                                 w->last_pos.inode, i->snapshot, i->count, count2);
1026                         return -BCH_ERR_internal_fsck_err;
1027                 }
1028
1029                 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_sectors_dirty),
1030                                 c, inode_i_sectors_wrong,
1031                                 "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
1032                                 w->last_pos.inode, i->snapshot,
1033                                 i->inode.bi_sectors, i->count)) {
1034                         i->inode.bi_sectors = i->count;
1035                         ret = fsck_write_inode(trans, &i->inode, i->snapshot);
1036                         if (ret)
1037                                 break;
1038                 }
1039         }
1040 fsck_err:
1041         bch_err_fn(c, ret);
1042         return ret ?: trans_was_restarted(trans, restart_count);
1043 }
1044
1045 struct extent_end {
1046         u32                     snapshot;
1047         u64                     offset;
1048         struct snapshots_seen   seen;
1049 };
1050
1051 struct extent_ends {
1052         struct bpos                     last_pos;
1053         DARRAY(struct extent_end)       e;
1054 };
1055
1056 static void extent_ends_reset(struct extent_ends *extent_ends)
1057 {
1058         darray_for_each(extent_ends->e, i)
1059                 snapshots_seen_exit(&i->seen);
1060         extent_ends->e.nr = 0;
1061 }
1062
1063 static void extent_ends_exit(struct extent_ends *extent_ends)
1064 {
1065         extent_ends_reset(extent_ends);
1066         darray_exit(&extent_ends->e);
1067 }
1068
1069 static void extent_ends_init(struct extent_ends *extent_ends)
1070 {
1071         memset(extent_ends, 0, sizeof(*extent_ends));
1072 }
1073
1074 static int extent_ends_at(struct bch_fs *c,
1075                           struct extent_ends *extent_ends,
1076                           struct snapshots_seen *seen,
1077                           struct bkey_s_c k)
1078 {
1079         struct extent_end *i, n = (struct extent_end) {
1080                 .offset         = k.k->p.offset,
1081                 .snapshot       = k.k->p.snapshot,
1082                 .seen           = *seen,
1083         };
1084
1085         n.seen.ids.data = kmemdup(seen->ids.data,
1086                               sizeof(seen->ids.data[0]) * seen->ids.size,
1087                               GFP_KERNEL);
1088         if (!n.seen.ids.data)
1089                 return -BCH_ERR_ENOMEM_fsck_extent_ends_at;
1090
1091         __darray_for_each(extent_ends->e, i) {
1092                 if (i->snapshot == k.k->p.snapshot) {
1093                         snapshots_seen_exit(&i->seen);
1094                         *i = n;
1095                         return 0;
1096                 }
1097
1098                 if (i->snapshot >= k.k->p.snapshot)
1099                         break;
1100         }
1101
1102         return darray_insert_item(&extent_ends->e, i - extent_ends->e.data, n);
1103 }
1104
1105 static int overlapping_extents_found(struct btree_trans *trans,
1106                                      enum btree_id btree,
1107                                      struct bpos pos1, struct snapshots_seen *pos1_seen,
1108                                      struct bkey pos2,
1109                                      bool *fixed,
1110                                      struct extent_end *extent_end)
1111 {
1112         struct bch_fs *c = trans->c;
1113         struct printbuf buf = PRINTBUF;
1114         struct btree_iter iter1, iter2 = { NULL };
1115         struct bkey_s_c k1, k2;
1116         int ret;
1117
1118         BUG_ON(bkey_le(pos1, bkey_start_pos(&pos2)));
1119
1120         bch2_trans_iter_init(trans, &iter1, btree, pos1,
1121                              BTREE_ITER_ALL_SNAPSHOTS|
1122                              BTREE_ITER_NOT_EXTENTS);
1123         k1 = bch2_btree_iter_peek_upto(&iter1, POS(pos1.inode, U64_MAX));
1124         ret = bkey_err(k1);
1125         if (ret)
1126                 goto err;
1127
1128         prt_str(&buf, "\n  ");
1129         bch2_bkey_val_to_text(&buf, c, k1);
1130
1131         if (!bpos_eq(pos1, k1.k->p)) {
1132                 prt_str(&buf, "\n  wanted\n  ");
1133                 bch2_bpos_to_text(&buf, pos1);
1134                 prt_str(&buf, "\n  ");
1135                 bch2_bkey_to_text(&buf, &pos2);
1136
1137                 bch_err(c, "%s: error finding first overlapping extent when repairing, got%s",
1138                         __func__, buf.buf);
1139                 ret = -BCH_ERR_internal_fsck_err;
1140                 goto err;
1141         }
1142
1143         bch2_trans_copy_iter(&iter2, &iter1);
1144
1145         while (1) {
1146                 bch2_btree_iter_advance(&iter2);
1147
1148                 k2 = bch2_btree_iter_peek_upto(&iter2, POS(pos1.inode, U64_MAX));
1149                 ret = bkey_err(k2);
1150                 if (ret)
1151                         goto err;
1152
1153                 if (bpos_ge(k2.k->p, pos2.p))
1154                         break;
1155         }
1156
1157         prt_str(&buf, "\n  ");
1158         bch2_bkey_val_to_text(&buf, c, k2);
1159
1160         if (bpos_gt(k2.k->p, pos2.p) ||
1161             pos2.size != k2.k->size) {
1162                 bch_err(c, "%s: error finding seconding overlapping extent when repairing%s",
1163                         __func__, buf.buf);
1164                 ret = -BCH_ERR_internal_fsck_err;
1165                 goto err;
1166         }
1167
1168         prt_printf(&buf, "\n  overwriting %s extent",
1169                    pos1.snapshot >= pos2.p.snapshot ? "first" : "second");
1170
1171         if (fsck_err(c, extent_overlapping,
1172                      "overlapping extents%s", buf.buf)) {
1173                 struct btree_iter *old_iter = &iter1;
1174                 struct disk_reservation res = { 0 };
1175
1176                 if (pos1.snapshot < pos2.p.snapshot) {
1177                         old_iter = &iter2;
1178                         swap(k1, k2);
1179                 }
1180
1181                 trans->extra_disk_res += bch2_bkey_sectors_compressed(k2);
1182
1183                 ret =   bch2_trans_update_extent_overwrite(trans, old_iter,
1184                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE,
1185                                 k1, k2) ?:
1186                         bch2_trans_commit(trans, &res, NULL, BCH_TRANS_COMMIT_no_enospc);
1187                 bch2_disk_reservation_put(c, &res);
1188
1189                 if (ret)
1190                         goto err;
1191
1192                 *fixed = true;
1193
1194                 if (pos1.snapshot == pos2.p.snapshot) {
1195                         /*
1196                          * We overwrote the first extent, and did the overwrite
1197                          * in the same snapshot:
1198                          */
1199                         extent_end->offset = bkey_start_offset(&pos2);
1200                 } else if (pos1.snapshot > pos2.p.snapshot) {
1201                         /*
1202                          * We overwrote the first extent in pos2's snapshot:
1203                          */
1204                         ret = snapshots_seen_add_inorder(c, pos1_seen, pos2.p.snapshot);
1205                 } else {
1206                         /*
1207                          * We overwrote the second extent - restart
1208                          * check_extent() from the top:
1209                          */
1210                         ret = -BCH_ERR_transaction_restart_nested;
1211                 }
1212         }
1213 fsck_err:
1214 err:
1215         bch2_trans_iter_exit(trans, &iter2);
1216         bch2_trans_iter_exit(trans, &iter1);
1217         printbuf_exit(&buf);
1218         return ret;
1219 }
1220
1221 static int check_overlapping_extents(struct btree_trans *trans,
1222                               struct snapshots_seen *seen,
1223                               struct extent_ends *extent_ends,
1224                               struct bkey_s_c k,
1225                               u32 equiv,
1226                               struct btree_iter *iter,
1227                               bool *fixed)
1228 {
1229         struct bch_fs *c = trans->c;
1230         int ret = 0;
1231
1232         /* transaction restart, running again */
1233         if (bpos_eq(extent_ends->last_pos, k.k->p))
1234                 return 0;
1235
1236         if (extent_ends->last_pos.inode != k.k->p.inode)
1237                 extent_ends_reset(extent_ends);
1238
1239         darray_for_each(extent_ends->e, i) {
1240                 if (i->offset <= bkey_start_offset(k.k))
1241                         continue;
1242
1243                 if (!ref_visible2(c,
1244                                   k.k->p.snapshot, seen,
1245                                   i->snapshot, &i->seen))
1246                         continue;
1247
1248                 ret = overlapping_extents_found(trans, iter->btree_id,
1249                                                 SPOS(iter->pos.inode,
1250                                                      i->offset,
1251                                                      i->snapshot),
1252                                                 &i->seen,
1253                                                 *k.k, fixed, i);
1254                 if (ret)
1255                         goto err;
1256         }
1257
1258         ret = extent_ends_at(c, extent_ends, seen, k);
1259         if (ret)
1260                 goto err;
1261
1262         extent_ends->last_pos = k.k->p;
1263 err:
1264         return ret;
1265 }
1266
1267 static int check_extent_overbig(struct btree_trans *trans, struct btree_iter *iter,
1268                                 struct bkey_s_c k)
1269 {
1270         struct bch_fs *c = trans->c;
1271         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1272         struct bch_extent_crc_unpacked crc;
1273         const union bch_extent_entry *i;
1274         unsigned encoded_extent_max_sectors = c->opts.encoded_extent_max >> 9;
1275
1276         bkey_for_each_crc(k.k, ptrs, crc, i)
1277                 if (crc_is_encoded(crc) &&
1278                     crc.uncompressed_size > encoded_extent_max_sectors) {
1279                         struct printbuf buf = PRINTBUF;
1280
1281                         bch2_bkey_val_to_text(&buf, c, k);
1282                         bch_err(c, "overbig encoded extent, please report this:\n  %s", buf.buf);
1283                         printbuf_exit(&buf);
1284                 }
1285
1286         return 0;
1287 }
1288
1289 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
1290                         struct bkey_s_c k,
1291                         struct inode_walker *inode,
1292                         struct snapshots_seen *s,
1293                         struct extent_ends *extent_ends)
1294 {
1295         struct bch_fs *c = trans->c;
1296         struct inode_walker_entry *i;
1297         struct printbuf buf = PRINTBUF;
1298         struct bpos equiv = k.k->p;
1299         int ret = 0;
1300
1301         equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1302
1303         ret = check_key_has_snapshot(trans, iter, k);
1304         if (ret) {
1305                 ret = ret < 0 ? ret : 0;
1306                 goto out;
1307         }
1308
1309         if (inode->last_pos.inode != k.k->p.inode) {
1310                 ret = check_i_sectors(trans, inode);
1311                 if (ret)
1312                         goto err;
1313         }
1314
1315         i = walk_inode(trans, inode, equiv, k.k->type == KEY_TYPE_whiteout);
1316         ret = PTR_ERR_OR_ZERO(i);
1317         if (ret)
1318                 goto err;
1319
1320         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1321         if (ret)
1322                 goto err;
1323
1324         if (k.k->type != KEY_TYPE_whiteout) {
1325                 if (fsck_err_on(!i, c, extent_in_missing_inode,
1326                                 "extent in missing inode:\n  %s",
1327                                 (printbuf_reset(&buf),
1328                                  bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1329                         goto delete;
1330
1331                 if (fsck_err_on(i &&
1332                                 !S_ISREG(i->inode.bi_mode) &&
1333                                 !S_ISLNK(i->inode.bi_mode),
1334                                 c, extent_in_non_reg_inode,
1335                                 "extent in non regular inode mode %o:\n  %s",
1336                                 i->inode.bi_mode,
1337                                 (printbuf_reset(&buf),
1338                                  bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1339                         goto delete;
1340
1341                 ret = check_overlapping_extents(trans, s, extent_ends, k,
1342                                                 equiv.snapshot, iter,
1343                                                 &inode->recalculate_sums);
1344                 if (ret)
1345                         goto err;
1346         }
1347
1348         /*
1349          * Check inodes in reverse order, from oldest snapshots to newest,
1350          * starting from the inode that matches this extent's snapshot. If we
1351          * didn't have one, iterate over all inodes:
1352          */
1353         if (!i)
1354                 i = inode->inodes.data + inode->inodes.nr - 1;
1355
1356         for (;
1357              inode->inodes.data && i >= inode->inodes.data;
1358              --i) {
1359                 if (i->snapshot > equiv.snapshot ||
1360                     !key_visible_in_snapshot(c, s, i->snapshot, equiv.snapshot))
1361                         continue;
1362
1363                 if (k.k->type != KEY_TYPE_whiteout) {
1364                         if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_size_dirty) &&
1365                                         k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9 &&
1366                                         !bkey_extent_is_reservation(k),
1367                                         c, extent_past_end_of_inode,
1368                                         "extent type past end of inode %llu:%u, i_size %llu\n  %s",
1369                                         i->inode.bi_inum, i->snapshot, i->inode.bi_size,
1370                                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1371                                 struct btree_iter iter2;
1372
1373                                 bch2_trans_copy_iter(&iter2, iter);
1374                                 bch2_btree_iter_set_snapshot(&iter2, i->snapshot);
1375                                 ret =   bch2_btree_iter_traverse(&iter2) ?:
1376                                         bch2_btree_delete_at(trans, &iter2,
1377                                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1378                                 bch2_trans_iter_exit(trans, &iter2);
1379                                 if (ret)
1380                                         goto err;
1381
1382                                 iter->k.type = KEY_TYPE_whiteout;
1383                         }
1384
1385                         if (bkey_extent_is_allocation(k.k))
1386                                 i->count += k.k->size;
1387                 }
1388
1389                 i->seen_this_pos = true;
1390         }
1391 out:
1392 err:
1393 fsck_err:
1394         printbuf_exit(&buf);
1395         bch_err_fn(c, ret);
1396         return ret;
1397 delete:
1398         ret = bch2_btree_delete_at(trans, iter, BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1399         goto out;
1400 }
1401
1402 /*
1403  * Walk extents: verify that extents have a corresponding S_ISREG inode, and
1404  * that i_size an i_sectors are consistent
1405  */
1406 int bch2_check_extents(struct bch_fs *c)
1407 {
1408         struct inode_walker w = inode_walker_init();
1409         struct snapshots_seen s;
1410         struct extent_ends extent_ends;
1411         struct disk_reservation res = { 0 };
1412
1413         snapshots_seen_init(&s);
1414         extent_ends_init(&extent_ends);
1415
1416         int ret = bch2_trans_run(c,
1417                 for_each_btree_key_commit(trans, iter, BTREE_ID_extents,
1418                                 POS(BCACHEFS_ROOT_INO, 0),
1419                                 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1420                                 &res, NULL,
1421                                 BCH_TRANS_COMMIT_no_enospc, ({
1422                         bch2_disk_reservation_put(c, &res);
1423                         check_extent(trans, &iter, k, &w, &s, &extent_ends) ?:
1424                         check_extent_overbig(trans, &iter, k);
1425                 })) ?:
1426                 check_i_sectors(trans, &w));
1427
1428         bch2_disk_reservation_put(c, &res);
1429         extent_ends_exit(&extent_ends);
1430         inode_walker_exit(&w);
1431         snapshots_seen_exit(&s);
1432
1433         bch_err_fn(c, ret);
1434         return ret;
1435 }
1436
1437 int bch2_check_indirect_extents(struct bch_fs *c)
1438 {
1439         struct disk_reservation res = { 0 };
1440
1441         int ret = bch2_trans_run(c,
1442                 for_each_btree_key_commit(trans, iter, BTREE_ID_reflink,
1443                                 POS_MIN,
1444                                 BTREE_ITER_PREFETCH, k,
1445                                 &res, NULL,
1446                                 BCH_TRANS_COMMIT_no_enospc, ({
1447                         bch2_disk_reservation_put(c, &res);
1448                         check_extent_overbig(trans, &iter, k);
1449                 })));
1450
1451         bch2_disk_reservation_put(c, &res);
1452         bch_err_fn(c, ret);
1453         return ret;
1454 }
1455
1456 static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
1457 {
1458         struct bch_fs *c = trans->c;
1459         u32 restart_count = trans->restart_count;
1460         int ret = 0;
1461         s64 count2;
1462
1463         darray_for_each(w->inodes, i) {
1464                 if (i->inode.bi_nlink == i->count)
1465                         continue;
1466
1467                 count2 = bch2_count_subdirs(trans, w->last_pos.inode, i->snapshot);
1468                 if (count2 < 0)
1469                         return count2;
1470
1471                 if (i->count != count2) {
1472                         bch_err(c, "fsck counted subdirectories wrong: got %llu should be %llu",
1473                                 i->count, count2);
1474                         i->count = count2;
1475                         if (i->inode.bi_nlink == i->count)
1476                                 continue;
1477                 }
1478
1479                 if (fsck_err_on(i->inode.bi_nlink != i->count,
1480                                 c, inode_dir_wrong_nlink,
1481                                 "directory %llu:%u with wrong i_nlink: got %u, should be %llu",
1482                                 w->last_pos.inode, i->snapshot, i->inode.bi_nlink, i->count)) {
1483                         i->inode.bi_nlink = i->count;
1484                         ret = fsck_write_inode(trans, &i->inode, i->snapshot);
1485                         if (ret)
1486                                 break;
1487                 }
1488         }
1489 fsck_err:
1490         bch_err_fn(c, ret);
1491         return ret ?: trans_was_restarted(trans, restart_count);
1492 }
1493
1494 static int check_dirent_target(struct btree_trans *trans,
1495                                struct btree_iter *iter,
1496                                struct bkey_s_c_dirent d,
1497                                struct bch_inode_unpacked *target,
1498                                u32 target_snapshot)
1499 {
1500         struct bch_fs *c = trans->c;
1501         struct bkey_i_dirent *n;
1502         struct printbuf buf = PRINTBUF;
1503         struct btree_iter bp_iter = { NULL };
1504         int ret = 0;
1505
1506         if (!target->bi_dir &&
1507             !target->bi_dir_offset) {
1508                 target->bi_dir          = d.k->p.inode;
1509                 target->bi_dir_offset   = d.k->p.offset;
1510
1511                 ret = __write_inode(trans, target, target_snapshot);
1512                 if (ret)
1513                         goto err;
1514         }
1515
1516         if (!inode_points_to_dirent(target, d)) {
1517                 struct bkey_s_c_dirent bp_dirent = dirent_get_by_pos(trans, &bp_iter,
1518                                       SPOS(target->bi_dir, target->bi_dir_offset, target_snapshot));
1519                 ret = bkey_err(bp_dirent);
1520                 if (ret && !bch2_err_matches(ret, ENOENT))
1521                         goto err;
1522
1523                 bool backpointer_exists = !ret;
1524                 ret = 0;
1525
1526                 bch2_bkey_val_to_text(&buf, c, d.s_c);
1527                 prt_newline(&buf);
1528                 if (backpointer_exists)
1529                         bch2_bkey_val_to_text(&buf, c, bp_dirent.s_c);
1530
1531                 if (fsck_err_on(S_ISDIR(target->bi_mode) && backpointer_exists,
1532                                 c, inode_dir_multiple_links,
1533                                 "directory %llu:%u with multiple links\n%s",
1534                                 target->bi_inum, target_snapshot, buf.buf)) {
1535                         ret = __remove_dirent(trans, d.k->p);
1536                         goto out;
1537                 }
1538
1539                 /*
1540                  * hardlinked file with nlink 0:
1541                  * We're just adjusting nlink here so check_nlinks() will pick
1542                  * it up, it ignores inodes with nlink 0
1543                  */
1544                 if (fsck_err_on(backpointer_exists && !target->bi_nlink,
1545                                 c, inode_multiple_links_but_nlink_0,
1546                                 "inode %llu:%u type %s has multiple links but i_nlink 0\n%s",
1547                                 target->bi_inum, target_snapshot, bch2_d_types[d.v->d_type], buf.buf)) {
1548                         target->bi_nlink++;
1549                         target->bi_flags &= ~BCH_INODE_unlinked;
1550
1551                         ret = __write_inode(trans, target, target_snapshot);
1552                         if (ret)
1553                                 goto err;
1554                 }
1555
1556                 if (fsck_err_on(!backpointer_exists,
1557                                 c, inode_wrong_backpointer,
1558                                 "inode %llu:%u has wrong backpointer:\n"
1559                                 "got       %llu:%llu\n"
1560                                 "should be %llu:%llu",
1561                                 target->bi_inum, target_snapshot,
1562                                 target->bi_dir,
1563                                 target->bi_dir_offset,
1564                                 d.k->p.inode,
1565                                 d.k->p.offset)) {
1566                         target->bi_dir          = d.k->p.inode;
1567                         target->bi_dir_offset   = d.k->p.offset;
1568
1569                         ret = __write_inode(trans, target, target_snapshot);
1570                         if (ret)
1571                                 goto err;
1572                 }
1573         }
1574
1575         if (fsck_err_on(d.v->d_type != inode_d_type(target),
1576                         c, dirent_d_type_wrong,
1577                         "incorrect d_type: got %s, should be %s:\n%s",
1578                         bch2_d_type_str(d.v->d_type),
1579                         bch2_d_type_str(inode_d_type(target)),
1580                         (printbuf_reset(&buf),
1581                          bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
1582                 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1583                 ret = PTR_ERR_OR_ZERO(n);
1584                 if (ret)
1585                         goto err;
1586
1587                 bkey_reassemble(&n->k_i, d.s_c);
1588                 n->v.d_type = inode_d_type(target);
1589
1590                 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1591                 if (ret)
1592                         goto err;
1593
1594                 d = dirent_i_to_s_c(n);
1595         }
1596
1597         if (fsck_err_on(d.v->d_type == DT_SUBVOL &&
1598                         target->bi_parent_subvol != le32_to_cpu(d.v->d_parent_subvol),
1599                         c, dirent_d_parent_subvol_wrong,
1600                         "dirent has wrong d_parent_subvol field: got %u, should be %u",
1601                         le32_to_cpu(d.v->d_parent_subvol),
1602                         target->bi_parent_subvol)) {
1603                 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1604                 ret = PTR_ERR_OR_ZERO(n);
1605                 if (ret)
1606                         goto err;
1607
1608                 bkey_reassemble(&n->k_i, d.s_c);
1609                 n->v.d_parent_subvol = cpu_to_le32(target->bi_parent_subvol);
1610
1611                 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1612                 if (ret)
1613                         goto err;
1614
1615                 d = dirent_i_to_s_c(n);
1616         }
1617 out:
1618 err:
1619 fsck_err:
1620         bch2_trans_iter_exit(trans, &bp_iter);
1621         printbuf_exit(&buf);
1622         bch_err_fn(c, ret);
1623         return ret;
1624 }
1625
1626 static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
1627                         struct bkey_s_c k,
1628                         struct bch_hash_info *hash_info,
1629                         struct inode_walker *dir,
1630                         struct inode_walker *target,
1631                         struct snapshots_seen *s)
1632 {
1633         struct bch_fs *c = trans->c;
1634         struct bkey_s_c_dirent d;
1635         struct inode_walker_entry *i;
1636         struct printbuf buf = PRINTBUF;
1637         struct bpos equiv;
1638         int ret = 0;
1639
1640         ret = check_key_has_snapshot(trans, iter, k);
1641         if (ret) {
1642                 ret = ret < 0 ? ret : 0;
1643                 goto out;
1644         }
1645
1646         equiv = k.k->p;
1647         equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1648
1649         ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1650         if (ret)
1651                 goto err;
1652
1653         if (k.k->type == KEY_TYPE_whiteout)
1654                 goto out;
1655
1656         if (dir->last_pos.inode != k.k->p.inode) {
1657                 ret = check_subdir_count(trans, dir);
1658                 if (ret)
1659                         goto err;
1660         }
1661
1662         BUG_ON(!btree_iter_path(trans, iter)->should_be_locked);
1663
1664         i = walk_inode(trans, dir, equiv, k.k->type == KEY_TYPE_whiteout);
1665         ret = PTR_ERR_OR_ZERO(i);
1666         if (ret < 0)
1667                 goto err;
1668
1669         if (dir->first_this_inode && dir->inodes.nr)
1670                 *hash_info = bch2_hash_info_init(c, &dir->inodes.data[0].inode);
1671         dir->first_this_inode = false;
1672
1673         if (fsck_err_on(!i, c, dirent_in_missing_dir_inode,
1674                         "dirent in nonexisting directory:\n%s",
1675                         (printbuf_reset(&buf),
1676                          bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1677                 ret = bch2_btree_delete_at(trans, iter,
1678                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1679                 goto out;
1680         }
1681
1682         if (!i)
1683                 goto out;
1684
1685         if (fsck_err_on(!S_ISDIR(i->inode.bi_mode),
1686                         c, dirent_in_non_dir_inode,
1687                         "dirent in non directory inode type %s:\n%s",
1688                         bch2_d_type_str(inode_d_type(&i->inode)),
1689                         (printbuf_reset(&buf),
1690                          bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1691                 ret = bch2_btree_delete_at(trans, iter, 0);
1692                 goto out;
1693         }
1694
1695         ret = hash_check_key(trans, bch2_dirent_hash_desc, hash_info, iter, k);
1696         if (ret < 0)
1697                 goto err;
1698         if (ret) {
1699                 /* dirent has been deleted */
1700                 ret = 0;
1701                 goto out;
1702         }
1703
1704         if (k.k->type != KEY_TYPE_dirent)
1705                 goto out;
1706
1707         d = bkey_s_c_to_dirent(k);
1708
1709         if (d.v->d_type == DT_SUBVOL) {
1710                 struct bch_inode_unpacked subvol_root;
1711                 u32 target_subvol = le32_to_cpu(d.v->d_child_subvol);
1712                 u32 target_snapshot;
1713                 u64 target_inum;
1714
1715                 ret = subvol_lookup(trans, target_subvol,
1716                                       &target_snapshot, &target_inum);
1717                 if (ret && !bch2_err_matches(ret, ENOENT))
1718                         goto err;
1719
1720                 if (fsck_err_on(ret, c, dirent_to_missing_subvol,
1721                                 "dirent points to missing subvolume %u",
1722                                 le32_to_cpu(d.v->d_child_subvol))) {
1723                         ret = __remove_dirent(trans, d.k->p);
1724                         goto err;
1725                 }
1726
1727                 ret = lookup_inode(trans, target_inum,
1728                                    &subvol_root, &target_snapshot);
1729                 if (ret && !bch2_err_matches(ret, ENOENT))
1730                         goto err;
1731
1732                 if (fsck_err_on(ret, c, subvol_to_missing_root,
1733                                 "subvolume %u points to missing subvolume root %llu",
1734                                 target_subvol,
1735                                 target_inum)) {
1736                         bch_err(c, "repair not implemented yet");
1737                         ret = -EINVAL;
1738                         goto err;
1739                 }
1740
1741                 if (fsck_err_on(subvol_root.bi_subvol != target_subvol,
1742                                 c, subvol_root_wrong_bi_subvol,
1743                                 "subvol root %llu has wrong bi_subvol field: got %u, should be %u",
1744                                 target_inum,
1745                                 subvol_root.bi_subvol, target_subvol)) {
1746                         subvol_root.bi_subvol = target_subvol;
1747                         ret = __write_inode(trans, &subvol_root, target_snapshot);
1748                         if (ret)
1749                                 goto err;
1750                 }
1751
1752                 ret = check_dirent_target(trans, iter, d, &subvol_root,
1753                                           target_snapshot);
1754                 if (ret)
1755                         goto err;
1756         } else {
1757                 ret = __get_visible_inodes(trans, target, s, le64_to_cpu(d.v->d_inum));
1758                 if (ret)
1759                         goto err;
1760
1761                 if (fsck_err_on(!target->inodes.nr,
1762                                 c, dirent_to_missing_inode,
1763                                 "dirent points to missing inode: (equiv %u)\n%s",
1764                                 equiv.snapshot,
1765                                 (printbuf_reset(&buf),
1766                                  bch2_bkey_val_to_text(&buf, c, k),
1767                                  buf.buf))) {
1768                         ret = __remove_dirent(trans, d.k->p);
1769                         if (ret)
1770                                 goto err;
1771                 }
1772
1773                 darray_for_each(target->inodes, i) {
1774                         ret = check_dirent_target(trans, iter, d,
1775                                                   &i->inode, i->snapshot);
1776                         if (ret)
1777                                 goto err;
1778                 }
1779         }
1780
1781         if (d.v->d_type == DT_DIR)
1782                 for_each_visible_inode(c, s, dir, equiv.snapshot, i)
1783                         i->count++;
1784
1785 out:
1786 err:
1787 fsck_err:
1788         printbuf_exit(&buf);
1789         bch_err_fn(c, ret);
1790         return ret;
1791 }
1792
1793 /*
1794  * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
1795  * validate d_type
1796  */
1797 int bch2_check_dirents(struct bch_fs *c)
1798 {
1799         struct inode_walker dir = inode_walker_init();
1800         struct inode_walker target = inode_walker_init();
1801         struct snapshots_seen s;
1802         struct bch_hash_info hash_info;
1803
1804         snapshots_seen_init(&s);
1805
1806         int ret = bch2_trans_run(c,
1807                 for_each_btree_key_commit(trans, iter, BTREE_ID_dirents,
1808                                 POS(BCACHEFS_ROOT_INO, 0),
1809                                 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
1810                                 k,
1811                                 NULL, NULL,
1812                                 BCH_TRANS_COMMIT_no_enospc,
1813                         check_dirent(trans, &iter, k, &hash_info, &dir, &target, &s)));
1814
1815         snapshots_seen_exit(&s);
1816         inode_walker_exit(&dir);
1817         inode_walker_exit(&target);
1818         bch_err_fn(c, ret);
1819         return ret;
1820 }
1821
1822 static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
1823                        struct bkey_s_c k,
1824                        struct bch_hash_info *hash_info,
1825                        struct inode_walker *inode)
1826 {
1827         struct bch_fs *c = trans->c;
1828         struct inode_walker_entry *i;
1829         int ret;
1830
1831         ret = check_key_has_snapshot(trans, iter, k);
1832         if (ret)
1833                 return ret;
1834
1835         i = walk_inode(trans, inode, k.k->p, k.k->type == KEY_TYPE_whiteout);
1836         ret = PTR_ERR_OR_ZERO(i);
1837         if (ret)
1838                 return ret;
1839
1840         if (inode->first_this_inode && inode->inodes.nr)
1841                 *hash_info = bch2_hash_info_init(c, &inode->inodes.data[0].inode);
1842         inode->first_this_inode = false;
1843
1844         if (fsck_err_on(!i, c, xattr_in_missing_inode,
1845                         "xattr for missing inode %llu",
1846                         k.k->p.inode))
1847                 return bch2_btree_delete_at(trans, iter, 0);
1848
1849         if (!i)
1850                 return 0;
1851
1852         ret = hash_check_key(trans, bch2_xattr_hash_desc, hash_info, iter, k);
1853 fsck_err:
1854         bch_err_fn(c, ret);
1855         return ret;
1856 }
1857
1858 /*
1859  * Walk xattrs: verify that they all have a corresponding inode
1860  */
1861 int bch2_check_xattrs(struct bch_fs *c)
1862 {
1863         struct inode_walker inode = inode_walker_init();
1864         struct bch_hash_info hash_info;
1865         int ret = 0;
1866
1867         ret = bch2_trans_run(c,
1868                 for_each_btree_key_commit(trans, iter, BTREE_ID_xattrs,
1869                         POS(BCACHEFS_ROOT_INO, 0),
1870                         BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
1871                         k,
1872                         NULL, NULL,
1873                         BCH_TRANS_COMMIT_no_enospc,
1874                 check_xattr(trans, &iter, k, &hash_info, &inode)));
1875         bch_err_fn(c, ret);
1876         return ret;
1877 }
1878
1879 static int check_root_trans(struct btree_trans *trans)
1880 {
1881         struct bch_fs *c = trans->c;
1882         struct bch_inode_unpacked root_inode;
1883         u32 snapshot;
1884         u64 inum;
1885         int ret;
1886
1887         ret = subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
1888         if (ret && !bch2_err_matches(ret, ENOENT))
1889                 return ret;
1890
1891         if (mustfix_fsck_err_on(ret, c, root_subvol_missing,
1892                                 "root subvol missing")) {
1893                 struct bkey_i_subvolume root_subvol;
1894
1895                 snapshot        = U32_MAX;
1896                 inum            = BCACHEFS_ROOT_INO;
1897
1898                 bkey_subvolume_init(&root_subvol.k_i);
1899                 root_subvol.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1900                 root_subvol.v.flags     = 0;
1901                 root_subvol.v.snapshot  = cpu_to_le32(snapshot);
1902                 root_subvol.v.inode     = cpu_to_le64(inum);
1903                 ret = bch2_btree_insert_trans(trans, BTREE_ID_subvolumes, &root_subvol.k_i, 0);
1904                 bch_err_msg(c, ret, "writing root subvol");
1905                 if (ret)
1906                         goto err;
1907         }
1908
1909         ret = lookup_inode(trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
1910         if (ret && !bch2_err_matches(ret, ENOENT))
1911                 return ret;
1912
1913         if (mustfix_fsck_err_on(ret, c, root_dir_missing,
1914                                 "root directory missing") ||
1915             mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode),
1916                                 c, root_inode_not_dir,
1917                                 "root inode not a directory")) {
1918                 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755,
1919                                 0, NULL);
1920                 root_inode.bi_inum = inum;
1921
1922                 ret = __write_inode(trans, &root_inode, snapshot);
1923                 bch_err_msg(c, ret, "writing root inode");
1924         }
1925 err:
1926 fsck_err:
1927         return ret;
1928 }
1929
1930 /* Get root directory, create if it doesn't exist: */
1931 int bch2_check_root(struct bch_fs *c)
1932 {
1933         int ret = bch2_trans_do(c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1934                 check_root_trans(trans));
1935         bch_err_fn(c, ret);
1936         return ret;
1937 }
1938
1939 struct pathbuf_entry {
1940         u64     inum;
1941         u32     snapshot;
1942 };
1943
1944 typedef DARRAY(struct pathbuf_entry) pathbuf;
1945
1946 static bool path_is_dup(pathbuf *p, u64 inum, u32 snapshot)
1947 {
1948         darray_for_each(*p, i)
1949                 if (i->inum     == inum &&
1950                     i->snapshot == snapshot)
1951                         return true;
1952         return false;
1953 }
1954
1955 static int path_down(struct bch_fs *c, pathbuf *p,
1956                      u64 inum, u32 snapshot)
1957 {
1958         int ret = darray_push(p, ((struct pathbuf_entry) {
1959                 .inum           = inum,
1960                 .snapshot       = snapshot,
1961         }));
1962
1963         if (ret)
1964                 bch_err(c, "fsck: error allocating memory for pathbuf, size %zu",
1965                         p->size);
1966         return ret;
1967 }
1968
1969 /*
1970  * Check that a given inode is reachable from the root:
1971  *
1972  * XXX: we should also be verifying that inodes are in the right subvolumes
1973  */
1974 static int check_path(struct btree_trans *trans,
1975                       pathbuf *p,
1976                       struct bch_inode_unpacked *inode,
1977                       u32 snapshot)
1978 {
1979         struct bch_fs *c = trans->c;
1980         int ret = 0;
1981
1982         snapshot = bch2_snapshot_equiv(c, snapshot);
1983         p->nr = 0;
1984
1985         while (!(inode->bi_inum == BCACHEFS_ROOT_INO &&
1986                  inode->bi_subvol == BCACHEFS_ROOT_SUBVOL)) {
1987                 struct btree_iter dirent_iter;
1988                 struct bkey_s_c_dirent d;
1989                 u32 parent_snapshot = snapshot;
1990
1991                 if (inode->bi_subvol) {
1992                         u64 inum;
1993
1994                         ret = subvol_lookup(trans, inode->bi_parent_subvol,
1995                                             &parent_snapshot, &inum);
1996                         if (ret)
1997                                 break;
1998                 }
1999
2000                 d = dirent_get_by_pos(trans, &dirent_iter,
2001                                       SPOS(inode->bi_dir, inode->bi_dir_offset,
2002                                            parent_snapshot));
2003                 ret = bkey_err(d.s_c);
2004                 if (ret && !bch2_err_matches(ret, ENOENT))
2005                         break;
2006
2007                 if (!ret && !dirent_points_to_inode(d, inode)) {
2008                         bch2_trans_iter_exit(trans, &dirent_iter);
2009                         ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
2010                 }
2011
2012                 if (bch2_err_matches(ret, ENOENT)) {
2013                         if (fsck_err(c,  inode_unreachable,
2014                                      "unreachable inode %llu:%u, type %s nlink %u backptr %llu:%llu",
2015                                      inode->bi_inum, snapshot,
2016                                      bch2_d_type_str(inode_d_type(inode)),
2017                                      inode->bi_nlink,
2018                                      inode->bi_dir,
2019                                      inode->bi_dir_offset))
2020                                 ret = reattach_inode(trans, inode, snapshot);
2021                         break;
2022                 }
2023
2024                 bch2_trans_iter_exit(trans, &dirent_iter);
2025
2026                 if (!S_ISDIR(inode->bi_mode))
2027                         break;
2028
2029                 ret = path_down(c, p, inode->bi_inum, snapshot);
2030                 if (ret) {
2031                         bch_err(c, "memory allocation failure");
2032                         return ret;
2033                 }
2034
2035                 snapshot = parent_snapshot;
2036
2037                 ret = lookup_inode(trans, inode->bi_dir, inode, &snapshot);
2038                 if (ret) {
2039                         /* Should have been caught in dirents pass */
2040                         if (!bch2_err_matches(ret, BCH_ERR_transaction_restart))
2041                                 bch_err(c, "error looking up parent directory: %i", ret);
2042                         break;
2043                 }
2044
2045                 if (path_is_dup(p, inode->bi_inum, snapshot)) {
2046                         /* XXX print path */
2047                         bch_err(c, "directory structure loop");
2048
2049                         darray_for_each(*p, i)
2050                                 pr_err("%llu:%u", i->inum, i->snapshot);
2051                         pr_err("%llu:%u", inode->bi_inum, snapshot);
2052
2053                         if (!fsck_err(c, dir_loop, "directory structure loop"))
2054                                 return 0;
2055
2056                         ret = remove_backpointer(trans, inode);
2057                         if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
2058                                 bch_err_msg(c, ret, "removing dirent");
2059                         if (ret)
2060                                 break;
2061
2062                         ret = reattach_inode(trans, inode, snapshot);
2063                         if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
2064                                 bch_err_msg(c, ret, "reattaching inode %llu", inode->bi_inum);
2065                         break;
2066                 }
2067         }
2068 fsck_err:
2069         bch_err_fn(c, ret);
2070         return ret;
2071 }
2072
2073 /*
2074  * Check for unreachable inodes, as well as loops in the directory structure:
2075  * After bch2_check_dirents(), if an inode backpointer doesn't exist that means it's
2076  * unreachable:
2077  */
2078 int bch2_check_directory_structure(struct bch_fs *c)
2079 {
2080         struct bch_inode_unpacked u;
2081         pathbuf path = { 0, };
2082         int ret;
2083
2084         ret = bch2_trans_run(c,
2085                 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes, POS_MIN,
2086                                           BTREE_ITER_INTENT|
2087                                           BTREE_ITER_PREFETCH|
2088                                           BTREE_ITER_ALL_SNAPSHOTS, k,
2089                                           NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
2090                         if (!bkey_is_inode(k.k))
2091                                 continue;
2092
2093                         BUG_ON(bch2_inode_unpack(k, &u));
2094
2095                         if (u.bi_flags & BCH_INODE_unlinked)
2096                                 continue;
2097
2098                         check_path(trans, &path, &u, iter.pos.snapshot);
2099                 })));
2100         darray_exit(&path);
2101
2102         bch_err_fn(c, ret);
2103         return ret;
2104 }
2105
2106 struct nlink_table {
2107         size_t          nr;
2108         size_t          size;
2109
2110         struct nlink {
2111                 u64     inum;
2112                 u32     snapshot;
2113                 u32     count;
2114         }               *d;
2115 };
2116
2117 static int add_nlink(struct bch_fs *c, struct nlink_table *t,
2118                      u64 inum, u32 snapshot)
2119 {
2120         if (t->nr == t->size) {
2121                 size_t new_size = max_t(size_t, 128UL, t->size * 2);
2122                 void *d = kvmalloc_array(new_size, sizeof(t->d[0]), GFP_KERNEL);
2123
2124                 if (!d) {
2125                         bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
2126                                 new_size);
2127                         return -BCH_ERR_ENOMEM_fsck_add_nlink;
2128                 }
2129
2130                 if (t->d)
2131                         memcpy(d, t->d, t->size * sizeof(t->d[0]));
2132                 kvfree(t->d);
2133
2134                 t->d = d;
2135                 t->size = new_size;
2136         }
2137
2138
2139         t->d[t->nr++] = (struct nlink) {
2140                 .inum           = inum,
2141                 .snapshot       = snapshot,
2142         };
2143
2144         return 0;
2145 }
2146
2147 static int nlink_cmp(const void *_l, const void *_r)
2148 {
2149         const struct nlink *l = _l;
2150         const struct nlink *r = _r;
2151
2152         return cmp_int(l->inum, r->inum);
2153 }
2154
2155 static void inc_link(struct bch_fs *c, struct snapshots_seen *s,
2156                      struct nlink_table *links,
2157                      u64 range_start, u64 range_end, u64 inum, u32 snapshot)
2158 {
2159         struct nlink *link, key = {
2160                 .inum = inum, .snapshot = U32_MAX,
2161         };
2162
2163         if (inum < range_start || inum >= range_end)
2164                 return;
2165
2166         link = __inline_bsearch(&key, links->d, links->nr,
2167                                 sizeof(links->d[0]), nlink_cmp);
2168         if (!link)
2169                 return;
2170
2171         while (link > links->d && link[0].inum == link[-1].inum)
2172                 --link;
2173
2174         for (; link < links->d + links->nr && link->inum == inum; link++)
2175                 if (ref_visible(c, s, snapshot, link->snapshot)) {
2176                         link->count++;
2177                         if (link->snapshot >= snapshot)
2178                                 break;
2179                 }
2180 }
2181
2182 noinline_for_stack
2183 static int check_nlinks_find_hardlinks(struct bch_fs *c,
2184                                        struct nlink_table *t,
2185                                        u64 start, u64 *end)
2186 {
2187         int ret = bch2_trans_run(c,
2188                 for_each_btree_key(trans, iter, BTREE_ID_inodes,
2189                                    POS(0, start),
2190                                    BTREE_ITER_INTENT|
2191                                    BTREE_ITER_PREFETCH|
2192                                    BTREE_ITER_ALL_SNAPSHOTS, k, ({
2193                         if (!bkey_is_inode(k.k))
2194                                 continue;
2195
2196                         /* Should never fail, checked by bch2_inode_invalid: */
2197                         struct bch_inode_unpacked u;
2198                         BUG_ON(bch2_inode_unpack(k, &u));
2199
2200                         /*
2201                          * Backpointer and directory structure checks are sufficient for
2202                          * directories, since they can't have hardlinks:
2203                          */
2204                         if (S_ISDIR(u.bi_mode))
2205                                 continue;
2206
2207                         if (!u.bi_nlink)
2208                                 continue;
2209
2210                         ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot);
2211                         if (ret) {
2212                                 *end = k.k->p.offset;
2213                                 ret = 0;
2214                                 break;
2215                         }
2216                         0;
2217                 })));
2218
2219         bch_err_fn(c, ret);
2220         return ret;
2221 }
2222
2223 noinline_for_stack
2224 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
2225                                      u64 range_start, u64 range_end)
2226 {
2227         struct snapshots_seen s;
2228
2229         snapshots_seen_init(&s);
2230
2231         int ret = bch2_trans_run(c,
2232                 for_each_btree_key(trans, iter, BTREE_ID_dirents, POS_MIN,
2233                                    BTREE_ITER_INTENT|
2234                                    BTREE_ITER_PREFETCH|
2235                                    BTREE_ITER_ALL_SNAPSHOTS, k, ({
2236                         ret = snapshots_seen_update(c, &s, iter.btree_id, k.k->p);
2237                         if (ret)
2238                                 break;
2239
2240                         if (k.k->type == KEY_TYPE_dirent) {
2241                                 struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
2242
2243                                 if (d.v->d_type != DT_DIR &&
2244                                     d.v->d_type != DT_SUBVOL)
2245                                         inc_link(c, &s, links, range_start, range_end,
2246                                                  le64_to_cpu(d.v->d_inum),
2247                                                  bch2_snapshot_equiv(c, d.k->p.snapshot));
2248                         }
2249                         0;
2250                 })));
2251
2252         snapshots_seen_exit(&s);
2253
2254         bch_err_fn(c, ret);
2255         return ret;
2256 }
2257
2258 static int check_nlinks_update_inode(struct btree_trans *trans, struct btree_iter *iter,
2259                                      struct bkey_s_c k,
2260                                      struct nlink_table *links,
2261                                      size_t *idx, u64 range_end)
2262 {
2263         struct bch_fs *c = trans->c;
2264         struct bch_inode_unpacked u;
2265         struct nlink *link = &links->d[*idx];
2266         int ret = 0;
2267
2268         if (k.k->p.offset >= range_end)
2269                 return 1;
2270
2271         if (!bkey_is_inode(k.k))
2272                 return 0;
2273
2274         BUG_ON(bch2_inode_unpack(k, &u));
2275
2276         if (S_ISDIR(u.bi_mode))
2277                 return 0;
2278
2279         if (!u.bi_nlink)
2280                 return 0;
2281
2282         while ((cmp_int(link->inum, k.k->p.offset) ?:
2283                 cmp_int(link->snapshot, k.k->p.snapshot)) < 0) {
2284                 BUG_ON(*idx == links->nr);
2285                 link = &links->d[++*idx];
2286         }
2287
2288         if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count,
2289                         c, inode_wrong_nlink,
2290                         "inode %llu type %s has wrong i_nlink (%u, should be %u)",
2291                         u.bi_inum, bch2_d_types[mode_to_type(u.bi_mode)],
2292                         bch2_inode_nlink_get(&u), link->count)) {
2293                 bch2_inode_nlink_set(&u, link->count);
2294                 ret = __write_inode(trans, &u, k.k->p.snapshot);
2295         }
2296 fsck_err:
2297         return ret;
2298 }
2299
2300 noinline_for_stack
2301 static int check_nlinks_update_hardlinks(struct bch_fs *c,
2302                                struct nlink_table *links,
2303                                u64 range_start, u64 range_end)
2304 {
2305         size_t idx = 0;
2306
2307         int ret = bch2_trans_run(c,
2308                 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
2309                                 POS(0, range_start),
2310                                 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
2311                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2312                         check_nlinks_update_inode(trans, &iter, k, links, &idx, range_end)));
2313         if (ret < 0) {
2314                 bch_err(c, "error in fsck walking inodes: %s", bch2_err_str(ret));
2315                 return ret;
2316         }
2317
2318         return 0;
2319 }
2320
2321 int bch2_check_nlinks(struct bch_fs *c)
2322 {
2323         struct nlink_table links = { 0 };
2324         u64 this_iter_range_start, next_iter_range_start = 0;
2325         int ret = 0;
2326
2327         do {
2328                 this_iter_range_start = next_iter_range_start;
2329                 next_iter_range_start = U64_MAX;
2330
2331                 ret = check_nlinks_find_hardlinks(c, &links,
2332                                                   this_iter_range_start,
2333                                                   &next_iter_range_start);
2334
2335                 ret = check_nlinks_walk_dirents(c, &links,
2336                                           this_iter_range_start,
2337                                           next_iter_range_start);
2338                 if (ret)
2339                         break;
2340
2341                 ret = check_nlinks_update_hardlinks(c, &links,
2342                                          this_iter_range_start,
2343                                          next_iter_range_start);
2344                 if (ret)
2345                         break;
2346
2347                 links.nr = 0;
2348         } while (next_iter_range_start != U64_MAX);
2349
2350         kvfree(links.d);
2351         bch_err_fn(c, ret);
2352         return ret;
2353 }
2354
2355 static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter,
2356                              struct bkey_s_c k)
2357 {
2358         struct bkey_s_c_reflink_p p;
2359         struct bkey_i_reflink_p *u;
2360
2361         if (k.k->type != KEY_TYPE_reflink_p)
2362                 return 0;
2363
2364         p = bkey_s_c_to_reflink_p(k);
2365
2366         if (!p.v->front_pad && !p.v->back_pad)
2367                 return 0;
2368
2369         u = bch2_trans_kmalloc(trans, sizeof(*u));
2370         int ret = PTR_ERR_OR_ZERO(u);
2371         if (ret)
2372                 return ret;
2373
2374         bkey_reassemble(&u->k_i, k);
2375         u->v.front_pad  = 0;
2376         u->v.back_pad   = 0;
2377
2378         return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_NORUN);
2379 }
2380
2381 int bch2_fix_reflink_p(struct bch_fs *c)
2382 {
2383         if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
2384                 return 0;
2385
2386         int ret = bch2_trans_run(c,
2387                 for_each_btree_key_commit(trans, iter,
2388                                 BTREE_ID_extents, POS_MIN,
2389                                 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|
2390                                 BTREE_ITER_ALL_SNAPSHOTS, k,
2391                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
2392                         fix_reflink_p_key(trans, &iter, k)));
2393         bch_err_fn(c, ret);
2394         return ret;
2395 }