Handle ePDG COOKIE challenges during IKE_SA_INIT (#65)

Some ePDGs return zero-Responder-SPI COOKIE challenges, and multiple resolved gateways can respond differently. Accept valid COOKIE responses, retry with COOKIE as the first payload, and prefer a gateway that completes SA negotiation.\n\nConstraint: RFC 7296 requires the COOKIE notification to be the first payload on the retry.\nRejected: Treat COOKIE as an ordinary or fatal notification | either drops a valid challenge or prevents the required retry.\nConfidence: high\nScope-risk: narrow\nReversibility: clean\nDirective: Keep relaxed zero-Responder-SPI matching limited to IKE_SA_INIT COOKIE responses.\nTested: go test ./internal/vowifi/ike ./internal/vowifi/...\nNot-tested: Live carrier authorization after IKE_AUTH.

Co-authored-by: Meng Meng <[email protected]>

internal/vowifi/ike/provider.go

@@ -188,34 +188,60 @@ func (provider *Provider) start(ctx context.Context, request vowifi.TunnelReques
 		makeNotify(notifyNATSource, sourceHash),
 		makeNotify(notifyNATDestination, destinationHash),
 	}
-	first, initBody, err := marshalPayloadChain(initPayloads)
-	if err != nil {
-		return nil, err
-	}
-	initRequest := ikeHeader{
-		InitiatorSPI: initiatorSPI,
-		NextPayload:  first,
-		Exchange:     exchangeIKEInit,
-		Flags:        flagInitiator,
-		MessageID:    0,
-	}.marshal(initBody)
-	initResponse, err := transport.RoundTrip(ctx, initRequest)
-	if err != nil {
-		return nil, err
-	}
-	responseHeader, responseBody, err := validateResponse(initResponse, initiatorSPI, [8]byte{}, exchangeIKEInit, 0)
-	if err != nil {
-		return nil, err
-	}
-	if responseHeader.ResponderSPI == [8]byte{} {
-		return nil, errors.New("ike: responder returned a zero SPI")
-	}
-	initResponsePayloads, err := parsePayloadChain(responseHeader.NextPayload, responseBody)
-	if err != nil {
-		return nil, err
-	}
-	if err := rejectFatalNotifications(initResponsePayloads); err != nil {
-		return nil, err
+	var (
+		initRequest          []byte
+		initResponse         []byte
+		responseHeader       ikeHeader
+		initResponsePayloads []payload
+		cookie               []byte
+	)
+	for attempt := 0; attempt < maxIKEInitCookieChallenges; attempt++ {
+		requestPayloads := append([]payload(nil), initPayloads...)
+		if len(cookie) > 0 {
+			requestPayloads = append([]payload{makeNotify(notifyCookie, cookie)}, requestPayloads...)
+		}
+		first, initBody, err := marshalPayloadChain(requestPayloads)
+		if err != nil {
+			return nil, err
+		}
+		initRequest = ikeHeader{
+			InitiatorSPI: initiatorSPI,
+			NextPayload:  first,
+			Exchange:     exchangeIKEInit,
+			Flags:        flagInitiator,
+			MessageID:    0,
+		}.marshal(initBody)
+		initResponse, err = transport.RoundTrip(ctx, initRequest)
+		if err != nil {
+			return nil, err
+		}
+		var responseBody []byte
+		responseHeader, responseBody, err = validateResponse(initResponse, initiatorSPI, [8]byte{}, exchangeIKEInit, 0)
+		if err != nil {
+			return nil, err
+		}
+		initResponsePayloads, err = parsePayloadChain(responseHeader.NextPayload, responseBody)
+		if err != nil {
+			return nil, err
+		}
+		if err := rejectFatalNotifications(initResponsePayloads); err != nil {
+			return nil, err
+		}
+		challenge, hasCookie, err := ikeInitCookie(initResponsePayloads)
+		if err != nil {
+			return nil, err
+		}
+		if hasCookie {
+			if attempt+1 == maxIKEInitCookieChallenges {
+				return nil, errors.New("ike: ePDG requested too many COOKIE challenges")
+			}
+			cookie = challenge
+			continue
+		}
+		if responseHeader.ResponderSPI == [8]byte{} {
+			return nil, errors.New("ike: responder returned a zero SPI")
+		}
+		break
 	}
 	saPayload, err := onePayload(initResponsePayloads, payloadSA)
 	if err != nil {
@@ -616,6 +642,29 @@ func (provider *Provider) start(ctx context.Context, request vowifi.TunnelReques
 	return session, nil
 }
 
+const maxIKEInitCookieChallenges = 2
+
+func ikeInitCookie(payloads []payload) ([]byte, bool, error) {
+	var cookie []byte
+	for _, item := range payloadsOfType(payloads, payloadNotify) {
+		kind, data, err := parseNotify(item)
+		if err != nil {
+			return nil, false, err
+		}
+		if kind != notifyCookie {
+			continue
+		}
+		if len(data) == 0 {
+			return nil, false, errors.New("ike: ePDG returned an empty COOKIE")
+		}
+		if cookie != nil {
+			return nil, false, errors.New("ike: ePDG returned multiple COOKIE notifications")
+		}
+		cookie = append([]byte(nil), data...)
+	}
+	return cookie, cookie != nil, nil
+}
+
 func buildInitialEAPAuth(
 	idi payload,
 	requestedIDr payload,

internal/vowifi/ike/provider_test.go

@@ -4,8 +4,10 @@ import (
 	"bytes"
 	"context"
 	"errors"
+	"fmt"
 	"io"
 	"net"
+	"strings"
 	"testing"
 	"time"
 
@@ -64,17 +66,20 @@ func (reader constantReader) Read(destination []byte) (int, error) {
 }
 
 type firstAuthCaptureTransport struct {
-	t           *testing.T
-	wantEAPOnly bool
-	wantGroup   uint16
-	calls       int
-	suite       negotiatedSuite
-	keys        ikeKeys
-	spii        [8]byte
-	spir        [8]byte
-	nonceI      []byte
-	nonceR      []byte
-	floated     bool
+	t               *testing.T
+	wantEAPOnly     bool
+	wantGroup       uint16
+	calls           int
+	suite           negotiatedSuite
+	keys            ikeKeys
+	spii            [8]byte
+	spir            [8]byte
+	nonceI          []byte
+	nonceR          []byte
+	floated         bool
+	cookieChallenge []byte
+	cookieSeen      bool
+	cookieLoop      bool
 }
 
 func (transport *firstAuthCaptureTransport) LocalAddr() *net.UDPAddr {
@@ -92,6 +97,24 @@ func (transport *firstAuthCaptureTransport) Float(context.Context) error {
 
 func (transport *firstAuthCaptureTransport) RoundTrip(_ context.Context, packet []byte) ([]byte, error) {
 	transport.calls++
+	if len(transport.cookieChallenge) > 0 {
+		switch transport.calls {
+		case 1:
+			return transport.answerIKECookie(packet)
+		case 2:
+			if err := transport.verifyIKECookie(packet); err != nil {
+				return nil, err
+			}
+			if transport.cookieLoop {
+				return transport.answerIKECookie(packet)
+			}
+			return transport.answerIKEInit(packet)
+		case 3:
+			return nil, transport.observeFirstAuth(packet)
+		default:
+			return nil, errors.New("test: unexpected exchange")
+		}
+	}
 	switch transport.calls {
 	case 1:
 		return transport.answerIKEInit(packet)
@@ -102,6 +125,69 @@ func (transport *firstAuthCaptureTransport) RoundTrip(_ context.Context, packet
 	}
 }
 
+func (transport *firstAuthCaptureTransport) answerIKECookie(packet []byte) ([]byte, error) {
+	header, _, err := parseIKEPacket(packet)
+	if err != nil {
+		return nil, err
+	}
+	first, body, err := marshalPayloadChain([]payload{
+		makeNotify(notifyCookie, transport.cookieChallenge),
+	})
+	if err != nil {
+		return nil, err
+	}
+	return ikeHeader{
+		InitiatorSPI: header.InitiatorSPI,
+		NextPayload:  first,
+		Exchange:     exchangeIKEInit,
+		Flags:        flagResponse,
+		MessageID:    0,
+	}.marshal(body), nil
+}
+
+func (transport *firstAuthCaptureTransport) verifyIKECookie(packet []byte) error {
+	header, body, err := parseIKEPacket(packet)
+	if err != nil {
+		return err
+	}
+	if header.Exchange != exchangeIKEInit || header.MessageID != 0 || header.Flags != flagInitiator {
+		return errors.New("test: invalid retried IKE_SA_INIT header")
+	}
+	payloads, err := parsePayloadChain(header.NextPayload, body)
+	if err != nil {
+		return err
+	}
+	if len(payloads) == 0 || payloads[0].Type != payloadNotify {
+		return errors.New("test: retried IKE_SA_INIT did not put COOKIE first")
+	}
+	firstKind, firstData, err := parseNotify(payloads[0])
+	if err != nil || firstKind != notifyCookie || !bytes.Equal(firstData, transport.cookieChallenge) {
+		return errors.New("test: first retried IKE_SA_INIT payload is not the expected COOKIE")
+	}
+	cookies := payloadsOfType(payloads, payloadNotify)
+	if len(cookies) != 3 {
+		return fmt.Errorf("test: retried IKE_SA_INIT has %d notify payloads, want 3", len(cookies))
+	}
+	found := false
+	for _, item := range cookies {
+		kind, data, err := parseNotify(item)
+		if err != nil {
+			return err
+		}
+		if kind == notifyCookie {
+			if !bytes.Equal(data, transport.cookieChallenge) {
+				return fmt.Errorf("test: cookie = %x, want %x", data, transport.cookieChallenge)
+			}
+			found = true
+		}
+	}
+	if !found {
+		return errors.New("test: retried IKE_SA_INIT did not carry COOKIE")
+	}
+	transport.cookieSeen = true
+	return nil
+}
+
 func (transport *firstAuthCaptureTransport) answerIKEInit(packet []byte) ([]byte, error) {
 	header, body, err := parseIKEPacket(packet)
 	if err != nil {
@@ -291,6 +377,93 @@ func TestProviderVodafoneFirstAuthIsEAPOnlyAndRequestsIMSAPN(t *testing.T) {
 	}
 }
 
+func TestProviderRetriesIKEInitAfterCookie(t *testing.T) {
+	capture := &firstAuthCaptureTransport{
+		t:               t,
+		wantEAPOnly:     true,
+		cookieChallenge: []byte{0x10, 0x20, 0x30, 0x40},
+	}
+	provider, err := NewProvider(Config{
+		Random:    constantReader{value: 0x42},
+		Timeout:   time.Second,
+		Installer: unusedInstaller{},
+		APN:       "ims",
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	provider.transportFactory = func(
+		context.Context,
+		transportConfig,
+		vowifi.ProxyRoute,
+		string,
+	) (datagramTransport, error) {
+		return capture, nil
+	}
+	aka := &testAKAProvider{}
+	_, err = provider.Start(context.Background(), vowifi.TunnelRequest{
+		DeviceID: "ec20-cookie",
+		Identity: vowifi.SIMIdentity{
+			ICCID:   "8944100000000000000",
+			IMSI:    "234150123456789",
+			HomeMCC: "234",
+			HomeMNC: "15",
+		},
+		EPDG: "epdg.epc.mnc015.mcc234.pub.3gppnetwork.org",
+		AKA:  aka,
+	})
+	if !errors.Is(err, errFirstAuthObserved) {
+		t.Fatalf("Start() error = %v, want capture sentinel", err)
+	}
+	if capture.calls != 3 || !capture.cookieSeen || capture.floated || aka.calls != 0 {
+		t.Fatalf("capture calls=%d cookie_seen=%v floated=%v AKA calls=%d", capture.calls, capture.cookieSeen, capture.floated, aka.calls)
+	}
+}
+
+func TestProviderBoundsRepeatedIKEInitCookieChallenges(t *testing.T) {
+	capture := &firstAuthCaptureTransport{
+		t:               t,
+		wantEAPOnly:     true,
+		cookieChallenge: []byte{0x10, 0x20, 0x30, 0x40},
+		cookieLoop:      true,
+	}
+	provider, err := NewProvider(Config{
+		Random:    constantReader{value: 0x42},
+		Timeout:   time.Second,
+		Installer: unusedInstaller{},
+		APN:       "ims",
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	provider.transportFactory = func(
+		context.Context,
+		transportConfig,
+		vowifi.ProxyRoute,
+		string,
+	) (datagramTransport, error) {
+		return capture, nil
+	}
+	aka := &testAKAProvider{}
+	_, err = provider.Start(context.Background(), vowifi.TunnelRequest{
+		DeviceID: "ec20-cookie-loop",
+		Identity: vowifi.SIMIdentity{
+			ICCID:   "8944100000000000000",
+			IMSI:    "234150123456789",
+			HomeMCC: "234",
+			HomeMNC: "15",
+		},
+		EPDG: "epdg.epc.mnc015.mcc234.pub.3gppnetwork.org",
+		AKA:  aka,
+	})
+	if err == nil || !strings.Contains(err.Error(), "too many COOKIE challenges") {
+		t.Fatalf("Start() error = %v, want bounded COOKIE error", err)
+	}
+	if capture.calls != maxIKEInitCookieChallenges || aka.calls != 0 {
+		t.Fatalf("capture calls=%d AKA calls=%d", capture.calls, aka.calls)
+	}
+}
+
 func TestProviderO2GermanyFirstAuthUsesStandardEAPAndRequestsIMSAPN(t *testing.T) {
 	capture := &firstAuthCaptureTransport{t: t, wantEAPOnly: false, wantGroup: dhMODP2048}
 	provider, err := NewProvider(Config{

internal/vowifi/ike/transport.go

@@ -576,10 +576,17 @@ func (transport *socks5UDP) RoundTrip(ctx context.Context, packet []byte) ([]byt
 	// Once a gateway answers, keep it pinned for the lifetime of the IKE SA.
 	if !transport.floated && requestHeader.Exchange == exchangeIKEInit && requestHeader.MessageID == 0 && len(transport.remotes) > 1 {
 		var lastErr error
+		var cookieResponse []byte
 		for _, candidate := range transport.remotes {
 			transport.remote = cloneUDPAddr(candidate)
 			response, attemptErr := transport.roundTripLocked(ctx, packet, requestHeader)
 			if attemptErr == nil {
+				if ikeInitResponseHasCookie(response) {
+					if cookieResponse == nil {
+						cookieResponse = append([]byte(nil), response...)
+					}
+					continue
+				}
 				return response, nil
 			}
 			lastErr = attemptErr
@@ -587,6 +594,9 @@ func (transport *socks5UDP) RoundTrip(ctx context.Context, packet []byte) ([]byt
 				return nil, attemptErr
 			}
 		}
+		if cookieResponse != nil {
+			return cookieResponse, nil
+		}
 		return nil, fmt.Errorf("ike: all %d resolved ePDG addresses timed out: %w", len(transport.remotes), lastErr)
 	}
 	return transport.roundTripLocked(ctx, packet, requestHeader)
@@ -824,11 +834,34 @@ func ikeResponseMatchesRequest(
 	}
 	var zeroSPI [8]byte
 	if request.ResponderSPI == zeroSPI {
-		return response.ResponderSPI != zeroSPI
+		if response.ResponderSPI != zeroSPI {
+			return true
+		}
+		return response.Exchange == exchangeIKEInit &&
+			response.MessageID == 0 &&
+			ikeInitResponseHasCookie(packet)
 	}
 	return response.ResponderSPI == request.ResponderSPI
 }
 
+func ikeInitResponseHasCookie(packet []byte) bool {
+	header, body, err := parseIKEPacket(packet)
+	if err != nil || header.Exchange != exchangeIKEInit || header.MessageID != 0 {
+		return false
+	}
+	payloads, err := parsePayloadChain(header.NextPayload, body)
+	if err != nil {
+		return false
+	}
+	for _, item := range payloadsOfType(payloads, payloadNotify) {
+		kind, data, err := parseNotify(item)
+		if err == nil && kind == notifyCookie && len(data) > 0 {
+			return true
+		}
+	}
+	return false
+}
+
 func marshalSOCKS5Datagram(remote *net.UDPAddr, payload []byte) ([]byte, error) {
 	if remote == nil || remote.IP == nil || remote.Port < 1 || remote.Port > 65535 {
 		return nil, errors.New("ike: invalid SOCKS5 UDP destination")

internal/vowifi/ike/transport_test.go

@@ -145,6 +145,18 @@ func TestSOCKS5InitialExchangeFallsBackAcrossResolvedEPDGAddresses(t *testing.T)
 		Exchange:     exchangeIKEInit,
 		Flags:        flagResponse,
 	}.marshal([]byte("response"))
+	cookieFirst, cookieBody, err := marshalPayloadChain([]payload{
+		makeNotify(notifyCookie, []byte{0x10, 0x20, 0x30, 0x40}),
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	cookieResponse := ikeHeader{
+		InitiatorSPI: requestHeader.InitiatorSPI,
+		NextPayload:  cookieFirst,
+		Exchange:     exchangeIKEInit,
+		Flags:        flagResponse,
+	}.marshal(cookieBody)
 	serverDone := make(chan error, 1)
 	go func() {
 		buffer := make([]byte, 2048)
@@ -160,6 +172,14 @@ func TestSOCKS5InitialExchangeFallsBackAcrossResolvedEPDGAddresses(t *testing.T)
 				return
 			}
 			if !destination.IP.Equal(second.IP) {
+				cookieWire, marshalErr := marshalSOCKS5Datagram(first, cookieResponse)
+				if marshalErr == nil {
+					_, marshalErr = relay.WriteToUDP(cookieWire, peer)
+				}
+				if marshalErr != nil {
+					serverDone <- marshalErr
+					return
+				}
 				continue
 			}
 			wire, marshalErr := marshalSOCKS5Datagram(second, response)
@@ -186,6 +206,31 @@ func TestSOCKS5InitialExchangeFallsBackAcrossResolvedEPDGAddresses(t *testing.T)
 	}
 }
 
+func TestIKEResponseMatchesCookieChallengeWithZeroResponderSPI(t *testing.T) {
+	request := ikeHeader{
+		InitiatorSPI: [8]byte{1, 2, 3, 4, 5, 6, 7, 8},
+		Exchange:     exchangeIKEInit,
+		Flags:        flagInitiator,
+		MessageID:    0,
+	}
+	first, body, err := marshalPayloadChain([]payload{
+		makeNotify(notifyCookie, []byte{0x10, 0x20, 0x30, 0x40}),
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	response := ikeHeader{
+		InitiatorSPI: request.InitiatorSPI,
+		Exchange:     exchangeIKEInit,
+		Flags:        flagResponse,
+		MessageID:    0,
+		NextPayload:  first,
+	}.marshal(body)
+	if !ikeResponseMatchesRequest(response, request) {
+		t.Fatal("IKE COOKIE response with zero Responder SPI was rejected")
+	}
+}
+
 func TestSOCKS5RoundTripSkipsStaleAndESPDatagrams(t *testing.T) {
 	relay, err := net.ListenUDP(
 		"udp",

internal/vowifi/ike/wire.go

@@ -59,6 +59,7 @@ const (
 	notifyMOBIKESupported = 16396
 	notifyNATSource       = 16388
 	notifyNATDestination  = 16389
+	notifyCookie          = 16390
 	notifyEAPOnlyAuth     = 16417
 	notifyDeviceIdentity  = 41101
 	notifyInvalidKE       = 17