1 // SPDX-License-Identifier: GPL-2.0
5 #include "btree_update.h"
12 #include "subvolume.h"
16 #include <linux/bsearch.h>
17 #include <linux/dcache.h> /* struct qstr */
19 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
21 static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum,
24 struct btree_iter iter;
29 for_each_btree_key(trans, iter, BTREE_ID_extents,
30 SPOS(inum, 0, snapshot), 0, k, ret) {
31 if (k.k->p.inode != inum)
34 if (bkey_extent_is_allocation(k.k))
38 bch2_trans_iter_exit(trans, &iter);
40 return ret ?: sectors;
43 static s64 bch2_count_subdirs(struct btree_trans *trans, u64 inum,
46 struct btree_iter iter;
48 struct bkey_s_c_dirent d;
52 for_each_btree_key(trans, iter, BTREE_ID_dirents,
53 SPOS(inum, 0, snapshot), 0, k, ret) {
54 if (k.k->p.inode != inum)
57 if (k.k->type != KEY_TYPE_dirent)
60 d = bkey_s_c_to_dirent(k);
61 if (d.v->d_type == DT_DIR)
65 bch2_trans_iter_exit(trans, &iter);
67 return ret ?: subdirs;
70 static int __snapshot_lookup_subvol(struct btree_trans *trans, u32 snapshot,
73 struct btree_iter iter;
77 bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots,
79 k = bch2_btree_iter_peek_slot(&iter);
84 if (k.k->type != KEY_TYPE_snapshot) {
85 bch_err(trans->c, "snapshot %u not fonud", snapshot);
90 *subvol = le32_to_cpu(bkey_s_c_to_snapshot(k).v->subvol);
92 bch2_trans_iter_exit(trans, &iter);
97 static int __subvol_lookup(struct btree_trans *trans, u32 subvol,
98 u32 *snapshot, u64 *inum)
100 struct bch_subvolume s;
103 ret = bch2_subvolume_get(trans, subvol, false, 0, &s);
105 *snapshot = le32_to_cpu(s.snapshot);
106 *inum = le64_to_cpu(s.inode);
110 static int subvol_lookup(struct btree_trans *trans, u32 subvol,
111 u32 *snapshot, u64 *inum)
113 return lockrestart_do(trans, __subvol_lookup(trans, subvol, snapshot, inum));
116 static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr,
117 struct bch_inode_unpacked *inode)
119 struct btree_iter iter;
123 bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
125 BTREE_ITER_ALL_SNAPSHOTS);
126 k = bch2_btree_iter_peek(&iter);
131 if (!k.k || bkey_cmp(k.k->p, POS(0, inode_nr))) {
136 ret = bch2_inode_unpack(k, inode);
138 if (ret && ret != -EINTR)
139 bch_err(trans->c, "error %i fetching inode %llu",
141 bch2_trans_iter_exit(trans, &iter);
145 static int __lookup_inode(struct btree_trans *trans, u64 inode_nr,
146 struct bch_inode_unpacked *inode,
149 struct btree_iter iter;
153 bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
154 SPOS(0, inode_nr, *snapshot), 0);
155 k = bch2_btree_iter_peek_slot(&iter);
160 ret = bkey_is_inode(k.k)
161 ? bch2_inode_unpack(k, inode)
164 *snapshot = iter.pos.snapshot;
166 if (ret && ret != -EINTR)
167 bch_err(trans->c, "error %i fetching inode %llu:%u",
168 ret, inode_nr, *snapshot);
169 bch2_trans_iter_exit(trans, &iter);
173 static int lookup_inode(struct btree_trans *trans, u64 inode_nr,
174 struct bch_inode_unpacked *inode,
177 return lockrestart_do(trans, __lookup_inode(trans, inode_nr, inode, snapshot));
180 static int __lookup_dirent(struct btree_trans *trans,
181 struct bch_hash_info hash_info,
182 subvol_inum dir, struct qstr *name,
183 u64 *target, unsigned *type)
185 struct btree_iter iter;
186 struct bkey_s_c_dirent d;
189 ret = bch2_hash_lookup(trans, &iter, bch2_dirent_hash_desc,
190 &hash_info, dir, name, 0);
194 d = bkey_s_c_to_dirent(bch2_btree_iter_peek_slot(&iter));
195 *target = le64_to_cpu(d.v->d_inum);
197 bch2_trans_iter_exit(trans, &iter);
201 static int __write_inode(struct btree_trans *trans,
202 struct bch_inode_unpacked *inode,
205 struct btree_iter iter;
208 bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
209 SPOS(0, inode->bi_inum, snapshot),
212 ret = bch2_btree_iter_traverse(&iter) ?:
213 bch2_inode_write(trans, &iter, inode);
214 bch2_trans_iter_exit(trans, &iter);
218 static int write_inode(struct btree_trans *trans,
219 struct bch_inode_unpacked *inode,
222 int ret = __bch2_trans_do(trans, NULL, NULL,
224 BTREE_INSERT_LAZY_RW,
225 __write_inode(trans, inode, snapshot));
227 bch_err(trans->c, "error in fsck: error %i updating inode", ret);
231 static int fsck_inode_rm(struct btree_trans *trans, u64 inum, u32 snapshot)
233 struct btree_iter iter = { NULL };
234 struct bkey_i_inode_generation delete;
235 struct bch_inode_unpacked inode_u;
239 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
240 SPOS(inum, 0, snapshot),
241 SPOS(inum, U64_MAX, snapshot),
243 bch2_btree_delete_range_trans(trans, BTREE_ID_dirents,
244 SPOS(inum, 0, snapshot),
245 SPOS(inum, U64_MAX, snapshot),
247 bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs,
248 SPOS(inum, 0, snapshot),
249 SPOS(inum, U64_MAX, snapshot),
254 bch2_trans_begin(trans);
256 bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
257 SPOS(0, inum, snapshot), BTREE_ITER_INTENT);
258 k = bch2_btree_iter_peek_slot(&iter);
264 if (!bkey_is_inode(k.k)) {
265 bch2_fs_inconsistent(trans->c,
266 "inode %llu:%u not found when deleting",
272 bch2_inode_unpack(k, &inode_u);
274 /* Subvolume root? */
275 if (inode_u.bi_subvol) {
276 ret = bch2_subvolume_delete(trans, inode_u.bi_subvol);
281 bkey_inode_generation_init(&delete.k_i);
282 delete.k.p = iter.pos;
283 delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
285 ret = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
286 bch2_trans_commit(trans, NULL, NULL,
287 BTREE_INSERT_NOFAIL);
289 bch2_trans_iter_exit(trans, &iter);
296 static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
298 struct bch_fs *c = trans->c;
299 struct btree_iter iter;
300 struct bch_inode_unpacked dir_inode;
301 struct bch_hash_info dir_hash_info;
304 ret = lookup_first_inode(trans, pos.inode, &dir_inode);
308 dir_hash_info = bch2_hash_info_init(c, &dir_inode);
310 bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_INTENT);
312 ret = bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
313 &dir_hash_info, &iter, 0);
314 bch2_trans_iter_exit(trans, &iter);
318 /* Get lost+found, create if it doesn't exist: */
319 static int lookup_lostfound(struct btree_trans *trans, u32 subvol,
320 struct bch_inode_unpacked *lostfound)
322 struct bch_fs *c = trans->c;
323 struct bch_inode_unpacked root;
324 struct bch_hash_info root_hash_info;
325 struct qstr lostfound_str = QSTR("lost+found");
326 subvol_inum root_inum = { .subvol = subvol };
332 ret = __subvol_lookup(trans, subvol, &snapshot, &root_inum.inum);
336 ret = __lookup_inode(trans, root_inum.inum, &root, &snapshot);
340 root_hash_info = bch2_hash_info_init(c, &root);
342 ret = __lookup_dirent(trans, root_hash_info, root_inum,
343 &lostfound_str, &inum, &d_type);
344 if (ret == -ENOENT) {
345 bch_notice(c, "creating lost+found");
346 goto create_lostfound;
349 if (ret && ret != -EINTR)
350 bch_err(c, "error looking up lost+found: %i", ret);
354 if (d_type != DT_DIR) {
355 bch_err(c, "error looking up lost+found: not a directory");
360 * The check_dirents pass has already run, dangling dirents
361 * shouldn't exist here:
363 return __lookup_inode(trans, inum, lostfound, &snapshot);
366 bch2_inode_init_early(c, lostfound);
368 ret = bch2_create_trans(trans, root_inum, &root,
369 lostfound, &lostfound_str,
370 0, 0, S_IFDIR|0700, 0, NULL, NULL,
371 (subvol_inum) { }, 0);
372 if (ret && ret != -EINTR)
373 bch_err(c, "error creating lost+found: %i", ret);
377 static int __reattach_inode(struct btree_trans *trans,
378 struct bch_inode_unpacked *inode,
381 struct bch_hash_info dir_hash;
382 struct bch_inode_unpacked lostfound;
389 ret = __snapshot_lookup_subvol(trans, inode_snapshot, &subvol);
393 ret = lookup_lostfound(trans, subvol, &lostfound);
397 if (S_ISDIR(inode->bi_mode)) {
398 lostfound.bi_nlink++;
400 ret = __write_inode(trans, &lostfound, U32_MAX);
405 dir_hash = bch2_hash_info_init(trans->c, &lostfound);
407 snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
408 name = (struct qstr) QSTR(name_buf);
410 ret = bch2_dirent_create(trans,
413 .inum = lostfound.bi_inum,
417 &name, inode->bi_inum, &dir_offset,
418 BCH_HASH_SET_MUST_CREATE);
422 inode->bi_dir = lostfound.bi_inum;
423 inode->bi_dir_offset = dir_offset;
425 return __write_inode(trans, inode, inode_snapshot);
428 static int reattach_inode(struct btree_trans *trans,
429 struct bch_inode_unpacked *inode,
432 int ret = __bch2_trans_do(trans, NULL, NULL,
433 BTREE_INSERT_LAZY_RW|
435 __reattach_inode(trans, inode, inode_snapshot));
437 bch_err(trans->c, "error %i reattaching inode %llu",
438 ret, inode->bi_inum);
445 static int remove_backpointer(struct btree_trans *trans,
446 struct bch_inode_unpacked *inode)
448 struct btree_iter iter;
452 bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents,
453 POS(inode->bi_dir, inode->bi_dir_offset), 0);
454 k = bch2_btree_iter_peek_slot(&iter);
458 if (k.k->type != KEY_TYPE_dirent) {
463 ret = __remove_dirent(trans, k.k->p);
465 bch2_trans_iter_exit(trans, &iter);
469 static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s, struct bpos pos)
471 pos.snapshot = snapshot_t(c, pos.snapshot)->equiv;
473 if (bkey_cmp(s->pos, pos))
477 /* Might get called multiple times due to lock restarts */
478 if (s->nr && s->d[s->nr - 1] == pos.snapshot)
481 return snapshots_seen_add(c, s, pos.snapshot);
485 * key_visible_in_snapshot - returns true if @id is a descendent of @ancestor,
486 * and @ancestor hasn't been overwritten in @seen
488 * That is, returns whether key in @ancestor snapshot is visible in @id snapshot
490 static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *seen,
491 u32 id, u32 ancestor)
495 BUG_ON(id > ancestor);
497 id = snapshot_t(c, id)->equiv;
498 ancestor = snapshot_t(c, ancestor)->equiv;
500 /* @ancestor should be the snapshot most recently added to @seen */
501 BUG_ON(!seen->nr || seen->d[seen->nr - 1] != ancestor);
502 BUG_ON(seen->pos.snapshot != ancestor);
507 if (!bch2_snapshot_is_ancestor(c, id, ancestor))
510 for (i = seen->nr - 2;
511 i >= 0 && seen->d[i] >= id;
513 if (bch2_snapshot_is_ancestor(c, id, seen->d[i]) &&
514 bch2_snapshot_is_ancestor(c, seen->d[i], ancestor))
521 * ref_visible - given a key with snapshot id @src that points to a key with
522 * snapshot id @dst, test whether there is some snapshot in which @dst is
525 * This assumes we're visiting @src keys in natural key order.
527 * @s - list of snapshot IDs already seen at @src
528 * @src - snapshot ID of src key
529 * @dst - snapshot ID of dst key
531 static int ref_visible(struct bch_fs *c, struct snapshots_seen *s,
535 ? key_visible_in_snapshot(c, s, dst, src)
536 : bch2_snapshot_is_ancestor(c, src, dst);
539 #define for_each_visible_inode(_c, _s, _w, _snapshot, _i) \
540 for (_i = (_w)->d; _i < (_w)->d + (_w)->nr && (_i)->snapshot <= (_snapshot); _i++)\
541 if (key_visible_in_snapshot(_c, _s, _i->snapshot, _snapshot))
543 struct inode_walker {
544 bool first_this_inode;
549 struct inode_walker_entry {
550 struct bch_inode_unpacked inode;
556 static void inode_walker_exit(struct inode_walker *w)
562 static struct inode_walker inode_walker_init(void)
564 return (struct inode_walker) { 0, };
567 static int inode_walker_realloc(struct inode_walker *w)
569 if (w->nr == w->size) {
570 size_t new_size = max_t(size_t, 8UL, w->size * 2);
571 void *d = krealloc(w->d, new_size * sizeof(w->d[0]),
583 static int add_inode(struct bch_fs *c, struct inode_walker *w,
584 struct bkey_s_c inode)
586 struct bch_inode_unpacked u;
589 ret = inode_walker_realloc(w);
593 BUG_ON(bch2_inode_unpack(inode, &u));
595 w->d[w->nr++] = (struct inode_walker_entry) {
597 .snapshot = snapshot_t(c, inode.k->p.snapshot)->equiv,
603 static int __walk_inode(struct btree_trans *trans,
604 struct inode_walker *w, struct bpos pos)
606 struct bch_fs *c = trans->c;
607 struct btree_iter iter;
609 unsigned i, ancestor_pos;
612 pos.snapshot = snapshot_t(c, pos.snapshot)->equiv;
614 if (pos.inode == w->cur_inum) {
615 w->first_this_inode = false;
616 goto lookup_snapshot;
621 for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, pos.inode),
622 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
623 if (k.k->p.offset != pos.inode)
626 if (bkey_is_inode(k.k))
629 bch2_trans_iter_exit(trans, &iter);
634 w->cur_inum = pos.inode;
635 w->first_this_inode = true;
637 for (i = 0; i < w->nr; i++)
638 if (bch2_snapshot_is_ancestor(c, pos.snapshot, w->d[i].snapshot))
642 BUG_ON(pos.snapshot > w->d[i].snapshot);
644 if (pos.snapshot != w->d[i].snapshot) {
647 while (i && w->d[i - 1].snapshot > pos.snapshot)
650 ret = inode_walker_realloc(w);
654 array_insert_item(w->d, w->nr, i, w->d[ancestor_pos]);
655 w->d[i].snapshot = pos.snapshot;
662 static int __get_visible_inodes(struct btree_trans *trans,
663 struct inode_walker *w,
664 struct snapshots_seen *s,
667 struct bch_fs *c = trans->c;
668 struct btree_iter iter;
674 for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, inum),
675 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
676 if (k.k->p.offset != inum)
679 if (!bkey_is_inode(k.k))
682 if (ref_visible(c, s, s->pos.snapshot, k.k->p.snapshot)) {
684 if (k.k->p.snapshot >= s->pos.snapshot)
688 bch2_trans_iter_exit(trans, &iter);
693 static int check_key_has_snapshot(struct btree_trans *trans,
694 struct btree_iter *iter,
697 struct bch_fs *c = trans->c;
701 if (mustfix_fsck_err_on(!snapshot_t(c, k.k->p.snapshot)->equiv, c,
702 "key in missing snapshot: %s",
703 (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)))
704 return bch2_btree_delete_at(trans, iter,
705 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?: 1;
710 static int hash_redo_key(struct btree_trans *trans,
711 const struct bch_hash_desc desc,
712 struct bch_hash_info *hash_info,
713 struct btree_iter *k_iter, struct bkey_s_c k)
715 bch_err(trans->c, "hash_redo_key() not implemented yet");
718 struct bkey_i *delete;
721 delete = bch2_trans_kmalloc(trans, sizeof(*delete));
723 return PTR_ERR(delete);
725 tmp = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
729 bkey_reassemble(tmp, k);
731 bkey_init(&delete->k);
732 delete->k.p = k_iter->pos;
733 return bch2_btree_iter_traverse(k_iter) ?:
734 bch2_trans_update(trans, k_iter, delete, 0) ?:
735 bch2_hash_set(trans, desc, hash_info, k_iter->pos.inode, tmp, 0);
739 static int hash_check_key(struct btree_trans *trans,
740 const struct bch_hash_desc desc,
741 struct bch_hash_info *hash_info,
742 struct btree_iter *k_iter, struct bkey_s_c hash_k)
744 struct bch_fs *c = trans->c;
745 struct btree_iter iter = { NULL };
751 if (hash_k.k->type != desc.key_type)
754 hash = desc.hash_bkey(hash_info, hash_k);
756 if (likely(hash == hash_k.k->p.offset))
759 if (hash_k.k->p.offset < hash)
762 for_each_btree_key(trans, iter, desc.btree_id, POS(hash_k.k->p.inode, hash),
763 BTREE_ITER_SLOTS, k, ret) {
764 if (!bkey_cmp(k.k->p, hash_k.k->p))
767 if (fsck_err_on(k.k->type == desc.key_type &&
768 !desc.cmp_bkey(k, hash_k), c,
769 "duplicate hash table keys:\n%s",
770 (bch2_bkey_val_to_text(&PBUF(buf), c,
772 ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0) ?: 1;
776 if (bkey_deleted(k.k)) {
777 bch2_trans_iter_exit(trans, &iter);
782 bch2_trans_iter_exit(trans, &iter);
785 if (fsck_err(c, "hash table key at wrong offset: btree %u inode %llu offset %llu, "
786 "hashed to %llu\n%s",
787 desc.btree_id, hash_k.k->p.inode, hash_k.k->p.offset, hash,
788 (bch2_bkey_val_to_text(&PBUF(buf), c, hash_k), buf)) == FSCK_ERR_IGNORE)
791 ret = hash_redo_key(trans, desc, hash_info, k_iter, hash_k);
793 bch_err(c, "hash_redo_key err %i", ret);
801 static int check_inode(struct btree_trans *trans,
802 struct btree_iter *iter,
803 struct bch_inode_unpacked *prev,
806 struct bch_fs *c = trans->c;
808 struct bch_inode_unpacked u;
809 bool do_update = false;
812 k = bch2_btree_iter_peek(iter);
820 ret = check_key_has_snapshot(trans, iter, k);
822 return ret < 0 ? ret : 0;
825 * if snapshot id isn't a leaf node, skip it - deletion in
826 * particular is not atomic, so on the internal snapshot nodes
827 * we can see inodes marked for deletion after a clean shutdown
829 if (bch2_snapshot_internal_node(c, k.k->p.snapshot))
832 if (!bkey_is_inode(k.k))
835 BUG_ON(bch2_inode_unpack(k, &u));
838 !(u.bi_flags & (BCH_INODE_I_SIZE_DIRTY|
839 BCH_INODE_I_SECTORS_DIRTY|
840 BCH_INODE_UNLINKED)))
843 if (prev->bi_inum != u.bi_inum)
846 if (fsck_err_on(prev->bi_hash_seed != u.bi_hash_seed ||
847 inode_d_type(prev) != inode_d_type(&u), c,
848 "inodes in different snapshots don't match")) {
849 bch_err(c, "repair not implemented yet");
853 if (u.bi_flags & BCH_INODE_UNLINKED &&
855 fsck_err(c, "filesystem marked clean, but inode %llu unlinked",
857 bch2_trans_unlock(trans);
860 ret = fsck_inode_rm(trans, u.bi_inum, iter->pos.snapshot);
862 bch_err(c, "error in fsck: error %i while deleting inode", ret);
866 if (u.bi_flags & BCH_INODE_I_SIZE_DIRTY &&
868 fsck_err(c, "filesystem marked clean, but inode %llu has i_size dirty",
870 bch_verbose(c, "truncating inode %llu", u.bi_inum);
872 bch2_trans_unlock(trans);
876 * XXX: need to truncate partial blocks too here - or ideally
877 * just switch units to bytes and that issue goes away
879 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
880 SPOS(u.bi_inum, round_up(u.bi_size, block_bytes(c)) >> 9,
882 POS(u.bi_inum, U64_MAX),
885 bch_err(c, "error in fsck: error %i truncating inode", ret);
890 * We truncated without our normal sector accounting hook, just
891 * make sure we recalculate it:
893 u.bi_flags |= BCH_INODE_I_SECTORS_DIRTY;
895 u.bi_flags &= ~BCH_INODE_I_SIZE_DIRTY;
899 if (u.bi_flags & BCH_INODE_I_SECTORS_DIRTY &&
901 fsck_err(c, "filesystem marked clean, but inode %llu has i_sectors dirty",
905 bch_verbose(c, "recounting sectors for inode %llu",
908 sectors = bch2_count_inode_sectors(trans, u.bi_inum, iter->pos.snapshot);
910 bch_err(c, "error in fsck: error %i recounting inode sectors",
915 u.bi_sectors = sectors;
916 u.bi_flags &= ~BCH_INODE_I_SECTORS_DIRTY;
920 if (u.bi_flags & BCH_INODE_BACKPTR_UNTRUSTED) {
923 u.bi_flags &= ~BCH_INODE_BACKPTR_UNTRUSTED;
928 ret = write_inode(trans, &u, iter->pos.snapshot);
930 bch_err(c, "error in fsck: error %i "
931 "updating inode", ret);
938 static int check_inodes(struct bch_fs *c, bool full)
940 struct btree_trans trans;
941 struct btree_iter iter;
942 struct bch_inode_unpacked prev = { 0 };
945 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
947 bch2_trans_iter_init(&trans, &iter, BTREE_ID_inodes, POS_MIN,
950 BTREE_ITER_ALL_SNAPSHOTS);
953 ret = __bch2_trans_do(&trans, NULL, NULL,
954 BTREE_INSERT_LAZY_RW|
956 check_inode(&trans, &iter, &prev, full));
959 } while (bch2_btree_iter_advance(&iter));
960 bch2_trans_iter_exit(&trans, &iter);
962 bch2_trans_exit(&trans);
966 static int check_subvol(struct btree_trans *trans,
967 struct btree_iter *iter)
970 struct bkey_s_c_subvolume subvol;
973 k = bch2_btree_iter_peek(iter);
981 if (k.k->type != KEY_TYPE_subvolume)
984 subvol = bkey_s_c_to_subvolume(k);
986 if (BCH_SUBVOLUME_UNLINKED(subvol.v)) {
987 ret = bch2_subvolume_delete(trans, iter->pos.offset);
988 if (ret && ret != -EINTR)
989 bch_err(trans->c, "error deleting subvolume %llu: %i",
990 iter->pos.offset, ret);
999 static int check_subvols(struct bch_fs *c)
1001 struct btree_trans trans;
1002 struct btree_iter iter;
1005 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1007 bch2_trans_iter_init(&trans, &iter, BTREE_ID_subvolumes,
1010 BTREE_ITER_PREFETCH);
1013 ret = __bch2_trans_do(&trans, NULL, NULL,
1014 BTREE_INSERT_LAZY_RW|
1015 BTREE_INSERT_NOFAIL,
1016 check_subvol(&trans, &iter));
1019 } while (bch2_btree_iter_advance(&iter));
1020 bch2_trans_iter_exit(&trans, &iter);
1022 bch2_trans_exit(&trans);
1027 * Checking for overlapping extents needs to be reimplemented
1030 static int fix_overlapping_extent(struct btree_trans *trans,
1031 struct bkey_s_c k, struct bpos cut_at)
1033 struct btree_iter iter;
1037 u = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
1038 ret = PTR_ERR_OR_ZERO(u);
1042 bkey_reassemble(u, k);
1043 bch2_cut_front(cut_at, u);
1047 * We don't want to go through the extent_handle_overwrites path:
1049 * XXX: this is going to screw up disk accounting, extent triggers
1050 * assume things about extent overwrites - we should be running the
1051 * triggers manually here
1053 bch2_trans_iter_init(trans, &iter, BTREE_ID_extents, u->k.p,
1054 BTREE_ITER_INTENT|BTREE_ITER_NOT_EXTENTS);
1056 BUG_ON(iter.flags & BTREE_ITER_IS_EXTENTS);
1057 ret = bch2_btree_iter_traverse(&iter) ?:
1058 bch2_trans_update(trans, &iter, u, BTREE_TRIGGER_NORUN) ?:
1059 bch2_trans_commit(trans, NULL, NULL,
1060 BTREE_INSERT_NOFAIL|
1061 BTREE_INSERT_LAZY_RW);
1062 bch2_trans_iter_exit(trans, &iter);
1067 static struct bkey_s_c_dirent dirent_get_by_pos(struct btree_trans *trans,
1068 struct btree_iter *iter,
1074 bch2_trans_iter_init(trans, iter, BTREE_ID_dirents, pos, 0);
1075 k = bch2_btree_iter_peek_slot(iter);
1077 if (!ret && k.k->type != KEY_TYPE_dirent)
1080 bch2_trans_iter_exit(trans, iter);
1081 return (struct bkey_s_c_dirent) { .k = ERR_PTR(ret) };
1084 return bkey_s_c_to_dirent(k);
1087 static bool inode_points_to_dirent(struct bch_inode_unpacked *inode,
1088 struct bkey_s_c_dirent d)
1090 return inode->bi_dir == d.k->p.inode &&
1091 inode->bi_dir_offset == d.k->p.offset;
1094 static bool dirent_points_to_inode(struct bkey_s_c_dirent d,
1095 struct bch_inode_unpacked *inode)
1097 return d.v->d_type == DT_SUBVOL
1098 ? le32_to_cpu(d.v->d_child_subvol) == inode->bi_subvol
1099 : le64_to_cpu(d.v->d_inum) == inode->bi_inum;
1102 static int inode_backpointer_exists(struct btree_trans *trans,
1103 struct bch_inode_unpacked *inode,
1106 struct btree_iter iter;
1107 struct bkey_s_c_dirent d;
1110 d = dirent_get_by_pos(trans, &iter,
1111 SPOS(inode->bi_dir, inode->bi_dir_offset, snapshot));
1112 ret = bkey_err(d.s_c);
1116 ret = dirent_points_to_inode(d, inode);
1117 bch2_trans_iter_exit(trans, &iter);
1121 static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
1123 struct bch_fs *c = trans->c;
1124 struct inode_walker_entry *i;
1125 int ret = 0, ret2 = 0;
1128 for (i = w->d; i < w->d + w->nr; i++) {
1129 if (i->inode.bi_sectors == i->count)
1132 count2 = lockrestart_do(trans,
1133 bch2_count_inode_sectors(trans, w->cur_inum, i->snapshot));
1135 if (i->count != count2) {
1136 bch_err(c, "fsck counted i_sectors wrong: got %llu should be %llu",
1139 if (i->inode.bi_sectors == i->count)
1143 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SECTORS_DIRTY), c,
1144 "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
1145 w->cur_inum, i->snapshot,
1146 i->inode.bi_sectors, i->count) == FSCK_ERR_IGNORE)
1149 i->inode.bi_sectors = i->count;
1150 ret = write_inode(trans, &i->inode, i->snapshot);
1159 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
1160 struct inode_walker *inode,
1161 struct snapshots_seen *s)
1163 struct bch_fs *c = trans->c;
1165 struct inode_walker_entry *i;
1169 k = bch2_btree_iter_peek(iter);
1177 ret = check_key_has_snapshot(trans, iter, k);
1179 return ret < 0 ? ret : 0;
1181 ret = snapshots_seen_update(c, s, k.k->p);
1185 if (k.k->type == KEY_TYPE_whiteout)
1188 if (inode->cur_inum != k.k->p.inode) {
1189 ret = check_i_sectors(trans, inode);
1194 if (bkey_cmp(prev.k->k.p, bkey_start_pos(k.k)) > 0) {
1198 bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(prev.k));
1199 bch2_bkey_val_to_text(&PBUF(buf2), c, k);
1201 if (fsck_err(c, "overlapping extents:\n%s\n%s", buf1, buf2))
1202 return fix_overlapping_extent(trans, k, prev.k->k.p) ?: -EINTR;
1205 ret = __walk_inode(trans, inode, k.k->p);
1209 if (fsck_err_on(ret == INT_MAX, c,
1210 "extent in missing inode:\n %s",
1211 (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)))
1212 return bch2_btree_delete_at(trans, iter,
1213 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1221 if (fsck_err_on(!S_ISREG(i->inode.bi_mode) &&
1222 !S_ISLNK(i->inode.bi_mode), c,
1223 "extent in non regular inode mode %o:\n %s",
1225 (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)))
1226 return bch2_btree_delete_at(trans, iter,
1227 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1229 if (!bch2_snapshot_internal_node(c, k.k->p.snapshot)) {
1230 for_each_visible_inode(c, s, inode, k.k->p.snapshot, i) {
1231 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SIZE_DIRTY) &&
1232 k.k->type != KEY_TYPE_reservation &&
1233 k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9, c,
1234 "extent type %u offset %llu past end of inode %llu, i_size %llu",
1235 k.k->type, k.k->p.offset, k.k->p.inode, i->inode.bi_size)) {
1237 return bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
1238 SPOS(k.k->p.inode, round_up(i->inode.bi_size, block_bytes(c)) >> 9,
1240 POS(k.k->p.inode, U64_MAX),
1246 if (bkey_extent_is_allocation(k.k))
1247 for_each_visible_inode(c, s, inode, k.k->p.snapshot, i)
1248 i->count += k.k->size;
1250 bch2_bkey_buf_reassemble(&prev, c, k);
1258 * Walk extents: verify that extents have a corresponding S_ISREG inode, and
1259 * that i_size an i_sectors are consistent
1262 static int check_extents(struct bch_fs *c)
1264 struct inode_walker w = inode_walker_init();
1265 struct snapshots_seen s;
1266 struct btree_trans trans;
1267 struct btree_iter iter;
1271 struct bkey_buf prev;
1272 bch2_bkey_buf_init(&prev);
1273 prev.k->k = KEY(0, 0, 0);
1275 snapshots_seen_init(&s);
1276 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1278 bch_verbose(c, "checking extents");
1280 bch2_trans_iter_init(&trans, &iter, BTREE_ID_extents,
1281 POS(BCACHEFS_ROOT_INO, 0),
1283 BTREE_ITER_PREFETCH|
1284 BTREE_ITER_ALL_SNAPSHOTS);
1287 ret = __bch2_trans_do(&trans, NULL, NULL,
1288 BTREE_INSERT_LAZY_RW|
1289 BTREE_INSERT_NOFAIL,
1290 check_extent(&trans, &iter, &w, &s));
1293 } while (bch2_btree_iter_advance(&iter));
1294 bch2_trans_iter_exit(&trans, &iter);
1296 bch2_bkey_buf_exit(&prev, c);
1298 inode_walker_exit(&w);
1299 bch2_trans_exit(&trans);
1300 snapshots_seen_exit(&s);
1305 static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
1307 struct bch_fs *c = trans->c;
1308 struct inode_walker_entry *i;
1309 int ret = 0, ret2 = 0;
1312 for (i = w->d; i < w->d + w->nr; i++) {
1313 if (i->inode.bi_nlink == i->count)
1316 count2 = lockrestart_do(trans,
1317 bch2_count_subdirs(trans, w->cur_inum, i->snapshot));
1319 if (i->count != count2) {
1320 bch_err(c, "fsck counted subdirectories wrong: got %llu should be %llu",
1323 if (i->inode.bi_nlink == i->count)
1327 if (fsck_err_on(i->inode.bi_nlink != i->count, c,
1328 "directory %llu:%u with wrong i_nlink: got %u, should be %llu",
1329 w->cur_inum, i->snapshot, i->inode.bi_nlink, i->count)) {
1330 i->inode.bi_nlink = i->count;
1331 ret = write_inode(trans, &i->inode, i->snapshot);
1341 static int check_dirent_target(struct btree_trans *trans,
1342 struct btree_iter *iter,
1343 struct bkey_s_c_dirent d,
1344 struct bch_inode_unpacked *target,
1345 u32 target_snapshot)
1347 struct bch_fs *c = trans->c;
1348 struct bkey_i_dirent *n;
1349 bool backpointer_exists = true;
1353 if (!target->bi_dir &&
1354 !target->bi_dir_offset) {
1355 target->bi_dir = d.k->p.inode;
1356 target->bi_dir_offset = d.k->p.offset;
1358 ret = __write_inode(trans, target, target_snapshot);
1363 if (!inode_points_to_dirent(target, d)) {
1364 ret = inode_backpointer_exists(trans, target, d.k->p.snapshot);
1368 backpointer_exists = ret;
1371 if (fsck_err_on(S_ISDIR(target->bi_mode) &&
1372 backpointer_exists, c,
1373 "directory %llu with multiple links",
1375 ret = __remove_dirent(trans, d.k->p);
1381 if (fsck_err_on(backpointer_exists &&
1382 !target->bi_nlink, c,
1383 "inode %llu has multiple links but i_nlink 0",
1386 target->bi_flags &= ~BCH_INODE_UNLINKED;
1388 ret = __write_inode(trans, target, target_snapshot);
1393 if (fsck_err_on(!backpointer_exists, c,
1394 "inode %llu:%u has wrong backpointer:\n"
1396 "should be %llu:%llu",
1397 target->bi_inum, target_snapshot,
1399 target->bi_dir_offset,
1402 target->bi_dir = d.k->p.inode;
1403 target->bi_dir_offset = d.k->p.offset;
1405 ret = __write_inode(trans, target, target_snapshot);
1411 if (fsck_err_on(d.v->d_type != inode_d_type(target), c,
1412 "incorrect d_type: got %s, should be %s:\n%s",
1413 bch2_d_type_str(d.v->d_type),
1414 bch2_d_type_str(inode_d_type(target)),
1415 (bch2_bkey_val_to_text(&PBUF(buf), c, d.s_c), buf))) {
1416 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1417 ret = PTR_ERR_OR_ZERO(n);
1421 bkey_reassemble(&n->k_i, d.s_c);
1422 n->v.d_type = inode_d_type(target);
1424 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1428 d = dirent_i_to_s_c(n);
1431 if (d.v->d_type == DT_SUBVOL &&
1432 target->bi_parent_subvol != le32_to_cpu(d.v->d_parent_subvol) &&
1433 (c->sb.version < bcachefs_metadata_version_subvol_dirent ||
1434 fsck_err(c, "dirent has wrong d_parent_subvol field: got %u, should be %u",
1435 le32_to_cpu(d.v->d_parent_subvol),
1436 target->bi_parent_subvol))) {
1437 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1438 ret = PTR_ERR_OR_ZERO(n);
1442 bkey_reassemble(&n->k_i, d.s_c);
1443 n->v.d_parent_subvol = cpu_to_le32(target->bi_parent_subvol);
1445 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1449 d = dirent_i_to_s_c(n);
1456 static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
1457 struct bch_hash_info *hash_info,
1458 struct inode_walker *dir,
1459 struct inode_walker *target,
1460 struct snapshots_seen *s)
1462 struct bch_fs *c = trans->c;
1464 struct bkey_s_c_dirent d;
1465 struct inode_walker_entry *i;
1469 k = bch2_btree_iter_peek(iter);
1477 ret = check_key_has_snapshot(trans, iter, k);
1479 return ret < 0 ? ret : 0;
1481 ret = snapshots_seen_update(c, s, k.k->p);
1485 if (k.k->type == KEY_TYPE_whiteout)
1488 if (dir->cur_inum != k.k->p.inode) {
1489 ret = check_subdir_count(trans, dir);
1494 ret = __walk_inode(trans, dir, k.k->p);
1498 if (fsck_err_on(ret == INT_MAX, c,
1499 "dirent in nonexisting directory:\n%s",
1500 (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)))
1501 return bch2_btree_delete_at(trans, iter,
1502 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1510 if (fsck_err_on(!S_ISDIR(i->inode.bi_mode), c,
1511 "dirent in non directory inode type %s:\n%s",
1512 bch2_d_type_str(inode_d_type(&i->inode)),
1513 (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)))
1514 return bch2_btree_delete_at(trans, iter, 0);
1516 if (dir->first_this_inode)
1517 *hash_info = bch2_hash_info_init(c, &dir->d[0].inode);
1519 ret = hash_check_key(trans, bch2_dirent_hash_desc,
1520 hash_info, iter, k);
1523 if (ret) /* dirent has been deleted */
1526 if (k.k->type != KEY_TYPE_dirent)
1529 d = bkey_s_c_to_dirent(k);
1531 if (d.v->d_type == DT_SUBVOL) {
1532 struct bch_inode_unpacked subvol_root;
1533 u32 target_subvol = le32_to_cpu(d.v->d_child_subvol);
1534 u32 target_snapshot;
1537 ret = __subvol_lookup(trans, target_subvol,
1538 &target_snapshot, &target_inum);
1539 if (ret && ret != -ENOENT)
1542 if (fsck_err_on(ret, c,
1543 "dirent points to missing subvolume %llu",
1544 le64_to_cpu(d.v->d_child_subvol)))
1545 return __remove_dirent(trans, d.k->p);
1547 ret = __lookup_inode(trans, target_inum,
1548 &subvol_root, &target_snapshot);
1549 if (ret && ret != -ENOENT)
1552 if (fsck_err_on(ret, c,
1553 "subvolume %u points to missing subvolume root %llu",
1556 bch_err(c, "repair not implemented yet");
1560 if (fsck_err_on(subvol_root.bi_subvol != target_subvol, c,
1561 "subvol root %llu has wrong bi_subvol field: got %u, should be %u",
1563 subvol_root.bi_subvol, target_subvol)) {
1564 subvol_root.bi_subvol = target_subvol;
1565 ret = __write_inode(trans, &subvol_root, target_snapshot);
1570 ret = check_dirent_target(trans, iter, d, &subvol_root,
1575 ret = __get_visible_inodes(trans, target, s, le64_to_cpu(d.v->d_inum));
1579 if (fsck_err_on(!target->nr, c,
1580 "dirent points to missing inode:\n%s",
1581 (bch2_bkey_val_to_text(&PBUF(buf), c,
1583 ret = __remove_dirent(trans, d.k->p);
1588 for (i = target->d; i < target->d + target->nr; i++) {
1589 ret = check_dirent_target(trans, iter, d,
1590 &i->inode, i->snapshot);
1596 if (d.v->d_type == DT_DIR)
1597 for_each_visible_inode(c, s, dir, d.k->p.snapshot, i)
1605 * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
1609 static int check_dirents(struct bch_fs *c)
1611 struct inode_walker dir = inode_walker_init();
1612 struct inode_walker target = inode_walker_init();
1613 struct snapshots_seen s;
1614 struct bch_hash_info hash_info;
1615 struct btree_trans trans;
1616 struct btree_iter iter;
1619 bch_verbose(c, "checking dirents");
1621 snapshots_seen_init(&s);
1622 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1624 bch2_trans_iter_init(&trans, &iter, BTREE_ID_dirents,
1625 POS(BCACHEFS_ROOT_INO, 0),
1627 BTREE_ITER_PREFETCH|
1628 BTREE_ITER_ALL_SNAPSHOTS);
1631 ret = __bch2_trans_do(&trans, NULL, NULL,
1632 BTREE_INSERT_LAZY_RW|
1633 BTREE_INSERT_NOFAIL,
1634 check_dirent(&trans, &iter, &hash_info,
1635 &dir, &target, &s));
1638 } while (bch2_btree_iter_advance(&iter));
1639 bch2_trans_iter_exit(&trans, &iter);
1641 bch2_trans_exit(&trans);
1642 snapshots_seen_exit(&s);
1643 inode_walker_exit(&dir);
1644 inode_walker_exit(&target);
1648 static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
1649 struct bch_hash_info *hash_info,
1650 struct inode_walker *inode)
1652 struct bch_fs *c = trans->c;
1656 k = bch2_btree_iter_peek(iter);
1664 ret = check_key_has_snapshot(trans, iter, k);
1668 ret = __walk_inode(trans, inode, k.k->p);
1672 if (fsck_err_on(ret == INT_MAX, c,
1673 "xattr for missing inode %llu",
1675 return bch2_btree_delete_at(trans, iter, 0);
1682 if (inode->first_this_inode)
1683 *hash_info = bch2_hash_info_init(c, &inode->d[0].inode);
1685 ret = hash_check_key(trans, bch2_xattr_hash_desc, hash_info, iter, k);
1691 * Walk xattrs: verify that they all have a corresponding inode
1694 static int check_xattrs(struct bch_fs *c)
1696 struct inode_walker inode = inode_walker_init();
1697 struct bch_hash_info hash_info;
1698 struct btree_trans trans;
1699 struct btree_iter iter;
1702 bch_verbose(c, "checking xattrs");
1704 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1706 bch2_trans_iter_init(&trans, &iter, BTREE_ID_xattrs,
1707 POS(BCACHEFS_ROOT_INO, 0),
1709 BTREE_ITER_PREFETCH|
1710 BTREE_ITER_ALL_SNAPSHOTS);
1713 ret = __bch2_trans_do(&trans, NULL, NULL,
1714 BTREE_INSERT_LAZY_RW|
1715 BTREE_INSERT_NOFAIL,
1716 check_xattr(&trans, &iter, &hash_info,
1720 } while (bch2_btree_iter_advance(&iter));
1721 bch2_trans_iter_exit(&trans, &iter);
1723 bch2_trans_exit(&trans);
1727 static int check_root_trans(struct btree_trans *trans)
1729 struct bch_fs *c = trans->c;
1730 struct bch_inode_unpacked root_inode;
1735 ret = __subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
1736 if (ret && ret != -ENOENT)
1739 if (mustfix_fsck_err_on(ret, c, "root subvol missing")) {
1740 struct bkey_i_subvolume root_subvol;
1743 inum = BCACHEFS_ROOT_INO;
1745 bkey_subvolume_init(&root_subvol.k_i);
1746 root_subvol.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1747 root_subvol.v.flags = 0;
1748 root_subvol.v.snapshot = cpu_to_le32(snapshot);
1749 root_subvol.v.inode = cpu_to_le64(inum);
1750 ret = __bch2_trans_do(trans, NULL, NULL,
1751 BTREE_INSERT_NOFAIL|
1752 BTREE_INSERT_LAZY_RW,
1753 __bch2_btree_insert(trans, BTREE_ID_subvolumes, &root_subvol.k_i));
1755 bch_err(c, "error writing root subvol: %i", ret);
1761 ret = __lookup_inode(trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
1762 if (ret && ret != -ENOENT)
1765 if (mustfix_fsck_err_on(ret, c, "root directory missing") ||
1766 mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode), c,
1767 "root inode not a directory")) {
1768 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755,
1770 root_inode.bi_inum = inum;
1772 ret = __write_inode(trans, &root_inode, snapshot);
1774 bch_err(c, "error writing root inode: %i", ret);
1781 /* Get root directory, create if it doesn't exist: */
1783 static int check_root(struct bch_fs *c)
1785 bch_verbose(c, "checking root directory");
1787 return bch2_trans_do(c, NULL, NULL,
1788 BTREE_INSERT_NOFAIL|
1789 BTREE_INSERT_LAZY_RW,
1790 check_root_trans(&trans));
1797 struct pathbuf_entry {
1803 static bool path_is_dup(struct pathbuf *p, u64 inum, u32 snapshot)
1805 struct pathbuf_entry *i;
1807 for (i = p->entries; i < p->entries + p->nr; i++)
1808 if (i->inum == inum &&
1809 i->snapshot == snapshot)
1815 static int path_down(struct pathbuf *p, u64 inum, u32 snapshot)
1817 if (p->nr == p->size) {
1818 size_t new_size = max_t(size_t, 256UL, p->size * 2);
1819 void *n = krealloc(p->entries,
1820 new_size * sizeof(p->entries[0]),
1830 p->entries[p->nr++] = (struct pathbuf_entry) {
1832 .snapshot = snapshot,
1838 * Check that a given inode is reachable from the root:
1840 * XXX: we should also be verifying that inodes are in the right subvolumes
1842 static int check_path(struct btree_trans *trans,
1844 struct bch_inode_unpacked *inode,
1847 struct bch_fs *c = trans->c;
1850 snapshot = snapshot_t(c, snapshot)->equiv;
1853 while (!(inode->bi_inum == BCACHEFS_ROOT_INO &&
1854 inode->bi_subvol == BCACHEFS_ROOT_SUBVOL)) {
1855 struct btree_iter dirent_iter;
1856 struct bkey_s_c_dirent d;
1857 u32 parent_snapshot = snapshot;
1859 if (inode->bi_subvol) {
1862 ret = subvol_lookup(trans, inode->bi_parent_subvol,
1863 &parent_snapshot, &inum);
1868 ret = lockrestart_do(trans,
1869 PTR_ERR_OR_ZERO((d = dirent_get_by_pos(trans, &dirent_iter,
1870 SPOS(inode->bi_dir, inode->bi_dir_offset,
1871 parent_snapshot))).k));
1872 if (ret && ret != -ENOENT)
1875 if (!ret && !dirent_points_to_inode(d, inode)) {
1876 bch2_trans_iter_exit(trans, &dirent_iter);
1880 if (ret == -ENOENT) {
1881 if (fsck_err(c, "unreachable inode %llu:%u, type %s nlink %u backptr %llu:%llu",
1882 inode->bi_inum, snapshot,
1883 bch2_d_type_str(inode_d_type(inode)),
1886 inode->bi_dir_offset))
1887 ret = reattach_inode(trans, inode, snapshot);
1891 bch2_trans_iter_exit(trans, &dirent_iter);
1893 if (!S_ISDIR(inode->bi_mode))
1896 ret = path_down(p, inode->bi_inum, snapshot);
1898 bch_err(c, "memory allocation failure");
1902 snapshot = parent_snapshot;
1904 ret = lookup_inode(trans, inode->bi_dir, inode, &snapshot);
1906 /* Should have been caught in dirents pass */
1907 bch_err(c, "error looking up parent directory: %i", ret);
1911 if (path_is_dup(p, inode->bi_inum, snapshot)) {
1912 struct pathbuf_entry *i;
1914 /* XXX print path */
1915 bch_err(c, "directory structure loop");
1917 for (i = p->entries; i < p->entries + p->nr; i++)
1918 pr_err("%llu:%u", i->inum, i->snapshot);
1919 pr_err("%llu:%u", inode->bi_inum, snapshot);
1921 if (!fsck_err(c, "directory structure loop"))
1924 ret = __bch2_trans_do(trans, NULL, NULL,
1925 BTREE_INSERT_NOFAIL|
1926 BTREE_INSERT_LAZY_RW,
1927 remove_backpointer(trans, inode));
1929 bch_err(c, "error removing dirent: %i", ret);
1933 ret = reattach_inode(trans, inode, snapshot);
1938 bch_err(c, "%s: err %i", __func__, ret);
1943 * Check for unreachable inodes, as well as loops in the directory structure:
1944 * After check_dirents(), if an inode backpointer doesn't exist that means it's
1948 static int check_directory_structure(struct bch_fs *c)
1950 struct btree_trans trans;
1951 struct btree_iter iter;
1953 struct bch_inode_unpacked u;
1954 struct pathbuf path = { 0, 0, NULL };
1957 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1959 for_each_btree_key(&trans, iter, BTREE_ID_inodes, POS_MIN,
1961 BTREE_ITER_PREFETCH|
1962 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
1963 if (!bkey_is_inode(k.k))
1966 ret = bch2_inode_unpack(k, &u);
1968 /* Should have been caught earlier in fsck: */
1969 bch_err(c, "error unpacking inode %llu: %i", k.k->p.offset, ret);
1973 if (u.bi_flags & BCH_INODE_UNLINKED)
1976 ret = check_path(&trans, &path, &u, iter.pos.snapshot);
1980 bch2_trans_iter_exit(&trans, &iter);
1982 BUG_ON(ret == -EINTR);
1984 kfree(path.entries);
1986 bch2_trans_exit(&trans);
1990 struct nlink_table {
2001 static int add_nlink(struct nlink_table *t, u64 inum, u32 snapshot)
2003 if (t->nr == t->size) {
2004 size_t new_size = max_t(size_t, 128UL, t->size * 2);
2005 void *d = kvmalloc(new_size * sizeof(t->d[0]), GFP_KERNEL);
2011 memcpy(d, t->d, t->size * sizeof(t->d[0]));
2019 t->d[t->nr++] = (struct nlink) {
2021 .snapshot = snapshot,
2027 static int nlink_cmp(const void *_l, const void *_r)
2029 const struct nlink *l = _l;
2030 const struct nlink *r = _r;
2032 return cmp_int(l->inum, r->inum) ?: cmp_int(l->snapshot, r->snapshot);
2035 static void inc_link(struct bch_fs *c, struct snapshots_seen *s,
2036 struct nlink_table *links,
2037 u64 range_start, u64 range_end, u64 inum, u32 snapshot)
2039 struct nlink *link, key = {
2040 .inum = inum, .snapshot = U32_MAX,
2043 if (inum < range_start || inum >= range_end)
2046 link = __inline_bsearch(&key, links->d, links->nr,
2047 sizeof(links->d[0]), nlink_cmp);
2051 while (link > links->d && link[0].inum == link[-1].inum)
2054 for (; link < links->d + links->nr && link->inum == inum; link++)
2055 if (ref_visible(c, s, snapshot, link->snapshot)) {
2057 if (link->snapshot >= snapshot)
2063 static int check_nlinks_find_hardlinks(struct bch_fs *c,
2064 struct nlink_table *t,
2065 u64 start, u64 *end)
2067 struct btree_trans trans;
2068 struct btree_iter iter;
2070 struct bch_inode_unpacked u;
2073 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2075 for_each_btree_key(&trans, iter, BTREE_ID_inodes,
2078 BTREE_ITER_PREFETCH|
2079 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2080 if (!bkey_is_inode(k.k))
2083 /* Should never fail, checked by bch2_inode_invalid: */
2084 BUG_ON(bch2_inode_unpack(k, &u));
2087 * Backpointer and directory structure checks are sufficient for
2088 * directories, since they can't have hardlinks:
2090 if (S_ISDIR(le16_to_cpu(u.bi_mode)))
2096 ret = add_nlink(t, k.k->p.offset, k.k->p.snapshot);
2098 *end = k.k->p.offset;
2104 bch2_trans_iter_exit(&trans, &iter);
2105 bch2_trans_exit(&trans);
2108 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
2114 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
2115 u64 range_start, u64 range_end)
2117 struct btree_trans trans;
2118 struct snapshots_seen s;
2119 struct btree_iter iter;
2121 struct bkey_s_c_dirent d;
2124 snapshots_seen_init(&s);
2126 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2128 for_each_btree_key(&trans, iter, BTREE_ID_dirents, POS_MIN,
2130 BTREE_ITER_PREFETCH|
2131 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2132 ret = snapshots_seen_update(c, &s, k.k->p);
2136 switch (k.k->type) {
2137 case KEY_TYPE_dirent:
2138 d = bkey_s_c_to_dirent(k);
2140 if (d.v->d_type != DT_DIR &&
2141 d.v->d_type != DT_SUBVOL)
2142 inc_link(c, &s, links, range_start, range_end,
2143 le64_to_cpu(d.v->d_inum),
2148 bch2_trans_iter_exit(&trans, &iter);
2151 bch_err(c, "error in fsck: btree error %i while walking dirents", ret);
2153 bch2_trans_exit(&trans);
2154 snapshots_seen_exit(&s);
2159 static int check_nlinks_update_hardlinks(struct bch_fs *c,
2160 struct nlink_table *links,
2161 u64 range_start, u64 range_end)
2163 struct btree_trans trans;
2164 struct btree_iter iter;
2166 struct bch_inode_unpacked u;
2167 struct nlink *link = links->d;
2170 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2172 for_each_btree_key(&trans, iter, BTREE_ID_inodes,
2173 POS(0, range_start),
2175 BTREE_ITER_PREFETCH|
2176 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2177 if (k.k->p.offset >= range_end)
2180 if (!bkey_is_inode(k.k))
2183 BUG_ON(bch2_inode_unpack(k, &u));
2185 if (S_ISDIR(le16_to_cpu(u.bi_mode)))
2191 while ((cmp_int(link->inum, k.k->p.offset) ?:
2192 cmp_int(link->snapshot, k.k->p.snapshot)) < 0) {
2194 BUG_ON(link >= links->d + links->nr);
2197 if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count, c,
2198 "inode %llu has wrong i_nlink (type %u i_nlink %u, should be %u)",
2199 u.bi_inum, mode_to_type(u.bi_mode),
2200 bch2_inode_nlink_get(&u), link->count)) {
2201 bch2_inode_nlink_set(&u, link->count);
2203 ret = write_inode(&trans, &u, k.k->p.snapshot);
2205 bch_err(c, "error in fsck: error %i updating inode", ret);
2209 bch2_trans_iter_exit(&trans, &iter);
2210 bch2_trans_exit(&trans);
2213 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
2219 static int check_nlinks(struct bch_fs *c)
2221 struct nlink_table links = { 0 };
2222 u64 this_iter_range_start, next_iter_range_start = 0;
2225 bch_verbose(c, "checking inode nlinks");
2228 this_iter_range_start = next_iter_range_start;
2229 next_iter_range_start = U64_MAX;
2231 ret = check_nlinks_find_hardlinks(c, &links,
2232 this_iter_range_start,
2233 &next_iter_range_start);
2235 ret = check_nlinks_walk_dirents(c, &links,
2236 this_iter_range_start,
2237 next_iter_range_start);
2241 ret = check_nlinks_update_hardlinks(c, &links,
2242 this_iter_range_start,
2243 next_iter_range_start);
2248 } while (next_iter_range_start != U64_MAX);
2255 static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter)
2258 struct bkey_s_c_reflink_p p;
2259 struct bkey_i_reflink_p *u;
2262 k = bch2_btree_iter_peek(iter);
2270 if (k.k->type != KEY_TYPE_reflink_p)
2273 p = bkey_s_c_to_reflink_p(k);
2275 if (!p.v->front_pad && !p.v->back_pad)
2278 u = bch2_trans_kmalloc(trans, sizeof(*u));
2279 ret = PTR_ERR_OR_ZERO(u);
2283 bkey_reassemble(&u->k_i, k);
2287 return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_NORUN);
2291 static int fix_reflink_p(struct bch_fs *c)
2293 struct btree_trans trans;
2294 struct btree_iter iter;
2298 if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
2301 bch_verbose(c, "fixing reflink_p keys");
2303 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2305 for_each_btree_key(&trans, iter, BTREE_ID_extents, POS_MIN,
2307 BTREE_ITER_PREFETCH|
2308 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2309 if (k.k->type == KEY_TYPE_reflink_p) {
2310 ret = __bch2_trans_do(&trans, NULL, NULL,
2311 BTREE_INSERT_NOFAIL|
2312 BTREE_INSERT_LAZY_RW,
2313 fix_reflink_p_key(&trans, &iter));
2318 bch2_trans_iter_exit(&trans, &iter);
2320 bch2_trans_exit(&trans);
2325 * Checks for inconsistencies that shouldn't happen, unless we have a bug.
2326 * Doesn't fix them yet, mainly because they haven't yet been observed:
2328 int bch2_fsck_full(struct bch_fs *c)
2330 return bch2_fs_snapshots_check(c) ?:
2331 check_inodes(c, true) ?:
2337 check_directory_structure(c) ?:
2342 int bch2_fsck_walk_inodes_only(struct bch_fs *c)
2344 return check_inodes(c, false);