Reusing Exit Nodes?

[Roger Dingledine]

On Thu, Jan 27, 2005 at 02:47:41AM -0600, Mike Perry wrote:
> So I've noticed my patch to reuse exit nodes that match certain
> domains hasn't showed up in CVS. Was there anything wrong with it?

Hi Mike,

Sorry about the delay on this. I've been trying to figure out how I
want this to work.

First, check out sec 3.11 of http://tor.eff.org/doc/control-spec.txt

The "mapaddress" functionality is aimed to solve a number of situations:

a) Alice wants to use a .onion or .exit address from a client that doesn't
speak socks4a. For example, we can intercept her gethostbyname() call,
return a fake address like 0.x.y.z, and then inform Tor via the Control
interface that 0.x.y.z should map to the string she asked to resolve.

b) Poor man's VPN. Let's say Alice runs a service somewhere that doesn't
have encryption or authentication -- like telnet. She can set up a
hidden service via Tor to let people connect and get encryption and
authentication, but connections to it will require the extra hassle of
hidden service foo. So instead she runs a Tor server on the same host
(or inside the trusted perimeter of the host), and she configures an
Address Mapping to internally rewrite any request for service.com to
service.com.aliceserver.exit. This gives two benefits: First, her local
applications can connect to it like normal and get end-to-end encryption
and auth for free. Second, we don't break applications that include the
destination address in the application-level. The HTTP "Host" header
is a good example of this -- if you type www.google.com.rodos.exit in
your Mozilla, then you break Google's virthosting because it only does
its geodns magic if you say Host: www.google.com. The alternative is to
write an application-level-aware proxy that rewrites the Host: line; yuck.

c) Inside Tor, mapaddress could do exactly what you're looking for here.
If Alice's destination matches certain configured names, or certain ports,
then once she picks an exit node it can establish an addressmap so she'll
pick it again later.

Of course, the controller interface probably needs more data than we
describe in the spec, such as an expiry time and maybe more. It would
probably make sense to implement the Tor side of things, plus being
able to specify it in the Tor config, and then somebody will deal with
configuring it via the controller later.

> Attached are patches against both 0.0.9.3 and CVS. Please do let me
> know if there is something wrong with the patch, I don't mind fixing
> it. It seems to work well for me, and I find it useful.

I think what you've got here is a good start, but I think if we
generalized it to do mapaddress we'd be in even better shape. What do
you think?

Thanks,
--Roger