IP backbone numbering/naming
Mike Lewinski
mike at rockynet.com
Sat Nov 16 16:56:06 UTC 2002
Jahowering at aol.com wrote:
> The DOS attack should be a real concern when using RFC 1918. A
> distributed) smurf attack, or one of it's derivatives, can cause the
> icmp echo replies to be sent to that src. address. Since the
> attackers just use blocks and blocks of spoofed addresses, you could
> become the sourced address victim. Of course, ingress and egress rfc
> 1918 filtering will prevent this...it's just something else to think
> about...
Well, those routes not being globally distributed mitigates that danger.
The reverse argument could actually be made- I'm unlikely to see any DoS
backscatter directed at the RFC1918 addresses, while the publics will
always see it.
I forgot one of the most important reasons we are migrating away from
this practice. We do source address filtering via RPF verification at
the edge, but it allows the /30 to leak because there is a valid route
for it internally. Meaning that deliberately spoofed packets can still
escape our network, and allowing just one per client is still too many.
The better answer to our DoS victim, of course, is an ACL for every hop
from the border at the border;that will be implemented as we complete
the purge of RFC1918.
Steve- the pain of renumbering is simply not worth it... take it from
someone who's been there and don't use 'em. Backbone links might be
easier to do than clients, but in the end you WILL end up renumbering
when it breaks something a client needs. Save yourself the hassle and do
it right from the start.
More information about the NANOG
mailing list