SignalR: how to enforce authentication/terminate hub connections server side

Question

This question is related to another thread, you can read here: Forms authentication with SignalR where I tried with the help and patience of the user dfowler to understand how enforce forms ASP .NET Forms Authentication on a SignalR Hub.

Description of the problem: I want that only authenticated users can connect the SignalR Hub and receive/send messages.

Intrusion scenario: the intruder can potentially capture/access the HTML and Javascripts of the web page accessing the temporary files on a client computer. This intruder can therefore know all the details (methods, hub names etc) required to set up/use a connection to the Hub. A proposed solution from dfowler is implementing IConnect:

You would implement IConnected and write the following code in Connect if(!Context.User.Identity.IsAuthenticated) throw new Exception("GTFO");

Therefore I tried with something like this

  public System.Threading.Tasks.Task Connect()
    {
        if (!Context.User.Identity.IsAuthenticated
            || !(Context.User.IsInRole("role1") || Context.User.IsInRole("role2")
            ))
            throw new Exception("User not authorized");
        return null;
    }

The problem, once tested, is that when the Connect method is being called, the connection has been already established and plain throwing the exception will not help (if I got if properly infact Connect should be used to send a message to the client at connection, throwing an exception will just yield in a welcome message not sent).

In facts, from my tests, the clients can still read all the messages (and also send them).

Now, approaches which come to my mind:

1. Perfect solution: reject or terminate the connection on server side: no clue how to do this in SignalR (I tried to find a method in the API but no luck)
2. check if the user is part of a group to avoid receiving/sending messages to him (but this is still prone to flooding/DOS attacks)
3. Sending a message to the client to disconnect: obviously does not help in case I am fighting an intruder.

Any other approach? Any way to terminate the connection on server side or should be accepted that the only real authentication is the one of the host webpage (leaving open the door to all the signalR client attacks?)

EDIT

Here is the sequence of client - server communication when I use the IConnect.Connect method throwing unconditionally an exception (browser IE9):

It looks like the foreverFrame fails but the longPolling fallback is being established and works anyway - this after throwing the error captured in the Javascript by the block

 if (connection.state === signalR.connectionState.connecting) {
            // Connection hasn't been started yet
            throw "SignalR: Connection has not been fully initialized. 
    Use .start().done() or .start().fail() to run logic after 
    the connection has started.";
        }

Answer 1

We have an issue where we need to allow blocking the connection altogether. Right now you'll have to guard each method. It's not the cleanest but for 1.0 alpha1 we'll have some mechanism for doing this.

One additional problem is that it's the same connection for all hubs so you can't reject a connection for a specific hub.

EDIT

Actually if you throw it does end the connection as far as my testing goes. What behavior are you seeing?