[Qemu-devel] Noob question on Linux networking

[Philip Prindeville]

Hi.

I was looking at the way that Qemu works under Linux, and was surprised
by the complexity required to configure networking.

I worked on a software harness on SunOS 3.5 17 years ago, and things seemed
to be a bit simpler then....  You could create a "software" network 
interface
cloned from a real physical interface (as long as it was a broadcast or 
multipoint
network) and create a new, unique address to that interface.

For instance, you could clone an ethernet interface, give it a new IP 
address,
and even set up new services running just on that interface...

This assumed that the Ethernet driver and hardware supported having
multiple MAC addresses that it recognized as its own (usually
implemented in a CAM as a key and mask... the mask would be applied
to the destination address and compared to the key... usually the mask
would be 48-bits of ones, i.e. match the entire key)... which most but
not all controllers could do.

These days, I think just about any controller will allow you do this,
even a lowly RealTek-8129.

So...  my question is, why not create a mac-level interface alias,
generate a separate unique mac address, and have Qemu use
this as the client OS's interface?

Then the client OS could see all of the same things as the
host OS, and they should even be able to ping each other
(well, it depends... some Ethernet interfaces "hear"
themselves when they transmit... some don't).

What would be involved in doing this?

Not sure that current Linux kernel drivers support
creating additional destination addresses through a
standardized interface...

I'll look into it.

-Philip