In some ways I think building a taxonomy of scenarios that could be called
route-leaks is useful,  but I do think there is a danger is overly fixating
on a precise definition of a commonly used term (i.e., arguing too much as
to what is or is not a route leak).

The real essence of the problem (in my mind) is that the robustness of the
global BGP system, as a practical matter makes assumptions about some
transitive (3+ AS path) properties of announcements, with out any/enough
well know transitive mechanisms in the protocol (other than AS_PATH) to
assist in implementing these assumptions, and or detecting that they have
been violated.

Sending a BGP update to a peer is an implicit authorization  to that peer
to route traffic from them towards the destination prefix.   But the
business models of the Internet are such that sending such authorization to
a peer is not saying I am willing (or engineered) to route traffic from any
and all other networks that my peer might distribute that update too.
 Often the business model / engineering dictate that the sending AS only
intended for that update to be distributed to some subset of the next hops
peers.

No strictly bi-lateral mechanism will assist a 3rd party to be able to
detect that they received an updated that was not the intent of the
announcing peer 2+ hops prior.

Previous attempts/ideas to completely prescribe solutions to specific
problems in this space seem to have failed due to relying on mechanisms
that can not be universally prescribed, or on broad access to data sets
that have privacy concerns.

Our motivation was to think of the problem in this context.   Could one add
simple mechanism to the protocol that would allow interested parties (both
BGP senders and receivers) to indicate some minimal, but useful intended
transitive properties of routes, and allow interested receivers to detect
in some scenarios that these intended properties have been violated.

We chose to look at a set of BGP robustness problems that we could find in
the recent literature and trade press.   We identified a simple signaling
mechanism, not present in the current protocol, that could enable
detection/mitigation of all of the scenarios we found in those documents.

Solutions to this by definition transitive problem that suggest that
everyone just bi-laterally do the right thing, certainly have the problem
that they are not robust to single AS, or implementation, failures
(implementation or configuration), nor are they robust against a single AS
purposefully violating the bi-lateral policy/mechanism.

So, another approach to thinking about the problem is to consider if there
are common BGP robustness problems that could benefit from a mechanism to
tag individual hops along the AS path with additional transitive semantic
information.

If so, what are those problems, what semantics would  tags need to address
the problem, and what properties should those tags have (e.g., security,
known/wellknown, transitivity, etc).

We identified a set of such transitive problems (often called as a "leak"
by most) and proposed a simple transitive tagging mechanism that could
assist in mitigating them.

Other transitive robustness problems maybe addressable with such a
mechanism.   Those problems may or may not be commonly known as "leaks".
Some might identify other scenarios that they might  call "leaks" that
would not be addressable by such mechanisms.

Clearly we need to bound the problem space (in particular, it seems clear
that it is not possible to address all bi-lateral policy violations with a
simple signaling mechanism) .... an there are some folks who refer to any
announcement that violates policy as a leak.

dougm




On Fri, Jul 25, 2014 at 5:23 PM, Christopher Morrow <
[email protected]> wrote:

> as in the room, I'd hope to focus on-list on:
>    "What is a route-leak?"
>
> I thought that a taxonomy:
>    "a scheme of classification."
>
> of what route-leaks/hijacks are would be helpful. It's possible that
> we could call all 'hijacks', 'mis-originations', 'valley-free
> violations'  the same thing: "Route Leaks"... but having a definition
> and a method to view these from either a direct neighbor OR (ideally)
> 2 as-hops away would be much better.
>
> Sriram/doug's document talks about 4 types of leak, of these 1, 2, 3
> all basically look the same to me, the original set is:
>    1 - prefix-hijack with path to legitimate origin
>    2 - u-turn with more specific prefix
>    3 - u-turn with full-prefix
>    4 - internal-prefix-leak
>
> To me this sounds like:
>    1, 2, 3 are all saying the same thing: "Someone leaked all or part
> of my prefix, but provided transit to me anyway"
>
> and 4 is mostly lost to me... though I suppose it might mean: "my peer
> leaked my prefix to his transits"
>
> The goal for this work really ought to be:
>   define route-leak
>
> Then the next follow-on is:
>     "Do operations folks view this thing as defined to be a problem?
> should someone fix that problem?"
>
> -Chris
>
>
> On Fri, Jul 25, 2014 at 3:36 PM, Tony Tauber <[email protected]> wrote:
> >
> > On Fri, Jul 25, 2014 at 3:27 PM, Jared Mauch <[email protected]>
> wrote:
> >>
> >> Communities are not sent by default (eg Cisco). Route leaks come for
> free
> >> on Cisco too.
> >
> >
> > Right, I meant "intentionally implemented" communities.
> > Anyway, so the idea is to introduce a new knob that's on by default and
> > which is more resistant to breaking off?
> >
> > Not so compelling to me as to warrant a wholesale protocol extension.
> >
> > Tony
> >
> >>
> >> On Jul 25, 2014, at 3:12 PM, Tony Tauber <[email protected]> wrote:
> >>
> >> How is this different than tagging with communities today?
> >> In either case, the provider's correct action on the semantics is needed
> >> (and can go awry through misconfiguration).
> >>
> >> Tony
> >>
> >
> >
> > _______________________________________________
> > GROW mailing list
> > [email protected]
> > https://www.ietf.org/mailman/listinfo/grow
> >
>
> _______________________________________________
> GROW mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/grow
>



-- 
DougM at Work
_______________________________________________
GROW mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/grow

Reply via email to