Hi,

Personally I I am very unsure of the wisdom of attempting to use a origination 
authority to infer validation of a proxy aggregation action, regardless of what 
lies inside the AS Set. It makes more sense to me to consider the requirements 
for validation of proxy aggregation as an initial step rather than this attempt 
to retrofit an existing mechanism that has different semantics to the problem. 
However, as I noted last week, I really am unsure whether this WG is chartered 
to look at requirements, and hence my question (still unanswered!) to the WG 
chairs on the charter.

regards,

   Geoff





On 13/04/2010, at 8:40 AM, Sriram, Kotikalapudi wrote:

> Comments inline.
> 
> Sriram
> 
>> -----Original Message-----
>> From: [email protected] [mailto:[email protected]] On Behalf Of 
>> Geoff Huston
>> Sent: Wednesday, April 07, 2010 7:07 PM
>> To: [email protected]
>> Subject: Re: [sidr] Enumerating all scenarios with AS-set (was: RE: Comment 
>> on draft-ietf-
>> sidr-roa-validation-05)
>> 
>> --- snip ---
>> 
>> The efforts to conflate security credentials associated with origination and 
>> proxy
>> aggregation have been less than successful, and as John Scudder pointed out 
>> its probably a
>> far more direct path to look at origination and proxy aggregation as 
>> distinct.  So, in that vein,
>> I'm still comfortable in using the text I proposed earlier in response to 
>> John's original
>> observation, namely:
>> 
>> "A route's "origin AS" is the final element of the route object's
>> AS_PATH attribute.  If the final AS_PATH element is an AS Set,
>> indicating that the route is an aggregate, then the origin AS
>> cannot be determined."
> 
> This approach (i.e., "the origin AS cannot be determined" interpretation 
> stated above)
> is vulnerable to attacks if the validation decision that goes with it is 
> "Unknown".
> (That would be the case if you give the user the benefit of the doubt.) 
> The adversary can insert his false origin as an AS_SET (artificially)
> and thus attempt to get away with it. But this type of attack can
> be prevented as explained below.
> 
> Why not look inside the AS_SET and attempt to validate the update? 
> If there is a single ASN in the AS_SET --
> occurring just once or repeated multiple times -- that is a straight forward 
> case to try 
> to test for validation. Danny's data shows 51 of 61 updates with AS_SETs are 
> like that.
> http://www.ietf.org/mail-archive/web/sidr/current/msg01515.html 
> 
> If there are two or more distinct ASNs in the AS_SET and there is no covering 
> ROA,
> then we go with "Unknown" as validation decision. 
> If there are two or more distinct ASNs in the AS_SET and there is a covering 
> ROA
> and the ASN in the ROA does not match any ASN in the AS_SET, then
> clearly we go with "Invalid" as validation decision.
> I enumerated other cases (9 cases in all) in my earlier email:
> http://www.ietf.org/mail-archive/web/sidr/current/msg01524.html
> I tried not to assign validation outcomes there -- I think one can determine 
> those 
> rationally following the principles of your current validation algorithm.
> 
> This approach (such as the one I proposed above) prevents an adversary 
> from taking advantage of AS_SET to try get past the validator. 
> And in most cases involving AS_SETs (referring again to Danny's data),
> we would be able to decisively evaluate the update as "Valid" or "Invalid" 
> when a covering ROA exists for the prefix.
> Not much shoehorning is involved as I see it
> (unless I am missing something).
> 
> Sriram
>> 
>> This wording, in effect, leaves the entire issue of proxy aggregation and 
>> the associated
>> security credentials that would allow relying parties to be assured of the 
>> integrity of the
>> proxy aggregation action to be validated by some to-be-determined mechanism 
>> that is not
>> necessarily a ROA. In thinking more about the original suggestion of John's 
>> it strikes me
>> that an effort to shoehorn the specification of security credentials 
>> relating to proxy
>> aggregation into the current constraints of a ROA is not going to be a 
>> productive one.
>> 
>>  Geoff
>> 

_______________________________________________
sidr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/sidr

Reply via email to