1 // SPDX-License-Identifier: GPL-2.0
5 #include "btree_cache.h"
6 #include "btree_update.h"
11 #include "fs-common.h"
20 #include <linux/bsearch.h>
21 #include <linux/dcache.h> /* struct qstr */
23 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
26 * XXX: this is handling transaction restarts without returning
27 * -BCH_ERR_transaction_restart_nested, this is not how we do things anymore:
29 static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum,
32 struct btree_iter iter;
37 for_each_btree_key_upto(trans, iter, BTREE_ID_extents,
38 SPOS(inum, 0, snapshot),
41 if (bkey_extent_is_allocation(k.k))
44 bch2_trans_iter_exit(trans, &iter);
46 return ret ?: sectors;
49 static s64 bch2_count_subdirs(struct btree_trans *trans, u64 inum,
52 struct btree_iter iter;
54 struct bkey_s_c_dirent d;
58 for_each_btree_key_upto(trans, iter, BTREE_ID_dirents,
59 SPOS(inum, 0, snapshot),
62 if (k.k->type != KEY_TYPE_dirent)
65 d = bkey_s_c_to_dirent(k);
66 if (d.v->d_type == DT_DIR)
69 bch2_trans_iter_exit(trans, &iter);
71 return ret ?: subdirs;
74 static int __snapshot_lookup_subvol(struct btree_trans *trans, u32 snapshot,
77 struct bch_snapshot s;
78 int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_snapshots,
82 *subvol = le32_to_cpu(s.subvol);
83 else if (bch2_err_matches(ret, ENOENT))
84 bch_err(trans->c, "snapshot %u not found", snapshot);
89 static int __subvol_lookup(struct btree_trans *trans, u32 subvol,
90 u32 *snapshot, u64 *inum)
92 struct bch_subvolume s;
95 ret = bch2_subvolume_get(trans, subvol, false, 0, &s);
97 *snapshot = le32_to_cpu(s.snapshot);
98 *inum = le64_to_cpu(s.inode);
102 static int subvol_lookup(struct btree_trans *trans, u32 subvol,
103 u32 *snapshot, u64 *inum)
105 return lockrestart_do(trans, __subvol_lookup(trans, subvol, snapshot, inum));
108 static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr,
109 struct bch_inode_unpacked *inode)
111 struct btree_iter iter;
115 bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
117 BTREE_ITER_ALL_SNAPSHOTS);
118 k = bch2_btree_iter_peek(&iter);
123 if (!k.k || !bkey_eq(k.k->p, POS(0, inode_nr))) {
124 ret = -BCH_ERR_ENOENT_inode;
128 ret = bch2_inode_unpack(k, inode);
130 bch_err_msg(trans->c, ret, "fetching inode %llu", inode_nr);
131 bch2_trans_iter_exit(trans, &iter);
135 static int __lookup_inode(struct btree_trans *trans, u64 inode_nr,
136 struct bch_inode_unpacked *inode,
139 struct btree_iter iter;
143 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
144 SPOS(0, inode_nr, *snapshot), 0);
149 ret = bkey_is_inode(k.k)
150 ? bch2_inode_unpack(k, inode)
151 : -BCH_ERR_ENOENT_inode;
153 *snapshot = iter.pos.snapshot;
155 bch_err_msg(trans->c, ret, "fetching inode %llu:%u", inode_nr, *snapshot);
156 bch2_trans_iter_exit(trans, &iter);
160 static int lookup_inode(struct btree_trans *trans, u64 inode_nr,
161 struct bch_inode_unpacked *inode,
164 return lockrestart_do(trans, __lookup_inode(trans, inode_nr, inode, snapshot));
167 static int __lookup_dirent(struct btree_trans *trans,
168 struct bch_hash_info hash_info,
169 subvol_inum dir, struct qstr *name,
170 u64 *target, unsigned *type)
172 struct btree_iter iter;
173 struct bkey_s_c_dirent d;
176 ret = bch2_hash_lookup(trans, &iter, bch2_dirent_hash_desc,
177 &hash_info, dir, name, 0);
181 d = bkey_s_c_to_dirent(bch2_btree_iter_peek_slot(&iter));
182 *target = le64_to_cpu(d.v->d_inum);
184 bch2_trans_iter_exit(trans, &iter);
188 static int __write_inode(struct btree_trans *trans,
189 struct bch_inode_unpacked *inode,
192 struct bkey_inode_buf *inode_p =
193 bch2_trans_kmalloc(trans, sizeof(*inode_p));
196 return PTR_ERR(inode_p);
198 bch2_inode_pack(inode_p, inode);
199 inode_p->inode.k.p.snapshot = snapshot;
201 return bch2_btree_insert_nonextent(trans, BTREE_ID_inodes,
203 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
206 static int fsck_write_inode(struct btree_trans *trans,
207 struct bch_inode_unpacked *inode,
210 int ret = commit_do(trans, NULL, NULL,
212 BTREE_INSERT_LAZY_RW,
213 __write_inode(trans, inode, snapshot));
215 bch_err_fn(trans->c, ret);
219 static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
221 struct bch_fs *c = trans->c;
222 struct btree_iter iter;
223 struct bch_inode_unpacked dir_inode;
224 struct bch_hash_info dir_hash_info;
227 ret = lookup_first_inode(trans, pos.inode, &dir_inode);
231 dir_hash_info = bch2_hash_info_init(c, &dir_inode);
233 bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_INTENT);
235 ret = bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
236 &dir_hash_info, &iter,
237 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
238 bch2_trans_iter_exit(trans, &iter);
244 /* Get lost+found, create if it doesn't exist: */
245 static int lookup_lostfound(struct btree_trans *trans, u32 subvol,
246 struct bch_inode_unpacked *lostfound)
248 struct bch_fs *c = trans->c;
249 struct bch_inode_unpacked root;
250 struct bch_hash_info root_hash_info;
251 struct qstr lostfound_str = QSTR("lost+found");
252 subvol_inum root_inum = { .subvol = subvol };
258 ret = __subvol_lookup(trans, subvol, &snapshot, &root_inum.inum);
262 ret = __lookup_inode(trans, root_inum.inum, &root, &snapshot);
266 root_hash_info = bch2_hash_info_init(c, &root);
268 ret = __lookup_dirent(trans, root_hash_info, root_inum,
269 &lostfound_str, &inum, &d_type);
270 if (bch2_err_matches(ret, ENOENT)) {
271 bch_notice(c, "creating lost+found");
272 goto create_lostfound;
279 if (d_type != DT_DIR) {
280 bch_err(c, "error looking up lost+found: not a directory");
281 return -BCH_ERR_ENOENT_not_directory;
285 * The bch2_check_dirents pass has already run, dangling dirents
286 * shouldn't exist here:
288 return __lookup_inode(trans, inum, lostfound, &snapshot);
291 bch2_inode_init_early(c, lostfound);
293 ret = bch2_create_trans(trans, root_inum, &root,
294 lostfound, &lostfound_str,
295 0, 0, S_IFDIR|0700, 0, NULL, NULL,
296 (subvol_inum) { }, 0);
297 bch_err_msg(c, ret, "creating lost+found");
301 static int __reattach_inode(struct btree_trans *trans,
302 struct bch_inode_unpacked *inode,
305 struct bch_hash_info dir_hash;
306 struct bch_inode_unpacked lostfound;
313 ret = __snapshot_lookup_subvol(trans, inode_snapshot, &subvol);
317 ret = lookup_lostfound(trans, subvol, &lostfound);
321 if (S_ISDIR(inode->bi_mode)) {
322 lostfound.bi_nlink++;
324 ret = __write_inode(trans, &lostfound, U32_MAX);
329 dir_hash = bch2_hash_info_init(trans->c, &lostfound);
331 snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
332 name = (struct qstr) QSTR(name_buf);
334 ret = bch2_dirent_create(trans,
337 .inum = lostfound.bi_inum,
341 &name, inode->bi_inum, &dir_offset,
342 BCH_HASH_SET_MUST_CREATE);
346 inode->bi_dir = lostfound.bi_inum;
347 inode->bi_dir_offset = dir_offset;
349 return __write_inode(trans, inode, inode_snapshot);
352 static int reattach_inode(struct btree_trans *trans,
353 struct bch_inode_unpacked *inode,
356 int ret = commit_do(trans, NULL, NULL,
357 BTREE_INSERT_LAZY_RW|
359 __reattach_inode(trans, inode, inode_snapshot));
360 bch_err_msg(trans->c, ret, "reattaching inode %llu", inode->bi_inum);
364 static int remove_backpointer(struct btree_trans *trans,
365 struct bch_inode_unpacked *inode)
367 struct btree_iter iter;
368 struct bkey_s_c_dirent d;
371 d = bch2_bkey_get_iter_typed(trans, &iter, BTREE_ID_dirents,
372 POS(inode->bi_dir, inode->bi_dir_offset), 0,
375 __remove_dirent(trans, d.k->p);
376 bch2_trans_iter_exit(trans, &iter);
380 struct snapshots_seen_entry {
385 struct snapshots_seen {
387 DARRAY(struct snapshots_seen_entry) ids;
390 static inline void snapshots_seen_exit(struct snapshots_seen *s)
392 darray_exit(&s->ids);
395 static inline void snapshots_seen_init(struct snapshots_seen *s)
397 memset(s, 0, sizeof(*s));
400 static int snapshots_seen_add_inorder(struct bch_fs *c, struct snapshots_seen *s, u32 id)
402 struct snapshots_seen_entry *i, n = {
404 .equiv = bch2_snapshot_equiv(c, id),
408 darray_for_each(s->ids, i) {
415 ret = darray_insert_item(&s->ids, i - s->ids.data, n);
417 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
422 static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s,
423 enum btree_id btree_id, struct bpos pos)
425 struct snapshots_seen_entry *i, n = {
427 .equiv = bch2_snapshot_equiv(c, pos.snapshot),
431 if (!bkey_eq(s->pos, pos))
435 s->pos.snapshot = n.equiv;
437 darray_for_each(s->ids, i) {
442 * We currently don't rigorously track for snapshot cleanup
443 * needing to be run, so it shouldn't be a fsck error yet:
445 if (i->equiv == n.equiv) {
446 bch_err(c, "snapshot deletion did not finish:\n"
447 " duplicate keys in btree %s at %llu:%llu snapshots %u, %u (equiv %u)\n",
448 bch2_btree_id_str(btree_id),
449 pos.inode, pos.offset,
450 i->id, n.id, n.equiv);
451 set_bit(BCH_FS_NEED_DELETE_DEAD_SNAPSHOTS, &c->flags);
452 return bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_delete_dead_snapshots);
456 ret = darray_push(&s->ids, n);
458 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
464 * key_visible_in_snapshot - returns true if @id is a descendent of @ancestor,
465 * and @ancestor hasn't been overwritten in @seen
467 * @c: filesystem handle
468 * @seen: list of snapshot ids already seen at current position
469 * @id: descendent snapshot id
470 * @ancestor: ancestor snapshot id
472 * Returns: whether key in @ancestor snapshot is visible in @id snapshot
474 static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *seen,
475 u32 id, u32 ancestor)
479 EBUG_ON(id > ancestor);
480 EBUG_ON(!bch2_snapshot_is_equiv(c, id));
481 EBUG_ON(!bch2_snapshot_is_equiv(c, ancestor));
483 /* @ancestor should be the snapshot most recently added to @seen */
484 EBUG_ON(ancestor != seen->pos.snapshot);
485 EBUG_ON(ancestor != seen->ids.data[seen->ids.nr - 1].equiv);
490 if (!bch2_snapshot_is_ancestor(c, id, ancestor))
494 * We know that @id is a descendant of @ancestor, we're checking if
495 * we've seen a key that overwrote @ancestor - i.e. also a descendent of
496 * @ascestor and with @id as a descendent.
498 * But we already know that we're scanning IDs between @id and @ancestor
499 * numerically, since snapshot ID lists are kept sorted, so if we find
500 * an id that's an ancestor of @id we're done:
503 for (i = seen->ids.nr - 2;
504 i >= 0 && seen->ids.data[i].equiv >= id;
506 if (bch2_snapshot_is_ancestor(c, id, seen->ids.data[i].equiv))
513 * ref_visible - given a key with snapshot id @src that points to a key with
514 * snapshot id @dst, test whether there is some snapshot in which @dst is
517 * @c: filesystem handle
518 * @s: list of snapshot IDs already seen at @src
519 * @src: snapshot ID of src key
520 * @dst: snapshot ID of dst key
521 * Returns: true if there is some snapshot in which @dst is visible
523 * Assumes we're visiting @src keys in natural key order
525 static bool ref_visible(struct bch_fs *c, struct snapshots_seen *s,
529 ? key_visible_in_snapshot(c, s, dst, src)
530 : bch2_snapshot_is_ancestor(c, src, dst);
533 static int ref_visible2(struct bch_fs *c,
534 u32 src, struct snapshots_seen *src_seen,
535 u32 dst, struct snapshots_seen *dst_seen)
537 src = bch2_snapshot_equiv(c, src);
538 dst = bch2_snapshot_equiv(c, dst);
542 swap(dst_seen, src_seen);
544 return key_visible_in_snapshot(c, src_seen, dst, src);
547 #define for_each_visible_inode(_c, _s, _w, _snapshot, _i) \
548 for (_i = (_w)->inodes.data; _i < (_w)->inodes.data + (_w)->inodes.nr && \
549 (_i)->snapshot <= (_snapshot); _i++) \
550 if (key_visible_in_snapshot(_c, _s, _i->snapshot, _snapshot))
552 struct inode_walker_entry {
553 struct bch_inode_unpacked inode;
559 struct inode_walker {
560 bool first_this_inode;
561 bool recalculate_sums;
562 struct bpos last_pos;
564 DARRAY(struct inode_walker_entry) inodes;
567 static void inode_walker_exit(struct inode_walker *w)
569 darray_exit(&w->inodes);
572 static struct inode_walker inode_walker_init(void)
574 return (struct inode_walker) { 0, };
577 static int add_inode(struct bch_fs *c, struct inode_walker *w,
578 struct bkey_s_c inode)
580 struct bch_inode_unpacked u;
582 BUG_ON(bch2_inode_unpack(inode, &u));
584 return darray_push(&w->inodes, ((struct inode_walker_entry) {
586 .snapshot = bch2_snapshot_equiv(c, inode.k->p.snapshot),
590 static int get_inodes_all_snapshots(struct btree_trans *trans,
591 struct inode_walker *w, u64 inum)
593 struct bch_fs *c = trans->c;
594 struct btree_iter iter;
596 u32 restart_count = trans->restart_count;
599 w->recalculate_sums = false;
602 for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, inum),
603 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
604 if (k.k->p.offset != inum)
607 if (bkey_is_inode(k.k))
610 bch2_trans_iter_exit(trans, &iter);
615 w->first_this_inode = true;
617 return trans_was_restarted(trans, restart_count);
620 static struct inode_walker_entry *
621 lookup_inode_for_snapshot(struct bch_fs *c, struct inode_walker *w,
622 u32 snapshot, bool is_whiteout)
624 struct inode_walker_entry *i;
626 snapshot = bch2_snapshot_equiv(c, snapshot);
628 darray_for_each(w->inodes, i)
629 if (bch2_snapshot_is_ancestor(c, snapshot, i->snapshot))
634 BUG_ON(snapshot > i->snapshot);
636 if (snapshot != i->snapshot && !is_whiteout) {
637 struct inode_walker_entry new = *i;
641 new.snapshot = snapshot;
644 bch_info(c, "have key for inode %llu:%u but have inode in ancestor snapshot %u",
645 w->last_pos.inode, snapshot, i->snapshot);
647 while (i > w->inodes.data && i[-1].snapshot > snapshot)
650 pos = i - w->inodes.data;
651 ret = darray_insert_item(&w->inodes, pos, new);
655 i = w->inodes.data + pos;
661 static struct inode_walker_entry *walk_inode(struct btree_trans *trans,
662 struct inode_walker *w, struct bpos pos,
665 if (w->last_pos.inode != pos.inode) {
666 int ret = get_inodes_all_snapshots(trans, w, pos.inode);
669 } else if (bkey_cmp(w->last_pos, pos)) {
670 struct inode_walker_entry *i;
672 darray_for_each(w->inodes, i)
673 i->seen_this_pos = false;
679 return lookup_inode_for_snapshot(trans->c, w, pos.snapshot, is_whiteout);
682 static int __get_visible_inodes(struct btree_trans *trans,
683 struct inode_walker *w,
684 struct snapshots_seen *s,
687 struct bch_fs *c = trans->c;
688 struct btree_iter iter;
694 for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum),
695 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
696 u32 equiv = bch2_snapshot_equiv(c, k.k->p.snapshot);
698 if (k.k->p.offset != inum)
701 if (!ref_visible(c, s, s->pos.snapshot, equiv))
704 if (bkey_is_inode(k.k))
707 if (equiv >= s->pos.snapshot)
710 bch2_trans_iter_exit(trans, &iter);
715 static int check_key_has_snapshot(struct btree_trans *trans,
716 struct btree_iter *iter,
719 struct bch_fs *c = trans->c;
720 struct printbuf buf = PRINTBUF;
723 if (mustfix_fsck_err_on(!bch2_snapshot_equiv(c, k.k->p.snapshot), c,
724 "key in missing snapshot: %s",
725 (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
726 ret = bch2_btree_delete_at(trans, iter,
727 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?: 1;
733 static int hash_redo_key(struct btree_trans *trans,
734 const struct bch_hash_desc desc,
735 struct bch_hash_info *hash_info,
736 struct btree_iter *k_iter, struct bkey_s_c k)
738 struct bkey_i *delete;
741 delete = bch2_trans_kmalloc(trans, sizeof(*delete));
743 return PTR_ERR(delete);
745 tmp = bch2_bkey_make_mut_noupdate(trans, k);
749 bkey_init(&delete->k);
750 delete->k.p = k_iter->pos;
751 return bch2_btree_iter_traverse(k_iter) ?:
752 bch2_trans_update(trans, k_iter, delete, 0) ?:
753 bch2_hash_set_snapshot(trans, desc, hash_info,
754 (subvol_inum) { 0, k.k->p.inode },
755 k.k->p.snapshot, tmp,
756 BCH_HASH_SET_MUST_CREATE,
757 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
758 bch2_trans_commit(trans, NULL, NULL,
760 BTREE_INSERT_LAZY_RW);
763 static int hash_check_key(struct btree_trans *trans,
764 const struct bch_hash_desc desc,
765 struct bch_hash_info *hash_info,
766 struct btree_iter *k_iter, struct bkey_s_c hash_k)
768 struct bch_fs *c = trans->c;
769 struct btree_iter iter = { NULL };
770 struct printbuf buf = PRINTBUF;
775 if (hash_k.k->type != desc.key_type)
778 hash = desc.hash_bkey(hash_info, hash_k);
780 if (likely(hash == hash_k.k->p.offset))
783 if (hash_k.k->p.offset < hash)
786 for_each_btree_key_norestart(trans, iter, desc.btree_id,
787 SPOS(hash_k.k->p.inode, hash, hash_k.k->p.snapshot),
788 BTREE_ITER_SLOTS, k, ret) {
789 if (bkey_eq(k.k->p, hash_k.k->p))
792 if (fsck_err_on(k.k->type == desc.key_type &&
793 !desc.cmp_bkey(k, hash_k), c,
794 "duplicate hash table keys:\n%s",
795 (printbuf_reset(&buf),
796 bch2_bkey_val_to_text(&buf, c, hash_k),
798 ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0) ?: 1;
802 if (bkey_deleted(k.k)) {
803 bch2_trans_iter_exit(trans, &iter);
808 bch2_trans_iter_exit(trans, &iter);
812 if (fsck_err(c, "hash table key at wrong offset: btree %s inode %llu offset %llu, hashed to %llu\n%s",
813 bch2_btree_id_str(desc.btree_id), hash_k.k->p.inode, hash_k.k->p.offset, hash,
814 (printbuf_reset(&buf),
815 bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf))) {
816 ret = hash_redo_key(trans, desc, hash_info, k_iter, hash_k);
820 ret = -BCH_ERR_transaction_restart_nested;
826 static int check_inode(struct btree_trans *trans,
827 struct btree_iter *iter,
829 struct bch_inode_unpacked *prev,
830 struct snapshots_seen *s,
833 struct bch_fs *c = trans->c;
834 struct bch_inode_unpacked u;
835 bool do_update = false;
838 ret = check_key_has_snapshot(trans, iter, k);
844 ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
848 if (!bkey_is_inode(k.k))
851 BUG_ON(bch2_inode_unpack(k, &u));
854 !(u.bi_flags & (BCH_INODE_I_SIZE_DIRTY|
855 BCH_INODE_I_SECTORS_DIRTY|
856 BCH_INODE_UNLINKED)))
859 if (prev->bi_inum != u.bi_inum)
862 if (fsck_err_on(prev->bi_hash_seed != u.bi_hash_seed ||
863 inode_d_type(prev) != inode_d_type(&u), c,
864 "inodes in different snapshots don't match")) {
865 bch_err(c, "repair not implemented yet");
869 if ((u.bi_flags & (BCH_INODE_I_SIZE_DIRTY|BCH_INODE_UNLINKED)) &&
870 bch2_key_has_snapshot_overwrites(trans, BTREE_ID_inodes, k.k->p)) {
871 struct bpos new_min_pos;
873 ret = bch2_propagate_key_to_snapshot_leaves(trans, iter->btree_id, k, &new_min_pos);
877 u.bi_flags &= ~BCH_INODE_I_SIZE_DIRTY|BCH_INODE_UNLINKED;
879 ret = __write_inode(trans, &u, iter->pos.snapshot);
880 bch_err_msg(c, ret, "in fsck updating inode");
884 if (!bpos_eq(new_min_pos, POS_MIN))
885 bch2_btree_iter_set_pos(iter, bpos_predecessor(new_min_pos));
889 if (u.bi_flags & BCH_INODE_UNLINKED &&
891 fsck_err(c, "filesystem marked clean, but inode %llu unlinked",
893 bch2_trans_unlock(trans);
896 ret = bch2_inode_rm_snapshot(trans, u.bi_inum, iter->pos.snapshot);
897 bch_err_msg(c, ret, "in fsck deleting inode");
901 if (u.bi_flags & BCH_INODE_I_SIZE_DIRTY &&
903 fsck_err(c, "filesystem marked clean, but inode %llu has i_size dirty",
905 bch_verbose(c, "truncating inode %llu", u.bi_inum);
907 bch2_trans_unlock(trans);
911 * XXX: need to truncate partial blocks too here - or ideally
912 * just switch units to bytes and that issue goes away
914 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
915 SPOS(u.bi_inum, round_up(u.bi_size, block_bytes(c)) >> 9,
917 POS(u.bi_inum, U64_MAX),
919 bch_err_msg(c, ret, "in fsck truncating inode");
924 * We truncated without our normal sector accounting hook, just
925 * make sure we recalculate it:
927 u.bi_flags |= BCH_INODE_I_SECTORS_DIRTY;
929 u.bi_flags &= ~BCH_INODE_I_SIZE_DIRTY;
933 if (u.bi_flags & BCH_INODE_I_SECTORS_DIRTY &&
935 fsck_err(c, "filesystem marked clean, but inode %llu has i_sectors dirty",
939 bch_verbose(c, "recounting sectors for inode %llu",
942 sectors = bch2_count_inode_sectors(trans, u.bi_inum, iter->pos.snapshot);
944 bch_err_msg(c, sectors, "in fsck recounting inode sectors");
948 u.bi_sectors = sectors;
949 u.bi_flags &= ~BCH_INODE_I_SECTORS_DIRTY;
953 if (u.bi_flags & BCH_INODE_BACKPTR_UNTRUSTED) {
956 u.bi_flags &= ~BCH_INODE_BACKPTR_UNTRUSTED;
961 ret = __write_inode(trans, &u, iter->pos.snapshot);
962 bch_err_msg(c, ret, "in fsck updating inode");
973 int bch2_check_inodes(struct bch_fs *c)
975 bool full = c->opts.fsck;
976 struct btree_trans *trans = bch2_trans_get(c);
977 struct btree_iter iter;
978 struct bch_inode_unpacked prev = { 0 };
979 struct snapshots_seen s;
983 snapshots_seen_init(&s);
985 ret = for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
987 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
988 NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
989 check_inode(trans, &iter, k, &prev, &s, full));
991 snapshots_seen_exit(&s);
992 bch2_trans_put(trans);
997 static struct bkey_s_c_dirent dirent_get_by_pos(struct btree_trans *trans,
998 struct btree_iter *iter,
1001 return bch2_bkey_get_iter_typed(trans, iter, BTREE_ID_dirents, pos, 0, dirent);
1004 static bool inode_points_to_dirent(struct bch_inode_unpacked *inode,
1005 struct bkey_s_c_dirent d)
1007 return inode->bi_dir == d.k->p.inode &&
1008 inode->bi_dir_offset == d.k->p.offset;
1011 static bool dirent_points_to_inode(struct bkey_s_c_dirent d,
1012 struct bch_inode_unpacked *inode)
1014 return d.v->d_type == DT_SUBVOL
1015 ? le32_to_cpu(d.v->d_child_subvol) == inode->bi_subvol
1016 : le64_to_cpu(d.v->d_inum) == inode->bi_inum;
1019 static int inode_backpointer_exists(struct btree_trans *trans,
1020 struct bch_inode_unpacked *inode,
1023 struct btree_iter iter;
1024 struct bkey_s_c_dirent d;
1027 d = dirent_get_by_pos(trans, &iter,
1028 SPOS(inode->bi_dir, inode->bi_dir_offset, snapshot));
1031 return bch2_err_matches(ret, ENOENT) ? 0 : ret;
1033 ret = dirent_points_to_inode(d, inode);
1034 bch2_trans_iter_exit(trans, &iter);
1038 static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
1040 struct bch_fs *c = trans->c;
1041 struct inode_walker_entry *i;
1042 u32 restart_count = trans->restart_count;
1046 darray_for_each(w->inodes, i) {
1047 if (i->inode.bi_sectors == i->count)
1050 count2 = bch2_count_inode_sectors(trans, w->last_pos.inode, i->snapshot);
1052 if (w->recalculate_sums)
1055 if (i->count != count2) {
1056 bch_err(c, "fsck counted i_sectors wrong for inode %llu:%u: got %llu should be %llu",
1057 w->last_pos.inode, i->snapshot, i->count, count2);
1058 return -BCH_ERR_internal_fsck_err;
1061 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SECTORS_DIRTY), c,
1062 "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
1063 w->last_pos.inode, i->snapshot,
1064 i->inode.bi_sectors, i->count)) {
1065 i->inode.bi_sectors = i->count;
1066 ret = fsck_write_inode(trans, &i->inode, i->snapshot);
1073 return ret ?: trans_was_restarted(trans, restart_count);
1079 struct snapshots_seen seen;
1082 struct extent_ends {
1083 struct bpos last_pos;
1084 DARRAY(struct extent_end) e;
1087 static void extent_ends_reset(struct extent_ends *extent_ends)
1089 struct extent_end *i;
1091 darray_for_each(extent_ends->e, i)
1092 snapshots_seen_exit(&i->seen);
1094 extent_ends->e.nr = 0;
1097 static void extent_ends_exit(struct extent_ends *extent_ends)
1099 extent_ends_reset(extent_ends);
1100 darray_exit(&extent_ends->e);
1103 static void extent_ends_init(struct extent_ends *extent_ends)
1105 memset(extent_ends, 0, sizeof(*extent_ends));
1108 static int extent_ends_at(struct bch_fs *c,
1109 struct extent_ends *extent_ends,
1110 struct snapshots_seen *seen,
1113 struct extent_end *i, n = (struct extent_end) {
1114 .offset = k.k->p.offset,
1115 .snapshot = k.k->p.snapshot,
1119 n.seen.ids.data = kmemdup(seen->ids.data,
1120 sizeof(seen->ids.data[0]) * seen->ids.size,
1122 if (!n.seen.ids.data)
1123 return -BCH_ERR_ENOMEM_fsck_extent_ends_at;
1125 darray_for_each(extent_ends->e, i) {
1126 if (i->snapshot == k.k->p.snapshot) {
1127 snapshots_seen_exit(&i->seen);
1132 if (i->snapshot >= k.k->p.snapshot)
1136 return darray_insert_item(&extent_ends->e, i - extent_ends->e.data, n);
1139 static int overlapping_extents_found(struct btree_trans *trans,
1140 enum btree_id btree,
1141 struct bpos pos1, struct snapshots_seen *pos1_seen,
1144 struct extent_end *extent_end)
1146 struct bch_fs *c = trans->c;
1147 struct printbuf buf = PRINTBUF;
1148 struct btree_iter iter1, iter2 = { NULL };
1149 struct bkey_s_c k1, k2;
1152 BUG_ON(bkey_le(pos1, bkey_start_pos(&pos2)));
1154 bch2_trans_iter_init(trans, &iter1, btree, pos1,
1155 BTREE_ITER_ALL_SNAPSHOTS|
1156 BTREE_ITER_NOT_EXTENTS);
1157 k1 = bch2_btree_iter_peek_upto(&iter1, POS(pos1.inode, U64_MAX));
1162 prt_str(&buf, "\n ");
1163 bch2_bkey_val_to_text(&buf, c, k1);
1165 if (!bpos_eq(pos1, k1.k->p)) {
1166 prt_str(&buf, "\n wanted\n ");
1167 bch2_bpos_to_text(&buf, pos1);
1168 prt_str(&buf, "\n ");
1169 bch2_bkey_to_text(&buf, &pos2);
1171 bch_err(c, "%s: error finding first overlapping extent when repairing, got%s",
1173 ret = -BCH_ERR_internal_fsck_err;
1177 bch2_trans_copy_iter(&iter2, &iter1);
1180 bch2_btree_iter_advance(&iter2);
1182 k2 = bch2_btree_iter_peek_upto(&iter2, POS(pos1.inode, U64_MAX));
1187 if (bpos_ge(k2.k->p, pos2.p))
1191 prt_str(&buf, "\n ");
1192 bch2_bkey_val_to_text(&buf, c, k2);
1194 if (bpos_gt(k2.k->p, pos2.p) ||
1195 pos2.size != k2.k->size) {
1196 bch_err(c, "%s: error finding seconding overlapping extent when repairing%s",
1198 ret = -BCH_ERR_internal_fsck_err;
1202 prt_printf(&buf, "\n overwriting %s extent",
1203 pos1.snapshot >= pos2.p.snapshot ? "first" : "second");
1205 if (fsck_err(c, "overlapping extents%s", buf.buf)) {
1206 struct btree_iter *old_iter = &iter1;
1207 struct disk_reservation res = { 0 };
1209 if (pos1.snapshot < pos2.p.snapshot) {
1214 trans->extra_journal_res += bch2_bkey_sectors_compressed(k2);
1216 ret = bch2_trans_update_extent_overwrite(trans, old_iter,
1217 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE,
1219 bch2_trans_commit(trans, &res, NULL,
1220 BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL);
1221 bch2_disk_reservation_put(c, &res);
1228 if (pos1.snapshot == pos2.p.snapshot) {
1230 * We overwrote the first extent, and did the overwrite
1231 * in the same snapshot:
1233 extent_end->offset = bkey_start_offset(&pos2);
1234 } else if (pos1.snapshot > pos2.p.snapshot) {
1236 * We overwrote the first extent in pos2's snapshot:
1238 ret = snapshots_seen_add_inorder(c, pos1_seen, pos2.p.snapshot);
1241 * We overwrote the second extent - restart
1242 * check_extent() from the top:
1244 ret = -BCH_ERR_transaction_restart_nested;
1249 bch2_trans_iter_exit(trans, &iter2);
1250 bch2_trans_iter_exit(trans, &iter1);
1251 printbuf_exit(&buf);
1255 static int check_overlapping_extents(struct btree_trans *trans,
1256 struct snapshots_seen *seen,
1257 struct extent_ends *extent_ends,
1260 struct btree_iter *iter,
1263 struct bch_fs *c = trans->c;
1264 struct extent_end *i;
1267 /* transaction restart, running again */
1268 if (bpos_eq(extent_ends->last_pos, k.k->p))
1271 if (extent_ends->last_pos.inode != k.k->p.inode)
1272 extent_ends_reset(extent_ends);
1274 darray_for_each(extent_ends->e, i) {
1275 if (i->offset <= bkey_start_offset(k.k))
1278 if (!ref_visible2(c,
1279 k.k->p.snapshot, seen,
1280 i->snapshot, &i->seen))
1283 ret = overlapping_extents_found(trans, iter->btree_id,
1284 SPOS(iter->pos.inode,
1293 ret = extent_ends_at(c, extent_ends, seen, k);
1297 extent_ends->last_pos = k.k->p;
1302 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
1304 struct inode_walker *inode,
1305 struct snapshots_seen *s,
1306 struct extent_ends *extent_ends)
1308 struct bch_fs *c = trans->c;
1309 struct inode_walker_entry *i;
1310 struct printbuf buf = PRINTBUF;
1311 struct bpos equiv = k.k->p;
1314 equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1316 ret = check_key_has_snapshot(trans, iter, k);
1318 ret = ret < 0 ? ret : 0;
1322 if (inode->last_pos.inode != k.k->p.inode) {
1323 ret = check_i_sectors(trans, inode);
1328 i = walk_inode(trans, inode, equiv, k.k->type == KEY_TYPE_whiteout);
1329 ret = PTR_ERR_OR_ZERO(i);
1333 ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1337 if (k.k->type != KEY_TYPE_whiteout) {
1338 if (fsck_err_on(!i, c,
1339 "extent in missing inode:\n %s",
1340 (printbuf_reset(&buf),
1341 bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1344 if (fsck_err_on(i &&
1345 !S_ISREG(i->inode.bi_mode) &&
1346 !S_ISLNK(i->inode.bi_mode), c,
1347 "extent in non regular inode mode %o:\n %s",
1349 (printbuf_reset(&buf),
1350 bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1353 ret = check_overlapping_extents(trans, s, extent_ends, k,
1354 equiv.snapshot, iter,
1355 &inode->recalculate_sums);
1361 * Check inodes in reverse order, from oldest snapshots to newest,
1362 * starting from the inode that matches this extent's snapshot. If we
1363 * didn't have one, iterate over all inodes:
1366 i = inode->inodes.data + inode->inodes.nr - 1;
1369 inode->inodes.data && i >= inode->inodes.data;
1371 if (i->snapshot > equiv.snapshot ||
1372 !key_visible_in_snapshot(c, s, i->snapshot, equiv.snapshot))
1375 if (k.k->type != KEY_TYPE_whiteout) {
1376 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SIZE_DIRTY) &&
1377 k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9 &&
1378 !bkey_extent_is_reservation(k), c,
1379 "extent type past end of inode %llu:%u, i_size %llu\n %s",
1380 i->inode.bi_inum, i->snapshot, i->inode.bi_size,
1381 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1382 struct btree_iter iter2;
1384 bch2_trans_copy_iter(&iter2, iter);
1385 bch2_btree_iter_set_snapshot(&iter2, i->snapshot);
1386 ret = bch2_btree_iter_traverse(&iter2) ?:
1387 bch2_btree_delete_at(trans, &iter2,
1388 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1389 bch2_trans_iter_exit(trans, &iter2);
1393 iter->k.type = KEY_TYPE_whiteout;
1396 if (bkey_extent_is_allocation(k.k))
1397 i->count += k.k->size;
1400 i->seen_this_pos = true;
1405 printbuf_exit(&buf);
1409 ret = bch2_btree_delete_at(trans, iter, BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1414 * Walk extents: verify that extents have a corresponding S_ISREG inode, and
1415 * that i_size an i_sectors are consistent
1417 int bch2_check_extents(struct bch_fs *c)
1419 struct inode_walker w = inode_walker_init();
1420 struct snapshots_seen s;
1421 struct btree_trans *trans = bch2_trans_get(c);
1422 struct btree_iter iter;
1424 struct extent_ends extent_ends;
1425 struct disk_reservation res = { 0 };
1428 snapshots_seen_init(&s);
1429 extent_ends_init(&extent_ends);
1431 ret = for_each_btree_key_commit(trans, iter, BTREE_ID_extents,
1432 POS(BCACHEFS_ROOT_INO, 0),
1433 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1435 BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL, ({
1436 bch2_disk_reservation_put(c, &res);
1437 check_extent(trans, &iter, k, &w, &s, &extent_ends);
1439 check_i_sectors(trans, &w);
1441 bch2_disk_reservation_put(c, &res);
1442 extent_ends_exit(&extent_ends);
1443 inode_walker_exit(&w);
1444 snapshots_seen_exit(&s);
1445 bch2_trans_put(trans);
1451 static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
1453 struct bch_fs *c = trans->c;
1454 struct inode_walker_entry *i;
1455 u32 restart_count = trans->restart_count;
1459 darray_for_each(w->inodes, i) {
1460 if (i->inode.bi_nlink == i->count)
1463 count2 = bch2_count_subdirs(trans, w->last_pos.inode, i->snapshot);
1467 if (i->count != count2) {
1468 bch_err(c, "fsck counted subdirectories wrong: got %llu should be %llu",
1471 if (i->inode.bi_nlink == i->count)
1475 if (fsck_err_on(i->inode.bi_nlink != i->count, c,
1476 "directory %llu:%u with wrong i_nlink: got %u, should be %llu",
1477 w->last_pos.inode, i->snapshot, i->inode.bi_nlink, i->count)) {
1478 i->inode.bi_nlink = i->count;
1479 ret = fsck_write_inode(trans, &i->inode, i->snapshot);
1486 return ret ?: trans_was_restarted(trans, restart_count);
1489 static int check_dirent_target(struct btree_trans *trans,
1490 struct btree_iter *iter,
1491 struct bkey_s_c_dirent d,
1492 struct bch_inode_unpacked *target,
1493 u32 target_snapshot)
1495 struct bch_fs *c = trans->c;
1496 struct bkey_i_dirent *n;
1497 bool backpointer_exists = true;
1498 struct printbuf buf = PRINTBUF;
1501 if (!target->bi_dir &&
1502 !target->bi_dir_offset) {
1503 target->bi_dir = d.k->p.inode;
1504 target->bi_dir_offset = d.k->p.offset;
1506 ret = __write_inode(trans, target, target_snapshot);
1511 if (!inode_points_to_dirent(target, d)) {
1512 ret = inode_backpointer_exists(trans, target, d.k->p.snapshot);
1516 backpointer_exists = ret;
1519 if (fsck_err_on(S_ISDIR(target->bi_mode) &&
1520 backpointer_exists, c,
1521 "directory %llu with multiple links",
1523 ret = __remove_dirent(trans, d.k->p);
1527 if (fsck_err_on(backpointer_exists &&
1528 !target->bi_nlink, c,
1529 "inode %llu type %s has multiple links but i_nlink 0",
1530 target->bi_inum, bch2_d_types[d.v->d_type])) {
1532 target->bi_flags &= ~BCH_INODE_UNLINKED;
1534 ret = __write_inode(trans, target, target_snapshot);
1539 if (fsck_err_on(!backpointer_exists, c,
1540 "inode %llu:%u has wrong backpointer:\n"
1542 "should be %llu:%llu",
1543 target->bi_inum, target_snapshot,
1545 target->bi_dir_offset,
1548 target->bi_dir = d.k->p.inode;
1549 target->bi_dir_offset = d.k->p.offset;
1551 ret = __write_inode(trans, target, target_snapshot);
1557 if (fsck_err_on(d.v->d_type != inode_d_type(target), c,
1558 "incorrect d_type: got %s, should be %s:\n%s",
1559 bch2_d_type_str(d.v->d_type),
1560 bch2_d_type_str(inode_d_type(target)),
1561 (printbuf_reset(&buf),
1562 bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
1563 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1564 ret = PTR_ERR_OR_ZERO(n);
1568 bkey_reassemble(&n->k_i, d.s_c);
1569 n->v.d_type = inode_d_type(target);
1571 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1575 d = dirent_i_to_s_c(n);
1578 if (d.v->d_type == DT_SUBVOL &&
1579 target->bi_parent_subvol != le32_to_cpu(d.v->d_parent_subvol) &&
1580 (c->sb.version < bcachefs_metadata_version_subvol_dirent ||
1581 fsck_err(c, "dirent has wrong d_parent_subvol field: got %u, should be %u",
1582 le32_to_cpu(d.v->d_parent_subvol),
1583 target->bi_parent_subvol))) {
1584 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1585 ret = PTR_ERR_OR_ZERO(n);
1589 bkey_reassemble(&n->k_i, d.s_c);
1590 n->v.d_parent_subvol = cpu_to_le32(target->bi_parent_subvol);
1592 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1596 d = dirent_i_to_s_c(n);
1601 printbuf_exit(&buf);
1606 static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
1608 struct bch_hash_info *hash_info,
1609 struct inode_walker *dir,
1610 struct inode_walker *target,
1611 struct snapshots_seen *s)
1613 struct bch_fs *c = trans->c;
1614 struct bkey_s_c_dirent d;
1615 struct inode_walker_entry *i;
1616 struct printbuf buf = PRINTBUF;
1620 ret = check_key_has_snapshot(trans, iter, k);
1622 ret = ret < 0 ? ret : 0;
1627 equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1629 ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1633 if (k.k->type == KEY_TYPE_whiteout)
1636 if (dir->last_pos.inode != k.k->p.inode) {
1637 ret = check_subdir_count(trans, dir);
1642 BUG_ON(!iter->path->should_be_locked);
1644 i = walk_inode(trans, dir, equiv, k.k->type == KEY_TYPE_whiteout);
1645 ret = PTR_ERR_OR_ZERO(i);
1649 if (dir->first_this_inode && dir->inodes.nr)
1650 *hash_info = bch2_hash_info_init(c, &dir->inodes.data[0].inode);
1651 dir->first_this_inode = false;
1653 if (fsck_err_on(!i, c,
1654 "dirent in nonexisting directory:\n%s",
1655 (printbuf_reset(&buf),
1656 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1657 ret = bch2_btree_delete_at(trans, iter,
1658 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1665 if (fsck_err_on(!S_ISDIR(i->inode.bi_mode), c,
1666 "dirent in non directory inode type %s:\n%s",
1667 bch2_d_type_str(inode_d_type(&i->inode)),
1668 (printbuf_reset(&buf),
1669 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1670 ret = bch2_btree_delete_at(trans, iter, 0);
1674 ret = hash_check_key(trans, bch2_dirent_hash_desc, hash_info, iter, k);
1678 /* dirent has been deleted */
1683 if (k.k->type != KEY_TYPE_dirent)
1686 d = bkey_s_c_to_dirent(k);
1688 if (d.v->d_type == DT_SUBVOL) {
1689 struct bch_inode_unpacked subvol_root;
1690 u32 target_subvol = le32_to_cpu(d.v->d_child_subvol);
1691 u32 target_snapshot;
1694 ret = __subvol_lookup(trans, target_subvol,
1695 &target_snapshot, &target_inum);
1696 if (ret && !bch2_err_matches(ret, ENOENT))
1699 if (fsck_err_on(ret, c,
1700 "dirent points to missing subvolume %u",
1701 le32_to_cpu(d.v->d_child_subvol))) {
1702 ret = __remove_dirent(trans, d.k->p);
1706 ret = __lookup_inode(trans, target_inum,
1707 &subvol_root, &target_snapshot);
1708 if (ret && !bch2_err_matches(ret, ENOENT))
1711 if (fsck_err_on(ret, c,
1712 "subvolume %u points to missing subvolume root %llu",
1715 bch_err(c, "repair not implemented yet");
1720 if (fsck_err_on(subvol_root.bi_subvol != target_subvol, c,
1721 "subvol root %llu has wrong bi_subvol field: got %u, should be %u",
1723 subvol_root.bi_subvol, target_subvol)) {
1724 subvol_root.bi_subvol = target_subvol;
1725 ret = __write_inode(trans, &subvol_root, target_snapshot);
1730 ret = check_dirent_target(trans, iter, d, &subvol_root,
1735 ret = __get_visible_inodes(trans, target, s, le64_to_cpu(d.v->d_inum));
1739 if (fsck_err_on(!target->inodes.nr, c,
1740 "dirent points to missing inode: (equiv %u)\n%s",
1742 (printbuf_reset(&buf),
1743 bch2_bkey_val_to_text(&buf, c, k),
1745 ret = __remove_dirent(trans, d.k->p);
1750 darray_for_each(target->inodes, i) {
1751 ret = check_dirent_target(trans, iter, d,
1752 &i->inode, i->snapshot);
1758 if (d.v->d_type == DT_DIR)
1759 for_each_visible_inode(c, s, dir, equiv.snapshot, i)
1765 printbuf_exit(&buf);
1771 * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
1774 int bch2_check_dirents(struct bch_fs *c)
1776 struct inode_walker dir = inode_walker_init();
1777 struct inode_walker target = inode_walker_init();
1778 struct snapshots_seen s;
1779 struct bch_hash_info hash_info;
1780 struct btree_trans *trans = bch2_trans_get(c);
1781 struct btree_iter iter;
1785 snapshots_seen_init(&s);
1787 ret = for_each_btree_key_commit(trans, iter, BTREE_ID_dirents,
1788 POS(BCACHEFS_ROOT_INO, 0),
1789 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
1792 BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
1793 check_dirent(trans, &iter, k, &hash_info, &dir, &target, &s));
1795 bch2_trans_put(trans);
1796 snapshots_seen_exit(&s);
1797 inode_walker_exit(&dir);
1798 inode_walker_exit(&target);
1803 static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
1805 struct bch_hash_info *hash_info,
1806 struct inode_walker *inode)
1808 struct bch_fs *c = trans->c;
1809 struct inode_walker_entry *i;
1812 ret = check_key_has_snapshot(trans, iter, k);
1816 i = walk_inode(trans, inode, k.k->p, k.k->type == KEY_TYPE_whiteout);
1817 ret = PTR_ERR_OR_ZERO(i);
1821 if (inode->first_this_inode && inode->inodes.nr)
1822 *hash_info = bch2_hash_info_init(c, &inode->inodes.data[0].inode);
1823 inode->first_this_inode = false;
1825 if (fsck_err_on(!i, c,
1826 "xattr for missing inode %llu",
1828 return bch2_btree_delete_at(trans, iter, 0);
1833 ret = hash_check_key(trans, bch2_xattr_hash_desc, hash_info, iter, k);
1840 * Walk xattrs: verify that they all have a corresponding inode
1842 int bch2_check_xattrs(struct bch_fs *c)
1844 struct inode_walker inode = inode_walker_init();
1845 struct bch_hash_info hash_info;
1846 struct btree_iter iter;
1850 ret = bch2_trans_run(c,
1851 for_each_btree_key_commit(trans, iter, BTREE_ID_xattrs,
1852 POS(BCACHEFS_ROOT_INO, 0),
1853 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
1856 BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
1857 check_xattr(trans, &iter, k, &hash_info, &inode)));
1862 static int check_root_trans(struct btree_trans *trans)
1864 struct bch_fs *c = trans->c;
1865 struct bch_inode_unpacked root_inode;
1870 ret = __subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
1871 if (ret && !bch2_err_matches(ret, ENOENT))
1874 if (mustfix_fsck_err_on(ret, c, "root subvol missing")) {
1875 struct bkey_i_subvolume root_subvol;
1878 inum = BCACHEFS_ROOT_INO;
1880 bkey_subvolume_init(&root_subvol.k_i);
1881 root_subvol.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1882 root_subvol.v.flags = 0;
1883 root_subvol.v.snapshot = cpu_to_le32(snapshot);
1884 root_subvol.v.inode = cpu_to_le64(inum);
1885 ret = commit_do(trans, NULL, NULL,
1886 BTREE_INSERT_NOFAIL|
1887 BTREE_INSERT_LAZY_RW,
1888 bch2_btree_insert_trans(trans, BTREE_ID_subvolumes,
1889 &root_subvol.k_i, 0));
1890 bch_err_msg(c, ret, "writing root subvol");
1896 ret = __lookup_inode(trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
1897 if (ret && !bch2_err_matches(ret, ENOENT))
1900 if (mustfix_fsck_err_on(ret, c, "root directory missing") ||
1901 mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode), c,
1902 "root inode not a directory")) {
1903 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755,
1905 root_inode.bi_inum = inum;
1907 ret = __write_inode(trans, &root_inode, snapshot);
1908 bch_err_msg(c, ret, "writing root inode");
1915 /* Get root directory, create if it doesn't exist: */
1916 int bch2_check_root(struct bch_fs *c)
1920 ret = bch2_trans_do(c, NULL, NULL,
1921 BTREE_INSERT_NOFAIL|
1922 BTREE_INSERT_LAZY_RW,
1923 check_root_trans(trans));
1928 struct pathbuf_entry {
1933 typedef DARRAY(struct pathbuf_entry) pathbuf;
1935 static bool path_is_dup(pathbuf *p, u64 inum, u32 snapshot)
1937 struct pathbuf_entry *i;
1939 darray_for_each(*p, i)
1940 if (i->inum == inum &&
1941 i->snapshot == snapshot)
1947 static int path_down(struct bch_fs *c, pathbuf *p,
1948 u64 inum, u32 snapshot)
1950 int ret = darray_push(p, ((struct pathbuf_entry) {
1952 .snapshot = snapshot,
1956 bch_err(c, "fsck: error allocating memory for pathbuf, size %zu",
1962 * Check that a given inode is reachable from the root:
1964 * XXX: we should also be verifying that inodes are in the right subvolumes
1966 static int check_path(struct btree_trans *trans,
1968 struct bch_inode_unpacked *inode,
1971 struct bch_fs *c = trans->c;
1974 snapshot = bch2_snapshot_equiv(c, snapshot);
1977 while (!(inode->bi_inum == BCACHEFS_ROOT_INO &&
1978 inode->bi_subvol == BCACHEFS_ROOT_SUBVOL)) {
1979 struct btree_iter dirent_iter;
1980 struct bkey_s_c_dirent d;
1981 u32 parent_snapshot = snapshot;
1983 if (inode->bi_subvol) {
1986 ret = subvol_lookup(trans, inode->bi_parent_subvol,
1987 &parent_snapshot, &inum);
1992 ret = lockrestart_do(trans,
1993 PTR_ERR_OR_ZERO((d = dirent_get_by_pos(trans, &dirent_iter,
1994 SPOS(inode->bi_dir, inode->bi_dir_offset,
1995 parent_snapshot))).k));
1996 if (ret && !bch2_err_matches(ret, ENOENT))
1999 if (!ret && !dirent_points_to_inode(d, inode)) {
2000 bch2_trans_iter_exit(trans, &dirent_iter);
2001 ret = -BCH_ERR_ENOENT_dirent_doesnt_match_inode;
2004 if (bch2_err_matches(ret, ENOENT)) {
2005 if (fsck_err(c, "unreachable inode %llu:%u, type %s nlink %u backptr %llu:%llu",
2006 inode->bi_inum, snapshot,
2007 bch2_d_type_str(inode_d_type(inode)),
2010 inode->bi_dir_offset))
2011 ret = reattach_inode(trans, inode, snapshot);
2015 bch2_trans_iter_exit(trans, &dirent_iter);
2017 if (!S_ISDIR(inode->bi_mode))
2020 ret = path_down(c, p, inode->bi_inum, snapshot);
2022 bch_err(c, "memory allocation failure");
2026 snapshot = parent_snapshot;
2028 ret = lookup_inode(trans, inode->bi_dir, inode, &snapshot);
2030 /* Should have been caught in dirents pass */
2031 bch_err(c, "error looking up parent directory: %i", ret);
2035 if (path_is_dup(p, inode->bi_inum, snapshot)) {
2036 struct pathbuf_entry *i;
2038 /* XXX print path */
2039 bch_err(c, "directory structure loop");
2041 darray_for_each(*p, i)
2042 pr_err("%llu:%u", i->inum, i->snapshot);
2043 pr_err("%llu:%u", inode->bi_inum, snapshot);
2045 if (!fsck_err(c, "directory structure loop"))
2048 ret = commit_do(trans, NULL, NULL,
2049 BTREE_INSERT_NOFAIL|
2050 BTREE_INSERT_LAZY_RW,
2051 remove_backpointer(trans, inode));
2053 bch_err(c, "error removing dirent: %i", ret);
2057 ret = reattach_inode(trans, inode, snapshot);
2066 * Check for unreachable inodes, as well as loops in the directory structure:
2067 * After bch2_check_dirents(), if an inode backpointer doesn't exist that means it's
2070 int bch2_check_directory_structure(struct bch_fs *c)
2072 struct btree_trans *trans = bch2_trans_get(c);
2073 struct btree_iter iter;
2075 struct bch_inode_unpacked u;
2076 pathbuf path = { 0, };
2079 for_each_btree_key(trans, iter, BTREE_ID_inodes, POS_MIN,
2081 BTREE_ITER_PREFETCH|
2082 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2083 if (!bkey_is_inode(k.k))
2086 ret = bch2_inode_unpack(k, &u);
2088 /* Should have been caught earlier in fsck: */
2089 bch_err(c, "error unpacking inode %llu: %i", k.k->p.offset, ret);
2093 if (u.bi_flags & BCH_INODE_UNLINKED)
2096 ret = check_path(trans, &path, &u, iter.pos.snapshot);
2100 bch2_trans_iter_exit(trans, &iter);
2101 bch2_trans_put(trans);
2107 struct nlink_table {
2118 static int add_nlink(struct bch_fs *c, struct nlink_table *t,
2119 u64 inum, u32 snapshot)
2121 if (t->nr == t->size) {
2122 size_t new_size = max_t(size_t, 128UL, t->size * 2);
2123 void *d = kvmalloc_array(new_size, sizeof(t->d[0]), GFP_KERNEL);
2126 bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
2128 return -BCH_ERR_ENOMEM_fsck_add_nlink;
2132 memcpy(d, t->d, t->size * sizeof(t->d[0]));
2140 t->d[t->nr++] = (struct nlink) {
2142 .snapshot = snapshot,
2148 static int nlink_cmp(const void *_l, const void *_r)
2150 const struct nlink *l = _l;
2151 const struct nlink *r = _r;
2153 return cmp_int(l->inum, r->inum) ?: cmp_int(l->snapshot, r->snapshot);
2156 static void inc_link(struct bch_fs *c, struct snapshots_seen *s,
2157 struct nlink_table *links,
2158 u64 range_start, u64 range_end, u64 inum, u32 snapshot)
2160 struct nlink *link, key = {
2161 .inum = inum, .snapshot = U32_MAX,
2164 if (inum < range_start || inum >= range_end)
2167 link = __inline_bsearch(&key, links->d, links->nr,
2168 sizeof(links->d[0]), nlink_cmp);
2172 while (link > links->d && link[0].inum == link[-1].inum)
2175 for (; link < links->d + links->nr && link->inum == inum; link++)
2176 if (ref_visible(c, s, snapshot, link->snapshot)) {
2178 if (link->snapshot >= snapshot)
2184 static int check_nlinks_find_hardlinks(struct bch_fs *c,
2185 struct nlink_table *t,
2186 u64 start, u64 *end)
2188 struct btree_trans *trans = bch2_trans_get(c);
2189 struct btree_iter iter;
2191 struct bch_inode_unpacked u;
2194 for_each_btree_key(trans, iter, BTREE_ID_inodes,
2197 BTREE_ITER_PREFETCH|
2198 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2199 if (!bkey_is_inode(k.k))
2202 /* Should never fail, checked by bch2_inode_invalid: */
2203 BUG_ON(bch2_inode_unpack(k, &u));
2206 * Backpointer and directory structure checks are sufficient for
2207 * directories, since they can't have hardlinks:
2209 if (S_ISDIR(u.bi_mode))
2215 ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot);
2217 *end = k.k->p.offset;
2223 bch2_trans_iter_exit(trans, &iter);
2224 bch2_trans_put(trans);
2227 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
2233 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
2234 u64 range_start, u64 range_end)
2236 struct btree_trans *trans = bch2_trans_get(c);
2237 struct snapshots_seen s;
2238 struct btree_iter iter;
2240 struct bkey_s_c_dirent d;
2243 snapshots_seen_init(&s);
2245 for_each_btree_key(trans, iter, BTREE_ID_dirents, POS_MIN,
2247 BTREE_ITER_PREFETCH|
2248 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2249 ret = snapshots_seen_update(c, &s, iter.btree_id, k.k->p);
2253 switch (k.k->type) {
2254 case KEY_TYPE_dirent:
2255 d = bkey_s_c_to_dirent(k);
2257 if (d.v->d_type != DT_DIR &&
2258 d.v->d_type != DT_SUBVOL)
2259 inc_link(c, &s, links, range_start, range_end,
2260 le64_to_cpu(d.v->d_inum),
2261 bch2_snapshot_equiv(c, d.k->p.snapshot));
2265 bch2_trans_iter_exit(trans, &iter);
2268 bch_err(c, "error in fsck: btree error %i while walking dirents", ret);
2270 bch2_trans_put(trans);
2271 snapshots_seen_exit(&s);
2275 static int check_nlinks_update_inode(struct btree_trans *trans, struct btree_iter *iter,
2277 struct nlink_table *links,
2278 size_t *idx, u64 range_end)
2280 struct bch_fs *c = trans->c;
2281 struct bch_inode_unpacked u;
2282 struct nlink *link = &links->d[*idx];
2285 if (k.k->p.offset >= range_end)
2288 if (!bkey_is_inode(k.k))
2291 BUG_ON(bch2_inode_unpack(k, &u));
2293 if (S_ISDIR(u.bi_mode))
2299 while ((cmp_int(link->inum, k.k->p.offset) ?:
2300 cmp_int(link->snapshot, k.k->p.snapshot)) < 0) {
2301 BUG_ON(*idx == links->nr);
2302 link = &links->d[++*idx];
2305 if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count, c,
2306 "inode %llu type %s has wrong i_nlink (%u, should be %u)",
2307 u.bi_inum, bch2_d_types[mode_to_type(u.bi_mode)],
2308 bch2_inode_nlink_get(&u), link->count)) {
2309 bch2_inode_nlink_set(&u, link->count);
2310 ret = __write_inode(trans, &u, k.k->p.snapshot);
2317 static int check_nlinks_update_hardlinks(struct bch_fs *c,
2318 struct nlink_table *links,
2319 u64 range_start, u64 range_end)
2321 struct btree_iter iter;
2326 ret = bch2_trans_run(c,
2327 for_each_btree_key_commit(trans, iter, BTREE_ID_inodes,
2328 POS(0, range_start),
2329 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
2330 NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
2331 check_nlinks_update_inode(trans, &iter, k, links, &idx, range_end)));
2333 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
2340 int bch2_check_nlinks(struct bch_fs *c)
2342 struct nlink_table links = { 0 };
2343 u64 this_iter_range_start, next_iter_range_start = 0;
2347 this_iter_range_start = next_iter_range_start;
2348 next_iter_range_start = U64_MAX;
2350 ret = check_nlinks_find_hardlinks(c, &links,
2351 this_iter_range_start,
2352 &next_iter_range_start);
2354 ret = check_nlinks_walk_dirents(c, &links,
2355 this_iter_range_start,
2356 next_iter_range_start);
2360 ret = check_nlinks_update_hardlinks(c, &links,
2361 this_iter_range_start,
2362 next_iter_range_start);
2367 } while (next_iter_range_start != U64_MAX);
2374 static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter,
2377 struct bkey_s_c_reflink_p p;
2378 struct bkey_i_reflink_p *u;
2381 if (k.k->type != KEY_TYPE_reflink_p)
2384 p = bkey_s_c_to_reflink_p(k);
2386 if (!p.v->front_pad && !p.v->back_pad)
2389 u = bch2_trans_kmalloc(trans, sizeof(*u));
2390 ret = PTR_ERR_OR_ZERO(u);
2394 bkey_reassemble(&u->k_i, k);
2398 return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_NORUN);
2401 int bch2_fix_reflink_p(struct bch_fs *c)
2403 struct btree_iter iter;
2407 if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
2410 ret = bch2_trans_run(c,
2411 for_each_btree_key_commit(trans, iter,
2412 BTREE_ID_extents, POS_MIN,
2413 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|
2414 BTREE_ITER_ALL_SNAPSHOTS, k,
2415 NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
2416 fix_reflink_p_key(trans, &iter, k)));