Splitting up pfinet [Was: Re: i want to help ! how ?]
On Tue, Mar 07, 2000 at 11:01:16AM +0100, Niels Möller wrote:
> Igor Khavkine <i_khavki@alcor.concordia.ca> writes:
>
> > Well, that's what I was asking about. Maybe a better idea would be to make
> > pfinet support a general tunneling interface that gives control of the
> > underlying protocol to a userspace daemon like ppp, slip, plip or whatever
> > else one can run in userspace.
>
> Note that pfinet is just as "userspace" as any other daemon. I suppose
> you really mean dividing up the work between several daemons? I would
> think that the first step in splitting pfinet into several processes
> is to think about the interfaces between the processes, and the
> responsibilities of each. (Of course, the result of that thinking might
> also be that the whole division thing was a bad idea).
>
You are right they are all userspace, i was thinking along the lines of
division of labor.
> One division could be:
>
> pfinet keeping track of ip packets, addresses, sockets (including
> all tcp processing, etc) and interfaces, and possibly some routing
> info.
>
> ppp and eth sending and receiving ip packets using the appropriate
> link-layer mechanisms
>
That's almost exactly what I was thinking. To me taking the interface support
out of pfinet seems like a good idea, that way support for different kind of
interfaces like ppp or VPN tunneling could be added separately. But I don't
know the answers to the basically same questions that you are asking:
* Is it possible for more then one "socket server" (like ipv4, ipv6, ipx) to
use the same interface?
* If so, would it be possible to marshal the packets coming over the interface
to appropriate daemons? If all these packets belong to the AF_INET domain, it
shouldn't be a problem, pfinet can sort them out. But is it possible to
receive packets belonging to other AF_* families?
> Bottom line: I think it would be a very valuable contribution to look
> carefully on the network stuff architecture, try to sort it out, and
> write up your conclusions.
>
> /Niels
I'm interested to know what the architects of the Hurd think of this idea
(Thomas, Roland, Marcus ?). I'll also try to get some free time to try and
understand the present networking code.
Igor
Reply to: