Re: [sidr] I-D Action: draft-ietf-sidr-lta-use-cases-00.txt

[Stephen Kent <[e-mail]>]

Tim,
> Looks like a good starting point to me. Though I had to parse the line 
> about unicorns twice.. 
just twice ;-) ? The text needs work to be clear, precise, and a lot 
less cutesy.
>> Local Trust Anchor Management: http://tools.ietf.org/html/draft-ietf-sidr-ltamgmt-08
> If I understood correctly it was the authors intent to replace the existing ltamgmt document with the new work?
Yes, with two new docs. David Mandelburg is submitting a doc to deal 
with the easiest cases,
and I think he will request a slot to present that doc, plus a doc on 
address space transfer.
>> Suspenders: http://tools.ietf.org/html/draft-kent-sidr-suspenders-00
> Fundamentally I think there is a problem in letting a child refer to a third party that can override its parent. I think it just doesn't fit in the hierarchical rpki, and hence all the complexity to deal with history, and trying to separate noise from signal. I appreciate that it's well intended and a lot of thought has gone into this, but in my opinion this is a very complicated way to deal with this.
I am a firm believer in the hierarchic PKI. But, that said, I do think 
we need a credible solution
to the concerns raised by folks who worry about errors or compelled 
actions by RIRs or ISPs. Suspenders
proposes a fallback strategy to address these concerns. But, if we can 
develop a simpler solution, after
we agree on use cases, that's great.
> What I would suggest instead is to go to the third party directly. I think we already have all the building blocks..
If one goal is to not undermine the hierarchy, then we want constraints 
on what a third party can assert.
Suspenders limits the third party to preserving the status quo; it can 
turn back the clock, but it can't
make arbitrary statement that will be accepted by RPs (if they follow 
the spec).
> This third party can publish a TAL containing resources that it claims to know better. They can then operate a normal CA and publish all the ROAs they see fit, or even act as parent CA using up-down. RPs could be configured to use both TAs and treat them as complementary (i.e. accept the ROAs from both), or exclusive (i.e. ignore the ROAs for the resources listed by third party under any other TA tree), or probably best even: alert the operator and let them choose and set defaults.
That is precisely the sort of design that has the potential to undermine 
the hierarchy. And it seems
to head in the direction of the awful Web PKI we have today.
> To deal with Carol's case, well-known third parties could be set up. If all is well they should have no content, but the key difference is that it would no longer be possible to do a *covert* attack on Carol. I understand that it's re-active rather than pro-active, but I think this is enough to make the attack moot: it's not very effective and it has drawbacks: it degrades trust and thereby security of internet infrastructure.
The attack on Carol is not covert; every RP will see it. The only 
question is what RPs should do
in response. Allowing Carol to publish info that makes it clear that she 
doesn't agree with the
changes to her RPKI data is probably the least dangerous way to provide 
RPs with the info from
which they can make a decision.
> Bob can just create a complementary TAL for the private space.
>
> Alice can create a TAL that takes precedence, and have her management's vision of the truth.
>
> All this needs some tooling, but I don't think it needs more standards.
>
The message I just posted noted that several of the use cases need 
clearer statements of the
problems being addressed, and a less colloquial tone. So I won't 
comments on how we should address
the other use cases until we have agreement on them.

Steve