UDP hole punching. Have server talk to client

Question

I been reading a lot on how to implement UDP hole punching but fore some reason I cannot make it work.

For those that are not familiar of what udp hole punching is here is my own definition:

The goal is to be able to transfer data between two clients (Client A and client B) with the help of a server. So client A connects to the server and sends its info. Client B does the same. The server has the nessesary info so that Client A is able to send data to Client B and vise versa . Therefore the server gives that info to both clients. Once both clients have that info about each other it is possible to start sending and receiving data between those clients without the help of the server.

My goal is to be able to do what I just described (udp hole punching). Before doing so I think it will be helpful to be able to connect from the server to the client. In order to do so I plan to send the server the info about the client. Once the server receives that info attempt to connect to the client from scratch. Once I am able to perform that I should have everything I need to start implementing the real udp hole punching.

Here is how I have things set up:

The top router has the server and bottom router connected to LAN ports. The bottom router (NAT) is connected to the top router via it's WAN port. And the client computer is connected to the bottom router to one of its LAN ports.

So in that connection the client is able to see the server but the server is not able to see the client.

So the algorithm I have done in pseudo code is:

• Client connects to server.
• Client send some UDP packages to the server in order to open some ports on the NAT
• Send information to the server on what ports the client is listening to.
• Once the server receives that info attempt to connect to the client from scratch.

Here is the implementation in code:

Server:

static void Main()
{     
    /* Part 1 receive data from client */
    UdpClient listener = new UdpClient(11000);
    IPEndPoint groupEP = new IPEndPoint(IPAddress.Any, 11000);
    string received_data;
    byte[] receive_byte_array = listener.Receive(ref groupEP);       
    received_data = Encoding.ASCII.GetString(receive_byte_array, 0, receive_byte_array.Length);

    // get info
    var ip = groupEP.Address.ToString();
    var port = groupEP.Port;

    /* Part 2 atempt to connect to client from scratch */
    // now atempt to send data to client from scratch once we have the info       
    Socket sendSocket = new Socket(AddressFamily.InterNetwork, SocketType.Dgram, ProtocolType.Udp);
    IPEndPoint endPointClient = new IPEndPoint(IPAddress.Parse(ip), port);
    sendSocket.SendTo(Encoding.ASCII.GetBytes("Hello"), endPointClient);
}

Client:

static void Main(string[] args)
{
    /* Part 1 send info to server */
    Socket sending_socket = new Socket(AddressFamily.InterNetwork, SocketType.Dgram,  ProtocolType.Udp);
    IPAddress send_to_address = IPAddress.Parse("192.168.0.132");
    IPEndPoint sending_end_point = new IPEndPoint(send_to_address, 11000);
    sending_socket.SendTo(Encoding.ASCII.GetBytes("Test"), sending_end_point);

    // get info
    var port = sending_socket.LocalEndPoint.ToString().Split(':')[1];

    /* Part 2 receive data from server */
    IPEndPoint groupEP = new IPEndPoint(IPAddress.Any, int.Parse(port));
    byte[] buffer = new byte[1024];
    sending_socket.Receive(buffer);
} 

For some reason it worked a few times! It works when the client receives data successfully on the line: sending_socket.Receive(buffer);

Things to note: If on the server on the second part I used the instance variable listner instead of creating the new variable sendSocket and send the bytes through that variable the client is able to receive the data being sent. Remember that the second part of the server is going to be implemented by a second client B that's why I am initializing variables again from scratch...

---

Edit:

Here is a different way of looking at the same problem. When I initialize a new object instead of using the same object the client does not receives the response.

I have a object of type UdpClient. I am able to send data with that object to the other peer. If I create another object of the same type with the same properties and attempt to send data it does not work! I might be missing to initialize some variables. I am able to set private variables with reflection so I should not have a problem. anyways here is the server code:

public static void Main()
{
    // wait for client to send data
    UdpClient listener = new UdpClient(11000);
    IPEndPoint groupEP = new IPEndPoint(IPAddress.Any, 11000);        
    byte[] receive_byte_array = listener.Receive(ref groupEP);

    // connect so that we are able to send data back
    listener.Connect(groupEP);

    byte[] dataToSend = new byte[] { 1, 2, 3, 4, 5 };

    // now let's atempt to reply back

    // this part does not work!
    UdpClient newClient = CopyUdpClient(listener, groupEP);
    newClient.Send(dataToSend, dataToSend.Length);

    // this part works!
    listener.Send(dataToSend, dataToSend.Length);
}

static UdpClient CopyUdpClient(UdpClient client, IPEndPoint groupEP)
{
    var ip = groupEP.Address.ToString();
    var port = groupEP.Port;
    var newUdpClient = new UdpClient(ip, port);
    return newUdpClient;
}

the client code basically sends data to the server and then waits for a response:

    string ipOfServer = "192.168.0.132";
    int portServerIsListeningOn = 11000;

    // send data to server
    Socket sending_socket = new Socket(AddressFamily.InterNetwork, SocketType.Dgram, ProtocolType.Udp);
    IPAddress send_to_address = IPAddress.Parse(ipOfServer);
    IPEndPoint sending_end_point = new IPEndPoint(send_to_address, portServerIsListeningOn);
    sending_socket.SendTo(Encoding.ASCII.GetBytes("Test"), sending_end_point);

    // get info
    var port = sending_socket.LocalEndPoint.ToString().Split(':')[1];

    // now wait for server to send data back
    IPEndPoint groupEP = new IPEndPoint(IPAddress.Any, int.Parse(port));
    byte[] buffer = new byte[1024];
    sending_socket.Receive(buffer); // <----- keeps waiting in here :(

note that the client is behind a router (NAT) otherwise I will not have this problem. The reason why I will like to copy udpClient is so that I can send that variable to another computer enabling the other computer to send data to the client.

So my question is why is the original object listener able to send data but newClient is not able to? The client keeps waiting at line sending_socket.Receive(buffer); even after the server executes the line: newClient.Send(dataToSend, dataToSend.Length);. the client successfully receives data when listener sends the data but not newClient. Why is this if both variables have the same destination IP and port? how do the variables differ?

Note: If the server and client are on the same network then the copy works and variable newClient is able to send data to the client. To simulate this problem the client must be behind a NAT (router). An example of such network may consist of two routers. let's call them router X and router Y. You also need a Server call that S. and a client C. so S can be connected to one of the LAN ports of X. C can be connected to one of the LAN ports of Y. Finally connect the WAN port of Y to one of the LAN ports of X.

Answer 1

Hmm, I think you are confusing several things here. For one thing, it's really called UDP hole punching. Let me try to explain how this should work.

NAT routers usually do port mapping when forwarding packets from the inside private network to the outside internet.

Say you created a UDP socket on a machine behind NAT, and sent a datagram to some external IP/port. When the IP packet carrying that datagram leaves the sending machine, its IP header has the source address field set to local not-globally-routable private IP address (like 192.168.1.15), and UDP header has the source port field set to whatever port was assigned to the socket (either explicitly via binding, or implicitly picked by the OS from the ephemeral ports). I'll call this source port number P1.

Then when the NAT router sends that packet out on the outside network, it overwrites the source IP address to its own external IP address (otherwise there's no way to route packets back), and often overwrites the source UDP port to some other value (maybe because some other host on the private network uses the same source port, which creates ambiguity). The mapping between the original source port and that new port number (let's label it P2) is preserved in the router to match return packets. This mapping might also be specific to the target IP address and target UDP port.

So now you have "punched a hole" in the router - UDP packets sent back to the router to port P2 are forwarded to internal machine on UDP port P1. Again, depending on NAT implementation, this could be restricted to only packets from the original target IP address and target UDP port.

For client-to-client communication you'll have to tell external IP/port of one to the other through the server, hoping that the NAT router maps same internal source ports to same external source ports. Then the clients will send packets to each other using those.

Hope this helps.