[ppml] Abstract of proposed Internet Draft for Best CurrentPractice (please comment)

Joe Provo ppml at rsuc.gweep.net
Tue Feb 18 15:23:07 EST 2003


[similar to a rant half-written and not sent during nanog]

On Tue, Feb 18, 2003 at 10:35:33AM -0700, John M. Brown wrote:
> no they don't  The ability to accept routes from a customer
> is strictly a matter between the service provider and
> its customer.
[snip - we'll use this context later]
> I can see a nice little DDOS vector here.  Happy Hacker
> tricks BGP into revoking EBAY's prefix, EBAY looses
> millions, sues RIR.
 
Straightforward denial of service, nothing distributed. And 
guess waht - the black hats are already (have been for a while) 
exploting longest-match as a way to impersonate routes/steal 
traffic/smokescreen their black-hatted-ness. How does this 
relate to RIRs? Through the side door:

- RIRs already have multihoming as a requirement for AS 
  allocations.
- Some RIRs have multihoming as address allocation 
  justification. 
...seems that the RIR's are NOT 'controlling the routing table' 
(but gosh, is there a problem with publishing allocation data 
in standardized machine-parsable format? RIPE seems to do it 
a-ok...) but DO have their fingers in the justification of 
space and -to a limited degree- how that space is to be 
utilized.

See it isn't the Vendor-customer relationship (well, except from 
clueless vendors who think any routes they propagate are instantly 
going to be "valid" because they are special/big/buy from 'all 
the tier1s'/etc) so much as the customer-vendor-rest of the net.

It is a Very Short step to dedicating and policing space for 
'officially blessed small multihoming'. This would
- be consistnt with RIR roles and previous actions (ASNs not 
  used for multihomed entities can be revoked; ISPs and other 
  LIRs are tacitly encouraged to revoke space that was 
  justified by multihoming when said multihoming doesn't 
  occur; etc)
- reduce deaggregation/holes in areas populated by aggregates
- give network operators additional tools to do *their* jobs 
  of filtering/fighting black-hatted-ness *without* making it
  the RIR's job (ie, i can filter against longest-match abuse 
  in space knpown to be populated with aggregates AND point
  any complainers to the Right Thing)
  

Do I want the RIRs managing the routing tables? No. Do I 
want registries to hold confirmed data, audit trails 
disambiguated, and everyone playing by the same rules? Yes.

joe, not enough coffee so i hope this is coherent

-- 
             RSUC / GweepNet / Spunk / FnB / Usenix / SAGE



More information about the ARIN-PPML mailing list