On Mon, Jul 14, 2003 at 12:37:26PM -0700, Aaron Suen wrote:

> I'm sure there are things I haven't thought through clearly enough when
> considering this topic, but it sounds like it will work to me.  And, though the
> payoff for a "typical" computer network is not spectacular, there are potential
> benefits for this kind of approach.  It could, effectively, render all current
> fragment handling methods obsolete (am I being too ambitious?).  

One point has already been mentioned: you might want the firewall to do
complete reassembly if your server is dealing poorly with fragmentation
attacks (if it keeps incomplete packets/fragments too long and runs out
of memory).

So I wouldn't replace full reassembly with your scheme, but add it as an
additional reassembly mode. I'm not sure yet whether this is a common
problem. What kind of services produce this kind of traffic? Obviously
NFS over UDP, but how many people allow untrusted peers to talk NFS
through a firewall? The added latency of full reassembly matters most
with slow uplinks (not local networks), so I wonder how many people do
NFS over UDP through slow uplinks to the internet :)

Anything else must have failed PMTU discovery, if it sends large IP
packets fragmented. We need a common, legitimate case that produces this
traffic. If the traffic is nearly always produced by an attack,
additional latency doesn't really matter...

Daniel

Reply via email to