[comp.protocols.tcp-ip] Reducing the risks when conencting to an internet

neil@stl.stc.co.uk (11/23/90)

We are starting the planning process to connect our large internal network
to an internet. We wish to reduce the risk of unwanted external vistors
as far as possible, whilst allowing our internal users access to other
hosts.

One possibility is to acquire a device to sit between our backbone and
the outside world. This device would be programmable so that I could
authorise connections on a per host, per port, per direction basis.

So, for example I have network 128.199, I could have an authorisation
file like:-

# Source		Dest		Comment
# host.port		host.port

*.smtp			mymailgw.smtp	Mail service inbound
mymailgw.smtp		*.smtp		Mail service outbound
128.199.200.*.ftp	*.ftp		FTP
128.199.200.*.telnet	*.telnet	Telnet


So, from the above, my mail gateway could send out to anybody, and anybody
can send to my mail gateway (and no where else). Any machine on my local
200 subnet can ftp or telnet to anywhere, but no inbound conenctions are
allowed.


Is this possible, or are we thinking in completely the wrong way ?
At this stage it isn't necessary that the device exists in the UK,
we can import if necessary.

The device probably needs to understand name servers as well.

We also have an eye on the future, so the Vendor would need to offer
support for OSI protocols.

Neil Todd			| ..In general, it is best to assume that the
PSI%234237100122::neil		| network is filled with malevolent entities
neil@stl.stc.co.uk		| that will send in packets designed to have
STC Technology Ltd		| the worst possible effect...

barmar@think.com (Barry Margolin) (11/25/90)

In article <3768@stl.stc.co.uk> neil@stl.stc.co.uk () writes:
>So, for example I have network 128.199, I could have an authorisation
>file like:-
>
># Source		Dest		Comment
># host.port		host.port
>
>*.smtp			mymailgw.smtp	Mail service inbound
>mymailgw.smtp		*.smtp		Mail service outbound
>128.199.200.*.ftp	*.ftp		FTP
>128.199.200.*.telnet	*.telnet	Telnet

The source port for connections is generally *not* the protocol's
well-known port.  The well-known port is normally only used as the
destination port.  The source port is usually a random port above 1024.

>Is this possible, or are we thinking in completely the wrong way ?
>At this stage it isn't necessary that the device exists in the UK,
>we can import if necessary.

cisco Gateway Servers can do packet filtering based on addresses and port
numbers.

>We also have an eye on the future, so the Vendor would need to offer
>support for OSI protocols.

cisco supports OSI protocols, although it doesn't yet seem to support them
in its filtering specifications.  It will probably come at some time,
thoug.
--
Barry Margolin, Thinking Machines Corp.

barmar@think.com
{uunet,harvard}!think!barmar

kmeyer@jerico.usc.edu (Kraig R. Meyer) (11/25/90)

In article <3768@stl.stc.co.uk>, neil@stl.stc.co.uk writes:
|>...We wish to reduce the risk of unwanted external vistors
|>as far as possible, whilst allowing our internal users access to other
|>hosts.
|>
|>One possibility is to acquire a device to sit between our backbone and
|>the outside world. This device would be programmable so that I could
|>authorise connections on a per host, per port, per direction basis.

Depending on how much isolation you want from the outside world, another
option (besides using filtering at the TCP/IP level in a router) is to
use a unix box as an application level gateway.  This is definitely an
inconvenience to your users, which may or may not be a good thing depending
on what your isolation goals are and how paranoid you are.

For example, you can configure sendmail to forward mail in both directions
and then only give accounts on the gateway unix machine to those people who 
you want to allow ftp/telnet/etc access to the outside network.  That makes
access on a per-person basis, rather than on a per-IP-address basis.
Filtering in the way you've described would not, for example, prevent an
internal user from spoofing IP addresses (if that is an issue).
---------------------------------------------------------------------------
| Kraig R. Meyer		   		          kmeyer@usc.edu  |
| University of Southern California                          Los Angeles  |
---------------------------------------------------------------------------