Re: peth0 received packet on own address

[Nivedita Singhvi <niv@us.ibm.com>]

Ian Pratt wrote:
>>I've been looking into an issue were in the message file I see: 
>>Apr 28 11:43:10 kdell kernel: peth0: received packet with  
>>own address as source address
>>
>>Trying to match the time stamp in the message file to a 
>>packet trace captured by ethereal, I found that the offending 
>>packets are of the type
>>ICMPv6 Multicast Listener Report Message v2, ICMPv6 Neighbor 
>>Solicitation, or ICMPv6 Router Solicitation packets.  It 
>>looks like these packets are sent every time a network 
>>interface is brought up.  In this case it is when peth0 is 
>>brought up from the network-bridge script since the MAC 
>>address is fe:ff:ff:ff:ff:ff.  When a peth0 comes up, all 
>>other machines with a peth0 up and running will receive the 
>>packets and log the messages.
>>
>>Is this an issue of concern?  
> 
> 
> It's not of particular concern, but it would be nice to make them stop.
> 
> Is it the 'ip link set dev X up' line that causes the packet to be sent?
> Does anyone know how to make them stop?

Not compiling in support for MLDv2, etc in the kernel
can make that stop, but distro kernels I believe turn that
on by default. But those broadcasts aren't the problem,
the fact is that the physical address needs to be unique
across the network. Someone might conceivably want to run
advanced routing functions on a guest domain, or on
multiple systems in the network, in which case it needs
to send out those packets.

We shouldn't be doing all of those things (peth0) on the
guest OSs - we had gone to some pains to make sure the
MAC was unique across the system - have we regressed in
that due to the -xen kernel or some other reason?

We also need to ensure that the MAC is unique on the subnet,
if this isn't a case of a guest domain having accidentally
a duplicate MAC.

I haven't caught up with the changes in the unstable kernel
in the last few weeks, looking into it..

thanks,
Nivedita