North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

RE: 69/8...this sucks

  • From: David Luyer
  • Date: Wed Mar 12 07:43:41 2003

Stephen J Wilcox wrote:
> On Wed, 12 Mar 2003, David Luyer wrote:
> > Iljitsch van Beijnum wrote:
> > > On Tue, 11 Mar 2003, Owen DeLong wrote:
> > > 
> > > > In short, it doesn't.  Longer answer, if the ISP configures 
> > > > his router correctly, he can actually refuse to accept
> > > > advertisements from other sessions that are longer versions
> > > > of prefixes received through this session.
> > > 
> > > How???
> > 
> > There is a technically possible (but rather twisted) way you
> > could not use the adverts, but not a way to refuse receiving
> > them that I know of.
> 
> I think youre mixing up with ingress filtering by prefix list 
> which you can 
> specify prefix length on and hence ignore longer (or smaller) matches.

The example I provided achieved both ingress and egress filtering
based on routes in a bogon BGP feed, in a way which would even
block when a more-specific route is in the provider's BGP table.
While it didn't actually prevent the routes being in the routing
table (as I said, it doesn't provide a way to stop receiving them),
it does prevent traffic from and to the bogon locations, which is
a significant part of the reason to use bogon lists.

However, yes, it has some deficiencies[1] compared with using the
static bogon lists for route filtering (and ingress/egress); it
does not prevent routing table bloat, and it does not prevent
traffic travelling across your WAN to the point of network egress
only to be dropped.

If you want to actually not receive into your network at all the
BGP routes which match bogons, as I stated earlier, there is no
way I know of to do this via a BGP feed.  The only way to do it
that I know of would be to use either a prefix list or a standard
ACL (you can do anything you can do with a prefix list with a
compiled extended ACL on BGP routes, it's just less clear to
read as an extended ACL).

Although, Owen DeLong has stated that it is possible, so maybe
we should wait for his response :-)

David.

[1] Apart from simply being a completely twisted design.