potential routing loops

Igor V. Semenyuk (iga@sovam.com)
Mon, 18 Aug 1997 06:30:11 +0400 (MSD)

Could anyone please explain me why Livingston is smart enough
to announce (over OSPF) a summary route for assigned block,
but is really stupid in not installing a reject (blackhole)
route for this block internally apparently causing a potential
routing loop?

This is a serious bug and it can be easiely used to mount a DoS
attack. If one pings an address in the assigned block which
happens not to be active at this moment the ping results in
30 duplicated packets between the portmaster and the next-hop router.
With 2ms average rtt over Ethernet you get 1500*8*1000/2=6Mbs with
1500 byte packets; and an attacker only needs to have only
1500*8*1000/30=400Kbs. Apply this 15x factor to your remote T1
POP (the rtt will be different but the 15 ratio will remain) -
4 incoming 1500 packets per second will saturate your T1.

I recall there had been some discussion of this matter on the list
but apparently nothing came out - in 3.7 there's still a routing loop.

Interesting thing is that if you add a static route for the assigned block
yourself (and point it to the pormaster's ethernet interface address)
the portmaster treats the route as a blackhole - no packets looping,
they just die there at portmaster.

One can also install blackhole summary route on a (smart) adjancent
router and run RIP, not OSPF (the portmaster will announce individual
routes in case of RIP).

But these are ugly workarounds of the very clear problem in ComOS.

-- 
Igor V. Semenyuk                    Internet: iga@sovam.com
SOVAM Teleport                      Phone:    +7 095 258 4170
Moscow, Russia                      Fax:      +7 095 258 4133