Re: [DNSOP] Resolver behaviour with multiple trust anchors

[Bob Harold <[e-mail]>]

On Thu, Nov 2, 2017 at 10:41 AM, Matt Larson <[e-mail]> wrote:

> The root KSK rollover project has given me a real appreciation for the
> brittleness of trust anchor configuration, even with RFC 5011. (Automated
> update support should get better over time, especially after the first KSK
> roll exposes problems, but right now it's pretty shaky, which is my
> informed opinion based on observation from the operational trenches.) In
> the real world, where trust anchors are going to go stale, an "Accept Any
> Success" (in the language of Appendix C) trust anchor policy is the safest
> operationally.
>
> I agree with Ed's observation that operators are, by and large, going to
> use the defaults in whatever implementation they're running, so defaults
> are important.
>
> Trust is always going to be a matter of local policy, so with DNSSEC
> there's never going to be a consistent output given a consistent input. The
> best we can do is give good guidance to implementors, ideally based on
> operational experience, to inform their choices for the default settings
> that operators will end up using.
>
> I think RFC 6840 gets it right: it acknowledges that trust anchor
> preference is a matter of local policy, but recommends an operationally
> safe default of "Accept Any Success".
>
> Why would one want a "Closest Encloser" trust anchor preference policy?
> I've heard two reasons in this thread:
>
> 1. The untrusted parent scenario: I submit there's no practical
> implication of distinguishing between the parent's control of the
> delegation and its control of the DS: the child zone is completely at the
> will of the parent zone, so if your parent has it in for you, you lose.
> This scenario is not sufficient motivation, in my opinion, to suggest
> "Closest Encloser" as a default policy.
>
> 2. In a split DNS context, reject answers that leak into the wrong view: I
> think using DNSSEC as a backstop to enforce split DNS policy is a dubious
> operational practice and likewise not sufficient motivation to suggest
> "Closest Encloser" as a default policy.
>
> In my perfect world, implementations would offer a knob to set "Closest
> Encloser" for consenting adults but default to "Accept Any Success".
>
> Matt
>

I generally agree with you, but wonder if there is a performance penalty to
searching every possible path before failing.  Is that a reasonable concern?

How many paths are there?  I can think of:
1. To the root
2. To a local trust anchor
3. To a DLV provider (gone as of Sept 30?)

If local trust anchors are checked before the root, and they are under
operator control, then maybe "Closest Encloser" is a reasonable default.  I
don't know where to fit DLV into that plan.

Also, if an operator does not configure DLV or local trust anchors, then is
root the only path?  So "Closest Encloser" and "Accept Any Success" are the
same?
Or am I not understanding something (no experience with this yet)?

-- 
Bob Harold