Re: [Xen-devel] [PATCH] [XM-TEST] Use explicit (IP id==packet size) on hping2 tests to avoid problems with IP id==0

[James Dykman <dykman@xxxxxxxxxx>]

niv@xxxxxxxxxxxxxxxxxxxxxxx wrote on 05/18/2006 05:32:26 PM:

> James Dykman wrote:
> > We have run into some xm-test network test failures in the tests that 
use 
> > hping2 that are caused by an IP id of 0:
> > 
> > - hping2 uses a packet socket, so must do its own fragmentation when 
> > sending messages >1500 bytes.
> > - When fragmenting, hping2 uses the low byte of the PID as the IP id 
in 
> > each of the fragmented IP headers.
> > - The kernel will assign an IP id if it sees the IP id ==  0 in the 
header 
> > that hping2 passes down 
> >    (net/ipv4/raw.c:raw_send_hdrinc())
> > So if the low byte of the PID is 0, every fragment will get assigned a 

> > unique IP id by the kernel, and 
> > reassembly will fail at the destination.
> > 
> > Jim
> > ----------------------------------
> > Use explicit (IP id==packet size) on hping2 tests to avoid problems 
with 
> > IP id==0
> 
> Jim, will this not be a cause potential corruption - if we have incoming
> fragments belonging to different packets with the same ipid
> there is no way for the reassembly code to distinguish them.
> Checksum failures might catch some, but that is no guarantee of
> safe delivery.
> 

That possibility exists in the tests as they are today... hping2 is 
assigning IP ids based on
its PID, completely independent of the kernel.

If you want to solve that problem TOO, then we need to replace hping2. Any 
suggestions?

Jim

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel