Re: [Gen-art] Gen-ART LC review of draft-ietf-sidr-roa-validation-10.txt

[Brian E Carpenter <[e-mail]>]

On 2011-04-12 14:15, Sandra Murphy wrote:
> 
> 
> On Sun, 10 Apr 2011, Brian E Carpenter wrote:
> 
>> On 2011-04-10 03:13, Geoff Huston wrote:
>>> On 09/04/2011, at 10:34 PM, Brian E Carpenter wrote:
>>>
>>>> Please see attached review.
>>>>
>>>> <draft-ietf-sidr-roa-validation-10-carpenter.txt>
>>>
>>>
>>>
>>> Thanks Brian,
>>>
>>> I'm not sure as to the appropriate form of response, but let me
>>> respond to the two minor issues you identified in this review (many
>>> thanks for the review, by the way)
>>
>> This is just fine... the gen-art archive is public, so replying here
>> is on the record.
>>
>>>> 3.  Applying Validation Outcomes to Route Selection
>>> ...
>>>>      "valid" is to be preferred over
>>>>      "unknown", which is to be preferred over
>>>>      "invalid".
>>> ...
>>>>   It is a matter of local routing policy as to the actions to be
>>>>   undertaken by a routing entity in processing those routes with
>>>>   "unknown" validity states.
>>>
>>> you commented that:
>>>
>>> That seems to leave open the possibility that an aggregated route (which
>>> is by definition "unknown") would be rejected. Assuming that the various
>>> separate routes that were aggregated together never reached this
>>> particular
>>> router, the result would be a black hole. At the least, it seems that
>>> this
>>> should be mentioned, even if it is an intentional possibility.
>>>
>>> But the next sentence in the document states:
>>>
>>>    Due to considerations of partial use of
>>>    ROAs in heterogeneous environments, such as in the public Internet,
>>>    it is advised that local policy settings should not result in
>>>    "unknown" validity state outcomes being considered as sufficient
>>>    grounds to reject a route outright from further consideration as a
>>>    local "best" route.
>>
>> Yes, indeed, and all I was thinking of was adding a sentence here saying
>> that the "unknown" category might include aggregates. It's a direct
>> implication
>> of the earlier text but I tend to assume that not all readers will
>> notice all implications.
> 
> I don't know that I quite understand your objection in the first place.
> Aggregates are not *by definition* "unknown" unless they are proxy
> aggregates.  Normal aggreages exist and are valid.  An ISP with a /16
> who suballocates /18s to customers and then propogates only the
> aggregate /16 can issue a ROA for the /16 itself and make the aggregate
> valid.
> 
> And proxy aggregates implies AS_SETs which as Geoff notes are being
> deprecated and are formally ruled out of bounds in the sidr work.

That was new to me, but in any case I was not raising an objection.
I was only suggesting that implementors reading the document might
miss this point when planning their policy. If they intend to respect
proxy aggregates (which is their choice whatever the IETF says) they
would need to allow for this rather than choosing to automatically
reject all 'unknown'.

> 
> And "unknown" routes don't necessarily result in a blackhole.  If
> "unknown" routes are acceptable by local policy, the traffic will get
> through.  The only way a definition of "unknown" would result in a
> blackhole would be if the local policy was so strict that it would
> reject its only route to a prefix.  Local shoot-foot policy decisions
> can occur already for many different reasons, this is nothing new.

Indeed. But again: my point was that readers might not think through
all the implications, so give them a warning. "Note that as indicated
above, "unknown" routes may include proxy aggregates."

>>
>>> Also, given the current proposal in the IDR WG to deprecate the use
>>> of AS Sets (and by implication deprecate the (rarely used if ever)
>>> practice of proxy aggregation, I am unsure of the need to call out
>>> proxy aggregation in this context.
>>
>> I won't get into that debate ;-)
>>
>>>> 5.  Route Validation Lifetime
>>>>
>>>>   The "lifetime" of a validation outcome refers to the time period
>>>>   during which the original validation outcome can be still applied.
>>>>   The implicit assumption here is that when the validation lifetime
>>>>   expires the routing object should be re-tested for validity.
>>>
>>> you commented that:
>>>
>>> OK, but shouldn't a previously "valid" route be downgraded to
>>> "unknown" after the lifetime expires and until the validity has
>>> been re-tested?
>>>
>>> Not necessarily. When a route is validated, the validation lifetime
>>> refers
>>> to the validation time of the EE cert used to sign that ROA. When the
>>> ROA is no longer valid the route should be re-tested for validity. It it
>>> possible that there is another ROA that still validates the route, or
>>> in the
>>> absence of the ROA that previously validated the route, the route may
>>> be considered invalid (i.e. there is an AS 0 ROA still extant that
>>> encompasses
>>> this prefix). For this reason the text specifically indicates that the
>>> appropriate action is to retest the route for validity in the context
>>> of the
>>> current local cache of valid ROAs.
>>
>> Yes, but I have no sense of the timing - would the time taken to
>> revalidate
>> the route leave a long enough gap for some kind of security exposure
>> while
>> an expired route was still marked as valid? Maybe I am worrying about
>> nothing.
> 
> Information necessary to revalidate the route should be available
> locally, and so should take little time.

OK, but that was not immediately obvious to me.

>>
>>>
>>>
>>> At this stage I am unsure if changes to the draft are warranted, as
>>> I believe that the issues you highlight here are addressed in the
>>> document as it stands.
>>
>> Sure, these were minor comments, unless your shepherd feels differently.
>>
> 
> I am not convinced that changes to the draft are necessary here.

That's your call.

   Brian

> 
>>   Brian
>>
> 
> --Sandy
> 
>