Re: [Qemu-devel] [PATCH] Set hostname in DHCP response

[Leonardo E. Reiter]

QEMU doesn't do a great job of supporting multiple user-net interfaces 
right now.  I was testing this the other day and getting some rather odd 
behavior (unexpected segfaults, etc.)  I didn't debug deep enough to 
figure out the specifics, but having a global state (like slirp is 
today) for multiple network providers is probably not a good idea.  I am 
contemplating writing a patch that will change the global state in slirp 
to an actual "opaque" state, even letting various slirp instances be 
configured to serve different subnets, etc.  This would be useful if for 
example you wanted to have 1 slirp interface access only "special" 
services, like smb, etc., and another act as a NAT to the internet.  In 
that case, in theory, stuff like VPN software in the guest OS should 
work without disrupting access to "special" services.  But for something 
like this to truly work, you need to have the ability to configure the 
subnet that slirp sets up _per NIC_, and that would of course mean 
removing the global state as the most obvious (and correct IMHO) solution.

Others will probably argue that you can use a combination of a single 
slirp NIC and a NAT tun/tap set up to accomplish the same basic thing 
today, without any modifications.  Still, the "zero" host configuration 
of slirp makes it very attractive on its own, so I still think the patch 
I'm referring to would be valuable.  I'd also implement an option to 
make slirp return the DNS/gateway in the DHCP packet for only 1 
interface, since guest OS's would probably not like having multiple 
default gateways.  This patch would be useful in a slightly different 
way even if you were trying to combine slirp with tun/tap, and having 
the guest OS configured for DHCP on all interfaces.  For example, you 
may want the guest OS to use the tun/tap interface as the default route, 
so telling the slirp interface to not return a gateway/DNS would 
probably still be a good idea even if you were only using 1 -net user. 
 That patch by itself would be trivial versus the global->local state 
patch I mentioned above, so I can see posting that one much sooner, if 
it even works as I'm expecting it to ;)

Regards,

Leo Reiter

Ed Swierk wrote:
> I agree, since the hostname is relevant only for user-net interfaces.
> An updated patch is attached.
>
> The only issue is that there's just a single, global hostname, not a
> hostname per user-net interface. On the other hand, I'm not sure if
> qemu supports multiple user-net interfaces in the first place. Does
> the following configuration make sense?
>
>     -net nic,vlan=0 -net user,vlan=0 -net nic,vlan=1 -net user,vlan=1
>
> --Ed

--
Leonardo E. Reiter
Vice President of Product Development, CTO

Win4Lin, Inc.
Virtual Computing from Desktop to Data Center
Main: +1 512 339 7979
Fax: +1 512 532 6501
http://www.win4lin.com