[Qemu-devel] network fails on windows

[Qemu-devel] network fails on windows

[Matthias Jung]

Hi,
i have found a bug: for long I was not able to use the network part of qemu, I 
tried almost everything to get a guest Windows LAN access, nothing worked. 
Neither tun nor user-net was working, no matter how hard i tried, and only 
Windows seemed to have this problem, linux guests always performed fine.
-isa didnt help either.
I was just on the verge of giving up until i tried one last thing:
I switched the mac address parameter. Til then I had used the mac  
a3:b2:23:54:7e:cf which is a valid mac and windows did detect the interface 
without flaws, but no network access was possible.
With 00:ff:7f:fc:36:d2 it suddenly worked, I only can imagine a bug in the 
ne2000 windows driver, or in windows (95,98,2000 tested) itself, not allowing 
such a mac. 
I had no mac collision at any time!

happy that its finally working,

Matthias Jung

Re: [Qemu-devel] network fails on windows

[Julian Seward]

I agree, there is something funny to do with networking with
WinXP/Win2K guests.  I use user-mode networking with the
current cvs head.  Sometimes XP is able to make a connection
to the outside world; yet if I just leave the guest machine
idle and try an hour later then it can't; try again 5 minutes
later and it's OK again.  It's very frustrating.  I have no idea
what's happening.  Also, even when it can communicate with 
other machines on my local network, it's really slow.

Wierd thing is, running Red Hat 7.3 as a guest, with user-mode
networking, the guest's connections to the outside world are
both reliable and fast.  Which corresponds with Matthias' observation.

Any ideas?  Has anyone else got any related observations?

J

On Saturday 17 July 2004 16:07, Matthias Jung wrote:
> Hi,
> i have found a bug: for long I was not able to use the network part of
> qemu, I tried almost everything to get a guest Windows LAN access, nothing
> worked. Neither tun nor user-net was working, no matter how hard i tried,
> and only Windows seemed to have this problem, linux guests always performed
> fine. -isa didnt help either.
> I was just on the verge of giving up until i tried one last thing:
> I switched the mac address parameter. Til then I had used the mac
> a3:b2:23:54:7e:cf which is a valid mac and windows did detect the interface
> without flaws, but no network access was possible.
> With 00:ff:7f:fc:36:d2 it suddenly worked, I only can imagine a bug in the
> ne2000 windows driver, or in windows (95,98,2000 tested) itself, not
> allowing such a mac.
> I had no mac collision at any time!
>
> happy that its finally working,
>
> Matthias Jung
>
>
> _______________________________________________
> Qemu-devel mailing list
> Qemu-devel@nongnu.org
> http://lists.nongnu.org/mailman/listinfo/qemu-devel

Re: [Qemu-devel] network fails on windows

[John R. Hogerhuis]

On Sat, 2004-07-17 at 08:21, Julian Seward wrote:
> I agree, there is something funny to do with networking with
> WinXP/Win2K guests.  I use user-mode networking with the
> current cvs head.  Sometimes XP is able to make a connection
> to the outside world; yet if I just leave the guest machine
> idle and try an hour later then it can't; try again 5 minutes
> later and it's OK again.  It's very frustrating.  I have no idea
> what's happening.  Also, even when it can communicate with 
> other machines on my local network, it's really slow.
> 
> Wierd thing is, running Red Hat 7.3 as a guest, with user-mode
> networking, the guest's connections to the outside world are
> both reliable and fast.  Which corresponds with Matthias' observation.
> 
> Any ideas?  Has anyone else got any related observations?
> 

If someone could post detailed tcpdump output somwehere the problem will
likely be apparent in the trace. When it comes to networking the trace
tells the story the best...

It would be good to have the TCP packets, ARP, DHCP, and any ICMP
packets in the trace. Try to filter out any noise from machines not
involved.

Anyway if someone posts it I'll take a look.

-- John.

Re: [Qemu-devel] network fails on windows

[Derek Fawcus]

On Sat, Jul 17, 2004 at 05:07:40PM +0200, Matthias Jung wrote:
> I was just on the verge of giving up until i tried one last thing:
> I switched the mac address parameter. Til then I had used the mac  
> a3:b2:23:54:7e:cf which is a valid mac

What do you mean 'valid'? - all values are 'valid'.

> and windows did detect the interface 
> without flaws, but no network access was possible.
> With 00:ff:7f:fc:36:d2 it suddenly worked, I only can imagine a bug in the 
> ne2000 windows driver, or in windows (95,98,2000 tested) itself, not allowing 
> such a mac. 

Well there would appear to be an important difference between those two MACs,
one is multicast,  one is unicast.

The first two bits of the MAC address (as transmitted on the wire) are the
group/unicast bit followed by the local/global bit.  These generally translate
into bits 0 and 1 of the first byte of the MAC address when written as above.

So the first address has A3 - which is multicast and local,  whereas the
second address has 00 which is unicast and global.  You may want to try
the first address again but with the first byte replaced as follows:

   a2 - A local unicast address
   a0 - A global unicast address

DF