On Wed, Mar 29, 2000 at 10:56:43AM +0100, Thomas Mangin wrote:
> Aaarrgggh ...
> 
> Thank you for the information.
> 
> Coda use a real DNS lookup or only the resolver (I guess it is a full DNS
> lookup) If so I will have to install locally two DNS servers, thing I
> would like to avoid ..

I am afraid it might be even worse than that.

A Coda server uses gethostbyname to resolve a name to it's ip-address,
and not only uses this to talk to the other server, but also records
this in RVM (in the volume location database, VLDB). In a way you are
lucky, because the 2 clients are behind the firewall, so if they get the
`internal network' ip-address, they can use it.

For most other applications it is better to keep the external addresses
in the DNS, and create specific host routes on the Coda servers to make
server-server traffic go over the internal net.

f.i. on serverA
serverA# route add serverB netmask 255.255.255.255 gw serverB-private

 or maybe this might work if the server accepts packets destined for
it's public address from the internal network (on eth1). A linux box
with 2.2/2.3 would probably be able to do this, which responds to arp
requests on the `wrong' interface.

serverA# route add serverB dev eth1

In your case you could add the internal server IP addresses to
/etc/hosts on the servers, then restart both servers, and then
regenerate the VLDB/VRDB.

SCM# bldvldb.sh
SCM# volutil makevrdb /vice/vol/VRList

Jan