1 // SPDX-License-Identifier: GPL-2.0
5 #include "btree_update.h"
13 #include "subvolume.h"
17 #include <linux/bsearch.h>
18 #include <linux/dcache.h> /* struct qstr */
20 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
23 * XXX: this is handling transaction restarts without returning
24 * -BCH_ERR_transaction_restart_nested, this is not how we do things anymore:
26 static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum,
29 struct btree_iter iter;
34 for_each_btree_key_upto(trans, iter, BTREE_ID_extents,
35 SPOS(inum, 0, snapshot),
38 if (bkey_extent_is_allocation(k.k))
41 bch2_trans_iter_exit(trans, &iter);
43 return ret ?: sectors;
46 static s64 bch2_count_subdirs(struct btree_trans *trans, u64 inum,
49 struct btree_iter iter;
51 struct bkey_s_c_dirent d;
55 for_each_btree_key_upto(trans, iter, BTREE_ID_dirents,
56 SPOS(inum, 0, snapshot),
59 if (k.k->type != KEY_TYPE_dirent)
62 d = bkey_s_c_to_dirent(k);
63 if (d.v->d_type == DT_DIR)
66 bch2_trans_iter_exit(trans, &iter);
68 return ret ?: subdirs;
71 static int __snapshot_lookup_subvol(struct btree_trans *trans, u32 snapshot,
74 struct btree_iter iter;
78 bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots,
80 k = bch2_btree_iter_peek_slot(&iter);
85 if (k.k->type != KEY_TYPE_snapshot) {
86 bch_err(trans->c, "snapshot %u not fonud", snapshot);
91 *subvol = le32_to_cpu(bkey_s_c_to_snapshot(k).v->subvol);
93 bch2_trans_iter_exit(trans, &iter);
98 static int __subvol_lookup(struct btree_trans *trans, u32 subvol,
99 u32 *snapshot, u64 *inum)
101 struct bch_subvolume s;
104 ret = bch2_subvolume_get(trans, subvol, false, 0, &s);
106 *snapshot = le32_to_cpu(s.snapshot);
107 *inum = le64_to_cpu(s.inode);
111 static int subvol_lookup(struct btree_trans *trans, u32 subvol,
112 u32 *snapshot, u64 *inum)
114 return lockrestart_do(trans, __subvol_lookup(trans, subvol, snapshot, inum));
117 static int lookup_first_inode(struct btree_trans *trans, u64 inode_nr,
118 struct bch_inode_unpacked *inode)
120 struct btree_iter iter;
124 bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
126 BTREE_ITER_ALL_SNAPSHOTS);
127 k = bch2_btree_iter_peek(&iter);
132 if (!k.k || !bkey_eq(k.k->p, POS(0, inode_nr))) {
137 ret = bch2_inode_unpack(k, inode);
139 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
140 bch_err(trans->c, "error fetching inode %llu: %s",
141 inode_nr, bch2_err_str(ret));
142 bch2_trans_iter_exit(trans, &iter);
146 static int __lookup_inode(struct btree_trans *trans, u64 inode_nr,
147 struct bch_inode_unpacked *inode,
150 struct btree_iter iter;
154 bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
155 SPOS(0, inode_nr, *snapshot), 0);
156 k = bch2_btree_iter_peek_slot(&iter);
161 ret = bkey_is_inode(k.k)
162 ? bch2_inode_unpack(k, inode)
165 *snapshot = iter.pos.snapshot;
167 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
168 bch_err(trans->c, "error fetching inode %llu:%u: %s",
169 inode_nr, *snapshot, bch2_err_str(ret));
170 bch2_trans_iter_exit(trans, &iter);
174 static int lookup_inode(struct btree_trans *trans, u64 inode_nr,
175 struct bch_inode_unpacked *inode,
178 return lockrestart_do(trans, __lookup_inode(trans, inode_nr, inode, snapshot));
181 static int __lookup_dirent(struct btree_trans *trans,
182 struct bch_hash_info hash_info,
183 subvol_inum dir, struct qstr *name,
184 u64 *target, unsigned *type)
186 struct btree_iter iter;
187 struct bkey_s_c_dirent d;
190 ret = bch2_hash_lookup(trans, &iter, bch2_dirent_hash_desc,
191 &hash_info, dir, name, 0);
195 d = bkey_s_c_to_dirent(bch2_btree_iter_peek_slot(&iter));
196 *target = le64_to_cpu(d.v->d_inum);
198 bch2_trans_iter_exit(trans, &iter);
202 static int __write_inode(struct btree_trans *trans,
203 struct bch_inode_unpacked *inode,
206 struct btree_iter iter;
209 bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
210 SPOS(0, inode->bi_inum, snapshot),
213 ret = bch2_btree_iter_traverse(&iter) ?:
214 bch2_inode_write(trans, &iter, inode);
215 bch2_trans_iter_exit(trans, &iter);
219 static int write_inode(struct btree_trans *trans,
220 struct bch_inode_unpacked *inode,
223 int ret = commit_do(trans, NULL, NULL,
225 BTREE_INSERT_LAZY_RW,
226 __write_inode(trans, inode, snapshot));
228 bch_err(trans->c, "error in fsck: error updating inode: %s",
233 static int fsck_inode_rm(struct btree_trans *trans, u64 inum, u32 snapshot)
235 struct bch_fs *c = trans->c;
236 struct btree_iter iter = { NULL };
237 struct bkey_i_inode_generation delete;
238 struct bch_inode_unpacked inode_u;
243 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
244 SPOS(inum, 0, snapshot),
245 SPOS(inum, U64_MAX, snapshot),
247 bch2_btree_delete_range_trans(trans, BTREE_ID_dirents,
248 SPOS(inum, 0, snapshot),
249 SPOS(inum, U64_MAX, snapshot),
251 bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs,
252 SPOS(inum, 0, snapshot),
253 SPOS(inum, U64_MAX, snapshot),
255 } while (ret == -BCH_ERR_transaction_restart_nested);
259 bch2_trans_begin(trans);
261 bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes,
262 SPOS(0, inum, snapshot), BTREE_ITER_INTENT);
263 k = bch2_btree_iter_peek_slot(&iter);
269 if (!bkey_is_inode(k.k)) {
270 bch2_fs_inconsistent(c,
271 "inode %llu:%u not found when deleting",
277 bch2_inode_unpack(k, &inode_u);
279 /* Subvolume root? */
280 if (inode_u.bi_subvol)
281 bch_warn(c, "deleting inode %llu marked as unlinked, but also a subvolume root!?", inode_u.bi_inum);
283 bkey_inode_generation_init(&delete.k_i);
284 delete.k.p = iter.pos;
285 delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
287 ret = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
288 bch2_trans_commit(trans, NULL, NULL,
289 BTREE_INSERT_NOFAIL);
291 bch2_trans_iter_exit(trans, &iter);
292 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
295 return ret ?: -BCH_ERR_transaction_restart_nested;
298 static int __remove_dirent(struct btree_trans *trans, struct bpos pos)
300 struct bch_fs *c = trans->c;
301 struct btree_iter iter;
302 struct bch_inode_unpacked dir_inode;
303 struct bch_hash_info dir_hash_info;
306 ret = lookup_first_inode(trans, pos.inode, &dir_inode);
310 dir_hash_info = bch2_hash_info_init(c, &dir_inode);
312 bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents, pos, BTREE_ITER_INTENT);
314 ret = bch2_hash_delete_at(trans, bch2_dirent_hash_desc,
315 &dir_hash_info, &iter,
316 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
317 bch2_trans_iter_exit(trans, &iter);
319 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
320 bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
324 /* Get lost+found, create if it doesn't exist: */
325 static int lookup_lostfound(struct btree_trans *trans, u32 subvol,
326 struct bch_inode_unpacked *lostfound)
328 struct bch_fs *c = trans->c;
329 struct bch_inode_unpacked root;
330 struct bch_hash_info root_hash_info;
331 struct qstr lostfound_str = QSTR("lost+found");
332 subvol_inum root_inum = { .subvol = subvol };
338 ret = __subvol_lookup(trans, subvol, &snapshot, &root_inum.inum);
342 ret = __lookup_inode(trans, root_inum.inum, &root, &snapshot);
346 root_hash_info = bch2_hash_info_init(c, &root);
348 ret = __lookup_dirent(trans, root_hash_info, root_inum,
349 &lostfound_str, &inum, &d_type);
350 if (ret == -ENOENT) {
351 bch_notice(c, "creating lost+found");
352 goto create_lostfound;
355 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
356 bch_err(c, "error looking up lost+found: %s", bch2_err_str(ret));
360 if (d_type != DT_DIR) {
361 bch_err(c, "error looking up lost+found: not a directory");
366 * The check_dirents pass has already run, dangling dirents
367 * shouldn't exist here:
369 return __lookup_inode(trans, inum, lostfound, &snapshot);
372 bch2_inode_init_early(c, lostfound);
374 ret = bch2_create_trans(trans, root_inum, &root,
375 lostfound, &lostfound_str,
376 0, 0, S_IFDIR|0700, 0, NULL, NULL,
377 (subvol_inum) { }, 0);
378 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
379 bch_err(c, "error creating lost+found: %s", bch2_err_str(ret));
383 static int __reattach_inode(struct btree_trans *trans,
384 struct bch_inode_unpacked *inode,
387 struct bch_hash_info dir_hash;
388 struct bch_inode_unpacked lostfound;
395 ret = __snapshot_lookup_subvol(trans, inode_snapshot, &subvol);
399 ret = lookup_lostfound(trans, subvol, &lostfound);
403 if (S_ISDIR(inode->bi_mode)) {
404 lostfound.bi_nlink++;
406 ret = __write_inode(trans, &lostfound, U32_MAX);
411 dir_hash = bch2_hash_info_init(trans->c, &lostfound);
413 snprintf(name_buf, sizeof(name_buf), "%llu", inode->bi_inum);
414 name = (struct qstr) QSTR(name_buf);
416 ret = bch2_dirent_create(trans,
419 .inum = lostfound.bi_inum,
423 &name, inode->bi_inum, &dir_offset,
424 BCH_HASH_SET_MUST_CREATE);
428 inode->bi_dir = lostfound.bi_inum;
429 inode->bi_dir_offset = dir_offset;
431 return __write_inode(trans, inode, inode_snapshot);
434 static int reattach_inode(struct btree_trans *trans,
435 struct bch_inode_unpacked *inode,
438 int ret = commit_do(trans, NULL, NULL,
439 BTREE_INSERT_LAZY_RW|
441 __reattach_inode(trans, inode, inode_snapshot));
443 bch_err(trans->c, "error reattaching inode %llu: %s",
444 inode->bi_inum, bch2_err_str(ret));
451 static int remove_backpointer(struct btree_trans *trans,
452 struct bch_inode_unpacked *inode)
454 struct btree_iter iter;
458 bch2_trans_iter_init(trans, &iter, BTREE_ID_dirents,
459 POS(inode->bi_dir, inode->bi_dir_offset), 0);
460 k = bch2_btree_iter_peek_slot(&iter);
464 if (k.k->type != KEY_TYPE_dirent) {
469 ret = __remove_dirent(trans, k.k->p);
471 bch2_trans_iter_exit(trans, &iter);
475 struct snapshots_seen_entry {
480 struct snapshots_seen {
482 DARRAY(struct snapshots_seen_entry) ids;
485 static inline void snapshots_seen_exit(struct snapshots_seen *s)
487 darray_exit(&s->ids);
490 static inline void snapshots_seen_init(struct snapshots_seen *s)
492 memset(s, 0, sizeof(*s));
495 static int snapshots_seen_add(struct bch_fs *c, struct snapshots_seen *s, u32 id)
497 struct snapshots_seen_entry *i, n = { id, id };
500 darray_for_each(s->ids, i) {
501 if (n.equiv < i->equiv)
504 if (i->equiv == n.equiv) {
505 bch_err(c, "%s(): adding duplicate snapshot", __func__);
510 ret = darray_insert_item(&s->ids, i - s->ids.data, n);
512 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
517 static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s,
518 enum btree_id btree_id, struct bpos pos)
520 struct snapshots_seen_entry *i, n = {
522 .equiv = bch2_snapshot_equiv(c, pos.snapshot),
526 if (!bkey_eq(s->pos, pos))
529 pos.snapshot = n.equiv;
532 darray_for_each(s->ids, i)
533 if (i->equiv == n.equiv) {
534 if (fsck_err_on(i->id != n.id, c,
535 "snapshot deletion did not run correctly:\n"
536 " duplicate keys in btree %s at %llu:%llu snapshots %u, %u (equiv %u)\n",
537 bch2_btree_ids[btree_id],
538 pos.inode, pos.offset,
539 i->id, n.id, n.equiv))
540 return -BCH_ERR_need_snapshot_cleanup;
545 ret = darray_push(&s->ids, n);
547 bch_err(c, "error reallocating snapshots_seen table (size %zu)",
554 * key_visible_in_snapshot - returns true if @id is a descendent of @ancestor,
555 * and @ancestor hasn't been overwritten in @seen
557 * That is, returns whether key in @ancestor snapshot is visible in @id snapshot
559 static bool key_visible_in_snapshot(struct bch_fs *c, struct snapshots_seen *seen,
560 u32 id, u32 ancestor)
563 u32 top = seen->ids.nr ? seen->ids.data[seen->ids.nr - 1].equiv : 0;
565 BUG_ON(id > ancestor);
566 BUG_ON(!bch2_snapshot_is_equiv(c, id));
567 BUG_ON(!bch2_snapshot_is_equiv(c, ancestor));
569 /* @ancestor should be the snapshot most recently added to @seen */
570 BUG_ON(ancestor != seen->pos.snapshot);
571 BUG_ON(ancestor != top);
576 if (!bch2_snapshot_is_ancestor(c, id, ancestor))
579 for (i = seen->ids.nr - 2;
580 i >= 0 && seen->ids.data[i].equiv >= id;
582 if (bch2_snapshot_is_ancestor(c, id, seen->ids.data[i].equiv) &&
583 bch2_snapshot_is_ancestor(c, seen->ids.data[i].equiv, ancestor))
590 * ref_visible - given a key with snapshot id @src that points to a key with
591 * snapshot id @dst, test whether there is some snapshot in which @dst is
594 * This assumes we're visiting @src keys in natural key order.
596 * @s - list of snapshot IDs already seen at @src
597 * @src - snapshot ID of src key
598 * @dst - snapshot ID of dst key
600 static int ref_visible(struct bch_fs *c, struct snapshots_seen *s,
604 ? key_visible_in_snapshot(c, s, dst, src)
605 : bch2_snapshot_is_ancestor(c, src, dst);
608 #define for_each_visible_inode(_c, _s, _w, _snapshot, _i) \
609 for (_i = (_w)->inodes.data; _i < (_w)->inodes.data + (_w)->inodes.nr && \
610 (_i)->snapshot <= (_snapshot); _i++) \
611 if (key_visible_in_snapshot(_c, _s, _i->snapshot, _snapshot))
613 struct inode_walker_entry {
614 struct bch_inode_unpacked inode;
619 struct inode_walker {
620 bool first_this_inode;
623 DARRAY(struct inode_walker_entry) inodes;
626 static void inode_walker_exit(struct inode_walker *w)
628 darray_exit(&w->inodes);
631 static struct inode_walker inode_walker_init(void)
633 return (struct inode_walker) { 0, };
636 static int add_inode(struct bch_fs *c, struct inode_walker *w,
637 struct bkey_s_c inode)
639 struct bch_inode_unpacked u;
641 BUG_ON(bch2_inode_unpack(inode, &u));
643 return darray_push(&w->inodes, ((struct inode_walker_entry) {
645 .snapshot = bch2_snapshot_equiv(c, inode.k->p.snapshot),
649 static int __walk_inode(struct btree_trans *trans,
650 struct inode_walker *w, struct bpos pos)
652 struct bch_fs *c = trans->c;
653 struct btree_iter iter;
655 u32 restart_count = trans->restart_count;
659 pos.snapshot = bch2_snapshot_equiv(c, pos.snapshot);
661 if (pos.inode == w->cur_inum) {
662 w->first_this_inode = false;
663 goto lookup_snapshot;
668 for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, pos.inode),
669 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
670 if (k.k->p.offset != pos.inode)
673 if (bkey_is_inode(k.k))
676 bch2_trans_iter_exit(trans, &iter);
681 w->cur_inum = pos.inode;
682 w->first_this_inode = true;
684 if (trans_was_restarted(trans, restart_count))
685 return -BCH_ERR_transaction_restart_nested;
688 for (i = 0; i < w->inodes.nr; i++)
689 if (bch2_snapshot_is_ancestor(c, pos.snapshot, w->inodes.data[i].snapshot))
693 BUG_ON(pos.snapshot > w->inodes.data[i].snapshot);
695 if (pos.snapshot != w->inodes.data[i].snapshot) {
696 struct inode_walker_entry e = w->inodes.data[i];
698 e.snapshot = pos.snapshot;
701 bch_info(c, "have key for inode %llu:%u but have inode in ancestor snapshot %u",
702 pos.inode, pos.snapshot, w->inodes.data[i].snapshot);
704 while (i && w->inodes.data[i - 1].snapshot > pos.snapshot)
707 ret = darray_insert_item(&w->inodes, i, e);
715 static int __get_visible_inodes(struct btree_trans *trans,
716 struct inode_walker *w,
717 struct snapshots_seen *s,
720 struct bch_fs *c = trans->c;
721 struct btree_iter iter;
727 for_each_btree_key_norestart(trans, iter, BTREE_ID_inodes, POS(0, inum),
728 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
729 u32 equiv = bch2_snapshot_equiv(c, k.k->p.snapshot);
731 if (k.k->p.offset != inum)
734 if (!ref_visible(c, s, s->pos.snapshot, equiv))
737 if (bkey_is_inode(k.k))
740 if (equiv >= s->pos.snapshot)
743 bch2_trans_iter_exit(trans, &iter);
748 static int check_key_has_snapshot(struct btree_trans *trans,
749 struct btree_iter *iter,
752 struct bch_fs *c = trans->c;
753 struct printbuf buf = PRINTBUF;
756 if (mustfix_fsck_err_on(!bch2_snapshot_equiv(c, k.k->p.snapshot), c,
757 "key in missing snapshot: %s",
758 (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
759 ret = bch2_btree_delete_at(trans, iter,
760 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?: 1;
766 static int hash_redo_key(struct btree_trans *trans,
767 const struct bch_hash_desc desc,
768 struct bch_hash_info *hash_info,
769 struct btree_iter *k_iter, struct bkey_s_c k)
771 struct bkey_i *delete;
774 delete = bch2_trans_kmalloc(trans, sizeof(*delete));
776 return PTR_ERR(delete);
778 tmp = bch2_bkey_make_mut(trans, k);
782 bkey_init(&delete->k);
783 delete->k.p = k_iter->pos;
784 return bch2_btree_iter_traverse(k_iter) ?:
785 bch2_trans_update(trans, k_iter, delete, 0) ?:
786 bch2_hash_set_snapshot(trans, desc, hash_info,
787 (subvol_inum) { 0, k.k->p.inode },
788 k.k->p.snapshot, tmp,
789 BCH_HASH_SET_MUST_CREATE,
790 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
791 bch2_trans_commit(trans, NULL, NULL,
793 BTREE_INSERT_LAZY_RW);
796 static int hash_check_key(struct btree_trans *trans,
797 const struct bch_hash_desc desc,
798 struct bch_hash_info *hash_info,
799 struct btree_iter *k_iter, struct bkey_s_c hash_k)
801 struct bch_fs *c = trans->c;
802 struct btree_iter iter = { NULL };
803 struct printbuf buf = PRINTBUF;
808 if (hash_k.k->type != desc.key_type)
811 hash = desc.hash_bkey(hash_info, hash_k);
813 if (likely(hash == hash_k.k->p.offset))
816 if (hash_k.k->p.offset < hash)
819 for_each_btree_key_norestart(trans, iter, desc.btree_id,
820 POS(hash_k.k->p.inode, hash),
821 BTREE_ITER_SLOTS, k, ret) {
822 if (bkey_eq(k.k->p, hash_k.k->p))
825 if (fsck_err_on(k.k->type == desc.key_type &&
826 !desc.cmp_bkey(k, hash_k), c,
827 "duplicate hash table keys:\n%s",
828 (printbuf_reset(&buf),
829 bch2_bkey_val_to_text(&buf, c, hash_k),
831 ret = bch2_hash_delete_at(trans, desc, hash_info, k_iter, 0) ?: 1;
835 if (bkey_deleted(k.k)) {
836 bch2_trans_iter_exit(trans, &iter);
841 bch2_trans_iter_exit(trans, &iter);
845 if (fsck_err(c, "hash table key at wrong offset: btree %s inode %llu offset %llu, hashed to %llu\n%s",
846 bch2_btree_ids[desc.btree_id], hash_k.k->p.inode, hash_k.k->p.offset, hash,
847 (printbuf_reset(&buf),
848 bch2_bkey_val_to_text(&buf, c, hash_k), buf.buf))) {
849 ret = hash_redo_key(trans, desc, hash_info, k_iter, hash_k);
851 bch_err(c, "hash_redo_key err %s", bch2_err_str(ret));
854 ret = -BCH_ERR_transaction_restart_nested;
860 static int check_inode(struct btree_trans *trans,
861 struct btree_iter *iter,
863 struct bch_inode_unpacked *prev,
864 struct snapshots_seen *s,
867 struct bch_fs *c = trans->c;
868 struct bch_inode_unpacked u;
869 bool do_update = false;
872 ret = check_key_has_snapshot(trans, iter, k);
878 ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
883 * if snapshot id isn't a leaf node, skip it - deletion in
884 * particular is not atomic, so on the internal snapshot nodes
885 * we can see inodes marked for deletion after a clean shutdown
887 if (bch2_snapshot_internal_node(c, k.k->p.snapshot))
890 if (!bkey_is_inode(k.k))
893 BUG_ON(bch2_inode_unpack(k, &u));
896 !(u.bi_flags & (BCH_INODE_I_SIZE_DIRTY|
897 BCH_INODE_I_SECTORS_DIRTY|
898 BCH_INODE_UNLINKED)))
901 if (prev->bi_inum != u.bi_inum)
904 if (fsck_err_on(prev->bi_hash_seed != u.bi_hash_seed ||
905 inode_d_type(prev) != inode_d_type(&u), c,
906 "inodes in different snapshots don't match")) {
907 bch_err(c, "repair not implemented yet");
911 if (u.bi_flags & BCH_INODE_UNLINKED &&
913 fsck_err(c, "filesystem marked clean, but inode %llu unlinked",
915 bch2_trans_unlock(trans);
918 ret = fsck_inode_rm(trans, u.bi_inum, iter->pos.snapshot);
919 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
920 bch_err(c, "error in fsck: error while deleting inode: %s",
925 if (u.bi_flags & BCH_INODE_I_SIZE_DIRTY &&
927 fsck_err(c, "filesystem marked clean, but inode %llu has i_size dirty",
929 bch_verbose(c, "truncating inode %llu", u.bi_inum);
931 bch2_trans_unlock(trans);
935 * XXX: need to truncate partial blocks too here - or ideally
936 * just switch units to bytes and that issue goes away
938 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
939 SPOS(u.bi_inum, round_up(u.bi_size, block_bytes(c)) >> 9,
941 POS(u.bi_inum, U64_MAX),
944 bch_err(c, "error in fsck: error truncating inode: %s",
950 * We truncated without our normal sector accounting hook, just
951 * make sure we recalculate it:
953 u.bi_flags |= BCH_INODE_I_SECTORS_DIRTY;
955 u.bi_flags &= ~BCH_INODE_I_SIZE_DIRTY;
959 if (u.bi_flags & BCH_INODE_I_SECTORS_DIRTY &&
961 fsck_err(c, "filesystem marked clean, but inode %llu has i_sectors dirty",
965 bch_verbose(c, "recounting sectors for inode %llu",
968 sectors = bch2_count_inode_sectors(trans, u.bi_inum, iter->pos.snapshot);
970 bch_err(c, "error in fsck: error recounting inode sectors: %s",
971 bch2_err_str(sectors));
975 u.bi_sectors = sectors;
976 u.bi_flags &= ~BCH_INODE_I_SECTORS_DIRTY;
980 if (u.bi_flags & BCH_INODE_BACKPTR_UNTRUSTED) {
983 u.bi_flags &= ~BCH_INODE_BACKPTR_UNTRUSTED;
988 ret = __write_inode(trans, &u, iter->pos.snapshot);
990 bch_err(c, "error in fsck: error updating inode: %s",
996 bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
1001 static int check_inodes(struct bch_fs *c, bool full)
1003 struct btree_trans trans;
1004 struct btree_iter iter;
1005 struct bch_inode_unpacked prev = { 0 };
1006 struct snapshots_seen s;
1010 snapshots_seen_init(&s);
1011 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1013 ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_inodes,
1015 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1016 NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
1017 check_inode(&trans, &iter, k, &prev, &s, full));
1019 bch2_trans_exit(&trans);
1020 snapshots_seen_exit(&s);
1022 bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
1027 * Checking for overlapping extents needs to be reimplemented
1030 static int fix_overlapping_extent(struct btree_trans *trans,
1031 struct bkey_s_c k, struct bpos cut_at)
1033 struct btree_iter iter;
1037 u = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
1038 ret = PTR_ERR_OR_ZERO(u);
1042 bkey_reassemble(u, k);
1043 bch2_cut_front(cut_at, u);
1047 * We don't want to go through the extent_handle_overwrites path:
1049 * XXX: this is going to screw up disk accounting, extent triggers
1050 * assume things about extent overwrites - we should be running the
1051 * triggers manually here
1053 bch2_trans_iter_init(trans, &iter, BTREE_ID_extents, u->k.p,
1054 BTREE_ITER_INTENT|BTREE_ITER_NOT_EXTENTS);
1056 BUG_ON(iter.flags & BTREE_ITER_IS_EXTENTS);
1057 ret = bch2_btree_iter_traverse(&iter) ?:
1058 bch2_trans_update(trans, &iter, u, BTREE_TRIGGER_NORUN) ?:
1059 bch2_trans_commit(trans, NULL, NULL,
1060 BTREE_INSERT_NOFAIL|
1061 BTREE_INSERT_LAZY_RW);
1062 bch2_trans_iter_exit(trans, &iter);
1067 static struct bkey_s_c_dirent dirent_get_by_pos(struct btree_trans *trans,
1068 struct btree_iter *iter,
1074 bch2_trans_iter_init(trans, iter, BTREE_ID_dirents, pos, 0);
1075 k = bch2_btree_iter_peek_slot(iter);
1077 if (!ret && k.k->type != KEY_TYPE_dirent)
1080 bch2_trans_iter_exit(trans, iter);
1081 return (struct bkey_s_c_dirent) { .k = ERR_PTR(ret) };
1084 return bkey_s_c_to_dirent(k);
1087 static bool inode_points_to_dirent(struct bch_inode_unpacked *inode,
1088 struct bkey_s_c_dirent d)
1090 return inode->bi_dir == d.k->p.inode &&
1091 inode->bi_dir_offset == d.k->p.offset;
1094 static bool dirent_points_to_inode(struct bkey_s_c_dirent d,
1095 struct bch_inode_unpacked *inode)
1097 return d.v->d_type == DT_SUBVOL
1098 ? le32_to_cpu(d.v->d_child_subvol) == inode->bi_subvol
1099 : le64_to_cpu(d.v->d_inum) == inode->bi_inum;
1102 static int inode_backpointer_exists(struct btree_trans *trans,
1103 struct bch_inode_unpacked *inode,
1106 struct btree_iter iter;
1107 struct bkey_s_c_dirent d;
1110 d = dirent_get_by_pos(trans, &iter,
1111 SPOS(inode->bi_dir, inode->bi_dir_offset, snapshot));
1112 ret = bkey_err(d.s_c);
1114 return ret == -ENOENT ? 0 : ret;
1116 ret = dirent_points_to_inode(d, inode);
1117 bch2_trans_iter_exit(trans, &iter);
1121 static int check_i_sectors(struct btree_trans *trans, struct inode_walker *w)
1123 struct bch_fs *c = trans->c;
1124 struct inode_walker_entry *i;
1125 u32 restart_count = trans->restart_count;
1129 darray_for_each(w->inodes, i) {
1130 if (i->inode.bi_sectors == i->count)
1133 count2 = bch2_count_inode_sectors(trans, w->cur_inum, i->snapshot);
1135 if (i->count != count2) {
1136 bch_err(c, "fsck counted i_sectors wrong: got %llu should be %llu",
1139 if (i->inode.bi_sectors == i->count)
1143 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SECTORS_DIRTY), c,
1144 "inode %llu:%u has incorrect i_sectors: got %llu, should be %llu",
1145 w->cur_inum, i->snapshot,
1146 i->inode.bi_sectors, i->count)) {
1147 i->inode.bi_sectors = i->count;
1148 ret = write_inode(trans, &i->inode, i->snapshot);
1155 bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
1156 if (!ret && trans_was_restarted(trans, restart_count))
1157 ret = -BCH_ERR_transaction_restart_nested;
1161 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
1163 struct inode_walker *inode,
1164 struct snapshots_seen *s)
1166 struct bch_fs *c = trans->c;
1167 struct inode_walker_entry *i;
1168 struct printbuf buf = PRINTBUF;
1172 ret = check_key_has_snapshot(trans, iter, k);
1174 ret = ret < 0 ? ret : 0;
1179 equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1181 ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1185 if (k.k->type == KEY_TYPE_whiteout)
1188 if (inode->cur_inum != k.k->p.inode) {
1189 ret = check_i_sectors(trans, inode);
1194 BUG_ON(!iter->path->should_be_locked);
1196 if (bkey_gt(prev.k->k.p, bkey_start_pos(k.k))) {
1200 bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(prev.k));
1201 bch2_bkey_val_to_text(&PBUF(buf2), c, k);
1203 if (fsck_err(c, "overlapping extents:\n%s\n%s", buf1, buf2)) {
1204 ret = fix_overlapping_extent(trans, k, prev.k->k.p)
1205 ?: -BCH_ERR_transaction_restart_nested;
1210 ret = __walk_inode(trans, inode, equiv);
1214 if (fsck_err_on(ret == INT_MAX, c,
1215 "extent in missing inode:\n %s",
1216 (printbuf_reset(&buf),
1217 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1218 ret = bch2_btree_delete_at(trans, iter,
1219 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1223 if (ret == INT_MAX) {
1228 i = inode->inodes.data + ret;
1231 if (fsck_err_on(!S_ISREG(i->inode.bi_mode) &&
1232 !S_ISLNK(i->inode.bi_mode), c,
1233 "extent in non regular inode mode %o:\n %s",
1235 (printbuf_reset(&buf),
1236 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1237 ret = bch2_btree_delete_at(trans, iter,
1238 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1243 * Check inodes in reverse order, from oldest snapshots to newest, so
1244 * that we emit the fewest number of whiteouts necessary:
1246 for (i = inode->inodes.data + inode->inodes.nr - 1;
1247 i >= inode->inodes.data;
1249 if (i->snapshot > equiv.snapshot ||
1250 !key_visible_in_snapshot(c, s, i->snapshot, equiv.snapshot))
1253 if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SIZE_DIRTY) &&
1254 k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9 &&
1255 !bkey_extent_is_reservation(k), c,
1256 "extent type past end of inode %llu:%u, i_size %llu\n %s",
1257 i->inode.bi_inum, i->snapshot, i->inode.bi_size,
1258 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1259 struct btree_iter iter2;
1261 bch2_trans_copy_iter(&iter2, iter);
1262 bch2_btree_iter_set_snapshot(&iter2, i->snapshot);
1263 ret = bch2_btree_iter_traverse(&iter2) ?:
1264 bch2_btree_delete_at(trans, &iter2,
1265 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1266 bch2_trans_iter_exit(trans, &iter2);
1270 if (i->snapshot != equiv.snapshot) {
1271 ret = snapshots_seen_add(c, s, i->snapshot);
1278 if (bkey_extent_is_allocation(k.k))
1279 for_each_visible_inode(c, s, inode, equiv.snapshot, i)
1280 i->count += k.k->size;
1282 bch2_bkey_buf_reassemble(&prev, c, k);
1288 printbuf_exit(&buf);
1290 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
1291 bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
1296 * Walk extents: verify that extents have a corresponding S_ISREG inode, and
1297 * that i_size an i_sectors are consistent
1300 static int check_extents(struct bch_fs *c)
1302 struct inode_walker w = inode_walker_init();
1303 struct snapshots_seen s;
1304 struct btree_trans trans;
1305 struct btree_iter iter;
1310 struct bkey_buf prev;
1311 bch2_bkey_buf_init(&prev);
1312 prev.k->k = KEY(0, 0, 0);
1314 snapshots_seen_init(&s);
1315 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1317 bch_verbose(c, "checking extents");
1319 ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_extents,
1320 POS(BCACHEFS_ROOT_INO, 0),
1321 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1323 BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
1324 check_extent(&trans, &iter, k, &w, &s));
1326 bch2_bkey_buf_exit(&prev, c);
1328 inode_walker_exit(&w);
1329 bch2_trans_exit(&trans);
1330 snapshots_seen_exit(&s);
1333 bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
1337 static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
1339 struct bch_fs *c = trans->c;
1340 struct inode_walker_entry *i;
1341 u32 restart_count = trans->restart_count;
1345 darray_for_each(w->inodes, i) {
1346 if (i->inode.bi_nlink == i->count)
1349 count2 = bch2_count_subdirs(trans, w->cur_inum, i->snapshot);
1353 if (i->count != count2) {
1354 bch_err(c, "fsck counted subdirectories wrong: got %llu should be %llu",
1357 if (i->inode.bi_nlink == i->count)
1361 if (fsck_err_on(i->inode.bi_nlink != i->count, c,
1362 "directory %llu:%u with wrong i_nlink: got %u, should be %llu",
1363 w->cur_inum, i->snapshot, i->inode.bi_nlink, i->count)) {
1364 i->inode.bi_nlink = i->count;
1365 ret = write_inode(trans, &i->inode, i->snapshot);
1372 bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
1373 if (!ret && trans_was_restarted(trans, restart_count))
1374 ret = -BCH_ERR_transaction_restart_nested;
1378 static int check_dirent_target(struct btree_trans *trans,
1379 struct btree_iter *iter,
1380 struct bkey_s_c_dirent d,
1381 struct bch_inode_unpacked *target,
1382 u32 target_snapshot)
1384 struct bch_fs *c = trans->c;
1385 struct bkey_i_dirent *n;
1386 bool backpointer_exists = true;
1387 struct printbuf buf = PRINTBUF;
1390 if (!target->bi_dir &&
1391 !target->bi_dir_offset) {
1392 target->bi_dir = d.k->p.inode;
1393 target->bi_dir_offset = d.k->p.offset;
1395 ret = __write_inode(trans, target, target_snapshot);
1400 if (!inode_points_to_dirent(target, d)) {
1401 ret = inode_backpointer_exists(trans, target, d.k->p.snapshot);
1405 backpointer_exists = ret;
1408 if (fsck_err_on(S_ISDIR(target->bi_mode) &&
1409 backpointer_exists, c,
1410 "directory %llu with multiple links",
1412 ret = __remove_dirent(trans, d.k->p);
1416 if (fsck_err_on(backpointer_exists &&
1417 !target->bi_nlink, c,
1418 "inode %llu type %s has multiple links but i_nlink 0",
1419 target->bi_inum, bch2_d_types[d.v->d_type])) {
1421 target->bi_flags &= ~BCH_INODE_UNLINKED;
1423 ret = __write_inode(trans, target, target_snapshot);
1428 if (fsck_err_on(!backpointer_exists, c,
1429 "inode %llu:%u has wrong backpointer:\n"
1431 "should be %llu:%llu",
1432 target->bi_inum, target_snapshot,
1434 target->bi_dir_offset,
1437 target->bi_dir = d.k->p.inode;
1438 target->bi_dir_offset = d.k->p.offset;
1440 ret = __write_inode(trans, target, target_snapshot);
1446 if (fsck_err_on(d.v->d_type != inode_d_type(target), c,
1447 "incorrect d_type: got %s, should be %s:\n%s",
1448 bch2_d_type_str(d.v->d_type),
1449 bch2_d_type_str(inode_d_type(target)),
1450 (printbuf_reset(&buf),
1451 bch2_bkey_val_to_text(&buf, c, d.s_c), buf.buf))) {
1452 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1453 ret = PTR_ERR_OR_ZERO(n);
1457 bkey_reassemble(&n->k_i, d.s_c);
1458 n->v.d_type = inode_d_type(target);
1460 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1464 d = dirent_i_to_s_c(n);
1467 if (d.v->d_type == DT_SUBVOL &&
1468 target->bi_parent_subvol != le32_to_cpu(d.v->d_parent_subvol) &&
1469 (c->sb.version < bcachefs_metadata_version_subvol_dirent ||
1470 fsck_err(c, "dirent has wrong d_parent_subvol field: got %u, should be %u",
1471 le32_to_cpu(d.v->d_parent_subvol),
1472 target->bi_parent_subvol))) {
1473 n = bch2_trans_kmalloc(trans, bkey_bytes(d.k));
1474 ret = PTR_ERR_OR_ZERO(n);
1478 bkey_reassemble(&n->k_i, d.s_c);
1479 n->v.d_parent_subvol = cpu_to_le32(target->bi_parent_subvol);
1481 ret = bch2_trans_update(trans, iter, &n->k_i, 0);
1485 d = dirent_i_to_s_c(n);
1490 printbuf_exit(&buf);
1492 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
1493 bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
1497 static int check_dirent(struct btree_trans *trans, struct btree_iter *iter,
1499 struct bch_hash_info *hash_info,
1500 struct inode_walker *dir,
1501 struct inode_walker *target,
1502 struct snapshots_seen *s)
1504 struct bch_fs *c = trans->c;
1505 struct bkey_s_c_dirent d;
1506 struct inode_walker_entry *i;
1507 struct printbuf buf = PRINTBUF;
1511 ret = check_key_has_snapshot(trans, iter, k);
1513 ret = ret < 0 ? ret : 0;
1518 equiv.snapshot = bch2_snapshot_equiv(c, k.k->p.snapshot);
1520 ret = snapshots_seen_update(c, s, iter->btree_id, k.k->p);
1524 if (k.k->type == KEY_TYPE_whiteout)
1527 if (dir->cur_inum != k.k->p.inode) {
1528 ret = check_subdir_count(trans, dir);
1533 BUG_ON(!iter->path->should_be_locked);
1535 ret = __walk_inode(trans, dir, equiv);
1539 if (fsck_err_on(ret == INT_MAX, c,
1540 "dirent in nonexisting directory:\n%s",
1541 (printbuf_reset(&buf),
1542 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1543 ret = bch2_btree_delete_at(trans, iter,
1544 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1548 if (ret == INT_MAX) {
1553 i = dir->inodes.data + ret;
1556 if (fsck_err_on(!S_ISDIR(i->inode.bi_mode), c,
1557 "dirent in non directory inode type %s:\n%s",
1558 bch2_d_type_str(inode_d_type(&i->inode)),
1559 (printbuf_reset(&buf),
1560 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1561 ret = bch2_btree_delete_at(trans, iter, 0);
1565 if (dir->first_this_inode)
1566 *hash_info = bch2_hash_info_init(c, &dir->inodes.data[0].inode);
1568 ret = hash_check_key(trans, bch2_dirent_hash_desc,
1569 hash_info, iter, k);
1573 /* dirent has been deleted */
1578 if (k.k->type != KEY_TYPE_dirent)
1581 d = bkey_s_c_to_dirent(k);
1583 if (d.v->d_type == DT_SUBVOL) {
1584 struct bch_inode_unpacked subvol_root;
1585 u32 target_subvol = le32_to_cpu(d.v->d_child_subvol);
1586 u32 target_snapshot;
1589 ret = __subvol_lookup(trans, target_subvol,
1590 &target_snapshot, &target_inum);
1591 if (ret && ret != -ENOENT)
1594 if (fsck_err_on(ret, c,
1595 "dirent points to missing subvolume %llu",
1596 le64_to_cpu(d.v->d_child_subvol))) {
1597 ret = __remove_dirent(trans, d.k->p);
1601 ret = __lookup_inode(trans, target_inum,
1602 &subvol_root, &target_snapshot);
1603 if (ret && ret != -ENOENT)
1606 if (fsck_err_on(ret, c,
1607 "subvolume %u points to missing subvolume root %llu",
1610 bch_err(c, "repair not implemented yet");
1615 if (fsck_err_on(subvol_root.bi_subvol != target_subvol, c,
1616 "subvol root %llu has wrong bi_subvol field: got %u, should be %u",
1618 subvol_root.bi_subvol, target_subvol)) {
1619 subvol_root.bi_subvol = target_subvol;
1620 ret = __write_inode(trans, &subvol_root, target_snapshot);
1625 ret = check_dirent_target(trans, iter, d, &subvol_root,
1630 ret = __get_visible_inodes(trans, target, s, le64_to_cpu(d.v->d_inum));
1634 if (fsck_err_on(!target->inodes.nr, c,
1635 "dirent points to missing inode: (equiv %u)\n%s",
1637 (printbuf_reset(&buf),
1638 bch2_bkey_val_to_text(&buf, c, k),
1640 ret = __remove_dirent(trans, d.k->p);
1645 darray_for_each(target->inodes, i) {
1646 ret = check_dirent_target(trans, iter, d,
1647 &i->inode, i->snapshot);
1653 if (d.v->d_type == DT_DIR)
1654 for_each_visible_inode(c, s, dir, equiv.snapshot, i)
1660 printbuf_exit(&buf);
1662 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
1663 bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
1668 * Walk dirents: verify that they all have a corresponding S_ISDIR inode,
1672 static int check_dirents(struct bch_fs *c)
1674 struct inode_walker dir = inode_walker_init();
1675 struct inode_walker target = inode_walker_init();
1676 struct snapshots_seen s;
1677 struct bch_hash_info hash_info;
1678 struct btree_trans trans;
1679 struct btree_iter iter;
1683 bch_verbose(c, "checking dirents");
1685 snapshots_seen_init(&s);
1686 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1688 ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_dirents,
1689 POS(BCACHEFS_ROOT_INO, 0),
1690 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
1693 BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
1694 check_dirent(&trans, &iter, k, &hash_info, &dir, &target, &s));
1696 bch2_trans_exit(&trans);
1697 snapshots_seen_exit(&s);
1698 inode_walker_exit(&dir);
1699 inode_walker_exit(&target);
1702 bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
1706 static int check_xattr(struct btree_trans *trans, struct btree_iter *iter,
1708 struct bch_hash_info *hash_info,
1709 struct inode_walker *inode)
1711 struct bch_fs *c = trans->c;
1714 ret = check_key_has_snapshot(trans, iter, k);
1718 ret = __walk_inode(trans, inode, k.k->p);
1722 if (fsck_err_on(ret == INT_MAX, c,
1723 "xattr for missing inode %llu",
1725 return bch2_btree_delete_at(trans, iter, 0);
1732 if (inode->first_this_inode)
1733 *hash_info = bch2_hash_info_init(c, &inode->inodes.data[0].inode);
1735 ret = hash_check_key(trans, bch2_xattr_hash_desc, hash_info, iter, k);
1737 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
1738 bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
1743 * Walk xattrs: verify that they all have a corresponding inode
1746 static int check_xattrs(struct bch_fs *c)
1748 struct inode_walker inode = inode_walker_init();
1749 struct bch_hash_info hash_info;
1750 struct btree_trans trans;
1751 struct btree_iter iter;
1755 bch_verbose(c, "checking xattrs");
1757 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1759 ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_xattrs,
1760 POS(BCACHEFS_ROOT_INO, 0),
1761 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS,
1764 BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
1765 check_xattr(&trans, &iter, k, &hash_info, &inode));
1767 bch2_trans_exit(&trans);
1770 bch_err(c, "%s(): error %s", __func__, bch2_err_str(ret));
1774 static int check_root_trans(struct btree_trans *trans)
1776 struct bch_fs *c = trans->c;
1777 struct bch_inode_unpacked root_inode;
1782 ret = __subvol_lookup(trans, BCACHEFS_ROOT_SUBVOL, &snapshot, &inum);
1783 if (ret && ret != -ENOENT)
1786 if (mustfix_fsck_err_on(ret, c, "root subvol missing")) {
1787 struct bkey_i_subvolume root_subvol;
1790 inum = BCACHEFS_ROOT_INO;
1792 bkey_subvolume_init(&root_subvol.k_i);
1793 root_subvol.k.p.offset = BCACHEFS_ROOT_SUBVOL;
1794 root_subvol.v.flags = 0;
1795 root_subvol.v.snapshot = cpu_to_le32(snapshot);
1796 root_subvol.v.inode = cpu_to_le64(inum);
1797 ret = commit_do(trans, NULL, NULL,
1798 BTREE_INSERT_NOFAIL|
1799 BTREE_INSERT_LAZY_RW,
1800 __bch2_btree_insert(trans, BTREE_ID_subvolumes, &root_subvol.k_i));
1802 bch_err(c, "error writing root subvol: %s", bch2_err_str(ret));
1808 ret = __lookup_inode(trans, BCACHEFS_ROOT_INO, &root_inode, &snapshot);
1809 if (ret && ret != -ENOENT)
1812 if (mustfix_fsck_err_on(ret, c, "root directory missing") ||
1813 mustfix_fsck_err_on(!S_ISDIR(root_inode.bi_mode), c,
1814 "root inode not a directory")) {
1815 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755,
1817 root_inode.bi_inum = inum;
1819 ret = __write_inode(trans, &root_inode, snapshot);
1821 bch_err(c, "error writing root inode: %s", bch2_err_str(ret));
1828 /* Get root directory, create if it doesn't exist: */
1830 static int check_root(struct bch_fs *c)
1832 bch_verbose(c, "checking root directory");
1834 return bch2_trans_do(c, NULL, NULL,
1835 BTREE_INSERT_NOFAIL|
1836 BTREE_INSERT_LAZY_RW,
1837 check_root_trans(&trans));
1840 struct pathbuf_entry {
1845 typedef DARRAY(struct pathbuf_entry) pathbuf;
1847 static bool path_is_dup(pathbuf *p, u64 inum, u32 snapshot)
1849 struct pathbuf_entry *i;
1851 darray_for_each(*p, i)
1852 if (i->inum == inum &&
1853 i->snapshot == snapshot)
1859 static int path_down(struct bch_fs *c, pathbuf *p,
1860 u64 inum, u32 snapshot)
1862 int ret = darray_push(p, ((struct pathbuf_entry) {
1864 .snapshot = snapshot,
1868 bch_err(c, "fsck: error allocating memory for pathbuf, size %zu",
1874 * Check that a given inode is reachable from the root:
1876 * XXX: we should also be verifying that inodes are in the right subvolumes
1878 static int check_path(struct btree_trans *trans,
1880 struct bch_inode_unpacked *inode,
1883 struct bch_fs *c = trans->c;
1886 snapshot = bch2_snapshot_equiv(c, snapshot);
1889 while (!(inode->bi_inum == BCACHEFS_ROOT_INO &&
1890 inode->bi_subvol == BCACHEFS_ROOT_SUBVOL)) {
1891 struct btree_iter dirent_iter;
1892 struct bkey_s_c_dirent d;
1893 u32 parent_snapshot = snapshot;
1895 if (inode->bi_subvol) {
1898 ret = subvol_lookup(trans, inode->bi_parent_subvol,
1899 &parent_snapshot, &inum);
1904 ret = lockrestart_do(trans,
1905 PTR_ERR_OR_ZERO((d = dirent_get_by_pos(trans, &dirent_iter,
1906 SPOS(inode->bi_dir, inode->bi_dir_offset,
1907 parent_snapshot))).k));
1908 if (ret && ret != -ENOENT)
1911 if (!ret && !dirent_points_to_inode(d, inode)) {
1912 bch2_trans_iter_exit(trans, &dirent_iter);
1916 if (ret == -ENOENT) {
1917 if (fsck_err(c, "unreachable inode %llu:%u, type %s nlink %u backptr %llu:%llu",
1918 inode->bi_inum, snapshot,
1919 bch2_d_type_str(inode_d_type(inode)),
1922 inode->bi_dir_offset))
1923 ret = reattach_inode(trans, inode, snapshot);
1927 bch2_trans_iter_exit(trans, &dirent_iter);
1929 if (!S_ISDIR(inode->bi_mode))
1932 ret = path_down(c, p, inode->bi_inum, snapshot);
1934 bch_err(c, "memory allocation failure");
1938 snapshot = parent_snapshot;
1940 ret = lookup_inode(trans, inode->bi_dir, inode, &snapshot);
1942 /* Should have been caught in dirents pass */
1943 bch_err(c, "error looking up parent directory: %i", ret);
1947 if (path_is_dup(p, inode->bi_inum, snapshot)) {
1948 struct pathbuf_entry *i;
1950 /* XXX print path */
1951 bch_err(c, "directory structure loop");
1953 darray_for_each(*p, i)
1954 pr_err("%llu:%u", i->inum, i->snapshot);
1955 pr_err("%llu:%u", inode->bi_inum, snapshot);
1957 if (!fsck_err(c, "directory structure loop"))
1960 ret = commit_do(trans, NULL, NULL,
1961 BTREE_INSERT_NOFAIL|
1962 BTREE_INSERT_LAZY_RW,
1963 remove_backpointer(trans, inode));
1965 bch_err(c, "error removing dirent: %i", ret);
1969 ret = reattach_inode(trans, inode, snapshot);
1974 bch_err(c, "%s: err %s", __func__, bch2_err_str(ret));
1979 * Check for unreachable inodes, as well as loops in the directory structure:
1980 * After check_dirents(), if an inode backpointer doesn't exist that means it's
1984 static int check_directory_structure(struct bch_fs *c)
1986 struct btree_trans trans;
1987 struct btree_iter iter;
1989 struct bch_inode_unpacked u;
1990 pathbuf path = { 0, };
1993 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
1995 for_each_btree_key(&trans, iter, BTREE_ID_inodes, POS_MIN,
1997 BTREE_ITER_PREFETCH|
1998 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
1999 if (!bkey_is_inode(k.k))
2002 ret = bch2_inode_unpack(k, &u);
2004 /* Should have been caught earlier in fsck: */
2005 bch_err(c, "error unpacking inode %llu: %i", k.k->p.offset, ret);
2009 if (u.bi_flags & BCH_INODE_UNLINKED)
2012 ret = check_path(&trans, &path, &u, iter.pos.snapshot);
2016 bch2_trans_iter_exit(&trans, &iter);
2020 bch2_trans_exit(&trans);
2024 struct nlink_table {
2035 static int add_nlink(struct bch_fs *c, struct nlink_table *t,
2036 u64 inum, u32 snapshot)
2038 if (t->nr == t->size) {
2039 size_t new_size = max_t(size_t, 128UL, t->size * 2);
2040 void *d = kvmalloc_array(new_size, sizeof(t->d[0]), GFP_KERNEL);
2043 bch_err(c, "fsck: error allocating memory for nlink_table, size %zu",
2049 memcpy(d, t->d, t->size * sizeof(t->d[0]));
2057 t->d[t->nr++] = (struct nlink) {
2059 .snapshot = snapshot,
2065 static int nlink_cmp(const void *_l, const void *_r)
2067 const struct nlink *l = _l;
2068 const struct nlink *r = _r;
2070 return cmp_int(l->inum, r->inum) ?: cmp_int(l->snapshot, r->snapshot);
2073 static void inc_link(struct bch_fs *c, struct snapshots_seen *s,
2074 struct nlink_table *links,
2075 u64 range_start, u64 range_end, u64 inum, u32 snapshot)
2077 struct nlink *link, key = {
2078 .inum = inum, .snapshot = U32_MAX,
2081 if (inum < range_start || inum >= range_end)
2084 link = __inline_bsearch(&key, links->d, links->nr,
2085 sizeof(links->d[0]), nlink_cmp);
2089 while (link > links->d && link[0].inum == link[-1].inum)
2092 for (; link < links->d + links->nr && link->inum == inum; link++)
2093 if (ref_visible(c, s, snapshot, link->snapshot)) {
2095 if (link->snapshot >= snapshot)
2101 static int check_nlinks_find_hardlinks(struct bch_fs *c,
2102 struct nlink_table *t,
2103 u64 start, u64 *end)
2105 struct btree_trans trans;
2106 struct btree_iter iter;
2108 struct bch_inode_unpacked u;
2111 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2113 for_each_btree_key(&trans, iter, BTREE_ID_inodes,
2116 BTREE_ITER_PREFETCH|
2117 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2118 if (!bkey_is_inode(k.k))
2121 /* Should never fail, checked by bch2_inode_invalid: */
2122 BUG_ON(bch2_inode_unpack(k, &u));
2125 * Backpointer and directory structure checks are sufficient for
2126 * directories, since they can't have hardlinks:
2128 if (S_ISDIR(le16_to_cpu(u.bi_mode)))
2134 ret = add_nlink(c, t, k.k->p.offset, k.k->p.snapshot);
2136 *end = k.k->p.offset;
2142 bch2_trans_iter_exit(&trans, &iter);
2143 bch2_trans_exit(&trans);
2146 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
2152 static int check_nlinks_walk_dirents(struct bch_fs *c, struct nlink_table *links,
2153 u64 range_start, u64 range_end)
2155 struct btree_trans trans;
2156 struct snapshots_seen s;
2157 struct btree_iter iter;
2159 struct bkey_s_c_dirent d;
2162 snapshots_seen_init(&s);
2164 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2166 for_each_btree_key(&trans, iter, BTREE_ID_dirents, POS_MIN,
2168 BTREE_ITER_PREFETCH|
2169 BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
2170 ret = snapshots_seen_update(c, &s, iter.btree_id, k.k->p);
2174 switch (k.k->type) {
2175 case KEY_TYPE_dirent:
2176 d = bkey_s_c_to_dirent(k);
2178 if (d.v->d_type != DT_DIR &&
2179 d.v->d_type != DT_SUBVOL)
2180 inc_link(c, &s, links, range_start, range_end,
2181 le64_to_cpu(d.v->d_inum),
2182 bch2_snapshot_equiv(c, d.k->p.snapshot));
2186 bch2_trans_iter_exit(&trans, &iter);
2189 bch_err(c, "error in fsck: btree error %i while walking dirents", ret);
2191 bch2_trans_exit(&trans);
2192 snapshots_seen_exit(&s);
2196 static int check_nlinks_update_inode(struct btree_trans *trans, struct btree_iter *iter,
2198 struct nlink_table *links,
2199 size_t *idx, u64 range_end)
2201 struct bch_fs *c = trans->c;
2202 struct bch_inode_unpacked u;
2203 struct nlink *link = &links->d[*idx];
2206 if (k.k->p.offset >= range_end)
2209 if (!bkey_is_inode(k.k))
2212 BUG_ON(bch2_inode_unpack(k, &u));
2214 if (S_ISDIR(le16_to_cpu(u.bi_mode)))
2220 while ((cmp_int(link->inum, k.k->p.offset) ?:
2221 cmp_int(link->snapshot, k.k->p.snapshot)) < 0) {
2222 BUG_ON(*idx == links->nr);
2223 link = &links->d[++*idx];
2226 if (fsck_err_on(bch2_inode_nlink_get(&u) != link->count, c,
2227 "inode %llu type %s has wrong i_nlink (%u, should be %u)",
2228 u.bi_inum, bch2_d_types[mode_to_type(u.bi_mode)],
2229 bch2_inode_nlink_get(&u), link->count)) {
2230 bch2_inode_nlink_set(&u, link->count);
2231 ret = __write_inode(trans, &u, k.k->p.snapshot);
2238 static int check_nlinks_update_hardlinks(struct bch_fs *c,
2239 struct nlink_table *links,
2240 u64 range_start, u64 range_end)
2242 struct btree_trans trans;
2243 struct btree_iter iter;
2248 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2250 ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_inodes,
2251 POS(0, range_start),
2252 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
2253 NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
2254 check_nlinks_update_inode(&trans, &iter, k, links, &idx, range_end));
2256 bch2_trans_exit(&trans);
2259 bch_err(c, "error in fsck: btree error %i while walking inodes", ret);
2267 static int check_nlinks(struct bch_fs *c)
2269 struct nlink_table links = { 0 };
2270 u64 this_iter_range_start, next_iter_range_start = 0;
2273 bch_verbose(c, "checking inode nlinks");
2276 this_iter_range_start = next_iter_range_start;
2277 next_iter_range_start = U64_MAX;
2279 ret = check_nlinks_find_hardlinks(c, &links,
2280 this_iter_range_start,
2281 &next_iter_range_start);
2283 ret = check_nlinks_walk_dirents(c, &links,
2284 this_iter_range_start,
2285 next_iter_range_start);
2289 ret = check_nlinks_update_hardlinks(c, &links,
2290 this_iter_range_start,
2291 next_iter_range_start);
2296 } while (next_iter_range_start != U64_MAX);
2303 static int fix_reflink_p_key(struct btree_trans *trans, struct btree_iter *iter,
2306 struct bkey_s_c_reflink_p p;
2307 struct bkey_i_reflink_p *u;
2310 if (k.k->type != KEY_TYPE_reflink_p)
2313 p = bkey_s_c_to_reflink_p(k);
2315 if (!p.v->front_pad && !p.v->back_pad)
2318 u = bch2_trans_kmalloc(trans, sizeof(*u));
2319 ret = PTR_ERR_OR_ZERO(u);
2323 bkey_reassemble(&u->k_i, k);
2327 return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_NORUN);
2331 static int fix_reflink_p(struct bch_fs *c)
2333 struct btree_trans trans;
2334 struct btree_iter iter;
2338 if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix)
2341 bch_verbose(c, "fixing reflink_p keys");
2343 bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
2345 ret = for_each_btree_key_commit(&trans, iter,
2346 BTREE_ID_extents, POS_MIN,
2347 BTREE_ITER_INTENT|BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
2348 NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
2349 fix_reflink_p_key(&trans, &iter, k));
2351 bch2_trans_exit(&trans);
2356 * Checks for inconsistencies that shouldn't happen, unless we have a bug.
2357 * Doesn't fix them yet, mainly because they haven't yet been observed:
2359 int bch2_fsck_full(struct bch_fs *c)
2363 ret = bch2_fs_check_snapshots(c) ?:
2364 bch2_fs_check_subvols(c) ?:
2365 bch2_delete_dead_snapshots(c) ?:
2366 check_inodes(c, true) ?:
2371 check_directory_structure(c) ?:
2375 if (bch2_err_matches(ret, BCH_ERR_need_snapshot_cleanup)) {
2376 set_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
2383 int bch2_fsck_walk_inodes_only(struct bch_fs *c)
2385 return bch2_fs_check_snapshots(c) ?:
2386 bch2_fs_check_subvols(c) ?:
2387 bch2_delete_dead_snapshots(c) ?:
2388 check_inodes(c, false);