OAuth2 Bearer Token Usage

No matter what, as an implementer, you need to always verify your understanding of a specification against other implementations.

I have immersed myself in the digital identity space for the past few years. A good chunk of this work involves reading (and sometimes creating) specifications, as you can imagine. It is critical that they be written in such a way that two independent parties can build interoperable implementations without relying on each other’s code. With this in mind, let’s have a brief chat about OAuth2 Bearer Token Usage with a focus on the token’s encoding.

But first, let’s have a brief talk about what OAuth2 is.

What Is OAuth 2.0?

OAuth2 is an authorization framework defined by RFC6749 outlining the overall flow of messages between three actors: a 'client,' a resource owner (RO), and an authorization server (AS). You might know the first two respectively as 'relying party' and 'user.' Those of you familiar with OpenID Connect also know the AS as the 'Identity Provider.'

At its heart, OAuth2 is all about a user authorizing a relying party to access their data hosted by an API protected by the authorization server. Note that it does not authorize the user themselves to access the API. The job of the AS is to collect and record the user’s consent to authorize the relying party's access.

You might have noticed the emphasis on framework above. That is because RFC6749 deliberately avoids normative text defining many implementation details. Stepping back a bit, all RFC6749 says is that there is a client that requests access to a resource protected by an authorization server and that the resource owner must approve this access. Once authorized, the client obtains an access token to consume the resource.

OAuth2 relies on the HTTP protocol and defines the basic structure of the messages flowing between its actors. Relevant to the topic at hand is the token_typeincluded in the response to the client. As per the RFC, this attribute 'provides the client with the information required to successfully utilize the access token to make a protected resource request.'

OAuth 2.0 Bearer Token Usage

RFC6750 is the normative specification for how to use OAuth 2.0 Bearer tokens.

What are 'Bearer Tokens?'

Recall the token_type attribute from above. It turns out that if the access token response indicates the token’s type is Bearer, then it is a 'bearer token' as defined in RFC6750, which means:

Any party in possession of the token can use it, and
It must be presented in a specific way (as defined in RFC6750).

This is, by far, the most common type of access token in use on the web today.

Great! I want to integrate social logins into my uber-mega website and disrupt a market overnight! Let’s get started!

The Misdirection

You have implemented one of the OAuth 2 grant types (aka 'flows') as a client, and the AS has issued a Bearer access_token to you. What now? How do we use this token?

Luckily for us, RFC6750 tells us exactly what to do! Or does it? Let’s explore my thought process on my first attempt at an implementation:

The client must format an Authorization HTTP header with the token in a certain way.
The syntax of bearer tokens includes a b64token:
`b64token = 1*( ALPHA / DIGIT / “-“ / “.” / “_” / “~” / “+” / “/” ) *”=”
This strongly suggests that Base64 encoding is involved in some way.
But, who encodes the access_token in Base64?
Recall that the access_token is usually opaque to the client.
Note that HTTP headers can have almost any US-ASCII character.
Also, recall that the access_token pretty much consists of all printable characters, a superset of Base64.
If the access_token is opaque to the client (I shouldn’t attempt to parse it), and it can also consist of invalid Base64 characters, then surely the client must Base64-encode the Bearer token, right?

But are we sure? Let’s double-check with RFC6750:

The syntax of the 'Authorization' header field for this scheme follows the usage of the Basic scheme defined in Section 2 of RFC2617.
Following through, we find that RFC2617 defines the Basic HTTP Authentication Scheme that also uses the Authorization HTTP header and Base64 to encode the credentials.

Putting it all together:

RFC6750 defines how to use OAuth 2.0 Bearer Tokens.
Must put the access_token in the Authorization header.
The syntax includes a character space identified byb64token.
This usage follows theBasicscheme in RFC2617.
RFC2617 uses Base64 encoding.

Great! All I have to do is encode the access_token in Base64 before putting it in the Authorization header. I’m ready to integrate my social logins!

Narrator: He was not ready for integration.

The Reality

Bearer tokens are laid bare in the Authorization header.

None of the existing implementations expect the access_token to be encoded in Base64 in the Authorization header. See, for example:

Microsoft Identity Platform
GitHub’s Authorizing OAuth Apps
An issue I filed with ORY Oathkeeper (only for me to subsequently realize my own confusion)

What gives? Did everyone else get it wrong? (because, of course, I interpreted the spec correctly!)

Lessons Learned

It is important that specifications have precise normative text around how messages are constructed and processed in order to be interoperable. If there are algorithms involved, specify them step-by-step.

It is important that normative text be labelled as such.

It is important to identify each role and their respective responsibilities and algorithms.

In my opinion, a good example showcasing the previous points is Web Authentication where:

The high-level architecture is clearly depicted in diagrams.
Non-normative sections are clearly labelled.
The interfaces are clearly defined.
Algorithms are explained in detail. Example: Create a new credential.

I’m still grappling with a real consolidation of RFC6750 with reality. If I squint just right, I can see that when RFC6750 says 'The syntax for Bearer credentials is as follows,' it was unnecessarily informing the client developer what the syntax of the token is. In hindsight, this seems to be a (rather terse) message meant for implementers of Authorization Servers. I think an improved version of this section would have been split into several parts, each directed at different audiences: one for developers of clients, another for developers of authorization servers, and another for developers of resource servers. However, the text in RFC6750 remains terse and mixes multiple implementation details that concern the different actors in a different manner.

Another improvement would be to rely less on examples and to provide normative descriptions of the (very simple) processing algorithms that construct and parse these messages. That would have cleared up most of the confusion in section 2.1, although the language itself could have used stronger wording. Indeed, the non-normative text in section 7.1 of RFC6749 has stronger wording than that in RFC6750!

No matter what, as an implementer: always verify your understanding of a specification against other implementations!

We Provide consulting, implementation, and management services on DevOps, DevSecOps, DataOps, Cloud, Automated Ops, Microservices, Infrastructure, and Security

Services offered by us: https://www.zippyops.com/services

Our Products: https://www.zippyops.com/products

Our Solutions: https://www.zippyops.com/solutions

For Demo, videos check out YouTube Playlist: https://www.youtube.com/watch?v=4FYvPooN_Tg&list=PLCJ3JpanNyCfXlHahZhYgJH9-rV6ouPro

If this seems interesting, please email us at [email protected] for a call.

Relevant Blogs:

Strengthening Your Web App Security: Preventing SQL Injections

Azure PAM: How to Manage Access With Azure Bastion and Azure PIM

Utilizing Apache Spark for Azure Network Security Group Management

Overcoming Challenges in UI/UX Application Modernization