Rusty Zickefoose wrote:
I've got to go with Pete K. on this one. In our current, cidr-ized world, it
is simply not possible for an upstream provider to determine what is, or is
not, a broadcast address in a downstream network. This is something that needs
to be implemented from the edge in, not from the core out.
I'll second the agreement.
I know my own subnetting strategy, and I also know what blocks I assign
to my customers. Although few of my customers are sophisticated enough
to subnet them on their own, even if we go do it for them (which is one
of the services we offer) this is not something that we record in the
database that will be used to build all our configurations.
I do have an access list deny for incoming destinations to *.*.*.255
since I do know that the only customer we have with larger than a /24
from us (via cw.net) also happens to have nothing larger than /26 in
their network. AFAIK, today, smurfers are only using *.*.*.255. They
would have to track a lot more information to use others, so for now
I can generally expect that deny to prevent us from being an amplifier.
As more and more *.*.*.255's get blocked, smurfer kiddies may look for
other broadcast addresses as well. It may come down to literally having
to build an access list from my assignment database. Of course those
smaller subnets will typically have fewer hosts to amplify from, but
when servers are carefully concentrated in a subnet, there can still
be a lot.
I cannot expect C&W to block *.*.*.255 incoming for me. Even though in
my case it would cause no problems, in the case of others, it can, and
they have no reason (or database) to know which is which. But when the
smurfers start using 127, 63, etc., that won't do any good, and I don't
want them blocking those all the way down to /30's (let's just cut the
IPv4 address space in half).
I do block outgoing sources with addresses other than our network blocks.
Thus we can't be the source of an amplified attack other than an attack
on our own network, or only amplified here, which limits it to our pipe
size, and makes tracking it (to here) very easy. Such blocking helps on
forgeries, too.
I would expect the backbones to do broadcast address blocking in their
own subnet space where a lot of broadcast replyable servers exist (surely
their Cisco routers aren't replying to broadcasts since that is easy to
turn off). But these can be thought of as an "edge", anyway, and they
do know the subnet mask there.