tech-net archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: Retiring src/sys/netiso and associated code

On Mon, Jan 28, 2013 at 11:25:03PM +0100, Joerg Sonnenberger wrote:
> On Mon, Jan 28, 2013 at 02:14:25PM -0800, John Nemeth wrote:
> > On Jun 20,  4:53pm, Joerg Sonnenberger wrote:
> > }
> > } a while ago the OSI options were removed from most default kernel
> > } configurations due to questionable security state of the code and
> > } general lack of testing. I don't think the situation has improved in any
> > } noticable way. It adds complexity for the routing code, which I want to
> > } work on. Not breaking it requires testing, which doesn't seem to happen.
> > } So who is actively using the OSI network stack with NetBSD and is in a
> > } position to test changes? Alternatively, I would like to finally let it
> > } rest in peace.
> > 
> >      Isn't it needed to support the IS-IS routing protocol?

gated uses the NETISO-provided sockets for IS-IS, as far as I know.
I don't know if it survives in the wild. Chris?

> Quagga contains at least parts of a userland-only implementation of the
> OSI protocol parts needed. As such, any work related to IS-IS should be
> focussing on that, I believe.

Most important real-world use of NETISO parts would certainly be IS-IS.
If we can provide that via Quagga userland-only, thats good.

That written:

I don't see how NETISO adds complexity to the routing code. We will
have at least two protocols for a long time anyway; three for a
while (there are people who use network printers over Appletalk,
and not only me).

Actually, I'd consider IPv4 non-contiguous netmask routing more alien
to IPv6 hierarchical routing than OSI. I know we discussed it years ago,
but did we actually decide to get rid of IPv4 hierarchical netmasks yet?

What OSI certainly adds, is *line count*. And some of it, unnecessarily;
e.g. nowadays we'd implement some of the tunneling options as a
few lines of GIF interface.

I have published testing code which should be enough for the datagram
socket and routing case:

What I never tried is to set up a lab where NetBSD routes DECnet-phase-V
(-OSI) machines. Should be easy with emulators nowadays.

There's been a report about problems with the packetstream (TP4)
code, which seems to hint at real-world uses.


Home | Main Index | Thread Index | Old Index