[Babel-users] v4-over-v6 implementation

[Dave Taht]

On Mon, Apr 20, 2020 at 1:45 AM Théophile Bastian
<theophile.bastian at ens.fr> wrote:
>
> > Well, it compiles, but how would you suggest going about testing it?
> > (in terms of setting up the routes in the linux kernel
> > to start with). Then I'd toss a bird implementation at it and see what
> > it breaks.
>
> My test setup is usually a network of containers, around variations of
> this topology:
>
> A --(v4+6)-- B --(v6)-- C --(v4+6)-- D
>
> where each NIC on a v4+6 link gets an IPv4 address. In this setup, A
> should be able to reach the IPv4 addresses of C and D.

Well for the record, I have a fairly large outdoor network with about
30+ radios on it normally, 6 uplinks to the
internet using source specific routing, and a couple vpns.

and a firewalled off lab where I can bring up a ton of devices on 7
architectures. The alignment bugfix in 1.9.2 we've seemingly had for
years is really bugging me as what its side effects could have been,
so I've been poking into that.

>
> > How long has the linux kernel had the ability to do this?
> > I would kind of prefer the ae numbers to be assigned sequentially
> > instead of out of the experimental range. > so what happens if you get this ae on an older kernel?
>
> I've been testing that yesterday on a Debian Stretch, which has a 4.9
> kernel (iirc). This results in errors in the logs, since babeld tries to
> insert v4-over-v6 routes, but the behaviour, as far as I was able to
> test, is correct: the route insertion is rejected by the kernel.
>
> However, relying on a kernel error looks a bit ugly to me, and I've put
> it on my to-do list to catch the error earlier.

It should not re-advertise the route if it cannot install it, IMHO.

> --
> Théophile Bastian



-- 
Make Music, Not War

Dave Täht
CTO, TekLibre, LLC
http://www.teklibre.com
Tel: 1-831-435-0729