fix(apiserver): never sign or accept JWTs with the shipped default secret (bug-1945)

Cause: config/app.yaml shipped secretKey "xce-api-key" and
resolveJWTSecret used it whenever XCE_JWT_SECRET was unset, so anyone
could forge tokens for any user on such a deployment.

Fix: resolveJWTSecret uses XCE_JWT_SECRET, else the config value, but
treats an empty value or the old public default as unset; then it
generates a random 256-bit per-process secret and InitConf logs a warning
(secret never logged): tokens do not survive a restart and apiservers
sharing one endpoint must set the same XCE_JWT_SECRET. The shipped
app.yaml no longer carries a secret. The controllers test secret is no
longer the old default.

Tests: TestResolveJWTSecret_NeverUsesKnownDefaultOrEmpty,
TestShippedAppYAML_NoJWTSecret, TestInitConf_NoJWTSecretGeneratesAndWarns,
TestInitConf_JWTSecretFromEnvNoWarning.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

src/frontend/apiserver/config/app.yaml

@@ -19,7 +19,12 @@ security:
 authentication:
   token:
     type: "jwt"
-    secretKey: "xce-api-key"
+    # JWT signing secret. Prefer the XCE_JWT_SECRET environment variable
+    # (overrides this value). If neither is set, or the value is the old
+    # public default "xce-api-key", the apiserver uses a random per-process
+    # secret and logs a warning: tokens then do not survive a restart, and
+    # several apiservers behind one endpoint must share XCE_JWT_SECRET.
+    secretKey: ""
     token_expire_time: 86400 # 1 day
 
 rate_limit:

src/frontend/apiserver/controllers/integration_test.go

@@ -2,7 +2,7 @@
 
 // Integration tests for XCE API server controllers (§2, §6, §8).
 //
-// Test signing secret: "xce-api-key" (test-only, never use in production).
+// Test signing secret: testJWTSecret (test-only, never use in production).
 // JWT tokens must embed "ip": "127.0.0.1" to match req.RemoteAddr set by httptest.
 package controllers
 
@@ -26,7 +26,7 @@ import (
 	"go.uber.org/zap"
 )
 
-const testJWTSecret = "xce-api-key"
+const testJWTSecret = "controllers-test-only-secret"
 
 // ========== Test Helpers ==========
 

src/frontend/apiserver/utils/config.go

@@ -1,6 +1,8 @@
 package utils
 
 import (
+	"crypto/rand"
+	"encoding/hex"
 	"errors"
 	"fmt"
 	"log"
@@ -164,10 +166,18 @@ func InitConf() error {
 		AppVersion: viper.GetString("app.version"),
 	}
 	// parse authentication section
+	configuredSecret := viper.GetString("authentication.token.secretKey")
+	jwtSecret := resolveJWTSecret(configuredSecret)
+	if jwtSecret != configuredSecret && jwtSecret != os.Getenv("XCE_JWT_SECRET") {
+		Warn("No JWT secret configured (XCE_JWT_SECRET or authentication.token.secretKey), " +
+			"or it is the publicly known default: using a random per-process secret. " +
+			"Tokens do not survive an apiserver restart, and apiservers sharing one endpoint " +
+			"must all set the same XCE_JWT_SECRET")
+	}
 	AuthenticationYamlConfig = &Authentication{
 		Token: TokenConfig{
 			Type:            viper.GetString("authentication.token.type"),
-			SecretKey:       resolveJWTSecret(viper.GetString("authentication.token.secretKey")),
+			SecretKey:       jwtSecret,
 			TokenExpireTime: int(viper.GetFloat64("authentication.token.token_expire_time")),
 		},
 	}
@@ -340,11 +350,31 @@ func resolveTrustedProxies(entries []string) ([]string, error) {
 	return proxies, nil
 }
 
-// resolveJWTSecret returns the JWT secret from XCE_JWT_SECRET env var if set and non-empty,
+// knownDefaultJWTSecret is the secret older app.yaml files shipped with; it is
-// otherwise falls back to the value from config file.
+// public, so it is treated as "no secret configured" (bug-1945).
+const knownDefaultJWTSecret = "xce-api-key"
+
+// usableJWTSecret reports whether s may sign tokens: non-empty and not the
+// publicly known default.
+func usableJWTSecret(s string) bool {
+	return s != "" && s != knownDefaultJWTSecret
+}
+
+// resolveJWTSecret returns the JWT secret: XCE_JWT_SECRET if usable, else
+// the config file value if usable, else a random per-process secret (the
+// caller detects that case because the result differs from both inputs).
 func resolveJWTSecret(configValue string) string {
-	if env := os.Getenv("XCE_JWT_SECRET"); env != "" {
+	if env := os.Getenv("XCE_JWT_SECRET"); usableJWTSecret(env) {
 		return env
 	}
-	return configValue
+	if usableJWTSecret(configValue) {
+		return configValue
+	}
+	buf := make([]byte, 32)
+	if _, err := rand.Read(buf); err != nil {
+		// crypto/rand never fails on supported platforms; refuse to fall
+		// back to a predictable secret if it ever does.
+		panic(fmt.Sprintf("cannot generate a JWT secret: %v", err))
+	}
+	return hex.EncodeToString(buf)
 }

src/frontend/apiserver/utils/config_test.go

@@ -40,6 +40,94 @@ func TestResolveJWTSecret_EmptyEnvUsesConfig(t *testing.T) {
 	}
 }
 
+// bug-1945: the publicly known default secret and an empty secret are never
+// used; a random per-process secret replaces them.
+func TestResolveJWTSecret_NeverUsesKnownDefaultOrEmpty(t *testing.T) {
+	cases := []struct {
+		name   string
+		env    string
+		config string
+	}{
+		{"default in config, no env", "", "xce-api-key"},
+		{"empty config, no env", "", ""},
+		{"default in env, empty config", "xce-api-key", ""},
+		{"default in env and config", "xce-api-key", "xce-api-key"},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Setenv("XCE_JWT_SECRET", tc.env)
+			got := resolveJWTSecret(tc.config)
+			if got == "" || got == "xce-api-key" {
+				t.Fatalf("resolveJWTSecret(%q) with env %q = %q: must be a generated secret", tc.config, tc.env, got)
+			}
+			if len(got) < 32 {
+				t.Errorf("generated secret too short: %d chars", len(got))
+			}
+			if again := resolveJWTSecret(tc.config); again == got {
+				t.Errorf("generated secrets must be random, got %q twice", got)
+			}
+		})
+	}
+}
+
+// bug-1945: the shipped app.yaml carries no JWT secret.
+func TestShippedAppYAML_NoJWTSecret(t *testing.T) {
+	shipped, err := os.ReadFile(filepath.Join("..", "config", "app.yaml"))
+	if err != nil {
+		t.Fatalf("read shipped app.yaml: %v", err)
+	}
+	v := viper.New()
+	v.SetConfigType("yaml")
+	if err := v.ReadConfig(strings.NewReader(string(shipped))); err != nil {
+		t.Fatalf("parse shipped app.yaml: %v", err)
+	}
+	if s := v.GetString("authentication.token.secretKey"); s != "" {
+		t.Errorf("shipped app.yaml must not set authentication.token.secretKey, has %q", s)
+	}
+}
+
+// bug-1945: InitConf on the shipped app.yaml without XCE_JWT_SECRET runs
+// with a random secret and logs a warning (never the secret itself).
+func TestInitConf_NoJWTSecretGeneratesAndWarns(t *testing.T) {
+	t.Setenv("XCE_JWT_SECRET", "")
+	logs := setupInitConf(t, "job:\n  max_array_size: 1000")
+	oldAuth := AuthenticationYamlConfig
+	t.Cleanup(func() { AuthenticationYamlConfig = oldAuth })
+	if err := InitConf(); err != nil {
+		t.Fatalf("InitConf: %v", err)
+	}
+	secret := AuthenticationYamlConfig.Token.SecretKey
+	if secret == "" || secret == "xce-api-key" {
+		t.Fatalf("SecretKey = %q: must be a generated secret", secret)
+	}
+	warns := logs.FilterLevelExact(zap.WarnLevel).FilterMessageSnippet("XCE_JWT_SECRET").All()
+	if len(warns) != 1 {
+		t.Fatalf("expected 1 JWT secret warning, got %d (all logs: %v)", len(warns), logs.All())
+	}
+	for _, e := range logs.All() {
+		if strings.Contains(e.Message, secret) {
+			t.Errorf("log message contains the secret: %q", e.Message)
+		}
+	}
+}
+
+// bug-1945: a configured secret is used as is, without a warning.
+func TestInitConf_JWTSecretFromEnvNoWarning(t *testing.T) {
+	t.Setenv("XCE_JWT_SECRET", "a-deployment-secret-of-sufficient-length")
+	logs := setupInitConf(t, "job:\n  max_array_size: 1000")
+	oldAuth := AuthenticationYamlConfig
+	t.Cleanup(func() { AuthenticationYamlConfig = oldAuth })
+	if err := InitConf(); err != nil {
+		t.Fatalf("InitConf: %v", err)
+	}
+	if got := AuthenticationYamlConfig.Token.SecretKey; got != "a-deployment-secret-of-sufficient-length" {
+		t.Errorf("SecretKey = %q, want the XCE_JWT_SECRET value", got)
+	}
+	if n := logs.FilterMessageSnippet("XCE_JWT_SECRET").Len(); n != 0 {
+		t.Errorf("unexpected JWT secret warning (%d)", n)
+	}
+}
+
 // bug-1948: security.trusted_proxies accepts IPs and CIDRs only; empty
 // means no trusted proxy.
 func TestResolveTrustedProxies(t *testing.T) {