I would have sent this earlier, but the net admin at Syracuse Univesity is a
dipshit.... (don't ask you don't even want to know.............)
If you guys want to make this networking idea work you should look into the raw
net devices provided by & to the kernel. This way you can make the host network
card redirect traffic for an arbitrary address to a virual network device hosted
by the kernel. This hardware-level data can then be forwarded to a program
running in user-space--for instance our network card emulation. Viola! Network
device/network data almost ready for the guest to use!!!
Kevin Lawton wrote:
> David C wrote:
> >
> > On Wed, 30 May 2001, Kevin Lawton wrote:
> > >
> > > .... Hack away. ...
> >
> > Which reminds me, this seems to be a good time to restate my idea of
> > networking. The guest os is provided an emulated network device that it
> > uses without having to write any drivers specific to plex86. On the host
> > end, the data sent through the virtual network card would be given to
> > a kernel module specifically written for plex86 (the monitor is already
> > being inserted; I don't see this as a big deal). The kernel module
> > would then emulate a network device on the host system that it can't
> > distinguish from a physical card put in. The host system can then act as
> > if it's on two networks and be a router using standard tools for data that
> > is coming from the guest. I know next to nothing about specifics of
> > assembly and low level drivers and kernel hacking, so I don't know how
> > this method's speed/performance compares to other ideas. This is just the
> > cleanest way I envision allowing the guest access to a network. There is
> > not really a special case on either the host or the guest.
>
> This sounds quite interesting. I haven't played with network
> drivers on Linux, but since we are free to define our own network
> card interface (which is really a redirect to the VM monitor),
> sounds like it'd be quite doable. As well as on other guests.
>
> Not only is it good in that all the standard tools/infrastructure
> work without modification, but it also saves a round trip for data
> packets (eliminates the extra hostOS -> host user -> hostOS trip),
> since packets stay at the kernel level.
>
> We would need to be able to move the network card emulation, and
> some minimal IO device interface stuff into the VM monitor.
>
> Special guestOS specific drivers could use the same network
> packet VM -> host interface as would the real network card emulation.
>
> Other people want to comment? I like this idea.
>
> -Kevin
>
> --
> =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
> Kevin Lawton [EMAIL PROTECTED]
> MandrakeSoft, Inc. Plex86 developer
> http://www.linux-mandrake.com/ http://www.plex86.org/
