bcachefs: better bi_subvol checking

5 files changed, 74 insertions, 2 deletions

diff --git a/fs/bcachefs/fsck.c b/fs/bcachefs/fsck.c
index 65616af69db7..027757b13526 100644
--- a/fs/bcachefs/fsck.c
+++ b/fs/bcachefs/fsck.c
@@ -923,6 +923,41 @@ static int check_inode(struct btree_trans *trans,
 		do_update = true;
 	}
 
+	if (fsck_err_on(u.bi_parent_subvol &&
+			(u.bi_subvol == 0 ||
+			 u.bi_subvol == BCACHEFS_ROOT_SUBVOL),
+			c, inode_bi_parent_subvol_nonzero,
+			"inode %llu:%u has subvol %u but nonzero parent subvol %u",
+			u.bi_inum, k.k->p.snapshot, u.bi_subvol, u.bi_parent_subvol)) {
+		u.bi_parent_subvol = 0;
+		do_update = true;
+	}
+
+	if (u.bi_subvol) {
+		struct bch_subvolume s;
+
+		ret = bch2_subvolume_get(trans, u.bi_subvol, false, 0, &s);
+		if (ret && !bch2_err_matches(ret, ENOENT))
+			goto err;
+
+		if (fsck_err_on(ret,
+				c, inode_bi_subvol_missing,
+				"inode %llu:%u bi_subvol points to missing subvolume %u",
+				u.bi_inum, k.k->p.snapshot, u.bi_subvol) ||
+		    fsck_err_on(le64_to_cpu(s.inode) != u.bi_inum ||
+				!bch2_snapshot_is_ancestor(c, le32_to_cpu(s.snapshot),
+							   k.k->p.snapshot),
+				c, inode_bi_subvol_wrong,
+				"inode %llu:%u points to subvol %u, but subvol points to %llu:%u",
+				u.bi_inum, k.k->p.snapshot, u.bi_subvol,
+				le64_to_cpu(s.inode),
+				le32_to_cpu(s.snapshot))) {
+			u.bi_subvol = 0;
+			u.bi_parent_subvol = 0;
+			do_update = true;
+		}
+	}
+
 	if (do_update) {
 		ret = __bch2_fsck_write_inode(trans, &u, iter->pos.snapshot);
 		bch_err_msg(c, ret, "in fsck updating inode");
diff --git a/fs/bcachefs/inode.c b/fs/bcachefs/inode.c
index e7ba169c4e54..dbe37ccc7519 100644
--- a/fs/bcachefs/inode.c
+++ b/fs/bcachefs/inode.c
@@ -324,7 +324,7 @@ int bch2_inode_unpack(struct bkey_s_c k,
 	return bch2_inode_unpack_slowpath(k, unpacked);
 }
 
-static int bch2_inode_peek_nowarn(struct btree_trans *trans,
+int bch2_inode_peek_nowarn(struct btree_trans *trans,
 		    struct btree_iter *iter,
 		    struct bch_inode_unpacked *inode,
 		    subvol_inum inum, unsigned flags)
diff --git a/fs/bcachefs/inode.h b/fs/bcachefs/inode.h
index b8da7ff8069d..9a9353c001c2 100644
--- a/fs/bcachefs/inode.h
+++ b/fs/bcachefs/inode.h
@@ -95,6 +95,8 @@ struct bkey_i *bch2_inode_to_v3(struct btree_trans *, struct bkey_i *);
 
 void bch2_inode_unpacked_to_text(struct printbuf *, struct bch_inode_unpacked *);
 
+int bch2_inode_peek_nowarn(struct btree_trans *, struct btree_iter *,
+		    struct bch_inode_unpacked *, subvol_inum, unsigned);
 int bch2_inode_peek(struct btree_trans *, struct btree_iter *,
 		    struct bch_inode_unpacked *, subvol_inum, unsigned);
 
diff --git a/fs/bcachefs/sb-errors_types.h b/fs/bcachefs/sb-errors_types.h
index dbfd91ab86cf..3ff83863413d 100644
--- a/fs/bcachefs/sb-errors_types.h
+++ b/fs/bcachefs/sb-errors_types.h
@@ -251,7 +251,11 @@
 	x(hash_table_key_wrong_offset,				243)	\
 	x(unlinked_inode_not_on_deleted_list,			244)	\
 	x(reflink_p_front_pad_bad,				245)	\
-	x(journal_entry_dup_same_device,			246)
+	x(journal_entry_dup_same_device,			246)	\
+	x(inode_bi_subvol_missing,				247)	\
+	x(inode_bi_subvol_wrong,				248)	\
+	x(inode_bi_parent_subvol_nonzero,			249)	\
+	x(inode_bi_parent_subvol_wrong,				250)
 
 enum bch_sb_error_id {
 #define x(t, n) BCH_FSCK_ERR_##t = n,
diff --git a/fs/bcachefs/subvolume.c b/fs/bcachefs/subvolume.c
index 7c67c28d3ef8..e7ee52c39990 100644
--- a/fs/bcachefs/subvolume.c
+++ b/fs/bcachefs/subvolume.c
@@ -42,6 +42,36 @@ static int check_subvol(struct btree_trans *trans,
 		return ret ?: -BCH_ERR_transaction_restart_nested;
 	}
 
+	struct bch_inode_unpacked inode;
+	struct btree_iter inode_iter = {};
+	ret = bch2_inode_peek_nowarn(trans, &inode_iter, &inode,
+				    (subvol_inum) { k.k->p.offset, le64_to_cpu(subvol.v->inode) },
+				    0);
+	bch2_trans_iter_exit(trans, &inode_iter);
+
+	if (ret && !bch2_err_matches(ret, ENOENT))
+		return ret;
+
+	if (fsck_err_on(ret, c, subvol_to_missing_root,
+			"subvolume %llu points to missing subvolume root %llu:%u",
+			k.k->p.offset, le64_to_cpu(subvol.v->inode),
+			le32_to_cpu(subvol.v->snapshot))) {
+		ret = bch2_subvolume_delete(trans, iter->pos.offset);
+		bch_err_msg(c, ret, "deleting subvolume %llu", iter->pos.offset);
+		return ret ?: -BCH_ERR_transaction_restart_nested;
+	}
+
+	if (fsck_err_on(inode.bi_subvol != subvol.k->p.offset,
+			c, subvol_root_wrong_bi_subvol,
+			"subvol root %llu:%u has wrong bi_subvol field: got %u, should be %llu",
+			inode.bi_inum, inode_iter.k.p.snapshot,
+			inode.bi_subvol, subvol.k->p.offset)) {
+		inode.bi_subvol = subvol.k->p.offset;
+		ret = __bch2_fsck_write_inode(trans, &inode, le32_to_cpu(subvol.v->snapshot));
+		if (ret)
+			goto err;
+	}
+
 	if (!BCH_SUBVOLUME_SNAP(subvol.v)) {
 		u32 snapshot_root = bch2_snapshot_root(c, le32_to_cpu(subvol.v->snapshot));
 		u32 snapshot_tree;
@@ -73,6 +103,7 @@ static int check_subvol(struct btree_trans *trans,
 		}
 	}
 
+err:
 fsck_err:
 	return ret;
 }