Reject config files with !-prefixed local YAML tags

CHANGELOG.md

@@ -8,6 +8,9 @@
 - Add `--stdin-filepath` flag to `buf format`, which reads a single `.proto` file from
   stdin and writes the formatted result to stdout. The path is not read from disk, and is
   only used to report parse errors and diffs.
+- Fix configuration files silently accepting unquoted values that start with `!`, such as
+  `ignore: [!foo/bar.proto]`. Previously they parsed as empty strings, now they rejected
+  with an error.
 - Fix lint comment ignores on proto2 `group` fields being ignored.
 - Improve the `buf curl` error message for methods that accept a single request message.
 - Deduplicate remote input fetches within a single command invocation, so that multiple

private/pkg/encoding/encoding.go

@@ -45,6 +45,9 @@ func UnmarshalYAMLStrict(data []byte, v any) error {
 	if len(data) == 0 {
 		return nil
 	}
+	if err := validateNoLocalYAMLTags(data); err != nil {
+		return err
+	}
 	yamlDecoder := NewYAMLDecoderStrict(bytes.NewReader(data))
 	return updateYAMLTypeError(yamlDecoder.Decode(v))
 }
@@ -191,6 +194,42 @@ func InterfaceSliceOrStringToStringSlice(in any) ([]string, error) {
 
 // *** PRIVATE ***
 
+// validateNoLocalYAMLTags returns an error if the YAML data contains a local
+// tag, that is a tag of the form "!name" as opposed to a standard "!!name" tag.
+// We never register custom tags that would be resolved with "!" so there is
+// no valid use for a value starting with a single exclamation mark. If we don't
+// detect and reject explicitly, they instead cause the tag to be trimmed from
+// the value, usually leaving an empty value behind that we use as the config.
+func validateNoLocalYAMLTags(data []byte) error {
+	var node yaml.Node
+	if err := yaml.Unmarshal(data, &node); err != nil {
+		return err
+	}
+	return walkYAMLNode(&node, func(node *yaml.Node) error {
+		if len(node.Tag) >= 2 && node.Tag[0] == '!' && node.Tag[1] != '!' {
+			return fmt.Errorf(
+				"yaml: line %d: unexpected tag %q: values that start with %q must be quoted",
+				node.Line,
+				node.Tag,
+				"!",
+			)
+		}
+		return nil
+	})
+}
+
+func walkYAMLNode(node *yaml.Node, f func(*yaml.Node) error) error {
+	if err := f(node); err != nil {
+		return err
+	}
+	for _, child := range node.Content {
+		if err := walkYAMLNode(child, f); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
 func updateYAMLTypeError(err error) error {
 	if err == nil {
 		return nil

private/pkg/encoding/encoding_test.go

@@ -42,3 +42,34 @@ func testInterfaceSliceOrStringToCommaSepString(t *testing.T, in any, expected s
 	require.NoError(t, err)
 	require.Equal(t, expected, v)
 }
+
+func TestUnmarshalYAMLStrictRejectsLocalTags(t *testing.T) {
+	t.Parallel()
+	type external struct {
+		Ignore []string `yaml:"ignore"`
+	}
+	// An unquoted value that starts with "!" is a YAML local tag with an
+	// empty value, which the decoder would otherwise silently drop.
+	var v external
+	err := UnmarshalYAMLStrict([]byte("ignore:\n  - foo\n  - !foo/bar.proto\n"), &v)
+	require.Error(t, err)
+	require.Contains(t, err.Error(), `line 3: unexpected tag "!foo/bar.proto"`)
+
+	v = external{}
+	err = UnmarshalYAMLStrict([]byte("ignore:\n  - !local bar\n"), &v)
+	require.Error(t, err)
+	require.Contains(t, err.Error(), `unexpected tag "!local"`)
+
+	// Standard tags, anchors, aliases, and quoted values that start with "!"
+	// are all fine.
+	v = external{}
+	err = UnmarshalYAMLStrict([]byte("ignore:\n  - !!str 123\n  - &a foo\n  - *a\n  - \"!foo/bar.proto\"\n"), &v)
+	require.NoError(t, err)
+	require.Equal(t, []string{"123", "foo", "foo", "!foo/bar.proto"}, v.Ignore)
+
+	// Non-strict unmarshalling is unchanged.
+	v = external{}
+	err = UnmarshalYAMLNonStrict([]byte("ignore:\n  - !foo/bar.proto\n"), &v)
+	require.NoError(t, err)
+	require.Equal(t, []string{""}, v.Ignore)
+}