[WIP] Signed JWT support

[aykevl]

Work-in-progress PR for verified JWT support in irmago.

Example JWT that I get out of it:

eyJhbGciOiJSUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiJQcml2YWN5IGJ5IERlc2lnbiBGb3VuZGF0aW9uIiwiaXNzIjoidmVyaWZpY2F0aW9uX3JlcXVlc3QiLCJpYXQiOjE1Mjg4MzM5MTksInNwcmVxdWVzdCI6eyJ2YWxpZGl0eSI6MTIwLCJyZXF1ZXN0Ijp7ImNvbnRleHQiOm51bGwsIm5vbmNlIjpudWxsLCJjb250ZW50IjpbeyJsYWJlbCI6IkZhbWlseSBuYW1lIiwiYXR0cmlidXRlcyI6WyJwYmRmLnBiZGYuaWRpbi5mYW1pbHluYW1lIl19LHsibGFiZWwiOiJEYXRlIG9mIGJpcnRoIiwiYXR0cmlidXRlcyI6WyJwYmRmLnBiZGYuaWRpbi5kYXRlb2ZiaXJ0aCJdfV19fX0.qHO5wVLdbmuukKWXD1YG_cmSR6mNKV6XbZp7a8MizY5Ve6KXI6HDjkRCUbxRI39SHxzJiYhi8Ar2Lim6ofK0MQxTEVdH4Vp-Lof6VDNvpHZwBZMD1FMz1Lw8jg0Q4WJ5PMpddz9PLf-zFdZCltAOB2RQwOdxkYzrqbyM1Xb57KkVodPKmgs_TFqWG8Ki40oMD6kYUEbbrTl_3VS1cxHZpxlBOqfNWEw4Hk_Kd1qBTLsIR_lwbnA8PyFOgto4Ovjtcft2Oj5vcdXZUePsC3vU1rRCMIPuqjqOtBZ7U1nkYKxiLvfzZC5xqQMfW3avxx8W-uDId5MZ8sxNkTdvqHCKIg

[sietseringers]

• Type should (in the case of disclosures) be the fixed string verification_request, see NewServiceProviderJwt()
• You can actually reuse NewServiceProviderJwt() to build the JWT before signing it
• We should have similar functions for signing and issuance (reusing NewSignatureRequestorJwt() and NewIdentityProviderJwt()

As to the serverName and humanName, that has now become a little messy:

• The variable humanName should be assigned to the field ServerName, whose name is inappropriate (I think JwtIssuer would be better
• The variable serverName should go into the JWT kid header.

[aykevl]

Ah, somehow I managed to miss NewServiceProviderJwt(). I've now reused this function.

We should have similar functions for signing and issuance [...]

Yes, I'll add them later. I would prefer getting this one right first.

The variable serverName should go into the JWT kid header.

Does this work? At least the Java implementation appears to use iss, not kid (verified by decoding the JWT for BIG name/birthdate disclosure).

[sietseringers]

Originally we indeed used only iss; this was (1) shown to the user in the IRMA app and (2) the API server used this to lookup the verifier's public key with which to verify its JWTs. At some point we split these into two, so now:

• The API server uses the kid aka serverName field in the header for the public key lookup
• iss aka humanName is what is shown to the user in the IRMA app.

For example, here is the header from a decoded IRMATube JWT obtained by clicking on a movie:

{
  "typ": "JWT",
  "alg": "RS256",
  "kid": "irmatube"
}

Body:

{
  "iat": 1529269092,
  "iss": "IRMATube",
  "sub": "verification_request",
  "sprequest": { "truncated" }
}

These are made by this PHP library.

[aykevl]

I've updated the PR with some bigger changes, to hopefully improve the structure of the code. An example JWT I've produced is:

eyJhbGciOiJSUzI1NiIsImtpZCI6ImR1byIsInR5cCI6IkpXVCJ9.eyJzdWIiOiJ2ZXJpZmljYXRpb25fcmVxdWVzdCIsImlzcyI6IlByaXZhY3kgYnkgRGVzaWduIEZvdW5kYXRpb24iLCJpYXQiOjE1MjkzMjQxNTgsInNwcmVxdWVzdCI6eyJ2YWxpZGl0eSI6MTIwLCJyZXF1ZXN0Ijp7ImNvbnRleHQiOm51bGwsIm5vbmNlIjpudWxsLCJjb250ZW50IjpbeyJsYWJlbCI6IkZhbWlseSBuYW1lIiwiYXR0cmlidXRlcyI6WyJwYmRmLnBiZGYuaWRpbi5mYW1pbHluYW1lIl19LHsibGFiZWwiOiJEYXRlIG9mIGJpcnRoIiwiYXR0cmlidXRlcyI6WyJwYmRmLnBiZGYuaWRpbi5kYXRlb2ZiaXJ0aCJdfV19fX0.cf6utb805i1TsjnuDqptKtHlyboUoGDyVADeknM0AGXEaFQYiEE3DIfyORo3t4ub6d8Cl3lB7LTFCHKYyIRINzEXgw-xOSO_SD1X6W9_VE8cilwKkvXO_QH3z0v_6ighQM-uFusLIsyGQLR8mVoMPOmx4H-By-fezQ5s_EwMI2EVqZv6t8FwiUIWs36hkI6kS4u_oFK1a1xBoPAabRDxi3kRio_Y07initV9Z5CqxNcYIdtSMzhu7uq_cbygTwOM7RlyZJEk27fed11GWZGC03054UPfi3qM6tRm4wY_7MNV4oCNZvUAZZRHyJMzehtn6RarUL51gzxb1cEdtewP-w

TODO:

• Implement this for IdentityProviderJwt (trivial)
• maybe rename serverName to humanName in some other parts of the code? Currently *Jwt.Requestor returns HumanName but then this variable is named ServerName in the callers, which is not consistent.
• comments

[confiks]

Could you please not force push branches, especially when they are tied to pull requests. It makes it very hard to figure out what the feedback is about. Just add new commits on top of each other. To solve merge conflicts you can merge master back in (instead of rebasing).

[aykevl]

Ok, will avoid that now. Some projects actually want you to force-push so that's why I did it.

EDIT: the commit is here: 0ec9427

[aykevl]

Any update on this? I need this for the new DUO issuer.

[sietseringers]

This is better now, thanks! Two things:

• The receiver name that I commented on
• The new dependency github.com/dgrijalva/jwt-go should be added to the Gopkg files (run dep ensure).

[sietseringers]

The receiver of *ServiceProviderJwt struct methods is earlier called jwt, and it is not idiomatic in Go to use different receiver names for the same struct. We should probably switch to claims on every method of ServiceProviderJwt and its parent and siblings (to avoid the name clash with the jwt package).

[aykevl]

I've also updated the other two receivers.

[aykevl]

I added the dependency, but avoided the inputs-digest line because it resulted in a merge conflict (and according to golang/dep#1224 that's probably safe).

[aykevl]

Added a method to verify and parse a disjunction JWT.

[aykevl]

Closed in favor of #21.