[Rats] a quick look at PKIX evidence

[Thomas Fossati <[e-mail]>]

Hi,

I read draft-ounsworth-rats-pkix-evidence-00 and have a few
observations which I wanted to share.

The TL;DR of this document is it provides an ASN.1-based encoding
format for attestation evidence produced by HSMs (and similar gizmo)
which reuses (and slightly extends) EAT semantics.

I think this is a laudable goal and personally support it.

Two tiny comments regarding the claims:
* Extensible EAT claims (e.g., Measurements, Manifests) have specific
extension mechanics, and may be tricky to deal with in another
context. (At a minimum, an extra coordination layer is probably
necessary.)
* Submods are associative arrays: not sure how ASN.1 deals with that
in non-clumsy ways (SEQUENCE OF SEQUENCE { name, CHOICE { value_1 ...
value_n } } ) ? :-)


The point where I stop agreeing with the direction the document is
headed to is section 6.

This is where it is explained how these evidence claims can be
re-published as attestation results.

I have two kinds of concerns with that:
1. privacy
2. architectural

The privacy recommendations in Section 8 are IMO not sufficient.   The
draft says that "the CA MUST have a configurable mechanism to control
re-publishing", but I think the subject here should be explicitly the
attester.  It should read: "the attester MUST be able to control
granularity (including opt-out)". But this is a bit tricky because
it's a requirement that trickles through any enrolment protocol (ACME,
etc.) - i.e., every conveyance protocol / API that allows
re-publishing must offer a way to opt-out altogether and/or select
what can be exposed.  Tough stuff.

The other concerning dimension is architectural: i.e., I don’t think
this is the right approach.

The problem is it creates a strong coupling between attesters and RPs.

Consequently, RP policy complexity grows unbounded because it needs to
take care of all possible attesters’ claims-set flavours.  (If you’ve
ever had to read through one of those seemingly infinite and
infinitely growing rego files you know what I am talking about.)
You may argue that the standardised claims-set put forward by the
draft is enough to create an homogenous attesters' space, but that's
not the case - cfr. EAT profiles.  Besides, note that this kind of
re-publishing could be, in principle, be used for attesters that share
the same info model, i.e., EAT-based attesters.

To counter the looming scalability disaster, the right re-publishing
model is instead one which provides abstract trust categories and a
standardised scoring system - e.g., AR4SI.

This way no privacy-related disclosures are ever possible and all RPs
can be instructed to use a normalised set of high-level attributes to
describe their policies, which scales optimally.

cheers, thanks!
t