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 bch_fs *c, 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]),
574 bch_err(c, "fsck: error allocating memory for inode_walker, size %zu",
586 static int add_inode(struct bch_fs *c, struct inode_walker *w,
587 struct bkey_s_c inode)
589 struct bch_inode_unpacked u;
592 ret = inode_walker_realloc(c, w);
596 BUG_ON(bch2_inode_unpack(inode, &u));
598 w->d[w->nr++] = (struct inode_walker_entry) {
600 .snapshot = snapshot_t(c, inode.k->p.snapshot)->equiv,
606 static int __walk_inode(struct btree_trans *trans,
607 struct inode_walker *w, struct bpos pos)
609 struct bch_fs *c = trans->c;
610 struct btree_iter iter;
612 unsigned i, ancestor_pos;
615 pos.snapshot = snapshot_t(c, pos.snapshot)->equiv;
617 if (pos.inode == w->cur_inum) {
618 w->first_this_inode = false;
619 goto lookup_snapshot;
624 for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, pos.inode),
625 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
626 if (k.k->p.offset != pos.inode)
629 if (bkey_is_inode(k.k))
632 bch2_trans_iter_exit(trans, &iter);
637 w->cur_inum = pos.inode;
638 w->first_this_inode = true;
640 for (i = 0; i < w->nr; i++)
641 if (bch2_snapshot_is_ancestor(c, pos.snapshot, w->d[i].snapshot))
645 BUG_ON(pos.snapshot > w->d[i].snapshot);
647 if (pos.snapshot != w->d[i].snapshot) {
650 while (i && w->d[i - 1].snapshot > pos.snapshot)
653 ret = inode_walker_realloc(c, w);
657 array_insert_item(w->d, w->nr, i, w->d[ancestor_pos]);
658 w->d[i].snapshot = pos.snapshot;
665 static int __get_visible_inodes(struct btree_trans *trans,
666 struct inode_walker *w,
667 struct snapshots_seen *s,
670 struct bch_fs *c = trans->c;
671 struct btree_iter iter;
677 for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, inum),
678 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
679 if (k.k->p.offset != inum)
682 if (!bkey_is_inode(k.k))
685 if (ref_visible(c, s, s->pos.snapshot, k.k->p.snapshot)) {
687 if (k.k->p.snapshot >= s->pos.snapshot)
691 bch2_trans_iter_exit(trans, &iter);
696 static int check_key_has_snapshot(struct btree_trans *trans,
697 struct btree_iter *iter,
700 struct bch_fs *c = trans->c;
701 struct printbuf buf = PRINTBUF;
704 if (mustfix_fsck_err_on(!snapshot_t(c, k.k->p.snapshot)->equiv, c,
705 "key in missing snapshot: %s",
706 (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
707 ret = bch2_btree_delete_at(trans, iter,
708 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?: 1;
714 static int hash_redo_key(struct btree_trans *trans,
715 const struct bch_hash_desc desc,
716 struct bch_hash_info *hash_info,
717 struct btree_iter *k_iter, struct bkey_s_c k)
719 bch_err(trans->c, "hash_redo_key() not implemented yet");
722 struct bkey_i *delete;
725 delete = bch2_trans_kmalloc(trans, sizeof(*delete));
727 return PTR_ERR(delete);
729 tmp = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
733 bkey_reassemble(tmp, k);
735 bkey_init(&delete->k);
736 delete->k.p = k_iter->pos;
737 return bch2_btree_iter_traverse(k_iter) ?:
738 bch2_trans_update(trans, k_iter, delete, 0) ?:
739 bch2_hash_set(trans, desc, hash_info, k_iter->pos.inode, tmp, 0);
743 static int hash_check_key(struct btree_trans *trans,
744 const struct bch_hash_desc desc,
745 struct bch_hash_info *hash_info,
746 struct btree_iter *k_iter, struct bkey_s_c hash_k)
748 struct bch_fs *c = trans->c;
749 struct btree_iter iter = { NULL };
750 struct printbuf buf = PRINTBUF;
755 if (hash_k.k->type != desc.key_type)
758 hash = desc.hash_bkey(hash_info, hash_k);
760 if (likely(hash == hash_k.k->p.offset))
763 if (hash_k.k->p.offset < hash)
766 for_each_btree_key(trans, iter, desc.btree_id, POS(hash_k.k->p.inode, hash),
767 BTREE_ITER_SLOTS, k, ret) {
768 if (!bkey_cmp(k.k->p, hash_k.k->p))
771 if (fsck_err_on(k.k->type == desc.key_type &&
772 !desc.cmp_bkey(k, hash_k), c,
773 "duplicate hash table keys:\n%s",
774 (printbuf_reset(&buf),
775 bch2_bkey_val_to_text(&buf, c, hash_k),
777 ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0) ?: 1;
781 if (bkey_deleted(k.k)) {
782 bch2_trans_iter_exit(trans, &iter);
788 bch2_trans_iter_exit(trans, &iter);
792 if (fsck_err(c, "hash table key at wrong offset: btree %u inode %llu offset %llu, "
793 "hashed to %llu\n%s",
794 desc.btree_id, hash_k.k->p.inode, hash_k.k->p.offset, hash,
795 (printbuf_reset(&buf),
796 bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf)) == FSCK_ERR_IGNORE)
799 ret = hash_redo_key(trans, desc, hash_info, k_iter, hash_k);
801 bch_err(c, "hash_redo_key err %i", ret);
809 static int check_inode(struct btree_trans *trans,
810 struct btree_iter *iter,
811 struct bch_inode_unpacked *prev,
814 struct bch_fs *c = trans->c;
816 struct bch_inode_unpacked u;
817 bool do_update = false;
820 k = bch2_btree_iter_peek(iter);
828 ret = check_key_has_snapshot(trans, iter, k);
830 return ret < 0 ? ret : 0;
833 * if snapshot id isn't a leaf node, skip it - deletion in
834 * particular is not atomic, so on the internal snapshot nodes
835 * we can see inodes marked for deletion after a clean shutdown
837 if (bch2_snapshot_internal_node(c, k.k->p.snapshot))
840 if (!bkey_is_inode(k.k))
843 BUG_ON(bch2_inode_unpack(k, &u));
846 !(u.bi_flags & (BCH_INODE_I_SIZE_DIRTY|
847 BCH_INODE_I_SECTORS_DIRTY|
848 BCH_INODE_UNLINKED)))
851 if (prev->bi_inum != u.bi_inum)
854 if (fsck_err_on(prev->bi_hash_seed != u.bi_hash_seed ||
855 inode_d_type(prev) != inode_d_type(&u), c,
856 "inodes in different snapshots don't match")) {
857 bch_err(c, "repair not implemented yet");
861 if (u.bi_flags & BCH_INODE_UNLINKED &&
863 fsck_err(c, "filesystem marked clean, but inode %llu unlinked",
865 bch2_trans_unlock(trans);
868 ret = fsck_inode_rm(trans, u.bi_inum, iter->pos.snapshot);
870 bch_err(c, "error in fsck: error %i while deleting inode", ret);
874 if (u.bi_flags & BCH_INODE_I_SIZE_DIRTY &&
876 fsck_err(c, "filesystem marked clean, but inode %llu has i_size dirty",
878 bch_verbose(c, "truncating inode %llu", u.bi_inum);
880 bch2_trans_unlock(trans);
884 * XXX: need to truncate partial blocks too here - or ideally
885 * just switch units to bytes and that issue goes away
887 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
888 SPOS(u.bi_inum, round_up(u.bi_size, block_bytes(c)) >> 9,
890 POS(u.bi_inum, U64_MAX),
893 bch_err(c, "error in fsck: error %i truncating inode", ret);
898 * We truncated without our normal sector accounting hook, just
899 * make sure we recalculate it:
901 u.bi_flags |= BCH_INODE_I_SECTORS_DIRTY;
903 u.bi_flags &= ~BCH_INODE_I_SIZE_DIRTY;
907 if (u.bi_flags & BCH_INODE_I_SECTORS_DIRTY &&
909 fsck_err(c, "filesystem marked clean, but inode %llu has i_sectors dirty",
913 bch_verbose(c, "recounting sectors for inode %llu",
916 sectors = bch2_count_inode_sectors(trans, u.bi_inum, iter->pos.snapshot);
918 bch_err(c, "error in fsck: error %i recounting inode sectors",
923 u.bi_sectors = sectors;
924 u.bi_flags &= ~BCH_INODE_I_SECTORS_DIRTY;
928 if (u.bi_flags & BCH_INODE_BACKPTR_UNTRUSTED) {
931 u.bi_flags &= ~BCH_INODE_BACKPTR_UNTRUSTED;
936 ret = write_inode(trans, &u, iter->pos.snapshot);
938 bch_err(c, "error in fsck: error %i "
939 "updating inode", ret);
946 static int check_inodes(struct bch_fs *c, bool full)
948 struct btree_trans trans;
949 struct btree_iter iter;
950 struct bch_inode_unpacked prev = { 0 };
953 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
955 bch2_trans_iter_init(&trans, &iter, BTREE_ID_inodes, POS_MIN,
958 BTREE_ITER_ALL_SNAPSHOTS);
961 ret = __bch2_trans_do(&trans, NULL, NULL,
962 BTREE_INSERT_LAZY_RW|
964 check_inode(&trans, &iter, &prev, full));
967 } while (bch2_btree_iter_advance(&iter));
968 bch2_trans_iter_exit(&trans, &iter);
970 bch2_trans_exit(&trans);
974 static int check_subvol(struct btree_trans *trans,
975 struct btree_iter *iter)
978 struct bkey_s_c_subvolume subvol;
981 k = bch2_btree_iter_peek(iter);
989 if (k.k->type != KEY_TYPE_subvolume)
992 subvol = bkey_s_c_to_subvolume(k);
994 if (BCH_SUBVOLUME_UNLINKED(subvol.v)) {
995 ret = bch2_subvolume_delete(trans, iter->pos.offset);
996 if (ret && ret != -EINTR)
997 bch_err(trans->c, "error deleting subvolume %llu: %i",
998 iter->pos.offset, ret);
1007 static int check_subvols(struct bch_fs *c)
1009 struct btree_trans trans;
1010 struct btree_iter iter;
1013 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1015 bch2_trans_iter_init(&trans, &iter, BTREE_ID_subvolumes,
1018 BTREE_ITER_PREFETCH);
1021 ret = __bch2_trans_do(&trans, NULL, NULL,
1022 BTREE_INSERT_LAZY_RW|
1023 BTREE_INSERT_NOFAIL,
1024 check_subvol(&trans, &iter));
1027 } while (bch2_btree_iter_advance(&iter));
1028 bch2_trans_iter_exit(&trans, &iter);
1030 bch2_trans_exit(&trans);
1035 * Checking for overlapping extents needs to be reimplemented
1038 static int fix_overlapping_extent(struct btree_trans *trans,
1039 struct bkey_s_c k, struct bpos cut_at)
1041 struct btree_iter iter;
1045 u = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
1046 ret = PTR_ERR_OR_ZERO(u);
1050 bkey_reassemble(u, k);
1051 bch2_cut_front(cut_at, u);
1055 * We don't want to go through the extent_handle_overwrites path:
1057 * XXX: this is going to screw up disk accounting, extent triggers
1058 * assume things about extent overwrites - we should be running the
1059 * triggers manually here
1061 bch2_trans_iter_init(trans, &iter, BTREE_ID_extents, u->k.p,
1062 BTREE_ITER_INTENT|BTREE_ITER_NOT_EXTENTS);
1064 BUG_ON(iter.flags & BTREE_ITER_IS_EXTENTS);
1065 ret = bch2_btree_iter_traverse(&iter) ?:
1066 bch2_trans_update(trans, &iter, u, BTREE_TRIGGER_NORUN) ?:
1067 bch2_trans_commit(trans, NULL, NULL,
1068 BTREE_INSERT_NOFAIL|
1069 BTREE_INSERT_LAZY_RW);
1070 bch2_trans_iter_exit(trans, &iter);
1075 static struct bkey_s_c_dirent dirent_get_by_pos(struct btree_trans *trans,
1076 struct btree_iter *iter,
1082 bch2_trans_iter_init(trans, iter, BTREE_ID_dirents, pos, 0);
1083 k = bch2_btree_iter_peek_slot(iter);
1085 if (!ret && k.k->type != KEY_TYPE_dirent)
1088 bch2_trans_iter_exit(trans, iter);
1089 return (struct bkey_s_c_dirent) { .k = ERR_PTR(ret) };
1092 return bkey_s_c_to_dirent(k);
1095 static bool inode_points_to_dirent(struct bch_inode_unpacked *inode,
1096 struct bkey_s_c_dirent d)
1098 return inode->bi_dir == d.k->p.inode &&
1099 inode->bi_dir_offset == d.k->p.offset;
1102 static bool dirent_points_to_inode(struct bkey_s_c_dirent d,
1103 struct bch_inode_unpacked *inode)
1105 return d.v->d_type == DT_SUBVOL
1106 ? le32_to_cpu(d.v->d_child_subvol) == inode->bi_subvol
1107 : le64_to_cpu(d.v->d_inum) == inode->bi_inum;
1110 static int inode_backpointer_exists(struct btree_trans *trans,
1111 struct bch_inode_unpacked *inode,
1114 struct btree_iter iter;
1115 struct bkey_s_c_dirent d;
1118 d = dirent_get_by_pos(trans, &iter,
1119 SPOS(inode->bi_dir, inode->bi_dir_offset, snapshot));
1120 ret = bkey_err(d.s_c);
1124 ret = dirent_points_to_inode(d, inode);
1125 bch2_trans_iter_exit(trans, &iter);
1129 static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
1131 struct bch_fs *c = trans->c;
1132 struct inode_walker_entry *i;
1133 int ret = 0, ret2 = 0;
1136 for (i = w->d; i < w->d + w->nr; i++) {
1137 if (i->inode.bi_sectors == i->count)
1140 count2 = lockrestart_do(trans,
1141 bch2_count_inode_sectors(trans, w->cur_inum, i->snapshot));
1143 if (i->count != count2) {
1144 bch_err(c, "fsck counted i_sectors wrong: got %llu should be %llu",
1147 if (i->inode.bi_sectors == i->count)
1151 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SECTORS_DIRTY), c,
1152 "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
1153 w->cur_inum, i->snapshot,
1154 i->inode.bi_sectors, i->count) == FSCK_ERR_IGNORE)
1157 i->inode.bi_sectors = i->count;
1158 ret = write_inode(trans, &i->inode, i->snapshot);
1167 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
1168 struct inode_walker *inode,
1169 struct snapshots_seen *s)
1171 struct bch_fs *c = trans->c;
1173 struct inode_walker_entry *i;
1174 struct printbuf buf = PRINTBUF;
1177 k = bch2_btree_iter_peek(iter);
1185 ret = check_key_has_snapshot(trans, iter, k);
1187 ret = ret < 0 ? ret : 0;
1191 ret = snapshots_seen_update(c, s, k.k->p);
1195 if (k.k->type == KEY_TYPE_whiteout)
1198 if (inode->cur_inum != k.k->p.inode) {
1199 ret = check_i_sectors(trans, inode);
1204 if (bkey_cmp(prev.k->k.p, bkey_start_pos(k.k)) > 0) {
1208 bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(prev.k));
1209 bch2_bkey_val_to_text(&PBUF(buf2), c, k);
1211 if (fsck_err(c, "overlapping extents:\n%s\n%s", buf1, buf2)) {
1212 ret = fix_overlapping_extent(trans, k, prev.k->k.p) ?: -EINTR;
1217 ret = __walk_inode(trans, inode, k.k->p);
1221 if (fsck_err_on(ret == INT_MAX, c,
1222 "extent in missing inode:\n %s",
1223 (printbuf_reset(&buf),
1224 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1225 ret = bch2_btree_delete_at(trans, iter,
1226 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1230 if (ret == INT_MAX) {
1238 if (fsck_err_on(!S_ISREG(i->inode.bi_mode) &&
1239 !S_ISLNK(i->inode.bi_mode), c,
1240 "extent in non regular inode mode %o:\n %s",
1242 (printbuf_reset(&buf),
1243 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1244 ret = bch2_btree_delete_at(trans, iter,
1245 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1249 if (!bch2_snapshot_internal_node(c, k.k->p.snapshot)) {
1250 for_each_visible_inode(c, s, inode, k.k->p.snapshot, i) {
1251 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SIZE_DIRTY) &&
1252 k.k->type != KEY_TYPE_reservation &&
1253 k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9, c,
1254 "extent type %u offset %llu past end of inode %llu, i_size %llu",
1255 k.k->type, k.k->p.offset, k.k->p.inode, i->inode.bi_size)) {
1257 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
1258 SPOS(k.k->p.inode, round_up(i->inode.bi_size, block_bytes(c)) >> 9,
1260 POS(k.k->p.inode, U64_MAX),
1267 if (bkey_extent_is_allocation(k.k))
1268 for_each_visible_inode(c, s, inode, k.k->p.snapshot, i)
1269 i->count += k.k->size;
1271 bch2_bkey_buf_reassemble(&prev, c, k);
1277 printbuf_exit(&buf);
1282 * Walk extents: verify that extents have a corresponding S_ISREG inode, and
1283 * that i_size an i_sectors are consistent
1286 static int check_extents(struct bch_fs *c)
1288 struct inode_walker w = inode_walker_init();
1289 struct snapshots_seen s;
1290 struct btree_trans trans;
1291 struct btree_iter iter;
1295 struct bkey_buf prev;
1296 bch2_bkey_buf_init(&prev);
1297 prev.k->k = KEY(0, 0, 0);
1299 snapshots_seen_init(&s);
1300 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1302 bch_verbose(c, "checking extents");
1304 bch2_trans_iter_init(&trans, &iter, BTREE_ID_extents,
1305 POS(BCACHEFS_ROOT_INO, 0),
1307 BTREE_ITER_PREFETCH|
1308 BTREE_ITER_ALL_SNAPSHOTS);
1311 ret = __bch2_trans_do(&trans, NULL, NULL,
1312 BTREE_INSERT_LAZY_RW|
1313 BTREE_INSERT_NOFAIL,
1314 check_extent(&trans, &iter, &w, &s));
1317 } while (bch2_btree_iter_advance(&iter));
1318 bch2_trans_iter_exit(&trans, &iter);
1320 bch2_bkey_buf_exit(&prev, c);
1322 inode_walker_exit(&w);
1323 bch2_trans_exit(&trans);
1324 snapshots_seen_exit(&s);
1329 static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
1331 struct bch_fs *c = trans->c;
1332 struct inode_walker_entry *i;
1333 int ret = 0, ret2 = 0;
1336 for (i = w->d; i < w->d + w->nr; i++) {
1337 if (i->inode.bi_nlink == i->count)
1340 count2 = bch2_count_subdirs(trans, w->cur_inum, i->snapshot);
1344 if (i->count != count2) {
1345 bch_err(c, "fsck counted subdirectories wrong: got %llu should be %llu",
1348 if (i->inode.bi_nlink == i->count)
1352 if (fsck_err_on(i->inode.bi_nlink != i->count, c,
1353 "directory %llu:%u with wrong i_nlink: got %u, should be %llu",
1354 w->cur_inum, i->snapshot, i->inode.bi_nlink, i->count)) {
1355 i->inode.bi_nlink = i->count;
1356 ret = write_inode(trans, &i->inode, i->snapshot);
1366 static int check_dirent_target(struct btree_trans *trans,
1367 struct btree_iter *iter,
1368 struct bkey_s_c_dirent d,
1369 struct bch_inode_unpacked *target,
1370 u32 target_snapshot)
1372 struct bch_fs *c = trans->c;
1373 struct bkey_i_dirent *n;
1374 bool backpointer_exists = true;
1375 struct printbuf buf = PRINTBUF;
1378 if (!target->bi_dir &&
1379 !target->bi_dir_offset) {
1380 target->bi_dir = d.k->p.inode;
1381 target->bi_dir_offset = d.k->p.offset;
1383 ret = __write_inode(trans, target, target_snapshot);
1388 if (!inode_points_to_dirent(target, d)) {
1389 ret = inode_backpointer_exists(trans, target, d.k->p.snapshot);
1393 backpointer_exists = ret;
1396 if (fsck_err_on(S_ISDIR(target->bi_mode) &&
1397 backpointer_exists, c,
1398 "directory %llu with multiple links",
1400 ret = __remove_dirent(trans, d.k->p);
1404 if (fsck_err_on(backpointer_exists &&
1405 !target->bi_nlink, c,
1406 "inode %llu has multiple links but i_nlink 0",
1409 target->bi_flags &= ~BCH_INODE_UNLINKED;
1411 ret = __write_inode(trans, target, target_snapshot);
1416 if (fsck_err_on(!backpointer_exists, c,
1417 "inode %llu:%u has wrong backpointer:\n"
1419 "should be %llu:%llu",
1420 target->bi_inum, target_snapshot,
1422 target->bi_dir_offset,
1425 target->bi_dir = d.k->p.inode;
1426 target->bi_dir_offset = d.k->p.offset;
1428 ret = __write_inode(trans, target, target_snapshot);
1434 if (fsck_err_on(d.v->d_type != inode_d_type(target), c,
1435 "incorrect d_type: got %s, should be %s:\n%s",
1436 bch2_d_type_str(d.v->d_type),
1437 bch2_d_type_str(inode_d_type(target)),
1438 (printbuf_reset(&buf),
1439 bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
1440 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1441 ret = PTR_ERR_OR_ZERO(n);
1445 bkey_reassemble(&n->k_i, d.s_c);
1446 n->v.d_type = inode_d_type(target);
1448 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1452 d = dirent_i_to_s_c(n);
1455 if (d.v->d_type == DT_SUBVOL &&
1456 target->bi_parent_subvol != le32_to_cpu(d.v->d_parent_subvol) &&
1457 (c->sb.version < bcachefs_metadata_version_subvol_dirent ||
1458 fsck_err(c, "dirent has wrong d_parent_subvol field: got %u, should be %u",
1459 le32_to_cpu(d.v->d_parent_subvol),
1460 target->bi_parent_subvol))) {
1461 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1462 ret = PTR_ERR_OR_ZERO(n);
1466 bkey_reassemble(&n->k_i, d.s_c);
1467 n->v.d_parent_subvol = cpu_to_le32(target->bi_parent_subvol);
1469 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1473 d = dirent_i_to_s_c(n);
1478 printbuf_exit(&buf);
1482 static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
1483 struct bch_hash_info *hash_info,
1484 struct inode_walker *dir,
1485 struct inode_walker *target,
1486 struct snapshots_seen *s)
1488 struct bch_fs *c = trans->c;
1490 struct bkey_s_c_dirent d;
1491 struct inode_walker_entry *i;
1492 struct printbuf buf = PRINTBUF;
1495 k = bch2_btree_iter_peek(iter);
1503 ret = check_key_has_snapshot(trans, iter, k);
1505 ret = ret < 0 ? ret : 0;
1509 ret = snapshots_seen_update(c, s, k.k->p);
1513 if (k.k->type == KEY_TYPE_whiteout)
1516 if (dir->cur_inum != k.k->p.inode) {
1517 ret = check_subdir_count(trans, dir);
1522 ret = __walk_inode(trans, dir, k.k->p);
1526 if (fsck_err_on(ret == INT_MAX, c,
1527 "dirent in nonexisting directory:\n%s",
1528 (printbuf_reset(&buf),
1529 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1530 ret = bch2_btree_delete_at(trans, iter,
1531 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1535 if (ret == INT_MAX) {
1543 if (fsck_err_on(!S_ISDIR(i->inode.bi_mode), c,
1544 "dirent in non directory inode type %s:\n%s",
1545 bch2_d_type_str(inode_d_type(&i->inode)),
1546 (printbuf_reset(&buf),
1547 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1548 ret = bch2_btree_delete_at(trans, iter, 0);
1552 if (dir->first_this_inode)
1553 *hash_info = bch2_hash_info_init(c, &dir->d[0].inode);
1555 ret = hash_check_key(trans, bch2_dirent_hash_desc,
1556 hash_info, iter, k);
1560 /* dirent has been deleted */
1565 if (k.k->type != KEY_TYPE_dirent)
1568 d = bkey_s_c_to_dirent(k);
1570 if (d.v->d_type == DT_SUBVOL) {
1571 struct bch_inode_unpacked subvol_root;
1572 u32 target_subvol = le32_to_cpu(d.v->d_child_subvol);
1573 u32 target_snapshot;
1576 ret = __subvol_lookup(trans, target_subvol,
1577 &target_snapshot, &target_inum);
1578 if (ret && ret != -ENOENT)
1581 if (fsck_err_on(ret, c,
1582 "dirent points to missing subvolume %llu",
1583 le64_to_cpu(d.v->d_child_subvol))) {
1584 ret = __remove_dirent(trans, d.k->p);
1588 ret = __lookup_inode(trans, target_inum,
1589 &subvol_root, &target_snapshot);
1590 if (ret && ret != -ENOENT)
1593 if (fsck_err_on(ret, c,
1594 "subvolume %u points to missing subvolume root %llu",
1597 bch_err(c, "repair not implemented yet");
1602 if (fsck_err_on(subvol_root.bi_subvol != target_subvol, c,
1603 "subvol root %llu has wrong bi_subvol field: got %u, should be %u",
1605 subvol_root.bi_subvol, target_subvol)) {
1606 subvol_root.bi_subvol = target_subvol;
1607 ret = __write_inode(trans, &subvol_root, target_snapshot);
1612 ret = check_dirent_target(trans, iter, d, &subvol_root,
1617 ret = __get_visible_inodes(trans, target, s, le64_to_cpu(d.v->d_inum));
1621 if (fsck_err_on(!target->nr, c,
1622 "dirent points to missing inode:\n%s",
1623 (printbuf_reset(&buf),
1624 bch2_bkey_val_to_text(&buf, c, k),
1626 ret = __remove_dirent(trans, d.k->p);
1631 for (i = target->d; i < target->d + target->nr; i++) {
1632 ret = check_dirent_target(trans, iter, d,
1633 &i->inode, i->snapshot);
1639 if (d.v->d_type == DT_DIR)
1640 for_each_visible_inode(c, s, dir, d.k->p.snapshot, i)
1646 printbuf_exit(&buf);
1651 * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
1655 static int check_dirents(struct bch_fs *c)
1657 struct inode_walker dir = inode_walker_init();
1658 struct inode_walker target = inode_walker_init();
1659 struct snapshots_seen s;
1660 struct bch_hash_info hash_info;
1661 struct btree_trans trans;
1662 struct btree_iter iter;
1665 bch_verbose(c, "checking dirents");
1667 snapshots_seen_init(&s);
1668 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1670 bch2_trans_iter_init(&trans, &iter, BTREE_ID_dirents,
1671 POS(BCACHEFS_ROOT_INO, 0),
1673 BTREE_ITER_PREFETCH|
1674 BTREE_ITER_ALL_SNAPSHOTS);
1677 ret = __bch2_trans_do(&trans, NULL, NULL,
1678 BTREE_INSERT_LAZY_RW|
1679 BTREE_INSERT_NOFAIL,
1680 check_dirent(&trans, &iter, &hash_info,
1681 &dir, &target, &s));
1684 } while (bch2_btree_iter_advance(&iter));
1685 bch2_trans_iter_exit(&trans, &iter);
1687 bch2_trans_exit(&trans);
1688 snapshots_seen_exit(&s);
1689 inode_walker_exit(&dir);
1690 inode_walker_exit(&target);
1694 static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
1695 struct bch_hash_info *hash_info,
1696 struct inode_walker *inode)
1698 struct bch_fs *c = trans->c;
1702 k = bch2_btree_iter_peek(iter);
1710 ret = check_key_has_snapshot(trans, iter, k);
1714 ret = __walk_inode(trans, inode, k.k->p);
1718 if (fsck_err_on(ret == INT_MAX, c,
1719 "xattr for missing inode %llu",
1721 return bch2_btree_delete_at(trans, iter, 0);
1728 if (inode->first_this_inode)
1729 *hash_info = bch2_hash_info_init(c, &inode->d[0].inode);
1731 ret = hash_check_key(trans, bch2_xattr_hash_desc, hash_info, iter, k);
1737 * Walk xattrs: verify that they all have a corresponding inode
1740 static int check_xattrs(struct bch_fs *c)
1742 struct inode_walker inode = inode_walker_init();
1743 struct bch_hash_info hash_info;
1744 struct btree_trans trans;
1745 struct btree_iter iter;
1748 bch_verbose(c, "checking xattrs");
1750 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1752 bch2_trans_iter_init(&trans, &iter, BTREE_ID_xattrs,
1753 POS(BCACHEFS_ROOT_INO, 0),
1755 BTREE_ITER_PREFETCH|
1756 BTREE_ITER_ALL_SNAPSHOTS);
1759 ret = __bch2_trans_do(&trans, NULL, NULL,
1760 BTREE_INSERT_LAZY_RW|
1761 BTREE_INSERT_NOFAIL,
1762 check_xattr(&trans, &iter, &hash_info,
1766 } while (bch2_btree_iter_advance(&iter));
1767 bch2_trans_iter_exit(&trans, &iter);
1769 bch2_trans_exit(&trans);
1773 static int check_root_trans(struct btree_trans *trans)
1775 struct bch_fs *c = trans->c;
1776 struct bch_inode_unpacked root_inode;
1781 ret = __subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
1782 if (ret && ret != -ENOENT)
1785 if (mustfix_fsck_err_on(ret, c, "root subvol missing")) {
1786 struct bkey_i_subvolume root_subvol;
1789 inum = BCACHEFS_ROOT_INO;
1791 bkey_subvolume_init(&root_subvol.k_i);
1792 root_subvol.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1793 root_subvol.v.flags = 0;
1794 root_subvol.v.snapshot = cpu_to_le32(snapshot);
1795 root_subvol.v.inode = cpu_to_le64(inum);
1796 ret = __bch2_trans_do(trans, NULL, NULL,
1797 BTREE_INSERT_NOFAIL|
1798 BTREE_INSERT_LAZY_RW,
1799 __bch2_btree_insert(trans, BTREE_ID_subvolumes, &root_subvol.k_i));
1801 bch_err(c, "error writing root subvol: %i", ret);
1807 ret = __lookup_inode(trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
1808 if (ret && ret != -ENOENT)
1811 if (mustfix_fsck_err_on(ret, c, "root directory missing") ||
1812 mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode), c,
1813 "root inode not a directory")) {
1814 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755,
1816 root_inode.bi_inum = inum;
1818 ret = __write_inode(trans, &root_inode, snapshot);
1820 bch_err(c, "error writing root inode: %i", ret);
1827 /* Get root directory, create if it doesn't exist: */
1829 static int check_root(struct bch_fs *c)
1831 bch_verbose(c, "checking root directory");
1833 return bch2_trans_do(c, NULL, NULL,
1834 BTREE_INSERT_NOFAIL|
1835 BTREE_INSERT_LAZY_RW,
1836 check_root_trans(&trans));
1843 struct pathbuf_entry {
1849 static bool path_is_dup(struct pathbuf *p, u64 inum, u32 snapshot)
1851 struct pathbuf_entry *i;
1853 for (i = p->entries; i < p->entries + p->nr; i++)
1854 if (i->inum == inum &&
1855 i->snapshot == snapshot)
1861 static int path_down(struct bch_fs *c, struct pathbuf *p,
1862 u64 inum, u32 snapshot)
1864 if (p->nr == p->size) {
1865 size_t new_size = max_t(size_t, 256UL, p->size * 2);
1866 void *n = krealloc(p->entries,
1867 new_size * sizeof(p->entries[0]),
1870 bch_err(c, "fsck: error allocating memory for pathbuf, size %zu",
1879 p->entries[p->nr++] = (struct pathbuf_entry) {
1881 .snapshot = snapshot,
1887 * Check that a given inode is reachable from the root:
1889 * XXX: we should also be verifying that inodes are in the right subvolumes
1891 static int check_path(struct btree_trans *trans,
1893 struct bch_inode_unpacked *inode,
1896 struct bch_fs *c = trans->c;
1899 snapshot = snapshot_t(c, snapshot)->equiv;
1902 while (!(inode->bi_inum == BCACHEFS_ROOT_INO &&
1903 inode->bi_subvol == BCACHEFS_ROOT_SUBVOL)) {
1904 struct btree_iter dirent_iter;
1905 struct bkey_s_c_dirent d;
1906 u32 parent_snapshot = snapshot;
1908 if (inode->bi_subvol) {
1911 ret = subvol_lookup(trans, inode->bi_parent_subvol,
1912 &parent_snapshot, &inum);
1917 ret = lockrestart_do(trans,
1918 PTR_ERR_OR_ZERO((d = dirent_get_by_pos(trans, &dirent_iter,
1919 SPOS(inode->bi_dir, inode->bi_dir_offset,
1920 parent_snapshot))).k));
1921 if (ret && ret != -ENOENT)
1924 if (!ret && !dirent_points_to_inode(d, inode)) {
1925 bch2_trans_iter_exit(trans, &dirent_iter);
1929 if (ret == -ENOENT) {
1930 if (fsck_err(c, "unreachable inode %llu:%u, type %s nlink %u backptr %llu:%llu",
1931 inode->bi_inum, snapshot,
1932 bch2_d_type_str(inode_d_type(inode)),
1935 inode->bi_dir_offset))
1936 ret = reattach_inode(trans, inode, snapshot);
1940 bch2_trans_iter_exit(trans, &dirent_iter);
1942 if (!S_ISDIR(inode->bi_mode))
1945 ret = path_down(c, p, inode->bi_inum, snapshot);
1947 bch_err(c, "memory allocation failure");
1951 snapshot = parent_snapshot;
1953 ret = lookup_inode(trans, inode->bi_dir, inode, &snapshot);
1955 /* Should have been caught in dirents pass */
1956 bch_err(c, "error looking up parent directory: %i", ret);
1960 if (path_is_dup(p, inode->bi_inum, snapshot)) {
1961 struct pathbuf_entry *i;
1963 /* XXX print path */
1964 bch_err(c, "directory structure loop");
1966 for (i = p->entries; i < p->entries + p->nr; i++)
1967 pr_err("%llu:%u", i->inum, i->snapshot);
1968 pr_err("%llu:%u", inode->bi_inum, snapshot);
1970 if (!fsck_err(c, "directory structure loop"))
1973 ret = __bch2_trans_do(trans, NULL, NULL,
1974 BTREE_INSERT_NOFAIL|
1975 BTREE_INSERT_LAZY_RW,
1976 remove_backpointer(trans, inode));
1978 bch_err(c, "error removing dirent: %i", ret);
1982 ret = reattach_inode(trans, inode, snapshot);
1987 bch_err(c, "%s: err %i", __func__, ret);
1992 * Check for unreachable inodes, as well as loops in the directory structure:
1993 * After check_dirents(), if an inode backpointer doesn't exist that means it's
1997 static int check_directory_structure(struct bch_fs *c)
1999 struct btree_trans trans;
2000 struct btree_iter iter;
2002 struct bch_inode_unpacked u;
2003 struct pathbuf path = { 0, 0, NULL };
2006 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2008 for_each_btree_key(&trans, iter, BTREE_ID_inodes, POS_MIN,
2010 BTREE_ITER_PREFETCH|
2011 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2012 if (!bkey_is_inode(k.k))
2015 ret = bch2_inode_unpack(k, &u);
2017 /* Should have been caught earlier in fsck: */
2018 bch_err(c, "error unpacking inode %llu: %i", k.k->p.offset, ret);
2022 if (u.bi_flags & BCH_INODE_UNLINKED)
2025 ret = check_path(&trans, &path, &u, iter.pos.snapshot);
2029 bch2_trans_iter_exit(&trans, &iter);
2031 BUG_ON(ret == -EINTR);
2033 kfree(path.entries);
2035 bch2_trans_exit(&trans);
2039 struct nlink_table {
2050 static int add_nlink(struct bch_fs *c, struct nlink_table *t,
2051 u64 inum, u32 snapshot)
2053 if (t->nr == t->size) {
2054 size_t new_size = max_t(size_t, 128UL, t->size * 2);
2055 void *d = kvmalloc(new_size * sizeof(t->d[0]), GFP_KERNEL);
2057 bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
2063 memcpy(d, t->d, t->size * sizeof(t->d[0]));
2071 t->d[t->nr++] = (struct nlink) {
2073 .snapshot = snapshot,
2079 static int nlink_cmp(const void *_l, const void *_r)
2081 const struct nlink *l = _l;
2082 const struct nlink *r = _r;
2084 return cmp_int(l->inum, r->inum) ?: cmp_int(l->snapshot, r->snapshot);
2087 static void inc_link(struct bch_fs *c, struct snapshots_seen *s,
2088 struct nlink_table *links,
2089 u64 range_start, u64 range_end, u64 inum, u32 snapshot)
2091 struct nlink *link, key = {
2092 .inum = inum, .snapshot = U32_MAX,
2095 if (inum < range_start || inum >= range_end)
2098 link = __inline_bsearch(&key, links->d, links->nr,
2099 sizeof(links->d[0]), nlink_cmp);
2103 while (link > links->d && link[0].inum == link[-1].inum)
2106 for (; link < links->d + links->nr && link->inum == inum; link++)
2107 if (ref_visible(c, s, snapshot, link->snapshot)) {
2109 if (link->snapshot >= snapshot)
2115 static int check_nlinks_find_hardlinks(struct bch_fs *c,
2116 struct nlink_table *t,
2117 u64 start, u64 *end)
2119 struct btree_trans trans;
2120 struct btree_iter iter;
2122 struct bch_inode_unpacked u;
2125 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2127 for_each_btree_key(&trans, iter, BTREE_ID_inodes,
2130 BTREE_ITER_PREFETCH|
2131 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2132 if (!bkey_is_inode(k.k))
2135 /* Should never fail, checked by bch2_inode_invalid: */
2136 BUG_ON(bch2_inode_unpack(k, &u));
2139 * Backpointer and directory structure checks are sufficient for
2140 * directories, since they can't have hardlinks:
2142 if (S_ISDIR(le16_to_cpu(u.bi_mode)))
2148 ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot);
2150 *end = k.k->p.offset;
2156 bch2_trans_iter_exit(&trans, &iter);
2157 bch2_trans_exit(&trans);
2160 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
2166 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
2167 u64 range_start, u64 range_end)
2169 struct btree_trans trans;
2170 struct snapshots_seen s;
2171 struct btree_iter iter;
2173 struct bkey_s_c_dirent d;
2176 snapshots_seen_init(&s);
2178 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2180 for_each_btree_key(&trans, iter, BTREE_ID_dirents, POS_MIN,
2182 BTREE_ITER_PREFETCH|
2183 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2184 ret = snapshots_seen_update(c, &s, k.k->p);
2188 switch (k.k->type) {
2189 case KEY_TYPE_dirent:
2190 d = bkey_s_c_to_dirent(k);
2192 if (d.v->d_type != DT_DIR &&
2193 d.v->d_type != DT_SUBVOL)
2194 inc_link(c, &s, links, range_start, range_end,
2195 le64_to_cpu(d.v->d_inum),
2200 bch2_trans_iter_exit(&trans, &iter);
2203 bch_err(c, "error in fsck: btree error %i while walking dirents", ret);
2205 bch2_trans_exit(&trans);
2206 snapshots_seen_exit(&s);
2211 static int check_nlinks_update_hardlinks(struct bch_fs *c,
2212 struct nlink_table *links,
2213 u64 range_start, u64 range_end)
2215 struct btree_trans trans;
2216 struct btree_iter iter;
2218 struct bch_inode_unpacked u;
2219 struct nlink *link = links->d;
2222 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2224 for_each_btree_key(&trans, iter, BTREE_ID_inodes,
2225 POS(0, range_start),
2227 BTREE_ITER_PREFETCH|
2228 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2229 if (k.k->p.offset >= range_end)
2232 if (!bkey_is_inode(k.k))
2235 BUG_ON(bch2_inode_unpack(k, &u));
2237 if (S_ISDIR(le16_to_cpu(u.bi_mode)))
2243 while ((cmp_int(link->inum, k.k->p.offset) ?:
2244 cmp_int(link->snapshot, k.k->p.snapshot)) < 0) {
2246 BUG_ON(link >= links->d + links->nr);
2249 if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count, c,
2250 "inode %llu has wrong i_nlink (type %u i_nlink %u, should be %u)",
2251 u.bi_inum, mode_to_type(u.bi_mode),
2252 bch2_inode_nlink_get(&u), link->count)) {
2253 bch2_inode_nlink_set(&u, link->count);
2255 ret = write_inode(&trans, &u, k.k->p.snapshot);
2257 bch_err(c, "error in fsck: error %i updating inode", ret);
2261 bch2_trans_iter_exit(&trans, &iter);
2262 bch2_trans_exit(&trans);
2265 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
2271 static int check_nlinks(struct bch_fs *c)
2273 struct nlink_table links = { 0 };
2274 u64 this_iter_range_start, next_iter_range_start = 0;
2277 bch_verbose(c, "checking inode nlinks");
2280 this_iter_range_start = next_iter_range_start;
2281 next_iter_range_start = U64_MAX;
2283 ret = check_nlinks_find_hardlinks(c, &links,
2284 this_iter_range_start,
2285 &next_iter_range_start);
2287 ret = check_nlinks_walk_dirents(c, &links,
2288 this_iter_range_start,
2289 next_iter_range_start);
2293 ret = check_nlinks_update_hardlinks(c, &links,
2294 this_iter_range_start,
2295 next_iter_range_start);
2300 } while (next_iter_range_start != U64_MAX);
2307 static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter)
2310 struct bkey_s_c_reflink_p p;
2311 struct bkey_i_reflink_p *u;
2314 k = bch2_btree_iter_peek(iter);
2322 if (k.k->type != KEY_TYPE_reflink_p)
2325 p = bkey_s_c_to_reflink_p(k);
2327 if (!p.v->front_pad && !p.v->back_pad)
2330 u = bch2_trans_kmalloc(trans, sizeof(*u));
2331 ret = PTR_ERR_OR_ZERO(u);
2335 bkey_reassemble(&u->k_i, k);
2339 return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_NORUN);
2343 static int fix_reflink_p(struct bch_fs *c)
2345 struct btree_trans trans;
2346 struct btree_iter iter;
2350 if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
2353 bch_verbose(c, "fixing reflink_p keys");
2355 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2357 for_each_btree_key(&trans, iter, BTREE_ID_extents, POS_MIN,
2359 BTREE_ITER_PREFETCH|
2360 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2361 if (k.k->type == KEY_TYPE_reflink_p) {
2362 ret = __bch2_trans_do(&trans, NULL, NULL,
2363 BTREE_INSERT_NOFAIL|
2364 BTREE_INSERT_LAZY_RW,
2365 fix_reflink_p_key(&trans, &iter));
2370 bch2_trans_iter_exit(&trans, &iter);
2372 bch2_trans_exit(&trans);
2377 * Checks for inconsistencies that shouldn't happen, unless we have a bug.
2378 * Doesn't fix them yet, mainly because they haven't yet been observed:
2380 int bch2_fsck_full(struct bch_fs *c)
2382 return bch2_fs_snapshots_check(c) ?:
2383 check_inodes(c, true) ?:
2389 check_directory_structure(c) ?:
2394 int bch2_fsck_walk_inodes_only(struct bch_fs *c)
2396 return check_inodes(c, false);