Re: KVM induced panic on 2.6.38[2367] & 2.6.39

[Patrick McHardy <kaber@trash.net>]

On 07.06.2011 20:31, Eric Dumazet wrote:
> Le mardi 07 juin 2011 à 17:35 +0200, Patrick McHardy a écrit :
> 
>> The main suspects would be NAT and TCPMSS. Did you also try whether
>> the crash occurs with only one of these these rules?
>>
>>> I've just compiled out CONFIG_BRIDGE_NETFILTER and can no longer access
>>> the address the way I was doing it, so that's a no-go for me.
>>
>> That's really weird since you're apparently not using any bridge
>> netfilter features. It shouldn't have any effect besides changing
>> at which point ip_tables is invoked. How are your network devices
>> configured (specifically any bridges)?
> 
> Something in the kernel does 
> 
> u16 *ptr = addr (given by kmalloc())
> 
> ptr[-1] = 0;
> 
> Could be an off-one error in a memmove()/memcopy() or loop...
> 
> I cant see a network issue here.

So far me neither, but netfilter appears to trigger the bug.

> I checked arch/x86/lib/memmove_64.S and it seems fine.

I was thinking it might be a missing skb_make_writable() combined
with vhost_net specifics in the netfilter code (TCPMSS and NAT are
both suspect), but was unable to find something. I also went
through the dst_metrics() conversion to see whether anything could
cause problems with the bridge fake_rttable, but also nothing
so far.