Re: [jose] Should we delete the "typ" header field

[<[e-mail]>]

My take on the "typ" header field:


-          "typ" must be optional because some application just knows what it is dealing with. It is a waste of space in this case.

-          "typ" is reserved header name and the values SHOULD be JWS, JWE, JWT. No MUST on the values.

-          For a general purpose JOSE library it makes sense to have "typ" and I would write it so that it  throws an exception on validateJWS( {"typ":"JWE", ...})
Especially if crit=["typ"].

I would like to keep (the first paragraph of)
https://tools.ietf.org/html/draft-ietf-jose-json-web-signature-11#section-4.1.8
as it is.

-Axel

From: [e-mail] [mailto:[e-mail]] On Behalf Of Nat Sakimura
Sent: Thursday, June 06, 2013 3:04 AM
To: Dick Hardt
Cc: Richard Barnes; Mike Jones; Brian Campbell; Jim Schaad; [e-mail]
Subject: Re: [jose] Should we delete the "typ" header field

It is giving the meaning to the signature - metadata about the signature, so I thought header would be more appropriate, besides, the payload is binary and I can avoid double base64 encoding that way.

2013/6/6 Dick Hardt <[e-mail]<mailto:[e-mail]>>
Why would you need that in the header? Why not in the payload?

On Wed, Jun 5, 2013 at 5:31 PM, Nat Sakimura <[e-mail]<mailto:[e-mail]>> wrote:
What I was thinking as an example was to have 'datetime' as a public header parameter to be used generally, and then differentiate between two cases of 'agreement' and 'timestamp'. It is the same signing operation over the original (e.g., a MS word document), but from an application point of view, it is a bit different. Former constitute an agreement while the later does not. The type=timestamp just signifies that the original document existed at the time indicated.

2013/6/5 Richard Barnes <[e-mail]<mailto:[e-mail]>>
I'm not sure I understand the use case here.  Let me look at a couple of options, and hopefully one of them will be what you meant :)

If it's just a signed timestamp you're talking about, then that's adequately described by cty="timestamp", typ="JWS".

If you want to sign something and add a timestamp in the header, then typ="timestamp" isn't accurate according to the current specification.  The type of the whole object would have to be something like "timestamped-content".  And in that case, an application can easily recognized that it's timestamped content by looking for the timestamp field in the header, so the "typ" isn't really helpful.


Moving from the example to the general question:

I think we finally have some common understanding on this list as to what Mike means by "typ" in the current document -- a label for the application-layer type of the overall object.  Thanks to Mike for clarifying that in this discussion.

The question is whether this is a sufficiently useful thing to have in the core spec.  Obviously, applications have to know what type of object they're dealing with when they get a JOSE object.  That includes (1) what type of content is in the payload, and (2) what application-layer fields are required to be in the header.  (1) is covered by "cty".  So "typ" is only useful in the case where:
A. The application is storing application-layer data in the JOSE header
B. The required application-layer fields can be different for different objects
C. The required application-layer fields can be different for different objects of the same content type
D. The required application-layer fields are not indicated by other application-layer context

Are there any known applications that meet these criteria?  JWT does not, for several reasons.
1. (!A) JWT does not store application-layer data in the JOSE header, only in the claims in the body
2. (!C) A JWT object can indicate with "cty" that it contains JWT claims or another JWT, either of which makes it obvious that this is a JWT
3. (!D) In the OAuth context, the Token Type value indicates what type of token is being presented, e.g., JWT

So there's not really any known use case for "typ" as currently specified.  On the other hand, Dick and James have pointed out that it can be useful to have "typ" indicate whether the object at hand is a JWE or JWS.  Which brings us back to:
-- Change the definition of "typ" so that it indicates the JOSE-layer type (JWE/JWS)
-- Remove "typ" altogether






On Wed, Jun 5, 2013 at 8:53 AM, Nat Sakimura <[e-mail]<mailto:[e-mail]>> wrote:
What about such as this?

cty: original
typ: timestamp

2013/6/5 Richard Barnes <[e-mail]<mailto:[e-mail]>>
It would be a good point, if it were true :)  In particular, this part: "[dropping 'typ'] is going to cause each and every application that uses JOSE to define their own field"

So far, I've heard of exactly one application of JOSE that requires "typ" in the way it is currently specified, namely JWT.

On the other hand, Dick's applications are apparently using it differently (and, IMO, properly) to distinguish JWE/JWS.  Then there are all the applications out there that are OK using application context and "cty" to recognize what type of object they have.  Apps using CMS have been doing this for ages.

So it's not at all clear to me that there's any other application that would use "typ" besides JWT.  And it's not clear to me that JWT needs it.

--Richard




On Fri, May 31, 2013 at 5:45 PM, Brian Campbell <[e-mail]<mailto:[e-mail]>> wrote:
Nat makes a good point here, I think.

On Thu, May 30, 2013 at 9:31 AM, Nat Sakimura <[e-mail]<mailto:[e-mail]>> wrote:
>From what I understand, both typ and cty are something to be consumed by the application and not JOSE processor. From the point of view of the JOSE processor, they should be ignored.

We could drop both fields from JOSE specs, but that is going to cause each and every application that uses JOSE to define their own field, which is a waste. That is why we are defining them here.

I would rather drop them than define JOSE processing semantics to these fields.

2013/5/31 Mike Jones <[e-mail]<mailto:[e-mail]>>
No, "cty" is used by the derived class to determine the type of the encapsulated field.  But that's not a complete description of the *entire object* - especially not the additional meaning imbued by the additional parameters the derived type may add to the JOSE header.  "typ" is there to provide the type of the entire object, including what you're calling the wrapper parts.

                                                            -- Mike

From: Richard Barnes [mailto:[e-mail]<mailto:[e-mail]>]
Sent: Thursday, May 30, 2013 7:58 AM

To: Mike Jones
Cc: Jim Schaad; [e-mail]<mailto:[e-mail]>; Dick Hardt
Subject: Re: [jose] Should we delete the "typ" header field

Isn't that requirement met by "cty"?  The only thing JOSE adds is a crypto wrapper around the real application content.  If you're an application, you know a JOSE object is the thing you want because it contains the content you want -- it's a JWT because it contains JWT claims.

Inheritance is the wrong metaphor.  This is encapsulation of application data:
if (jws.valid && jws.cty == "application/jwt_claims") {
    jwtClaims = jws.content;
}

--Richard

On Thu, May 30, 2013 at 10:43 AM, Mike Jones <[e-mail]<mailto:[e-mail]>> wrote:
Thanks for sharing the S/MIME details.  Although I was actually making the analogy to MIME - not S/MIME.  Like many analogies, it's imperfect, but I believe still illustrative.

The reason that the analogy isn't perfect is that the JOSE data structures are used to build application-specific data structures that are legal JOSE data structures but also have additional properties - including additional header fields with specific semantics.  (When we agreed to ignore not-understood header fields we let that horse out of the barn.)  For instance, Dick Hardt uses JWEs with issuer and audience fields in the headers, so they can be used by routing software.

Think of JOSE as the base class and the application types built using it as derived classes.  JWT is a derived class.  Dick's structures are a derived class.  These derived classes sometimes need names.  That's what "typ" is for.

                                                            -- Mike

From: Richard Barnes [mailto:[e-mail]<mailto:[e-mail]>]
Sent: Thursday, May 30, 2013 7:34 AM

To: Mike Jones
Cc: Jim Schaad; [e-mail]<mailto:[e-mail]>; Dick Hardt
Subject: Re: [jose] Should we delete the "typ" header field

You're mixing up "typ" and "cty".  If you want to make the analogy to S/MIME, "cty" is the equivalent to Content-Type inside the protected MIME body; "typ" is the content-type on the outer MIME header.  Pasting in an example:

-----BEGIN-----
Content-Type: application/pkcs7-mime; smime-type=signed-data;
     name=smime.p7m
Content-Transfer-Encoding: base64
Content-Disposition: attachment; filename=smime.p7m

567GhIGfHfYT6ghyHhHUujpfyF4f8HHGTrfvhJhjH776tbB9HG4VQbnj7
77n8HHGT9HG4VQpfyF467GhIGfHfYT6rfvbnj756tbBghyHhHUujhJhjH
HUujhJh4VQpfyF467GhIGfHfYGTrfvbnjT6jH7756tbB9H7n8HHGghyHh
6YT64V0GhIGfHfQbnj75
-----END-----
<http://tools.ietf.org/html/rfc3851#section-3.4.2>

The outer Content-Type, which is analogous to "typ", MUST be application/pkcs7-mime, with a parameter indicating the type of CMS object.  This is the same as requiring "typ" to be JWE or JWS.  The inner Content-Type (ASN.1/base64 encoded in the example) can be anything, just like "cty".

--Richard




On Thu, May 30, 2013 at 12:53 AM, Mike Jones <[e-mail]<mailto:[e-mail]>> wrote:
Requiring that the "typ" value be only "JWS" or "JWE" would be analogous to the MIME spec requiring that the Content-Type: field be only "text/plain" or "message/external-body".  It would render it useless.

                                                            -- Mike

From: [e-mail]<mailto:[e-mail]> [mailto:[e-mail]<mailto:[e-mail]>] On Behalf Of Richard Barnes
Sent: Wednesday, May 29, 2013 8:03 PM
To: Mike Jones
Cc: Jim Schaad; [e-mail]<mailto:[e-mail]>; Dick Hardt

Subject: Re: [jose] Should we delete the "typ" header field

If this is the level of "type" you're referring to, I think we should drop it from the spec.  It's an application-layer thing that the app can add or not according to its wishes.

I'm with Dick on this.  I think we should either have a mandatory indicator of what type of JOSE object this, or nothing at all.   If the former, the allowable values are "JWE" and "JWS".  The "+JSON" options are non-sensical -- the app needs to know what it's parsing before it gets this header.  While I have a preference for the former (for clarity), the latter approach is also OK with me, since the MIME types are specific to JWE/JWS.

Another approach here would be to address the JSON and compact forms separately.  The JSON form has no need of "typ" at all, since the type of the object is completely clear from what fields are there, e.g., "recipients" vs. "signatures".  For the compact form, we could do something like James's "E."/"S." prefix idea, which you need because the dot-separated components have different meanings and no field names to indicate this.

--Richard

On Wed, May 29, 2013 at 8:30 PM, Mike Jones <[e-mail]<mailto:[e-mail]>> wrote:
A standard library is unlikely to know the meanings of all possible "typ" values - and more to the point, it doesn't have to.  It's the application's job to determine that "this blob is a JOSE object" and then pass it to a standard library, which will then ignore the "typ" value.

A standard JOSE library won't know what "typ": "JWT" means.  It won't know what "typ": "BCGovToken" is, should the BC Government want to declare that it's using a token with particular characteristics.  It won't know what "typ": "XMPP" is, should XMPP want to declare that it's using a JOSE data structure with particular characteristics.  Etc.

All these values can be registered in the registry and used by applications that understand them.  That's the application's job - not the library's job.  The "typ" field is just there so that applications have a standard place to make any such declarations that they may need.

                                                                -- Mike

From: Dick Hardt [mailto:[e-mail]<mailto:[e-mail]>]
Sent: Wednesday, May 29, 2013 5:18 PM
To: Mike Jones
Cc: Jim Schaad; [e-mail]<mailto:[e-mail]>

Subject: Re: [jose] Should we delete the "typ" header field

I'd prefer to be able to use standard libraries for creating and parsing tokens, and not specialized libraries dependent on the use case.

I strongly think we either drop "typ" or make it required.

On Wed, May 29, 2013 at 5:03 PM, Mike Jones <[e-mail]<mailto:[e-mail]>> wrote:
It's fine for your application to specify that it's required for your use case.  Not applications need it, so they shouldn't be forced to pay the space penalty of an unnecessary field.

                                                                -- Mike

From: [e-mail]<mailto:[e-mail]> [mailto:[e-mail]<mailto:[e-mail]>] On Behalf Of Dick Hardt
Sent: Wednesday, May 29, 2013 4:56 PM

To: Jim Schaad
Cc: [e-mail]<mailto:[e-mail]>
Subject: Re: [jose] Should we delete the "typ" header field

I use it all the time and my code would barf if it was not there.

I think it should be required rather than be a hint if it is going ot be there.

On Wed, May 29, 2013 at 4:40 PM, Jim Schaad <[e-mail]<mailto:[e-mail]>> wrote:
I think the values just changed

However the way you are using it would be an argument to say that it should be a required field.  Are you just using it as a hint if it exists and then looking at the rest of the fields if it is not present?

Jim


From: Dick Hardt [mailto:[e-mail]<mailto:[e-mail]>]
Sent: Wednesday, May 29, 2013 3:49 PM
To: Jim Schaad
Cc: [e-mail]<mailto:[e-mail]>
Subject: Re: [jose] Should we delete the "typ" header field

Well, I have been using, but now realize the spec changed or I was confused.

I had been setting "typ" to be either "JWE" or "JWS" depending on the type of token I was creating or parsing as it was easier than looking at "alg"

As currently defined, I don't see value in "typ".

-- Dick


On Wed, May 29, 2013 at 3:02 PM, Jim Schaad <[e-mail]<mailto:[e-mail]>> wrote:
In reading the documents, I am trying to understand the justification for having the "typ" header parameter in the JOSE documents.

The purpose of the field is to hold the type of the object.  In the past, I believe that values which should now be placed in the cty field (such as "JWT") were placed in this field as well.  However the parameter is optional and an implementation cannot rely on its being present.  This means that for all practical purposes all of the code to determine the value of the type field from the values of the alg and enc fields.  If the field was mandatory then this code would disappear at a fairly small space cost and I can understand why the parameter would be present.

Can anybody justify why this field should be present in the document - or should it just disappear?

Jim


_______________________________________________
jose mailing list
[e-mail]<mailto:[e-mail]>
https://www.ietf.org/mailman/listinfo/jose



--
-- Dick



--
-- Dick

_______________________________________________
jose mailing list
[e-mail]<mailto:[e-mail]>
https://www.ietf.org/mailman/listinfo/jose



--
-- Dick

_______________________________________________
jose mailing list
[e-mail]<mailto:[e-mail]>
https://www.ietf.org/mailman/listinfo/jose




_______________________________________________
jose mailing list
[e-mail]<mailto:[e-mail]>
https://www.ietf.org/mailman/listinfo/jose



--
Nat Sakimura (=nat)
Chairman, OpenID Foundation
http://nat.sakimura.org/
@_nat_en

_______________________________________________
jose mailing list
[e-mail]<mailto:[e-mail]>
https://www.ietf.org/mailman/listinfo/jose





--
Nat Sakimura (=nat)
Chairman, OpenID Foundation
http://nat.sakimura.org/
@_nat_en




--
Nat Sakimura (=nat)
Chairman, OpenID Foundation
http://nat.sakimura.org/
@_nat_en



--
-- Dick



--
Nat Sakimura (=nat)
Chairman, OpenID Foundation
http://nat.sakimura.org/
@_nat_en