Re: [Rats] New RATS Architecture document

[Henk Birkholz <[e-mail]>]

I tried to be as quick as possible - hence the typos (I seem to be 
missing a lot of "be"s...).

A document that uses normative requirements terminology is standards 
track. Only one section of this I-D uses these - the authors tried not 
to be over-perscriptive, but these are quite essential in order to 
enable basic interoparbility between different solutions. As we cannot 
go through all open issues the authors were able to collected thanks to 
the replies on the list (and thank you all for all the great feedback) 
tomorrow, I think this is one of the issue we can discuss. If you could 
vet the one place where normative language is used for tomorrows meeting 
and highlight the (sub-)issues in there - that would be great!

Viele Grüße,

Henk

On 08.10.19 00:32, Giridhar Mandyam wrote:
> Hi Henk,
> 
> Thanks for the quick response.
> 
> I will focus on the standards-track versus informational aspect of the document.  For the RAts chairs, please plan on addressing this topic in tomorrow's virtual meeting.
> 
>> Does that mean the document is not ready for adoption because it is intended to be standards track or does this mean in order to be standards track there are things missing or wrong?
> 
> I believe the document should be informational.  I have not identified any aspect of the current document which corresponds to an interoperable and testable specification.  For instance - there are no protocols being defined, fixed message formats, or mandatory verification procedures.
> 
>> In order to enable interoperability, a very tight scope of normative prerequisites is imposed on the architectural constituents of RATS by the architecture I-D in the Normative Prerequisite section (that use BCP14 requirement notation).
> 
> So does this mean that any RAts deployment that does not strictly conform with the architecture outlined in this document would be considered non-compliant?  I hope this architectural model is sufficiently broad to cover all the corner cases then.  So far in what I've observed in my review, I am not sure that will be an easy task.
> 
>> Additionally, the architecture I-D of the "other" WG that is concerned with security posture of ToE - the SACM WG - is an example for also not being an informative I-D.
> 
> And I cannot understand why the SACM architecture is not informational either.  I do on the other hand understand why the WebRTC sec arch document (https://tools.ietf.org/html/draft-ietf-rtcweb-security-arch-20) is normative - there is an extension of SDP (https://tools.ietf.org/html/draft-ietf-rtcweb-security-arch-20#section-5).
> 
>> As far as I can see the normative prerequisites do not conflict with EAT, but support the architectural parts of EAT (which move into the RAT architecture I-D in any case).
> 
> The architectural aspects of EAT will certainly move into this doc, but they are non-normative - see Sec. 1.3 of https://www.ietf.org/id/draft-ietf-rats-eat-01.txt. Note the introductory text:  " At least the following three participants exist in all EAT operating models.  Some operating models have additional participants."
> 
> -Giri
> 
> -----Original Message-----
> From: RATS <[e-mail]> On Behalf Of Henk Birkholz
> Sent: Monday, October 7, 2019 3:06 PM
> To: Giridhar Mandyam <[e-mail]>; [e-mail]
> Subject: Re: [Rats] New RATS Architecture document
> 
> -------------------------------------------------------------------------
> CAUTION: This email originated from outside of the organization.
> -------------------------------------------------------------------------
> 
> Hi Giri,
> hi list,
> 
> please find comments and replies in-line.
> 
> On 07.10.19 23:02, Giridhar Mandyam wrote:
>> I have read through the latest version of the document.  At this time I do not think this document is ready to be the RAts architecture specirfication, and would object to its adoption as a Working Group draft.  I do believe the group needs an architecture document, however.  So if the editors can address my primary concern - a) below -  I would be happy to withdraw my objection.
> 
> Does that mean the document is not ready for adoption because it is intended to be standards track or does this mean in order to be standards track there are things missing or wrong?
> 
>>
>> -Giri Mandyam
>>
>> a) I have asked the editors twice on the mailing lists for a justification as to why this is a standards track draft and not an informative one.  I have not received an answer, so I assume that there is no reason.  Note that comparable documents such as the TEEP architecture document and SUIT architecture document are informative.
> 
> In order to enable interoperability, a very tight scope of normative prerequisites is imposed on the architectural constituents of RATS by the architecture I-D in the Normative Prerequisite section (that uses
> BCP14 requirement notation).
> 
> Additionally, the architecture I-D of the "other" WG that is concerned with security posture of ToE - the SACM WG - is an example for also not being an informative I-D.
> 
> As far as I can see the normative prerequisites do not conflict with EAT, but support the architectural parts of EAT (which move into the RAT architecture I-D in any case).
> 
>>
>> b)  "... this document provides normative guidance how to use, create or adopt network protocols that facilitate RATS".  So in the context of EAT, does this mean that guidance provided in this document supersedes the EAT, CWT, and JWT specifications?
> 
> It complements EAT with corresponding definitions. I do not see how it would be possible to supersede CWT or JWT with respect to remote attestation procedures. That seems to a totally different context or am I missing something very obvious here?
> 
>>
>> b)  "... trustworthiness claims are accompanied by a proof of veracity.  Typically, this proof is a cryptographic expression such as a digital signature or a message digest".  This is a misleading statement, as it could be interpreted to cover self-signing (see [3] for example), and I don't believe self-signing conveys any indication of trustworthiness.
> 
> While in a given scope, putting trust in self-signing can absolutely
> warranted, I think, I also think that was not the intended background
> for that text passage. Will be fixed by elaborating on that.
> 
> Self-signing (while not being a signing fool due to, e.g., missing key
> semantics), should be further discussed though. Again, this text was not
> phrased with self-signing in mind, I think.
> 
>>
>> c) As per h) of my previous review [4], I still cannotAndroid Key Attestation figure out how the concept of delegated attestation (e.g. Sec. 5.4 of [5]) would fit into this architecture.  For an informational document, it may not be strictly necessary to specify an architecture that can support delegated attestation since an informational document would likely not be expected to cover all scenarios.  Since this is being proposed as a normative specification however, we should ensure that it does not disallow what could be common remote attestation deployments.
> 
> Agreed, the text is lacking in this regard. Every proposed text with
> respect to "Android Key Attestation" is welcomed. We did the best we
> could based on the use cases I-D.
> 
>>
>> For instance, as per the diagram in Figure 1, all evidence must be conveyed directly to the verifier and the verifier in turn produces "attestation results" to the RP.  But "attestation results" are the "output from the appraisal of Evidence, Known-Good-Values and Endorsements".  I take this to mean that attestation results cannot be evidence themselves.  However, the delegated attestation model in [5] allows for the appraiser (approximately equivalent to the RP as per the RATS architecture doc) to compose "separate layered or orthogonal attestations, involving different proxies, in order to make a judgment".
> 
> Figure 1 is the most simple example. We are already working on moving
> that to section 1.1. It is also not using normative language. No RATS
> message must move exactly along those edges, as there can be various
> other ways to compose these workflows (e.g. Dave's work-flows).
> 
>>
>> As per Sec. 4.4 RATS Principals, the multiplicity property may allow for multiple verifiers to take on the role of attestation proxies.  But multiplicity explicitly calls out that the instances of the principals take on the "same RATS roles".  If there are multiple proxies, then it seems that they would be assessing different subsets of evidence and therefore would be occupying different roles.
> 
> "Same role" does not imply "same task/duty". It is a class of duties.
> E.g. appraisal of evidence can be split into multiple RATS principals,
> all taking on the Verifier Role, but appraising different portions of
> the evidence. The current phrasing seems to be lacking here - this
> should be more clear, if it reads like "different subsets of evidence
> require different roles with respect to appraisal of evidence". That is
> most certainty not the intent.
> 
>>
>> d) RATS roles:  composition:  "RATS Principals possessing different RATS Roles can be combined in to a singleton ...".  I can understand some aspects of this property, e.g. combining the RP and Verifier.  But why would an attester and/or asserter be combined with a verifier and RP?  Do you know of an example?
> 
> The Opal Storage Specification might be a good example. And RIV is
> another use case, where this can be the case, I think.
> 
>>
>> f) {Sorry to jump back}  Sec. 3.1.  "A Computing Environment with the capability of remote attestation ... is separate from other Attested Computing Environments (about which attestation evidence is created ...".  I could not tell if this definition allowed for solutions such as Android SafetyNet [6], where the rich execution environment (REE) is producing evidence about itself.  Is the intention of the RATS Architecture spec that such self-attestations not be considered valid for remote attestation?
> 
> The definition of attestation in
> [https://www.w3.org/TR/webauthn/#attestation] states: it "is employed to
> attest to the provenance of an authenticator and the data it emits".
> While this is very authentication-centric, first of all, your reference
> [6] also states that the Android SafetyNet Attestation Statement Format:
> "does not provide information regarding provenance of the authenticator
> and its associated data". As a result "platform-provided authenticators
> SHOULD make use of the Android Key Attestation when available, even if
> the SafetyNet API is also present."
> 
> As "statement" in that context seems to be most semantically related to
> what "Claims" are in RATS, it seems that actually - no, that does not
> seem to be evidence, but claim sets. In combination with Android Key
> Attestation, these WebAuth statements seem to become evidence.
> 
> But I am not an expert on authentication-centric attestation statements.
> 
>>
>> g) Sec. 3.2.  "The RATS architecture anticipates recursive trustworthiness properties and the need for termination".   What is being terminated in this context?
> 
> If no attesting computational environment can create believable evidence
> about a system component in a hierarchy of system components in, e.g., a
> composite device - that is a point where recursive trustworthiness
> properties terminate (and external endorsement are required). Ned might
> be able to provide a better example.
> 
>>
>> h) Sec. 3.2.  nit.  "For example, design reviews, manufacturing process audits and physical security." This is not a complete sentence.
> 
> This is not a complete sentence. Will be fixed.
> 
>>
>> i)  Sec. 3.3.  "RATS architecture defines attestation Roles ..., the messages they exchange, ...".  Don't specifications such as EAT already define messaging in part?  If so, what would the RATS architecture be defining in addition?
> 
> That is defined in the section RATS messages.
> 
>>
>> [1] https://datatracker.ietf.org/doc/draft-ietf-teep-architecture/.
>> [2] https://datatracker.ietf.org/doc/draft-ietf-suit-architecture/.
>> [3]  https://www.w3.org/TR/webauthn/#self-attestation.
>> [4] https://mailarchive.ietf.org/arch/msg/rats/9iIrnD9YixKCuwOGKtvGZIzd_bU.
>> [5] J. Guttman et al.  “”Attestation:  Evidence and Trust”.  MITRE Technical Report MTR080072.  March 2008.  https://www.mitre.org/sites/default/files/pdf/07_0186.pdf.
>> [6] https://www.w3.org/TR/webauthn/#android-safetynet-attestation.
> 
> IHTH as start!
> 
> Viele Grüße,
> 
> Henk
> 
>>
>>
>> -----Original Message-----
>> From: RATS <[e-mail]> On Behalf Of Giridhar Mandyam
>> Sent: Tuesday, September 10, 2019 9:29 AM
>> To: [e-mail]
>> Subject: Re: [Rats] New RATS Architecture document
>>
>> -------------------------------------------------------------------------
>> CAUTION: This email originated from outside of the organization.
>> -------------------------------------------------------------------------
>>
>> Henk and other editors of the architecture doc,
>>
>> I appreciate the effort put into this latest revision, and I will review it with a fresh eye.
>>
>> I do have one overarching question:  why is this document listed as 'Standards Track' (as opposed to 'Informational')?  I had expressed my concerns about this in an earlier review of the doc:  https://mailarchive.ietf.org/arch/msg/rats/9iIrnD9YixKCuwOGKtvGZIzd_bU.
>>
>> Since the editors' team is composed of experienced IETF'ers, I assume that there are solid reasons for doing so.  If you all have stated them before, I hope you do not mind me requesting that you lay out your reasons again because I seemed to have missed it.
>>
>> -Giri
>>
>> -----Original Message-----
>> From: RATS <[e-mail]> On Behalf Of Henk Birkholz
>> Sent: Tuesday, September 10, 2019 6:13 AM
>> To: [e-mail]
>> Subject: [Rats] New RATS Architecture document
>>
>> -------------------------------------------------------------------------
>> CAUTION: This email originated from outside of the organization.
>> -------------------------------------------------------------------------
>>
>> Hi all,
>>
>> we created a fully revised architecture document that maps and represents the state of the current discussion and the material presented at the last IETF meeting.
>>
>> The current Editor's version can be found here:
>>
>>> https://ietf-rats.github.io/draft-birkholz-rats-architecture/draft-bir
>>> kholz-rats-architecture.html
>>
>> We will submit a new version the day after the RATS virtual interim.
>>
>> TL;DR
>> Below you can find a list of essential changes & a list of action items still to be addressed.
>>
>>
>> This version of the RATS Architecture document:
>>
>> * does not define or uses the terms "root(s) of trust" (RoT) or "Trust Anchor" (TA) at the moment. (Note: It is a fact that the Asserter Role _is_ a TA for the Verifier Role and that an Attester Role _could_ rely on RoTs. But - this content will not go into the main body of this document),
>>
>> * does define RATS Roles, Messages, and Principals formerly known as "Actors" (borrowing heavily from ABLP),
>>
>> * provides an even more "base" interaction model diagram for the RATS Roles than presented in the last IETF meeting slide deck:
>>
>>> https://datatracker.ietf.org/meeting/105/materials/slides-105-rats-ses
>>> sb-rats-architecture-interaction-model-challange-response-yang-module-
>>> information-module-00.pdf
>>
>> * introduces a framework for "level of confidence" in the trustworthiness of an Attester and the endorsement of the protection characteristics of its "Attesting Computing Context", allowing for other entities to use this framework and fill it with, e.g., openly defined levels of confidence metrics,
>>
>> * is not based on the primitive of "trust" but the concept of "trustworthiness" as illustrated by the RATS charter,
>>
>> * simplifies the definitions of Attester and Verifier that seemed to have caused some unfortunate confusion following the proposal of Giri and starting with commonly-accepted definitions and then justify why they may need to be modified,
>>
>> * differentiates between the Attesting Computing Environment and the Attested Computing Environment better, which both are components of an Attester,
>>
>> * uses the "Claim" concept as a building block to compose Evidence, Known-Good-Values and Endorsements. Conversely, the "Assertion" concept is dropped in this proposal (as initially suggested by Laurence, IIRC?).
>> (Note: this was done to simplify the discussion about the information model. Please also note: Due to the {J|C}WT definition of "Claim", a key/value pair is implied, which is already a data model decision and not mandated by an information model), and
>>
>> * analogously, now uses the term Known-Good-Values instead of Attestation Assertions.
>>
>>
>> For future versions the authors intent:
>>
>> * to elaborate on the use of RATS Principals, including more exemplary
>> diagrams of RATS Role composition and interaction between RATS
>> Principals based on the use case document (and by that address a unified
>> mapping to TEEP, RIV, and models that use EAT),
>>
>> * to shift some of the focus on technical-trust as proposed by Thomas.
>> (the Endorsements provided by an Asserter are a first step into that
>> direction),
>>
>> * still not to define the roots of trust terms nor invent new words for
>> them :) But - start to reference them on a minimal level and define a
>> base set of primitives they can provide in order to describe what they
>> actually are and can do in the context of RATS as proposed by Ira, Simon
>> and Thomas,
>>
>> * to introduce and define a concise scope for layered attestation,
>> addressing, e.g., the staging of Computing Environments and the
>> (un-)availability of an Attesting Computing Environment at certain
>> points of time, or, another example given, addressing the
>> differentiation of an attested boot sequence of an Attester and an
>> Attester running TEEs or rich systems for years,
>>
>> * to address the change of Roles of a Principal over time as proposed by
>> Ian,
>>
>> * to move the remaining architectural sections in the EAT draft into the
>> RATS Architecture draft, and
>>
>> * to shift some of the focus on the out-of-band trust establishment in
>> order to illustrate a coherent RATS ecosystem (e.g. the provisioning of
>> key material is not include in the "base diagram" anymore for now - this
>> will be more elaborated on in future section of the architecture).
>>
>>
>> Viele Grüße,
>>
>> Henk
>>
>>
>>
>>
>>
>>
>> _______________________________________________
>> RATS mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/rats
>> _______________________________________________
>> RATS mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/rats
>> _______________________________________________
>> RATS mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/rats
>>
> 
> _______________________________________________
> RATS mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/rats
> _______________________________________________
> RATS mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/rats
>