[netstack] Implement IP(V6)_RECVERR socket option.

PiperOrigin-RevId: 348055514

3 files changed, 127 insertions, 13 deletions

diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go
index 9faab4b9e..e5e247342 100644
--- a/pkg/tcpip/transport/packet/endpoint.go
+++ b/pkg/tcpip/transport/packet/endpoint.go
@@ -366,6 +366,13 @@ func (ep *endpoint) LastError() *tcpip.Error {
 	return err
 }
 
+// UpdateLastError implements tcpip.SocketOptionsHandler.UpdateLastError.
+func (ep *endpoint) UpdateLastError(err *tcpip.Error) {
+	ep.lastErrorMu.Lock()
+	ep.lastError = err
+	ep.lastErrorMu.Unlock()
+}
+
 // GetSockOpt implements tcpip.Endpoint.GetSockOpt.
 func (ep *endpoint) GetSockOpt(opt tcpip.GettableSocketOption) *tcpip.Error {
 	return tcpip.ErrNotSupported
diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go
index bb0795f78..2128206d7 100644
--- a/pkg/tcpip/transport/tcp/endpoint.go
+++ b/pkg/tcpip/transport/tcp/endpoint.go
@@ -1303,6 +1303,15 @@ func (e *endpoint) LastError() *tcpip.Error {
 	return e.lastErrorLocked()
 }
 
+// UpdateLastError implements tcpip.SocketOptionsHandler.UpdateLastError.
+func (e *endpoint) UpdateLastError(err *tcpip.Error) {
+	e.LockUser()
+	e.lastErrorMu.Lock()
+	e.lastError = err
+	e.lastErrorMu.Unlock()
+	e.UnlockUser()
+}
+
 // Read reads data from the endpoint.
 func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, *tcpip.Error) {
 	e.LockUser()
@@ -2708,6 +2717,41 @@ func (e *endpoint) enqueueSegment(s *segment) bool {
 	return true
 }
 
+func (e *endpoint) onICMPError(err *tcpip.Error, id stack.TransportEndpointID, errType byte, errCode byte, extra uint32, pkt *stack.PacketBuffer) {
+	// Update last error first.
+	e.lastErrorMu.Lock()
+	e.lastError = err
+	e.lastErrorMu.Unlock()
+
+	// Update the error queue if IP_RECVERR is enabled.
+	if e.SocketOptions().GetRecvError() {
+		e.SocketOptions().QueueErr(&tcpip.SockError{
+			Err:       err,
+			ErrOrigin: header.ICMPOriginFromNetProto(pkt.NetworkProtocolNumber),
+			ErrType:   errType,
+			ErrCode:   errCode,
+			ErrInfo:   extra,
+			// Linux passes the payload with the TCP header. We don't know if the TCP
+			// header even exists, it may not for fragmented packets.
+			Payload: pkt.Data.ToView(),
+			Dst: tcpip.FullAddress{
+				NIC:  pkt.NICID,
+				Addr: id.RemoteAddress,
+				Port: id.RemotePort,
+			},
+			Offender: tcpip.FullAddress{
+				NIC:  pkt.NICID,
+				Addr: id.LocalAddress,
+				Port: id.LocalPort,
+			},
+			NetProto: pkt.NetworkProtocolNumber,
+		})
+	}
+
+	// Notify of the error.
+	e.notifyProtocolGoroutine(notifyError)
+}
+
 // HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket.
 func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, pkt *stack.PacketBuffer) {
 	switch typ {
@@ -2722,16 +2766,10 @@ func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.C
 		e.notifyProtocolGoroutine(notifyMTUChanged)
 
 	case stack.ControlNoRoute:
-		e.lastErrorMu.Lock()
-		e.lastError = tcpip.ErrNoRoute
-		e.lastErrorMu.Unlock()
-		e.notifyProtocolGoroutine(notifyError)
+		e.onICMPError(tcpip.ErrNoRoute, id, byte(header.ICMPv4DstUnreachable), byte(header.ICMPv4HostUnreachable), extra, pkt)
 
 	case stack.ControlNetworkUnreachable:
-		e.lastErrorMu.Lock()
-		e.lastError = tcpip.ErrNetworkUnreachable
-		e.lastErrorMu.Unlock()
-		e.notifyProtocolGoroutine(notifyError)
+		e.onICMPError(tcpip.ErrNetworkUnreachable, id, byte(header.ICMPv6DstUnreachable), byte(header.ICMPv6NetworkUnreachable), extra, pkt)
 	}
 }
 
diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go
index 8e16c8435..d919fa011 100644
--- a/pkg/tcpip/transport/udp/endpoint.go
+++ b/pkg/tcpip/transport/udp/endpoint.go
@@ -226,6 +226,13 @@ func (e *endpoint) LastError() *tcpip.Error {
 	return err
 }
 
+// UpdateLastError implements tcpip.SocketOptionsHandler.UpdateLastError.
+func (e *endpoint) UpdateLastError(err *tcpip.Error) {
+	e.lastErrorMu.Lock()
+	e.lastError = err
+	e.lastErrorMu.Unlock()
+}
+
 // Abort implements stack.TransportEndpoint.Abort.
 func (e *endpoint) Abort() {
 	e.Close()
@@ -511,6 +518,20 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c
 	}
 	if len(v) > header.UDPMaximumPacketSize {
 		// Payload can't possibly fit in a packet.
+		so := e.SocketOptions()
+		if so.GetRecvError() {
+			so.QueueLocalErr(
+				tcpip.ErrMessageTooLong,
+				route.NetProto,
+				header.UDPMaximumPacketSize,
+				tcpip.FullAddress{
+					NIC:  route.NICID(),
+					Addr: route.RemoteAddress,
+					Port: dstPort,
+				},
+				v,
+			)
+		}
 		return 0, nil, tcpip.ErrMessageTooLong
 	}
 
@@ -1338,15 +1359,63 @@ func (e *endpoint) HandlePacket(id stack.TransportEndpointID, pkt *stack.PacketB
 	}
 }
 
+func (e *endpoint) onICMPError(err *tcpip.Error, id stack.TransportEndpointID, errType byte, errCode byte, extra uint32, pkt *stack.PacketBuffer) {
+	// Update last error first.
+	e.lastErrorMu.Lock()
+	e.lastError = err
+	e.lastErrorMu.Unlock()
+
+	// Update the error queue if IP_RECVERR is enabled.
+	if e.SocketOptions().GetRecvError() {
+		// Linux passes the payload without the UDP header.
+		var payload []byte
+		udp := header.UDP(pkt.Data.ToView())
+		if len(udp) >= header.UDPMinimumSize {
+			payload = udp.Payload()
+		}
+
+		e.SocketOptions().QueueErr(&tcpip.SockError{
+			Err:       err,
+			ErrOrigin: header.ICMPOriginFromNetProto(pkt.NetworkProtocolNumber),
+			ErrType:   errType,
+			ErrCode:   errCode,
+			ErrInfo:   extra,
+			Payload:   payload,
+			Dst: tcpip.FullAddress{
+				NIC:  pkt.NICID,
+				Addr: id.RemoteAddress,
+				Port: id.RemotePort,
+			},
+			Offender: tcpip.FullAddress{
+				NIC:  pkt.NICID,
+				Addr: id.LocalAddress,
+				Port: id.LocalPort,
+			},
+			NetProto: pkt.NetworkProtocolNumber,
+		})
+	}
+
+	// Notify of the error.
+	e.waiterQueue.Notify(waiter.EventErr)
+}
+
 // HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket.
 func (e *endpoint) HandleControlPacket(id stack.TransportEndpointID, typ stack.ControlType, extra uint32, pkt *stack.PacketBuffer) {
 	if typ == stack.ControlPortUnreachable {
 		if e.EndpointState() == StateConnected {
-			e.lastErrorMu.Lock()
-			e.lastError = tcpip.ErrConnectionRefused
-			e.lastErrorMu.Unlock()
-
-			e.waiterQueue.Notify(waiter.EventErr)
+			var errType byte
+			var errCode byte
+			switch pkt.NetworkProtocolNumber {
+			case header.IPv4ProtocolNumber:
+				errType = byte(header.ICMPv4DstUnreachable)
+				errCode = byte(header.ICMPv4PortUnreachable)
+			case header.IPv6ProtocolNumber:
+				errType = byte(header.ICMPv6DstUnreachable)
+				errCode = byte(header.ICMPv6PortUnreachable)
+			default:
+				panic(fmt.Sprintf("unsupported net proto for infering ICMP type and code: %d", pkt.NetworkProtocolNumber))
+			}
+			e.onICMPError(tcpip.ErrConnectionRefused, id, errType, errCode, extra, pkt)
 			return
 		}
 	}