Hi Tim,

On Sat, 13 Jun 2020, Tim Johnson wrote:

> 
>
>       On Jun 11, 2020, at 6:39 PM, Levente Uzonyi <leves at caesar.elte.hu> wrote:
>
>       It is useful for SSO.  But for automated workflows, where human intervention is undesired, there is another type of OAuth2 which may be called "server to server" or apparently "two-legged OAuth":
>       https://developers.google.com/identity/protocols/oauth2/service-account
>       I tried and failed to implement this last year.  Where I failed was in computing the JSON Web Signature (JWS).  I could generate JWTs successfully*, but JWSs for Google require "SHA256withRSA (also known as
>       RSASSA-PKCS1-V1_5-SIGN with the SHA-256 hash function)" which requires one very specific algorithm missing from SqueakSSL and which I simply could not engineer on my own at the time (or perhaps ever — I
>       might just not be
>       smart enough ;) ).  I was able to get as far as crafting the JSON in Squeak and then signing using Python's implementation of the algorithm and it would work.
>       If we could get server-to-server OAuth2 using WebClient, that would also allow us to, say, connect to Google Drive directly from Squeak, or be a client of Google Cloud Platform / Compute Engine, etc.  I
>       think that would be
>       very cool.
>       Sadly, I seem to recall this type of OAuth (and thus this algorithm) would also be necessary for GitHub/GitLab.
> 
> 
> The Cryptography package seems to have this stuff implemented, though I haven't verified the results. Here's how to use it:
> Let's say privateKey is an RSAPrivateKey with your private key (See class side methods and RSAPrivateKeyFileReader how to initialize it) and message is your serialized json to sign. Then
> 
> privateKey v15SignMessageHash: (HashFunction newSHA256 digestInfoAsn1DerEncodingFromMessage: message)
> 
> should return the signature.
> 
> 
> Thank you Levente.  When I mentioned algorithms I was hoping you might respond.  :) 
> 
> I found my old JWT code from last year on GitHub:  https://github.com/tcj/beaufort      (Oh no, it contains an old email address of mine in one of its tests...)
> 
> I loaded Cryptography into an image so I could run my tests again, and noticed that the Cryptography package has seen a lot of work this year.  Thanks for that.
> 
> So:  the tests I wrote last year (with a small change) pass now.  This could be very good news and may be worth further investigation.  Thank you again!  I will update my tests (& remove my old email address...?) and upload a
> new version of this package.
> 
> I am reminded that the issue I'd encountered last year was that HS256 could work, but RS256 could not.  Now, unit tests are passing for both.  (My BfJWTRFCTest>>#testConversion is not passing so I'll need to look into that.)

It fails because ByteArray >> #base64Encoded adds new line characters to 
the resulting string. That should be changed in the Trunk.

> 
> Here is the gist of my JWT signing code:
> 
> BfJWT>>#signedWith: aKey
> | headerAndClaims |
> self secret: aKey.
> headerAndClaims := self headerAndClaims.
> ^ '{1}.{2}' format: { headerAndClaims . self signatureFrom: headerAndClaims }
> 
> ... #signatureFrom: is implemented differently for HS256 versus RS256 subclasses of BfJWT.  My RS256 encoding was like this:
> 
> BfRS256 signatureFrom: aString
> | signedMessage hashed privateKey |
> privateKey := (Pkcs12PrivateKeyFileReader fromFile: 'timj-project-mar-2019-a94d67a8d0c8.p12') asPrivateKey.
> signedMessage := privateKey signMessage: aString.
> hashed := SHA256 new hashStream: signedMessage readStream.

You're using #signMessage: which does SHA1withRSA, then apply SHA256 on 
the result. The change I suggested will do SHA256withRSA.

> ^ hashed base64UrlEncoded
> 
> A year ago, I could have described how & why it wasn't working... but that information has left my brain now.
> 
> For what it's worth, Norbert Hartl's JSONWebToken also lacks support for this RS256 encoding format.
> 
> https://github.com/noha/JSONWebToken/commit/4a4d20eaa6e84e2676a577f74bc6e24c1ead0047
> 
> It seems to be a common issue around the internet that people find RS256 very difficult.

It's indeed more complex. RS256 requires RSA while HS256 is simply a HMAC 
(AFAIK). But we have RSA implemented in Cryptography, and even though the 
API could be more user-friendly and have more documentation, it can still 
be done with a single line of code.


Levente

> 
> Thanks again,
> Tim
> 
> 
> 
> 
>