Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!uwm.edu!zaphod.mps.ohio-state.edu!lavaca.uh.edu!uhnix1!sugar!ficc!ihsan
From: ihsan@ficc.ferranti.com (jaleel ihsan)
Newsgroups: comp.protocols.tcp-ip
Subject: Listening to a specific IP address (was: How long before I can reopen a closed socket?
Message-ID: <3L94GT5@ccs.ferranti.com>
Date: 27 Jun 90 04:06:10 GMT
References: <12599677104.19.LYNCH@A.ISI.EDU>
Organization: Ferranti International Controls Corporation
Lines: 29

> 
> Jaleel,  You can certainly build a server to do exactly as you
> described.  It would only listen for specific ports from specific
> addresses.  The algorithm for selecting port numbers for TCP and
> UDP that are not the "well known port numbers" is entirely up
> to the hosts in question.  You are looking for a way to restrict
> service and this is one (weak) way to do so.  "Weak" in the security
> sense.  
> 
> Dan

Thanks for the information, Dan.

You see, we are working with Intel's RMX real-time executive, which I
do not think supports fork'ing.  But even if it did, our application
design does not support it.  So I was thinking of having a dedicated 
and a specific server for each one of the limited number of clients.

But how to channel a client to the proper server ?

A server could bind the listening socket to a well known local port (all
servers use the same well known local port so that multiple servers are
transparent to the clients), local address, a wild-card remote port, and 
a specific remote IP address.

Theoretically, it should (!!!) work. Can any one shoot some holes in it ?
Would any authority at the IAB care to comment ? Practically, ... well ...
minor details,  minor details !!!

Jaleel