Implement IP_TTL.

Also change the default TTL to 64 to match Linux.

PiperOrigin-RevId: 273430341

9 files changed, 131 insertions, 23 deletions

diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go
index a3a910d41..e17e737e4 100644
--- a/pkg/tcpip/transport/icmp/endpoint.go
+++ b/pkg/tcpip/transport/icmp/endpoint.go
@@ -84,6 +84,7 @@ type endpoint struct {
 	// NIC.
 	regNICID tcpip.NICID
 	route    stack.Route `state:"manual"`
+	ttl      uint8
 }
 
 func newEndpoint(stack *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
@@ -296,10 +297,10 @@ func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c
 
 	switch e.netProto {
 	case header.IPv4ProtocolNumber:
-		err = send4(route, e.id.LocalPort, v)
+		err = send4(route, e.id.LocalPort, v, e.ttl)
 
 	case header.IPv6ProtocolNumber:
-		err = send6(route, e.id.LocalPort, v)
+		err = send6(route, e.id.LocalPort, v, e.ttl)
 	}
 
 	if err != nil {
@@ -314,8 +315,15 @@ func (e *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) {
 	return 0, tcpip.ControlMessages{}, nil
 }
 
-// SetSockOpt sets a socket option. Currently not supported.
+// SetSockOpt sets a socket option.
 func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
+	switch o := opt.(type) {
+	case tcpip.TTLOption:
+		e.mu.Lock()
+		e.ttl = uint8(o)
+		e.mu.Unlock()
+	}
+
 	return nil
 }
 
@@ -362,12 +370,18 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
 		*o = 0
 		return nil
 
+	case *tcpip.TTLOption:
+		e.rcvMu.Lock()
+		*o = tcpip.TTLOption(e.ttl)
+		e.rcvMu.Unlock()
+		return nil
+
 	default:
 		return tcpip.ErrUnknownProtocolOption
 	}
 }
 
-func send4(r *stack.Route, ident uint16, data buffer.View) *tcpip.Error {
+func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Error {
 	if len(data) < header.ICMPv4MinimumSize {
 		return tcpip.ErrInvalidEndpointState
 	}
@@ -389,10 +403,10 @@ func send4(r *stack.Route, ident uint16, data buffer.View) *tcpip.Error {
 	icmpv4.SetChecksum(0)
 	icmpv4.SetChecksum(^header.Checksum(icmpv4, header.Checksum(data, 0)))
 
-	return r.WritePacket(nil /* gso */, hdr, data.ToVectorisedView(), header.ICMPv4ProtocolNumber, r.DefaultTTL())
+	return r.WritePacket(nil /* gso */, hdr, data.ToVectorisedView(), header.ICMPv4ProtocolNumber, ttl, ttl == 0 /* useDefaultTTL */)
 }
 
-func send6(r *stack.Route, ident uint16, data buffer.View) *tcpip.Error {
+func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Error {
 	if len(data) < header.ICMPv6EchoMinimumSize {
 		return tcpip.ErrInvalidEndpointState
 	}
@@ -412,7 +426,7 @@ func send6(r *stack.Route, ident uint16, data buffer.View) *tcpip.Error {
 	icmpv6.SetChecksum(0)
 	icmpv6.SetChecksum(^header.Checksum(icmpv6, header.Checksum(data, 0)))
 
-	return r.WritePacket(nil /* gso */, hdr, data.ToVectorisedView(), header.ICMPv6ProtocolNumber, r.DefaultTTL())
+	return r.WritePacket(nil /* gso */, hdr, data.ToVectorisedView(), header.ICMPv6ProtocolNumber, ttl, ttl == 0 /* useDefaultTTL */)
 }
 
 func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress, allowMismatch bool) (tcpip.NetworkProtocolNumber, *tcpip.Error) {
diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go
index a02731a5d..8ae132efa 100644
--- a/pkg/tcpip/transport/raw/endpoint.go
+++ b/pkg/tcpip/transport/raw/endpoint.go
@@ -332,7 +332,7 @@ func (ep *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64,
 			break
 		}
 		hdr := buffer.NewPrependable(len(payloadBytes) + int(route.MaxHeaderLength()))
-		if err := route.WritePacket(nil /* gso */, hdr, buffer.View(payloadBytes).ToVectorisedView(), ep.transProto, route.DefaultTTL()); err != nil {
+		if err := route.WritePacket(nil /* gso */, hdr, buffer.View(payloadBytes).ToVectorisedView(), ep.transProto, 0, true /* useDefaultTTL */); err != nil {
 			return 0, nil, err
 		}
 
diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go
index 3ae4a5426..b8b4bcee8 100644
--- a/pkg/tcpip/transport/tcp/accept.go
+++ b/pkg/tcpip/transport/tcp/accept.go
@@ -439,7 +439,7 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) {
 				TSEcr: opts.TSVal,
 				MSS:   uint16(mss),
 			}
-			sendSynTCP(&s.route, s.id, header.TCPFlagSyn|header.TCPFlagAck, cookie, s.sequenceNumber+1, ctx.rcvWnd, synOpts)
+			sendSynTCP(&s.route, s.id, e.ttl, header.TCPFlagSyn|header.TCPFlagAck, cookie, s.sequenceNumber+1, ctx.rcvWnd, synOpts)
 			e.stack.Stats().TCP.ListenOverflowSynCookieSent.Increment()
 		}
 
diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go
index 21038a65a..1d6e7f5f3 100644
--- a/pkg/tcpip/transport/tcp/connect.go
+++ b/pkg/tcpip/transport/tcp/connect.go
@@ -238,6 +238,7 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error {
 	h.state = handshakeSynRcvd
 	h.ep.mu.Lock()
 	h.ep.state = StateSynRecv
+	ttl := h.ep.ttl
 	h.ep.mu.Unlock()
 	synOpts := header.TCPSynOptions{
 		WS:    int(h.effectiveRcvWndScale()),
@@ -251,7 +252,10 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error {
 		SACKPermitted: rcvSynOpts.SACKPermitted,
 		MSS:           h.ep.amss,
 	}
-	sendSynTCP(&s.route, h.ep.id, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts)
+	if ttl == 0 {
+		ttl = s.route.DefaultTTL()
+	}
+	sendSynTCP(&s.route, h.ep.id, ttl, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts)
 
 	return nil
 }
@@ -296,7 +300,7 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error {
 			SACKPermitted: h.ep.sackPermitted,
 			MSS:           h.ep.amss,
 		}
-		sendSynTCP(&s.route, h.ep.id, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts)
+		sendSynTCP(&s.route, h.ep.id, h.ep.ttl, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts)
 		return nil
 	}
 
@@ -460,7 +464,7 @@ func (h *handshake) execute() *tcpip.Error {
 			synOpts.WS = -1
 		}
 	}
-	sendSynTCP(&h.ep.route, h.ep.id, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts)
+	sendSynTCP(&h.ep.route, h.ep.id, h.ep.ttl, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts)
 	for h.state != handshakeCompleted {
 		switch index, _ := s.Fetch(true); index {
 		case wakerForResend:
@@ -469,7 +473,7 @@ func (h *handshake) execute() *tcpip.Error {
 				return tcpip.ErrTimeout
 			}
 			rt.Reset(timeOut)
-			sendSynTCP(&h.ep.route, h.ep.id, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts)
+			sendSynTCP(&h.ep.route, h.ep.id, h.ep.ttl, h.flags, h.iss, h.ackNum, h.rcvWnd, synOpts)
 
 		case wakerForNotification:
 			n := h.ep.fetchNotifications()
@@ -579,9 +583,9 @@ func makeSynOptions(opts header.TCPSynOptions) []byte {
 	return options[:offset]
 }
 
-func sendSynTCP(r *stack.Route, id stack.TransportEndpointID, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts header.TCPSynOptions) *tcpip.Error {
+func sendSynTCP(r *stack.Route, id stack.TransportEndpointID, ttl uint8, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size, opts header.TCPSynOptions) *tcpip.Error {
 	options := makeSynOptions(opts)
-	err := sendTCP(r, id, buffer.VectorisedView{}, r.DefaultTTL(), flags, seq, ack, rcvWnd, options, nil)
+	err := sendTCP(r, id, buffer.VectorisedView{}, ttl, flags, seq, ack, rcvWnd, options, nil)
 	putOptions(options)
 	return err
 }
@@ -629,7 +633,7 @@ func sendTCP(r *stack.Route, id stack.TransportEndpointID, data buffer.Vectorise
 		r.Stats().TCP.ResetsSent.Increment()
 	}
 
-	return r.WritePacket(gso, hdr, data, ProtocolNumber, ttl)
+	return r.WritePacket(gso, hdr, data, ProtocolNumber, ttl, ttl == 0 /* useDefaultTTL */)
 }
 
 // makeOptions makes an options slice.
@@ -678,7 +682,7 @@ func (e *endpoint) sendRaw(data buffer.VectorisedView, flags byte, seq, ack seqn
 		sackBlocks = e.sack.Blocks[:e.sack.NumBlocks]
 	}
 	options := e.makeOptions(sackBlocks)
-	err := sendTCP(&e.route, e.id, data, e.route.DefaultTTL(), flags, seq, ack, rcvWnd, options, e.gso)
+	err := sendTCP(&e.route, e.id, data, e.ttl, flags, seq, ack, rcvWnd, options, e.gso)
 	putOptions(options)
 	return err
 }
diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go
index f9d5e0085..83d92b3e1 100644
--- a/pkg/tcpip/transport/tcp/endpoint.go
+++ b/pkg/tcpip/transport/tcp/endpoint.go
@@ -228,6 +228,7 @@ type endpoint struct {
 	isRegistered      bool
 	boundNICID        tcpip.NICID `state:"manual"`
 	route             stack.Route `state:"manual"`
+	ttl               uint8
 	v6only            bool
 	isConnectNotified bool
 	// TCP should never broadcast but Linux nevertheless supports enabling/
@@ -1116,6 +1117,12 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
 		e.v6only = v != 0
 		return nil
 
+	case tcpip.TTLOption:
+		e.mu.Lock()
+		e.ttl = uint8(v)
+		e.mu.Unlock()
+		return nil
+
 	case tcpip.KeepaliveEnabledOption:
 		e.keepalive.Lock()
 		e.keepalive.enabled = v != 0
@@ -1313,6 +1320,12 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
 		}
 		return nil
 
+	case *tcpip.TTLOption:
+		e.mu.Lock()
+		*o = tcpip.TTLOption(e.ttl)
+		e.mu.Unlock()
+		return nil
+
 	case *tcpip.TCPInfoOption:
 		*o = tcpip.TCPInfoOption{}
 		e.mu.RLock()
diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go
index 089826a88..a86123829 100644
--- a/pkg/tcpip/transport/tcp/tcp_test.go
+++ b/pkg/tcpip/transport/tcp/tcp_test.go
@@ -1752,6 +1752,34 @@ func TestSendGreaterThanMTU(t *testing.T) {
 	testBrokenUpWrite(t, c, maxPayload)
 }
 
+func TestSetTTL(t *testing.T) {
+	for _, wantTTL := range []uint8{1, 2, 50, 64, 128, 254, 255} {
+		t.Run(fmt.Sprintf("TTL:%d", wantTTL), func(t *testing.T) {
+			c := context.New(t, 65535)
+			defer c.Cleanup()
+
+			var err *tcpip.Error
+			c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{})
+			if err != nil {
+				t.Fatalf("NewEndpoint failed: %v", err)
+			}
+
+			if err := c.EP.SetSockOpt(tcpip.TTLOption(wantTTL)); err != nil {
+				t.Fatalf("SetSockOpt failed: %v", err)
+			}
+
+			if err := c.EP.Connect(tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrConnectStarted {
+				t.Fatalf("Unexpected return value from Connect: %v", err)
+			}
+
+			// Receive SYN packet.
+			b := c.GetPacket()
+
+			checker.IPv4(t, b, checker.TTL(wantTTL))
+		})
+	}
+}
+
 func TestActiveSendMSSLessThanMTU(t *testing.T) {
 	const maxPayload = 100
 	c := context.New(t, 65535)
diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go
index 52f5af777..8ae83437e 100644
--- a/pkg/tcpip/transport/udp/endpoint.go
+++ b/pkg/tcpip/transport/udp/endpoint.go
@@ -83,6 +83,7 @@ type endpoint struct {
 	route          stack.Route `state:"manual"`
 	dstPort        uint16
 	v6only         bool
+	ttl            uint8
 	multicastTTL   uint8
 	multicastAddr  tcpip.Address
 	multicastNICID tcpip.NICID
@@ -374,12 +375,16 @@ func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c
 		return 0, nil, tcpip.ErrMessageTooLong
 	}
 
-	ttl := route.DefaultTTL()
+	ttl := e.ttl
+	useDefaultTTL := ttl == 0
+
 	if header.IsV4MulticastAddress(route.RemoteAddress) || header.IsV6MulticastAddress(route.RemoteAddress) {
 		ttl = e.multicastTTL
+		// Multicast allows a 0 TTL.
+		useDefaultTTL = false
 	}
 
-	if err := sendUDP(route, buffer.View(v).ToVectorisedView(), e.id.LocalPort, dstPort, ttl); err != nil {
+	if err := sendUDP(route, buffer.View(v).ToVectorisedView(), e.id.LocalPort, dstPort, ttl, useDefaultTTL); err != nil {
 		return 0, nil, err
 	}
 	return int64(len(v)), nil, nil
@@ -414,6 +419,11 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error {
 
 		e.v6only = v != 0
 
+	case tcpip.TTLOption:
+		e.mu.Lock()
+		e.ttl = uint8(v)
+		e.mu.Unlock()
+
 	case tcpip.MulticastTTLOption:
 		e.mu.Lock()
 		e.multicastTTL = uint8(v)
@@ -628,6 +638,12 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
 		}
 		return nil
 
+	case *tcpip.TTLOption:
+		e.mu.Lock()
+		*o = tcpip.TTLOption(e.ttl)
+		e.mu.Unlock()
+		return nil
+
 	case *tcpip.MulticastTTLOption:
 		e.mu.Lock()
 		*o = tcpip.MulticastTTLOption(e.multicastTTL)
@@ -694,7 +710,7 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
 
 // sendUDP sends a UDP segment via the provided network endpoint and under the
 // provided identity.
-func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort uint16, ttl uint8) *tcpip.Error {
+func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort uint16, ttl uint8, useDefaultTTL bool) *tcpip.Error {
 	// Allocate a buffer for the UDP header.
 	hdr := buffer.NewPrependable(header.UDPMinimumSize + int(r.MaxHeaderLength()))
 
@@ -720,7 +736,7 @@ func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort u
 	// Track count of packets sent.
 	r.Stats().UDP.PacketsSent.Increment()
 
-	return r.WritePacket(nil /* gso */, hdr, data, ProtocolNumber, ttl)
+	return r.WritePacket(nil /* gso */, hdr, data, ProtocolNumber, ttl, useDefaultTTL)
 }
 
 func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress, allowMismatch bool) (tcpip.NetworkProtocolNumber, *tcpip.Error) {
diff --git a/pkg/tcpip/transport/udp/protocol.go b/pkg/tcpip/transport/udp/protocol.go
index f5cc932dd..0b24bbc85 100644
--- a/pkg/tcpip/transport/udp/protocol.go
+++ b/pkg/tcpip/transport/udp/protocol.go
@@ -130,7 +130,7 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans
 		pkt.SetType(header.ICMPv4DstUnreachable)
 		pkt.SetCode(header.ICMPv4PortUnreachable)
 		pkt.SetChecksum(header.ICMPv4Checksum(pkt, payload))
-		r.WritePacket(nil /* gso */, hdr, payload, header.ICMPv4ProtocolNumber, r.DefaultTTL())
+		r.WritePacket(nil /* gso */, hdr, payload, header.ICMPv4ProtocolNumber, 0, true /* useDefaultTTL */)
 
 	case header.IPv6AddressSize:
 		if !r.Stack().AllowICMPMessage() {
@@ -164,7 +164,7 @@ func (p *protocol) HandleUnknownDestinationPacket(r *stack.Route, id stack.Trans
 		pkt.SetType(header.ICMPv6DstUnreachable)
 		pkt.SetCode(header.ICMPv6PortUnreachable)
 		pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, payload))
-		r.WritePacket(nil /* gso */, hdr, payload, header.ICMPv6ProtocolNumber, r.DefaultTTL())
+		r.WritePacket(nil /* gso */, hdr, payload, header.ICMPv6ProtocolNumber, 0, true /* useDefaultTTL */)
 	}
 	return true
 }
diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go
index d1ba8d2ce..faa728b68 100644
--- a/pkg/tcpip/transport/udp/udp_test.go
+++ b/pkg/tcpip/transport/udp/udp_test.go
@@ -1207,6 +1207,39 @@ func TestTTL(t *testing.T) {
 	}
 }
 
+func TestSetTTL(t *testing.T) {
+	for _, flow := range []testFlow{unicastV4, unicastV4in6, unicastV6, unicastV6Only, broadcast, broadcastIn6} {
+		t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) {
+			for _, wantTTL := range []uint8{1, 2, 50, 64, 128, 254, 255} {
+				t.Run(fmt.Sprintf("TTL:%d", wantTTL), func(t *testing.T) {
+					c := newDualTestContext(t, defaultMTU)
+					defer c.cleanup()
+
+					c.createEndpointForFlow(flow)
+
+					if err := c.ep.SetSockOpt(tcpip.TTLOption(wantTTL)); err != nil {
+						c.t.Fatalf("SetSockOpt failed: %v", err)
+					}
+
+					var p stack.NetworkProtocol
+					if flow.isV4() {
+						p = ipv4.NewProtocol()
+					} else {
+						p = ipv6.NewProtocol()
+					}
+					ep, err := p.NewEndpoint(0, tcpip.AddressWithPrefix{}, nil, nil, nil)
+					if err != nil {
+						t.Fatal(err)
+					}
+					ep.Close()
+
+					testWrite(c, flow, checker.TTL(wantTTL))
+				})
+			}
+		})
+	}
+}
+
 func TestMulticastInterfaceOption(t *testing.T) {
 	for _, flow := range []testFlow{multicastV4, multicastV4in6, multicastV6, multicastV6Only} {
 		t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) {