Re: [jose] New Version Notification for draft-barnes-jose-jsms-00.txt

[Richard L. Barnes <[e-mail]>]

Hey Mike,

Thanks for having some back-and-forth with me on this.  I think this is a good discussion to have right now.  I would like to discuss some high-level points, and then I'll respond inline to your precise points.

tl;dr: In order to meet its deliverables, the WG needs to define JSON-based formats for integrity and confidentiality protections.  JWE and JWS don't fit that requirement, so we should re-consider their adoption and call for new candidates.

Major points:
1. JOSE is JSON
2. JSON does not imply canonicalization
3. Compactness and URL-safeness are still OK, as derivatives

Part 1: JOSE is JSON

The JOSE WG charter calls out four deliverables, three of which are explicitly required to be JSON:
-- "JSON-structured integrity protection"
-- "JSON-structured encryption"
-- "public keys as JSON-structured objects"
(The fourth is just a list of algorithms, which should be JSON as well, just as RFC 3370 is ASN.1.)

In light of these requirements, JWE and JWS are clearly not suitable candidates to fulfill the working group's milestones.  (At least in their current form.)  So ISTM that we need to step back and re-evaluate alternatives for the three above milestones, especially the first two.  (JWK is not as far off.)

As JSON-based mechanisms, though, the "JSON serialization" drafts are pretty half-hearted.  Because JWS and JWE include the cryptographic parameters within the integrity protection, they have to have the headers included as base64-encoded strings.  That obviously adds a lot of complexity to encoding and decoding -- for example, you have to keep the encoded header around even after you've decoded it! -- and just plain more octets.  The support for multiple recipients and multiple signatures is both unweildy (matching up parallel indices for headers and signatures) and inefficient (multiple integrity check values in encrypted objects).

By contrast, JSMS takes advantage of the cryptographic wisdom that went into CMS, while making the syntax pure JSON and as simple as possible.  Multiple signatures and multiple recipients are supported with absolute minimum overhead.  So I think it's worth consideration by the working group.


Part 2: JSON !=> canonicalization

I would like to dispel the myth that a JSON-based format requires canonicalization.  In JWE/JWS, the JSON header requires "canonicalization" (via base64url encoding) because it's included within the integrity protection.  I'm not aware of any reason for including the cryptographic parameters in the integrity check, instead of only integrity-protecting the content itself.  Indeed, if there were an attack based on non-integrity-protected cryptographic parameters, it would be a pretty big deal, since neither CMS nor IKE nor provides this protection.  (TLS does provide a signature over the algorithms, but only as a way to bind the public key to the session.)

Indeed CMS doesn't require canonicalization at all, except for attributes that are included in the protected content (signedAttrs / authenticatedAttrs), which need to be DER-encoded.  In JSMS, we don't allow those parameters, so there is no need for canonicalization at all.  All the "canonicalization" that's needed is to base64url-encode the protected content.  In that regard, JSMS is even simpler than JWE/JWS.


Part 3: Compactness and URL-safeness are still OK, as derivatives

I have absolutely no objection to defining a compact, URL-safe format.  Indeed, I did so in the JSMS draft, in Section 5.  Encoding the example from Section 3.1 of the JWS spec, we get the following JSMS object:

{ 
    "v":1,
    "t":"au",
    "a":"hs256",
    "c":"eyJpc3MiOiJqb2...0cnVlfQ",
    "mac":"pwM0gJNMLFxn8G1-juQFLrJtMEOR8_FeKIH3wIM_OAo",
    "ki":"rSIilRk8Ego"
}

If we remove the decorative spaces and base64url-encode, we only use 88 octets more than JWS (267 vs. 179).  If we take this a step further and define an explicit compact format for AuthenticatedData, we can emulate JWS by moving the content and MAC fields outside and dot-combining.  That avoids double-base64url-encoding, and only takes up 11 more octets than a JWS (190 vs 179).  (I've put the code to generate these in the Github project linked earlier.)

So it's fine to have a compact, URL-safe format for an objective, but it makes *much* more sense to derive that from a full-size, usable, JSON format than the other way around.

Responses to specific points inline below.


> Fair enough.  I'll briefly respond to each in turn.  Before doing so, I'll say that I'm writing in the spirit of trying to increase mutual understanding and consensus, rather than taking the point of view that the status quo is the only reasonable outcome.
> 
> -- Not being JSON
> 
> This was by design, so that the results are URL-safe.  This is required, for instance, for them to be used as OAuth Bearer Tokens.  At the request of the working group, closely related pure-JSON drafts were also published.  See:
> 
> http://tools.ietf.org/html/draft-jones-json-web-signature-json-serialization-01 and
> http://tools.ietf.org/html/draft-jones-json-web-encryption-json-serialization-01
> 
> Use them if you prefer if they better suit your use cases.  They could be adopted as working group documents, if there's interest.

See above.  These are super-clunky, and the need to be the main focus of the WG, not an afterthought.


> -- Not supporting multiple recipients
> 
> The JSON serialization drafts above do, again, in response to use cases presented by the working group.


See above.  Because of the unnecessary inclusion of the header in the integrity check, this is far less efficient than it should be.


> -- Lacking clear distinctions between signing and MAC
> 
> Functionally, these are very similar operations, at least from an implementation perspective, so it's not clear that making a clear distinction would help either developers or adoption in any practical way.  They achieve overlapping, but admittedly sometimes different goals.  I believe that developers can be trusted to use the algorithms that achieve their applications' goals without making a hard distinction between the two.  But I'd be interested in hearing your thoughts on why you believe a brighter-line distinction might be useful or necessary.

It's important if you want to do MACs with anything other than pre-shared keys.  If you want to do anything more advanced, you need to have wrapped keys, in which case an object with a MAC starts to look a lot more like a JWE than a JWS.


> -- Lacking clear processing instructions for different key establishment modalities
> 
> This seems like an area where the existing drafts could be refined to add descriptions for additional establishment modalities, possibly based upon your ideas and descriptions, should the working group decide to do so.  What additional modalities do you have use cases for that you believe are not achievable or straightforward with the existing drafts?  The existing modalities supported are largely those that existing developers had told us they needed and draw both from the precursor draft-jones-json-web-* documents and draft-rescorla-jsms.
> 
> Bear in mind that the balance that the current drafts try to strike is keeping the Mandatory to Implement (MTI) feature footprint small enough that developers, including Web developers, are likely to adopt the specs, while still providing enough functionality to make them generally useful.  As in any balance, it's often possible to shift things a bit and still maintain the balance, but I'd like people to keep the ubiquitous adoption goal front-and-center when proposing additional functionality.
> 
> Ubiquitous deployment and usefulness trump theoretical completeness.  (Just look at XMLDSIG and XMLENC to understand that!) 

I agree that we need to find a reasonable MTI.  The current JSMS draft doesn't really specify which parts are MTI, beyond basic things like base64url encoding and ignoring unknown fields.

My main concern is that JWS and JWE don't really tell me which fields need to be present for which use case.  If I'm putting a MAC on a JWS vs. a signature, do I need "jwk" or "jku" fields to be populated?  Whereas in JSMS, I think these things are much clearer due to the separation of the use cases (Signed/Authenticated/Encrypted) and the isolation of signatures and wrapped keys into sub-objects.


> -- Lack of clear typing of objects
> 
> Most JSON documents aren't clearly typed; instead, developers and implementations understand from context what the meaning of each field is.  XML is clearly typed, but developers are voting with their feet by moving to JSON.
> 
> What practical advantages resulting from overlaying a typing infrastructure on JSON do you believe are worth the potential costs of doing so?  BTW, I'm not opposed to having a JSON schema description of the JOSE JSON data structures like JSMS did in http://tools.ietf.org/html/draft-rescorla-jsms-00#appendix-A, but I'll remark that this an informative tool, not a normative one.

I actualy don't have a hugely urgent requirement here either.  It's something that others in the WG had requested, which is trivial to add in JSMS, but not in JWE/JWS -- again because of the unnecessary integrity protection of the header.


> -- Lack of versioning for future extensibility
> 
> An optional version field could be added if the working group desires, however the last time the working group discussed it, they decided not to do so.  Rather, specs like this effectively become "versioned" when revised in a way that new fields are added.  If you understand the new fields, you support that new version.  (Note that this is independent of whether there is a version number field, which according to some participants in the previous discussion, they regarded as therefore being redundant information.)
> 
> I'm not religious about not having a version number field, but often "less is more".

Indeed, I'm not religious either.  It just seems like a general good idea to state this directly as opposed to having parsers divine it based on which fields are present.  To quote RFC 5652:
"
   Each of the major data structures includes a version number as the
   first item in the data structure.  The version numbers are intended
   to avoid ASN.1 decode errors.  Some implementations do not check the
   version number prior to attempting a decode, and if a decode error
   occurs, then the version number is checked as part of the error
   handling routine.  This is a reasonable approach; it places error
   processing outside of the fast path.  This approach is also forgiving
   when an incorrect version number is used by the sender.
"


Best,
--Richard





> 
> 				Best wishes,
> 				-- Mike
> 
> -----Original Message-----
> From: Richard L. Barnes [mailto:[e-mail]] 
> Sent: Friday, June 15, 2012 3:00 PM
> To: Mike Jones
> Cc: [e-mail]
> Subject: Re: [jose] Fwd: New Version Notification for draft-barnes-jose-jsms-00.txt
> 
> Ok, so maybe I should have phrased my lead-in differently.  IMO, the current JW* specs are deficient in several ways, e.g.:
> -- Not being JSON
> -- Not supporting multiple recipients
> -- Lacking clear distinctions between signing and MAC
> -- Lacking clear processing instructions for different key establishment modalities
> -- Lack of clear typing of objects
> -- Lack of versioning for future extensibility
> 
> Looking at the current specs, it didn't seem to me that it would be simple to adapt them to address these deficiencies.  So I just did a re-write.  If you think they can be addressed within the current framework, please explain.  I just found it simpler to start from whole cloth.
> 
> Cheers,
> --Richard
> 
> 
> 
> On Jun 15, 2012, at 5:43 PM, Mike Jones wrote:
> 
>> Hi Richard,
>> 
>> If you're puzzled by specific design decisions in the JOSE documents, I encourage you to ask about them on the list.  The documents have significant history going back to mid-2010 when convergence discussions started among people with four different JSON signing & encryption proposals.  JWT and the current JOSE specs and implementations are the result.  I realize that not all the JOSE participants were part of those discussions from the beginning (although many were) and so teasing out the rationale for specific decisions might help people better understand the choices made, and hopefully thereby increase consensus behind the eventual working group choices.
>> 
>> If the list that you're puzzled by is the bulleted list below, maybe we can start a discussion on that basis.
>> 
>> 				Best wishes,
>> 				-- Mike
>> 
>> -----Original Message-----
>> From: [e-mail] [mailto:[e-mail]] On Behalf 
>> Of Richard L. Barnes
>> Sent: Friday, June 15, 2012 2:08 PM
>> To: [e-mail]
>> Subject: [jose] Fwd: New Version Notification for 
>> draft-barnes-jose-jsms-00.txt
>> 
>> Dear JOSE WG,
>> 
>> Like Mr. Manger, once I finally got around to doing a thorough review of the JW* specs, I was kind of puzzled by some of the design decisions.  So I also went through and wrote an alternative format, this time taking a more of a cue from CMS.  A link to the document is below.
>> 
>> Summary of differences:
>> 
>>  o  JSMS is pure JSON, whereas in JWE and JWS only the header
>>     parameters are represented in JSON.
>> 
>>  o  JSMS allows for full algorithm agility in key agreement, while JWE
>>     only allows ECDH-ES.
>> 
>>  o  JSMS supports multiple recipients for EncryptedData and
>>     AuthenticatedData objects via the inclusion of multiple WrappedKey
>>     objects.  Sending a JWE to multiple recipients requires re-
>>     encryption of the entire object for each recipient.
>> 
>>  o  The signature and MAC functions of the JWS object are separated
>>     into SignedData and AuthenticatedData JSMS objects.
>> 
>>  o  JSMS parameters are not integrity-protected, as they are in JWE
>>     and JWS.
>> 
>>  o  The "typ" and "zip" parameters are not defined in JSMS, but could
>>     be added without significant change.
>> 
>>  o  JSMS requires that recipients MUST ignore unknown header
>>     parameters, in order to facilitate extensibility.
>> 
>> I've also thrown together an initial JavaScript implementation (based on NodeJS for some python crypto glue), which is on github:
>> <https://github.com/bifurcation/jsms>
>> 
>> Comments are very welcome.  
>> 
>> Thanks,
>> --Richard
>> 
>> 
>> 
>> Begin forwarded message:
>> 
>>> From: [e-mail]
>>> Subject: New Version Notification for draft-barnes-jose-jsms-00.txt
>>> Date: June 15, 2012 4:56:09 PM EDT
>>> To: [e-mail]
>>> 
>>> 
>>> A new version of I-D, draft-barnes-jose-jsms-00.txt has been 
>>> successfully submitted by Richard Barnes and posted to the IETF 
>>> repository.
>>> 
>>> Filename:	 draft-barnes-jose-jsms
>>> Revision:	 00
>>> Title:		 JavaScript Message Security Format
>>> Creation date:	 2012-06-15
>>> WG ID:		 Individual Submission
>>> Number of pages: 25
>>> URL:             http://www.ietf.org/internet-drafts/draft-barnes-jose-jsms-00.txt
>>> Status:          http://datatracker.ietf.org/doc/draft-barnes-jose-jsms
>>> Htmlized:        http://tools.ietf.org/html/draft-barnes-jose-jsms-00
>>> 
>>> 
>>> Abstract:
>>> Many applications require the ability to send cryptographically  
>>> secured messages.  While the IETF has defined a number of formats for  
>>> such messages (e.g.  CMS) those formats use encodings which are not  
>>> easy to use in modern applications.  This document describes the  
>>> JavaScript Message Security format (JSMS), a new cryptographic  
>>> message format which is based on JavaScript Object Notation (JSON)  
>>> and thus is easy for many applications to generate and parse.
>>> 
>>> 
>>> 
>>> 
>>> The IETF Secretariat
>> 
>> _______________________________________________
>> jose mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/jose
>> 
>> 
>> _______________________________________________
>> jose mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/jose
> 
> 
> 
> _______________________________________________
> jose mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/jose