Two ISPs, One NAT'ed Internal Subnet, Firewall Policys
Daniel Bartlett
bartlett.d at gmail.com
Tue Sep 21 09:11:09 UTC 2004
Hi,
> Presumably you have separate (possibly dynamic) IP addresses from the two
> ISPs, and intend to use the connections for outbound (client) traffic,
> and not inbound (e.g., webserving) traffic? If so, an active/passive
> configuration can be rather simple.
We are dealing with inbound mail(smtp/pop3), http & https traffic. One
ISP is a static subnet of 8 IPs the other dynamic(1 IP). I have been
thinking that getting a dynamic DNS account and updating on the
failover(and regularly for the dynamic connection).
>
> > Updating IPtables at the same time.
>
> If the connections to the ISPs are configured to use two different
> interfaces, the netfilter configuration can be static; have separate
> SNAT rules for each interface, and update the default route instead,
> which amounts to
>
> /sbin/ip route replace default via $GATEWAY dev $DEV
I think this would work if not for the inbound traffic, if a remote
host opens a connection on one IP and the data goes back on the other
packet headers will mean the packet just gets dropped.(Although a
mangle rule might work here...if maybe quite confusing to setup)
>
> That's the simple part. The more interesting part is detecting the "dead"
> gateway, for some definition of "dead". In the typical external ADSL
> or cable modem configuration, there can be a failure of communication
> between the Linux firewall and the ADSL/cable router, between the
> ADSL/cable router and the ISP, and between the ISP and the wider Internet
> (usually due to routing screwups, etc., at the ISP). So detecting whether
> the local gateway (i.e., the ADSL/cable router) is alive is of only
> marginal utility; one usually wants to detect reachability of the wider
> Internet, via pinging highly-available sites, or an equivalent method.
>
> Then there is the issue of DNS resolution. For many clients, if the ISP's
> DNS servers are not working, the route to the internet is again of marginal
> utility. One can configure DNS to use the nameservers of both ISP's, though
> that doesn't help with certain Byzantine failures (that seem to occur in
> real life), where one ISP's nameserver returns nonsense. For this and
> other reasons, it is generally desirable to give priority to the DNS server
> of the ISP that you are routing through, and a more active approach to
> DNS server monitoring is often used.
The DNS issue i was thinking of setting up a caching DNS server that
had its configs updated on the connection failing, ie for the ISP
nameservers.
>
> > I was wondering if anyone could point me in a direction in this. I
> > have looked at the failoverd daemon but as it's not supported anymore
> > i was thinking there might be somthing newer.
>
> These topics are oft-discussed on the Linux Advanced Routing and
> Traffic Control (lartc.org) mail list,
>
> http://mailman.ds9a.nl/pipermail/lartc/
>
> Julian Anastasov and others have invested considerable effort in
> multiple *active* gateways (in contrast to an active/passive failover
> configuration). See
>
> http://www.ssi.bg/~ja/#routes
>
> It is generally agreed that multiple active gateways are easier to
> manage with Julian's patches. The patches effectively allow one to
> statically configure a useful routing policy which otherwise would need
> to be implemented with dynamic configuration in userspace. For your
> simple active/passive failover configuration, the patches are unnecessary.
>
> A tool that is useful with the multiple-ISP configuration is TCP cutter,
>
> http://www.lowth.com/cutter/
>
> which is used to forcibly abort TCP sessions through a NAT gateway.
I'll take a look at all of these, thanks.
Kind regards,
Daniel.
--
Daniel Bartlett
London, UK
More information about the fedora-list
mailing list