>> 
>> If that is the case, having a set of policy objects expressing AS 
>> relationship should do the same
>> thing  and more with less overhead? (yes, I know that data integrity becomes 
>> an issue, but data
>> integrity is always an issue.)


I was deliberately keeping away from participating in this conversation, but I 
have researched this topic in some depth in recent times 
(http://www.potaroo.net/papers/BGP_Security_Literature_Review.pdf), and I think 
that I have some familiarity with this topic that may be helpful to others who 
are also watching this and trying to understand what is being debated here.

What I see here is a dichotomy between two distinct architectural approaches, 
and the conversation appears to flip from one to the other when we discuss path 
validation. 

One view is that we are securing the operation of a particular protocol, namely 
BGP. This protocol operates in a standard distance vector mode of 
AShop-by-AShop, where information is propagated outward from the causal point. 
If you want to secure this protocol, then there are two fundamental attributes 
of the protocol's transitive data objects that are securable, namely the 
address prefix and the AS path. The AS path is constructed incrementally in 
this protocol, and the most direct mechanism that would allow it to be secured 
is for each AS to be able to demonstrate the validity of what it received and 
what it added. I.e. incremental signing of elements of the AS path as the basic 
unit of routing information propagates across the inter-AS network. To the 
extent that this is an accurate reflection of the underlying inter AS 
connection topology, EBGP multi-hop notwithstanding, this is an accurate 
reflection of a subset of the interAS topology, and it allows a BGP speaker the 
ability to validate, in the context of the operation of BGP the protocol, 
information that is being propagated by BGP.

The other view I see is a reflection of the observation that BGP, like any 
other inter-domain routing protocol, is a mechanism that "discovers" the 
inter-AS topology of the network. If you already "know" the "real" topology via 
some other mechanism, then you can use this information as a filter for 
information being passed to you in the routing protocol. i.e. In this instance 
if you had complete information of all "valid" AS adjacencies then you could 
use this as an information filter for AS paths that you are being presented 
with via BGP. If an AS path contains an assertion of an AS adjacency that is 
not in your local "map then you have grounds to doubt that the information you 
are be presented with in this particular update is in fact "valid"

Superficially, this second approach is seductively attractive: its simple to 
conceptualise, relatively lightweight and probably fast in terms of processing 
overheads, and makes the minimum of assumptions as to the way the protocol 
itself operates. This decoupling of the propagation of the security credentials 
from the protocol operations also appears to be consistent with the approach 
being taken to securing origination - the security credentials are flooded via 
some other mechanism and they are used locally to maintain a local route filter 
which is then used to pass / reject incoming route updates in BGP based on the 
implicit AS adjacencies that are being asserted in the AS Path.

But this is a superficial attraction as I see it. This decoupled approach still 
admits lies in routing. It limits the extent to which one can lie, but it still 
admits lies. 

However, I don't think this aspect of the issue of AS Plausibility vs AS Path 
Provability is the core of the issue _for me_ (others see it differently and 
want a stricter condition). More importantly for me at any rate, is that this 
decoupling of topology information from the operation of the protocol 
potentially stops the operation of the routing protocol to be a topology 
discovery tool. If AS 1 and AS2 bring up a peering session, then no one else 
will believe that adjacency (or the routes that use that adjacency) until the 
security credentials of the pairwise adjacency of AS1 and AS2 has already 
flooded the network. And how does this association get flooded? Equally, how do 
you inform everyone else that a peering session has been torn down? Or that 
this AS adjacency has some policy constraints that are intended to limit the 
scope of its visibility? As long as you are prepared to limit your concept of 
the scope of a routing protocol to be one of best path computation over a 
relatively fixed topology that all players know of in advance then this idea of 
a decoupled approach to propagation of AS adjacencies and the operation of the 
protocol itself could well be a viable approach to securing the outcomes of the 
operation of the protocol, or at least limiting the range of potential ways to 
lie in the protocol. But if you also want some level of dynamic discovery of 
the underlying topology as well as path computation in the scope of the 
operation of the protocol, then you are asking a lot of this form of decoupling 
of securing the underlying inter-AS topology from the operation of the routing 
protocol, because at some point you need to distribute the information of a new 
AS adjacency at the same time as you want to pass the routing protocol packet. 
I.e. If you want to secure what BGP is telling you in the AS Path then it makes 
sense to use a mechanism that is directly associated with the manner of the 
construction of the AS Path itself, within BGP itself. And if you couple the 
security credentials to the protocol not only do you get dynamic behaviour of 
topology discovery, you also get AS Provability in so far as you are now back 
to being able to validate that the path of propagation of the update matches 
the path describes in the update.

(I note that the same argument of using in-band security credentials to enable 
dynamic discovery of routing information could be applied to origination 
security credentials and their manner of distribution, and I also note that the 
WG decided not to go down that path for securing origination. But having made 
that decision, I don't think that this decision for the manner of securing 
origination necessarily precludes the WG from using the BGP protocol mechanisms 
to bind the AS Path and the AS Path security credentials together. I do however 
think that sBGP, which as I recall bundles both origination and path security 
credentials into the BGP protocol payload together with the routing data that 
the credentials relate to, still strikes me as an approach that has an 
attractive level of architectural consistency. :-) )

The above is all my own opinion, naturally! Now to clamber off this soap box 
and head back to hide in my corner!

regards,

  Geoff


Attachment: smime.p7s
Description: S/MIME cryptographic signature

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

Reply via email to