[openbeosnetteam] Re: New stack
- From: "Waldemar Kornewald" <Waldemar.Kornewald@xxxxxx>
- To: <openbeosnetteam@xxxxxxxxxxxxx>
- Date: Mon, 21 Jul 2003 15:55:26 +0200
> > Hi Philippe,
> > does your new stack allow these global dynamic/static attributes that
> > I
> > suggested some time ago?
>
> Yes, it will, that's the (Nathan's original) idea, but it's not
> implemented yet.
I did not want to say that the whole thing was my idea. Nathan did the main
part. I only suggested some extensions (or did Nathan have that idea with
dynamic/static attributes before me, too?). Anyway, what counts is that we
have it. ;)
Do you plan to rewrite the stack in C++ when we get shared kernel libraries
(or at some later point)? I think you said we do not want to force people to
use C++. But it looks much nicer. ;)
> I'll start to add attributs support first to net buffers, next are net
> layers *nodes*, so that we could store
> interfaces addresses, kind, nameserves, packets parsed infos, etc...
Did you plan to support dynamic and static attributes? Static attributes
should be saved on shutdown or immediately when they are added while dynamic
attributes are not saved permanently. PPP connections would add DNS servers
as dynamic attributes because normally they are only needed while the
connection exists.
> > I could really need some way to set up nameservers from the kernel.
>
> And editing /etc/resolv.conf is not an option?
Editing a file from kernel-land? I would not like to do that.
Additionally, the nameservers for PPP connections should not stay after
reboot (or after we disconnect). The former is impossible to implement
because after a crash we cannot clean that file (or we can use a comment to
indicate that these are dynamic entries so that we can delete them on
boot-up). The latter is not too complicated, but it still requires code (and
thus work) to find remove the entries.
> Currently, all DNS stuff is in libnet.so, maybe we should move some of
> this in the stack itself, via
> a DNS protocol module and a special socket type?
> That will allow network file system to have name services from kernel
> too...
I did not think of a DNS protocol module. Instead, the protocol itself
should do that. An IPv4 nameserver should be added to the IPv4 module.
The DNS module could work, too.
Are there any cases where moving the nameserver from its protocol to the DNS
module is better? I think it is sufficient to add the entries as attributes
to the protocol.
> > Or do you think it would be better to have a ppp_server application
> > and an
> > interface that just routes all packets?
>
> No, I don't like the idea of having another userland network server.
> The final objectif is:
> - a modular stack running in kernel space, frontend driver included
> - libraries to access/control the stack
> - userland apps, graphical and command-line based.
Agreed. :)
> > Bandwidth is increasing and in some years we will have very fast dial
> > -up
> > connections for everyone. A kernel implementation is much faster.
>
> Yep, I like better this way, as serial dialup will become less and less
> common...
Good. That it what I have thought, too.
Waldemar
Other related posts:
