On Thu, 3 Jan 2002, Lennert Buytenhek wrote:

> Having TCP/IP or not is probably not the problem.  If W2K answers ARP
> requests on this interface you're still fscked.

True. But I'd have thought disabling TCP/IP should disable ARP too. If the 
interface was configured, it'd also have a different IP

> Can you try connecting to sandy for the first time from an external host
> while having a tcpdump running on that external host (which has to be on
> the same subnet)?  Please try to find out:
> 
>       What MAC address sandy sends in ARP replies
>       What MAC address that ARP reply seems to be coming from.

The ARP replies very frequently don't seem to get to the external ethernet
at all, but when they do they come from 0:50:56:c0:0:1 (sandy's MAC
address on the rocky-sandy internal link) and advertise that same MAC
address. tcpdump on the rocky-sandy internal link also shows the same 
thing happening.

I've come up with a possible theory for what could be happening, but my
knowledge of precisely how ethernet works is a bit thin. Is it usual for
ethernet cards to echo packets sent by the machine they are in back to
that same machine? If the answer is no, then it could be caused by the
VMWare "bridging" (I'll call it slaving from now on to avoid confusion).  
Recall that the external ethernet interface on rocky is actually slaved to
sandy's external ethernet interface. Suppose the slaving decides which
frames to send to rocky by looking at the source MAC address; if it's the
MAC address of the slave device then it doesn't send them, and if it isn't
it does. Then every packet that gets sent out by rocky onto the external
interface which doesn't have its own MAC address will get echoed back to 
it; in particular this will confuse the bridge code into thinking that 
0:50:56:c0:0:1 is on the wrong port. Does that sound plausible?

Cheers,

Ganesh

_______________________________________________
Bridge mailing list
[EMAIL PROTECTED]
http://www.math.leidenuniv.nl/mailman/listinfo/bridge

Reply via email to