engineering --> ddos and flooding

Mailman List Mirror (Read Only)
1

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Sorry but IMESHO null routing a /32 during a DoS attacck doesn't exactly
strike me as engineering. It is more like dealing with the attack in
real-time. To mean engineering would mean desinging networks
to be resistant to DDoS and flooding in the first plsce.

To that end no NSP should ever allow spoofed IP addresses outside of
their network. (not just RFC 1918 addresses but valid IPs that don't
belong to that NSP)

  e.e if I'm have a circut from C&W nd I try to spoof a packet
  eith a source address of 216.35.172.135 it should be dropped as
  close to the edge of C&W's network as possible.

  note on RFC 1918 addresses: These should never get past customer
  edge routers IMESHO.

Two NSPs should rate limit DoS traffic (ICMP & SYNs) within their
networks in such a way that it can never DoS a T-1 (or E-1 if you are
not in the US). [note: I'm not sure if ciso's are up for this workload
since I primarily work with Juniper.]

Three NSPs should lock down their gear so it can't be used for DoS
amplification. I mean a clueless customer getting burned is one thing
but we should expect & demand more from NSPs.

[note: my primary responsiblity in life is security not networking.I
know just enough networking to be dangerous and annoying to our
Networking dept. :slight_smile: ]

> Assuming not adding the extra connection, this means that upstream
> prefix filtering, so that one can't mistakenly inject 255 /24s
> rather than a single /16, would go out the window. Now think about
> /32s and what the routing tables will start to look like. Now
> onsider that the upstream would also want to send to its upstream
> Tier-1 the NULLROUTE /32 as well so that his bandwidth is not eaten
> up as well and we have asituation whereby routing table size will
> triple in size every year.

This is a stop gap measure for customer networks. Those null routed
/32s are not meant to be permanently advertised, they are meant to
free the customer's pipe from smurf/fraggle until the SP can do
something about it. What would be the point of permanently
blackholing a host on your network?

One more problem...what if your mail/web server is the target of the
attack you have just taken that resource effectively off-line. No need
to continue the DoS you've done the work of the attacker.

I would imagine that most tier 1's are going to filter anything longer
than a /24 whether you advertise it or not. The question isn't about
route table size, it is whether your SP will go the extra mile to give
you a proactive option to deal with attack and has someone clueful to
implement it that will take responsibility for it (not that it is
hard).

This is a very limited measure that only helps in a very particular
situation for a small subset of customer networks. I think it is a
very useful tool for that particular situation... it is not meant as a
principle that SP networks should apply to their upstream as well.

Paul Johnson
Ratoath, Co. Meath
Republic of Ireland
- -----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.0.1 (GNU/Linux)
Comment: pgpenvelope - http://www.uiuc.edu/ph/www/ftobin/resources.html

iD8DBQE7G7QUEn8vUhkIGTURAuuFAJ0Ub5lBQWG4r/WisSOTaOPNPm0/nACg8fY3
qcEayLo0vyDGlVDSFbWMKlE=
=K9Xz
- -----END PGP SIGNATURE-----

2

Paul Johnson wrote:

To that end no NSP should ever allow spoofed IP addresses outside of
their network. (not just RFC 1918 addresses but valid IPs that don't
belong to that NSP)

Agreed, RFC 2827 should be the norm. Maybe some day it will be.

This is a stop gap measure for customer networks. Those null routed
/32s are not meant to be permanently advertised, they are meant to
free the customer's pipe from smurf/fraggle until the SP can do
something about it. What would be the point of permanently
blackholing a host on your network?

One more problem...what if your mail/web server is the target of the
attack you have just taken that resource effectively off-line. No need
to continue the DoS you've done the work of the attacker.

You've got to balance how badly the attack hurts against how badly this
method of stopping the attack hurts. If someone launches a distributed
attack against my mail server, and it eats up all of the bandwidth I've
bought from my upstreams, then I've got zero connectivity. You can be sure
I'd be in favor of taking my mail service offline to keep the rest of the
network running while I pursue the issue with the upstreams. Ditto if
someone launches a massive attack against one of my customers and it has the
unfortunate side effect of destroying my connectivity (and that of all of my
other customers.) A filter of this type is never a long-term solution, but
if it can get me even partially online, that's better than nothing.

Mark

3

Hmm.. I'd be *REALLY* unhappy if our upstream decided to rate-limit SYN
packets to prevent a DoS of a T-1, since the smallest pipe we have
deployed is in the OC-3 category.

The problem is that a *distributed* DOS effectively bypasses this sort
of check - you have (for instance) 1000 zombie machines, each contributing
only a few packets per second. So none of THEM gets filtered. Each ISP
may have only 3-4 zombies, so even aggregated they don't trigger a filter.

Nothing trips a filter, until it gets loose inside a Tier-1, with traffic
converging on one outbound pipe to the victim from 8 or 10 different
peering points. And at THAT point, it's too late.

        Valdis Kletnieks
        Operating Systems Analyst
        Virginia Tech

4

Except RPF.

-Dan