Re: Performance question for JSON-LD with vc.js

Orie, no disagreement on some of those items (esp. #1).  To that end, I have been working with ETSI (https://www.etsi.org/) on JAdES (https://portal.etsi.org/webapp/WorkProgram/Report_WorkItem.asp?WKI_ID=52897) which will significantly improve the limitations of JWT signatures in general.

Leonard

From: Orie Steele <orie@transmute.industries>
Date: Friday, April 3, 2020 at 11:38 AM
To: Oliver Terbu <oliver.terbu@consensys.net>
Cc: Gregg Kellogg <gregg@greggkellogg.net>, Dmitri Zagidulin <dzagidulin@gmail.com>, Leonard Rosenthol <lrosenth@adobe.com>, Credentials Community Group <public-credentials@w3.org>
Subject: Re: Performance question for JSON-LD with vc.js

I want to raise some concerns with the JWT Proof Format for VCs...

1. No Standard Way to Lookup Verification Keys

Lack of clear binding between iss / kid and verificationMethod... in other words... how are keys used for JWTs meant to be expressed in the JWT Header... we've seen this continuously trip developers, who think that kid should either look like: "did:example:123#fingerprint" or "fingerprint"... but then software needs to know how to resolve the verification key material... this is not OAuth, we don't have https://YOUR_DOMAIN/.well-known/jwks.json... there is essentially no standard way of looking up the key used to sign a VC-JWT.

2. Many Equivalent Spec Compliant Representations

there are many ways a VC JWT might have a proof applied, and they are not aligned with how the ld proof may look... this makes the proof formats essentially not interoperable, and given the optionality defined here: "For backward compatibility with JWT processors, the following JWT-registered claim names MUST be used instead of, or in addition to, their respective standard verifiable credential counterparts: [exp, iss, nbf, jti, sub, aud]"... there are lots of equivalent VC-JWT representations, verifiers are required to support all of them?

3. No Processing of JSON-LD

As Oliver mentioned, there is no actual processing of the JSON-LD embedded in the JWT.... so it's trivial to issue and verify credentials which are not valid JSON-LD, just add properties that are not defined in the context... You won't get any warnings or errors, until someone actually tries to use the JSON-LD... which they are likely to do at some point, since they see an `@context`... In other words, the fact that there is no performance hit for VC-JWT context parsing indicates a security vulnerability... unless people believe it's a good thing that there is no validation of the VC JSON-LD... which seems crazy to me... why include an `@context`, if you don't intend to rely on it.

That being said, if these issues were addressed, VC-JWT stands to the be an incredibly valuable part of the ecosystem, with some really significant benefits:

A. Single Verification Key for Multiple Proof Formats

Assuming a solution to 1, you could register a single JWK key, and use it for both LD and JWT VC Proofs.

B. Most Widely Supported Tooling

JOSE is by far the most widely supported web compatible crypto standard on the planet. Has built in support in browsers, including with software isolation for keys: https://developer.mozilla.org/en-US/docs/Web/API/Crypto/subtle<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdeveloper.mozilla.org%2Fen-US%2Fdocs%2FWeb%2FAPI%2FCrypto%2Fsubtle&data=02%7C01%7Clrosenth%40adobe.com%7C856286f2b1a04163f03d08d7d7e502c8%7Cfa7b1b5a7b34438794aed2c178decee1%7C0%7C0%7C637215250909169444&sdata=KdJO1YZToxFOsHwgSQ1LgNxu5w9MfvKI%2BNZV%2F%2Bjl2jk%3D&reserved=0>

Support in almost every programming language.

Used in almost all modern identity software (OAuth, OIDC)

In conclusion, I personally would not recommend VC-JWT until the issues above are addressed... I really wish they were addressed because of the reasons above as well.

I'm happy to help get these issues addressed, if that's possible to do.

OS



On Fri, Apr 3, 2020 at 5:08 AM Oliver Terbu <oliver.terbu@consensys.net<mailto:oliver.terbu@consensys.net>> wrote:
I have to correct a typo. It is certainly considered to be BAD practice.

Oliver

On Fri, Apr 3, 2020 at 12:03 PM Oliver Terbu <oliver.terbu@consensys.net<mailto:oliver.terbu@consensys.net>> wrote:
It might be probably worth pointing out that a non-JSON-LD and non-LD-Proof version of Verifiable Credentials is the JWT proof format: https://w3c.github.io/vc-data-model/#proof-formats<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fw3c.github.io%2Fvc-data-model%2F%23proof-formats&data=02%7C01%7Clrosenth%40adobe.com%7C856286f2b1a04163f03d08d7d7e502c8%7Cfa7b1b5a7b34438794aed2c178decee1%7C0%7C0%7C637215250909179441&sdata=yHH1jXT2Ma%2B98nhv1Qom683j776%2FDeHHNg0hD5rqC1M%3D&reserved=0>. It is kind of hidden in the W3C VC standard. For JWT VCs, you don't need JSON-LD processing. The only thing that is needed as per W3C VC spec is to add the correct @context values in the VC.

An implementation can be found here: https://github.com/uport-project/daf<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fuport-project%2Fdaf&data=02%7C01%7Clrosenth%40adobe.com%7C856286f2b1a04163f03d08d7d7e502c8%7Cfa7b1b5a7b34438794aed2c178decee1%7C0%7C0%7C637215250909179441&sdata=OQ17ybaOkBqWK4mwUwLfZ1ZgoQF2HsPCYCATmpgzVLU%3D&reserved=0>

On a separate note, we were recently debating about the security implications of remote context fetching in W3C CCG and that it is considered to be practice but had to admit that there is nothing *normative* in the JSON-LD spec and W3C VC spec that prevents people from doing that.

Oliver

On Fri, Apr 3, 2020 at 2:05 AM Gregg Kellogg <gregg@greggkellogg.net<mailto:gregg@greggkellogg.net>> wrote:
On Apr 2, 2020, at 3:07 PM, Dmitri Zagidulin <dzagidulin@gmail.com<mailto:dzagidulin@gmail.com>> wrote:

> There is *NOTHING* in the JSON-LD specification, in the definition of context (https://json-ld.org/spec/latest/json-ld/#the-context<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fjson-ld.org%2Fspec%2Flatest%2Fjson-ld%2F%23the-context&data=02%7C01%7Clrosenth%40adobe.com%7C856286f2b1a04163f03d08d7d7e502c8%7Cfa7b1b5a7b34438794aed2c178decee1%7C0%7C0%7C637215250909189434&sdata=w9nXBxcTyMjlsrQa3K3bOx3h7HxqRKP0oT0Lullg1QU%3D&reserved=0>) that mandates that the URI/IRI be (a) resolvable or (b) that any such resolution be machine readable.

The proper reference is to https://www.w3.org/TR/json-ld11/#the-context<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.w3.org%2FTR%2Fjson-ld11%2F%23the-context&data=02%7C01%7Clrosenth%40adobe.com%7C856286f2b1a04163f03d08d7d7e502c8%7Cfa7b1b5a7b34438794aed2c178decee1%7C0%7C0%7C637215250909189434&sdata=V9wupDNcsA%2BrtSzLJ52neFnxg2eeeu0wmDAecQb1KCE%3D&reserved=0>, not the CG draft. It is at Candidate Recommendation on the way to Proposed Recommendation soon. Also, that section is not normative.

In any case, there is no normative requirement that contexts be dereferencable; in many cases, this is an anti-pattern, and for the case of VC (IIRC) it is intended to act as an identifier that a conformant application running in JSON-LD mode, would use to provide the context that might otherwise be dereferenced from that location, possibly using a custom document loader.

This is the case for other specs as well, where there is either the inability to dereference remote contexts at runtime, or a desire to avoid such traffic for other reasons.

It might be worth bringing up that concern to the JSON-LD CG for the next version of the spec, over at https://github.com/w3c/json-ld-syntax/<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fw3c%2Fjson-ld-syntax%2F&data=02%7C01%7Clrosenth%40adobe.com%7C856286f2b1a04163f03d08d7d7e502c8%7Cfa7b1b5a7b34438794aed2c178decee1%7C0%7C0%7C637215250909199431&sdata=%2BZ0edh%2BK%2BGSawxPym%2BHxN7tOaGlzrSOqET4GGWhcEwE%3D&reserved=0>

Further down, in that same section, the wording "In JSON-LD documents, contexts may also be specified inline. This has the advantage that documents can be processed even in the absence of a connection to the Web." is the part that points out that contexts need to be resolvable (and, again, in actual implementation it's recommended that they're stored locally, like npm packages, and not fetched over the web).

An inline context would be an object value for @context within the main document. Generally, when there may be a large external context, or the content is closely managed, using the reference as an identifier which is resolved by a custom document loader is probably the best way to go. This also allows a document loader to notice attempts to access other external contexts, which could be specified in a document, and either whitelist or blacklist depending on your needs.

Gregg


On Thu, Apr 2, 2020 at 5:43 PM Leonard Rosenthol <lrosenth@adobe.com<mailto:lrosenth@adobe.com>> wrote:
Thinking of a context that way, Dmitri, is not correct.

There is *NOTHING* in the JSON-LD specification, in the definition of context (https://json-ld.org/spec/latest/json-ld/#the-context<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fjson-ld.org%2Fspec%2Flatest%2Fjson-ld%2F%23the-context&data=02%7C01%7Clrosenth%40adobe.com%7C856286f2b1a04163f03d08d7d7e502c8%7Cfa7b1b5a7b34438794aed2c178decee1%7C0%7C0%7C637215250909199431&sdata=o9z0MdUUmh5I7IkxtbrhbqxT0MY3toT6NreGpo%2BZ7Yc%3D&reserved=0>) that mandates that the URI/IRI be (a) resolvable or (b) that any such resolution be machine readable.  Same is true with the VC spec (https://www.w3.org/TR/vc-data-model/#contexts<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.w3.org%2FTR%2Fvc-data-model%2F%23contexts&data=02%7C01%7Clrosenth%40adobe.com%7C856286f2b1a04163f03d08d7d7e502c8%7Cfa7b1b5a7b34438794aed2c178decee1%7C0%7C0%7C637215250909209426&sdata=vxR82MzpoFhor0Wx%2BAiUPXOOqHrKGN4LfSmG%2FqaNE6s%3D&reserved=0>).

I agree that it is recommended to do that (https://www.w3.org/TR/vc-data-model/#contexts<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.w3.org%2FTR%2Fvc-data-model%2F%23contexts&data=02%7C01%7Clrosenth%40adobe.com%7C856286f2b1a04163f03d08d7d7e502c8%7Cfa7b1b5a7b34438794aed2c178decee1%7C0%7C0%7C637215250909209426&sdata=vxR82MzpoFhor0Wx%2BAiUPXOOqHrKGN4LfSmG%2FqaNE6s%3D&reserved=0>, paragraph 4, under @context) – no question about it.  However, it is *not* a requirement (MUST) and as such using a context in that way still produces a 100% valid LD doc and VC.

Standards are written with very clear language for a reason and they should be implemented accordingly.

Leonard

From: Dmitri Zagidulin <dzagidulin@gmail.com<mailto:dzagidulin@gmail.com>>
Reply-To: "dzagidulin@gmail.com<mailto:dzagidulin@gmail.com>" <dzagidulin@gmail.com<mailto:dzagidulin@gmail.com>>
Date: Thursday, April 2, 2020 at 1:00 PM
To: Leonard Rosenthol <lrosenth@adobe.com<mailto:lrosenth@adobe.com>>, Credentials Community Group <public-credentials@w3.org<mailto:public-credentials@w3.org>>
Subject: Re: Performance question for JSON-LD with vc.js

Hi Leonard,

So, you can think of a context as exactly functionally equivalent to an NPM package (or Maven, or ruby Gem, or whatever).
Which means that, by definition, is has to be machine readable (it's a common source of confusion to confuse contexts and human-readable vocab documentation -- think of the human-readable part as a README file for an npm package).
Does it have to be accessible - sure, but just like with NPM packages, you have the control of /when/ it's accessible.
You /could/, in theory, override your require() functions so that it loads NPM packages from the web, in your code. But even with caching, that would be a wildly impractical approach, full of performance and security issues.
Instead, what most developers do (and what we recommend to do with contexts), is to fix them at /build time/. That is, version them, bundle them locally with your code, and only allow your document loaders to interface with those local versions.

Does that make more sense?






On Thu, Apr 2, 2020 at 12:41 PM Leonard Rosenthol <lrosenth@adobe.com<mailto:lrosenth@adobe.com>> wrote:
Dmitri – that also forces all contexts to be (a) accessible and (b) machine readable…neither of which a mandatory requirement of either JSON-LD or VC itself.

Leonard

From: Dmitri Zagidulin <dzagidulin@gmail.com<mailto:dzagidulin@gmail.com>>
Reply-To: "dzagidulin@gmail.com<mailto:dzagidulin@gmail.com>" <dzagidulin@gmail.com<mailto:dzagidulin@gmail.com>>
Date: Thursday, April 2, 2020 at 11:39 AM
To: Anil Lewis <anillewi@ca.ibm.com<mailto:anillewi@ca.ibm.com>>, Credentials Community Group <public-credentials@w3.org<mailto:public-credentials@w3.org>>
Subject: Re: Performance question for JSON-LD with vc.js
Resent-From: <public-credentials@w3.org<mailto:public-credentials@w3.org>>
Resent-Date: Thursday, April 2, 2020 at 11:38 AM

Hi Anil,

> When using vc.js, I have observed that if the pre-configured contexts are not loaded, I am unable to even sign/verify the credential. I might have to come up with our own processing code to circumvent this issue.

For security reasons, the vc-js library /does not/ allow loading of arbitrary contexts from the web. You have to explicitly allow-list them, by providing a document loader function tailored to your usecase.
Absolutely happy to walk you through this, please feel free to open an issue on https://github.com/digitalbazaar/vc-js/issues<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fdigitalbazaar%2Fvc-js%2Fissues&data=02%7C01%7Clrosenth%40adobe.com%7C856286f2b1a04163f03d08d7d7e502c8%7Cfa7b1b5a7b34438794aed2c178decee1%7C0%7C0%7C637215250909219415&sdata=0c%2Fs6DvYNfGwrFpB9yk6T55uhrYLjjmlgVO3oOdAWRA%3D&reserved=0>!



--
ORIE STEELE
Chief Technical Officer
www.transmute.industries

[Image removed by sender.]<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.transmute.industries%2F&data=02%7C01%7Clrosenth%40adobe.com%7C856286f2b1a04163f03d08d7d7e502c8%7Cfa7b1b5a7b34438794aed2c178decee1%7C0%7C0%7C637215250909219415&sdata=uYnkYggeM1u1F5zHG0fhU4QsJAZSuxn49unvF1O6XXQ%3D&reserved=0>