XEN and ipq_read

XEN and ipq_read

[plamen ..]

 Hi all,

I'm using Ubuntu Hardy, Xen version 3.2.1-rc1-pre, Dom0 kernel 2.6.24-27-xen, PV DomU kernel 2.6.24-27-xen. 

I'm setting DomU as a router having iptables 1.3.8. I put an IDS system Snort in inline mode (IPS) on the router, which is configured to retrieve specific packets from kernel (iptables ... -j QUEUE and ip_queue module). At first snort started to report errors on each received packet. After a little bit of debugging and doing a sample application to test ipq_read() I found that raw data sent from kernel contains about 24 bytes more than expected. The additional bytes are in the meta data structure before the real packet content. This breaks raw data parsing. After a little bit of additional debugging I noticed that this happens only on Xen DomU VMs. On Dom0 it work fine, on other servers not running Xen it works also fine. 

Currently I'm about to install rtr DomU as HVM and I think it will work fine, but I don't want to leave it like this in production. 

Is there any reason in xen kernel to break sending packets from kernel to user space through the ip_queue module ? If so is there any way to work around this issue ?

Thanks in advance,
Plamen

Re: XEN and ipq_read

[Pasi Kärkkäinen]

On Tue, Apr 27, 2010 at 11:31:33AM +0300, plamen .. wrote:
>  Hi all,
> 
> I'm using Ubuntu Hardy, Xen version 3.2.1-rc1-pre, Dom0 kernel 2.6.24-27-xen, PV DomU kernel 2.6.24-27-xen. 
> 
> I'm setting DomU as a router having iptables 1.3.8. I put an IDS system Snort in inline mode (IPS) on the router, which is configured to retrieve specific packets from kernel (iptables ... -j QUEUE and ip_queue module). At first snort started to report errors on each received packet. After a little bit of debugging and doing a sample application to test ipq_read() I found that raw data sent from kernel contains about 24 bytes more than expected. The additional bytes are in the meta data structure before the real packet content. This breaks raw data parsing. After a little bit of additional debugging I noticed that this happens only on Xen DomU VMs. On Dom0 it work fine, on other servers not running Xen it works also fine. 
> 
> Currently I'm about to install rtr DomU as HVM and I think it will work fine, but I don't want to leave it like this in production. 
> 
> Is there any reason in xen kernel to break sending packets from kernel to user space through the ip_queue module ? If so is there any way to work around this issue ?
> 

Did you try disabling all network offloading settings in the domU? 
(and if that doesn't help, then also in all interfaces/bridges/vifs on dom0).

Other than that you might want to upgrade your Xen and kernels, they're pretty old
and known to have problems/bugs. 

(Only the kernel versions should affect packet processing though).

-- Pasi