system-config-network & Zeroconf

[Michael E Brown]

On Thu, Mar 06, 2008 at 08:15:12PM +0100, Lennart Poettering wrote:
> On Thu, 06.03.08 13:51, John W. Linville (linville at redhat.com) wrote:
> 
> > > I'm sorry, but that's just wrong. Setting a net route on a broadcast
> > > device will cause ARP request for the IP addresses in that network
> > > to be broadcast on that segment. Nothing more, nothing less. A host has
> > > to respond to these requests for routing to occur (most likely though
> > > proxy arp). The only thing being sent with a broadcast MAC are the ARP
> > > requests, but those are always sent this way.
> > 
> > Ralf is, of course, correct. [1]  In a sense the only purpose of
> > the routing table is to control which IP address gets ARPed when
> > sending-out a frame.  Setting the default route to "dev eth0" just
> > means you ARP for any address.
> > 
> > Just to make sure, I replicated this environment on my local LAN.
> > Simply setting the default route as "dev eth0" left me with a laptop
> > that could only reach the local LAN.  Turning on proxy arp at my NAT
> > router enabled me to communicate with the Internet.  I encourage you
> > to replicate my experiment. :-)
> > 
> > It's possible that there is some other setting that turns-on the
> > behavior you describe.  But if there is, I don't know about it.
> > 
> > Thanks,
> > 
> > John
> > 
> > [1] Conveniently, that means I am correct as well. :-)
> 
> Of course, you two are absolutely correct. My bad. I guess I shouldn't
> claim things without checking them first ;-)
> 
> But still, enabling this kind of routing on the gateway is just a
> matter of doing:
> 
> echo 1 > /proc/sys/net/ipv4/conf/eth0/proxy_arp
> echo 1 > /proc/sys/net/ipv4/conf/eth0/forwarding
> 
> Right?

This is a really evil thing to do. I once had somebody misconfigure a
netmask on one of the dell.com inbound mail servers to 0.0.0.0, which is
effectively the same thing. Nobody could figure out why it had all sorts
of wierd errors but overall network connectivity looked like it worked
fine.

Turns out the ARP table kept overflowing, which caused interesting
problems until I came around and looked at it.
--
Michael