On Wed, Mar 21, 2012 at 5:17 PM, Eric Osterweil <[email protected]> wrote:
> Hey Chris,
>
> On Mar 21, 2012, at 5:06 PM, Christopher Morrow wrote:
>
>> On Wed, Mar 21, 2012 at 5:04 PM, Eric Osterweil <[email protected]> 
>> wrote:
>>>
>>> On Mar 21, 2012, at 4:57 PM, Christopher Morrow wrote:
>>>
>>>> On Wed, Mar 21, 2012 at 3:19 PM, Eric Osterweil <[email protected]> 
>>>> wrote:
>>>>
>>>>> How about we turn this around with a simple question:
>>>>> Suppose two different feasible paths are being evaluated for a single 
>>>>> prefix/origin pair and one was delivered via a signed bgpsec update, and 
>>>>> the other was delivered via an unsigned update.  What 
>>>>> annotations/influencers/considerations/etc. does the bgpsec design 
>>>>> suggest when the router is making its path selection between these two?
>>>>>
>>>>
>>>> ideall, I think, the process is something like:
>>>>  1) first path gets evaluated, is it 'good' (next-hop reachable, not
>>>> discarded by prefix-list/etc)
>>>>  2) second path gets evaluated, is it 'good' (same as above +
>>>> origin-validate + path-validation)
>>>>
>>>> If both 1 and 2 come out 'ok', assume each path was received from a
>>>> different peer on the same device:
>>>>  1) 1 - 2 - 701 - you
>>>>  2) 1 - 3 - 7018 - you
>>>>
>>>> If you decide that your network's policy is 'prefer signed and valid
>>>> over unsigned' paths, you'd pick #2. It could be that your method of
>>>> doing this is saying:
>>>>  route-map inbound-bgp permit 10
>>>>    if route-signed and verified
>>>>    set localpref 100
>>>>
>>>> Which ideally would let localpref on the #1 route stay normal == 80
>>>> and magically you'd just use the 'right' (from your perspective) path.
>>>> (signed and verified).
>>>>
>>>> of course, in the beginning of this you may choose a more conservative:
>>>>   route-map inbound-bgp permit 10
>>>>      if route-signed and verified
>>>>        set community you:verified
>>>>
>>>> so you could measure the deployment state of your customers + 
>>>> rest-o-interwebs.
>>>>
>>>> In the end, I think 'bgpsec suggests' that the operator would make
>>>> some decision... ideally the same decision across the network.
>>>>
>>>> does that make sense? (the policy is deliberately short and simple,
>>>> obviously real policy is longer/stranger, but the end result is the
>>>> same)
>>>
>>> Indeed, this makes sense, thanks for the detailed response.  However, this 
>>> also illustrates an evolution of semantics under non-attack circumstances. 
>>> :)
>>
>> expand please.
>
>
> Sure, there is now a new element in the BGP update (the sig(s)) whose 
> presence, nature, or absence causes the interpretation of the update to 
> differ.  While I still disagree with the earlier caveat that all bets are off 
> during an attack, that statement clearly does cover this case where one 
> update simply isn't signed.  That's what I meant, though I didn't mean to be 
> cavalier when I said it before... Sorry if it came off as a barb. :)
>

I didn't take it as a barb:) you seemed to have in mind a problem, you
showed it above as well, i was curious what the problem was.

You highlight a problem that, of course, looks to me like a dnssec
sort of thing? "Hey, this domain is supposed to have sigs, it didn't?"

I think in this case you could change the simple example above as:
  1) unsigned path, everything 'ok'
  2) unsigned path, everything 'ok' except there is a prefix with ROA
data and no sigs

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

Reply via email to