path: root/yuck.md

# Overview

This a description of two authentication and authorization protocols,
and a sketch of acceptance criteria for an implementation of them.

This is very much work in progress.


## Concepts

Some basic concepts in this document:

* **identity** – data about who you are to tell you apart from
  everyone else
* **authentication** – proving your identity
* **authorization** – giving you permission to do something

FIXME: These could do with citations.

## The protocols: OAuth and OpenID Connect

[OpenID Connect]: https://openid.net/specs/openid-connect-core-1_0.html
[OAuth]: https://tools.ietf.org/html/rfc6749

The [OAuth][] 2.0 protocol is for authorization, not authentication, and
assumes an already existing way to authenticate users. It's mainly for
giving a service or application permission to do something on your
behalf.

The [OpenID Connect][] 1.0 (OIDC) protocol is for authenticating yourself
to one service or application by using a third party service. This
allows one authentication service (or identity provider) be used for
any number of other services or applications. Further, since the
identity provider can keep a login session open independently of the
other services and applications, this provides a single sign-on
experience.

We discuss here only these specific versions of these protocols, and
even then only subsets chosen based mainly from the point of security.

## Entities involved in the protocols

The protocols involves the following entities:

* the **end user**, who is trying to do something; also known as the
  resource owner
* the **web browser**, used by the user; might be a mobile or command
  line application instead of a browser as such; also know as the user
  agent
* the **application**, which the user uses to do things, and as part
  of that access resources; also know as the requesting party and the
  facade
* the **resource provider**, where the resources are, and which allows
  access to them via a web API
* the **identity provider** (IDP), which authenticates the user

The protocol specifications use different terminology, to be more
generic. The above have been chosen to make this document easier to
understand.




# The OAuth 2.0 protocol

The OAuth2 protocol is a way for an end user to allow one service
controlled access to their data on another service. It does not
authenticate the end user in any way.

As an example, imagine if Alice uses an email service, but would also
like to make use of another service that typesets and prints emails
into really nice, impressive, beautiful, heirloom quality, leather
bound hardcover books. The book service needs to be able to read
Alice's emails, but not delete them or send email as Alice.

In a kinder world than ours Alice could just give their email password
to the book service, but in our world, this needs to be done in a more
complicated way. Otherwise someone at the book service would abuse
access to Alice's email account by deleting all the email, or worse.

The gist of OAuth2 is that Alice can tell their email service that the
book service is allowed to read all her correspondence, but do nothing
else.

In a simplistic way, Alice logs into the email service, and asks for
an _access token_, then passes that onto the book service. The book
service logs into the email service, and gives the access token to
gain access to Alice's emails. The email service knows that the access
token only allows read-only access: no deletion or sending.

However, such a system would be cumbersome to use. Alice would have to
manually navigate to the email service's access token generation page,
copy the token, and have a way to communicate the token to the book
service. This is too much manual work, with too many steps, and too
much can go wrong.

Instead, Alice logs into the book service, and tells them which email
service to get emails from. The book service redirects Alice's to the
email service in a way that tells the email service that a) an access
is required b) by the book service c) for read-only access d) to
Alice's emails. The mail service checks that Alice is already logged
in, or else asks Alice to log in, and that Alice is OK with giving the
book service the access requested. If all that goes well, the email
service generates the token, and redirects Alice's web browser back to
the book service in such a way that the token is carried along. The
book service can now access the email service API with the access
token, and get what they need to print the books.

[@rfc6749] describes the OAuth2 protocol in detail. This chapter
condenses that into a shorter, and opinionated description.

## OAuth2 protocol variants (grant types)

OAuth2 allows for several different kinds of use cases, by providing
different ways to get an "authorization grant". An authorization grant
is a thing that provides access to a protected resource (see section
1.3 in [@rfc6749]).

There are four authorization grant types:

* authorization code
  - this is a slightly simplified description: the spec seems to allow
    less secure variants, but we'll get to that later
  - application redirects end user's web browser to the identity
    provider, which returns an authorization code to the application
  - application gives authorization code to resource provider, which
    gives back an access token
  - browser never gets end the token
  - the authorization code is random, temporary, short-lived, can only
    be used once, and can be tied to a specific resource provider
  - this allows the application to be authenticated as well as the
    end user
  - this is the most secure grant type and should be used if possible
* implicit
  - application is typically a JavaScript application running in the
    browser
  - access token is given directly to the browser
  - this is vulnerable to browser insecurities, which are legion
  - application can't be authenticated in any useful sense
  - don't use this, due to low security
* resource owner password credentials
  - client has the end user's username and password
  - never use this, as it leaks credentials and is inherently insecure
* client credentials
  - client has its own credentials
  - can be suitable when the client itself is the resource owner (such
    as for server-to-server communication) or the end user
    authorization has been arranged beforehand using some other way

In this document we only discuss the authorization code grant, as it's
the only one of these we can recommend from a security point of view.

## Overview of a basic OAuth2 transaction

~~~dot
digraph "oauth2" {
    alice [label="Alice"];
	booksite [label="Book service"];
	email [label="Email service"];

	alice -> booksite [label="1."];
	booksite -> email [label="2."];
	email -> alice [label="3."];
	alice -> email [label="4."];
	email -> booksite [label="5."];
	booksite -> email [label="6."];
	booksite -> alice [label="7."];
}
~~~

The above diagram represents the steps of Alice getting her book of
emails, at a very high level. There are many steps.

1. Alice asks the book service for a book of her emails.
2. Book service asks email service for an access token.
3. Email service asks Alice if book service may access her emails.
4. Alice tells email service "sure".
5. Email service gives book service an access token.
6. Book service uses access token to download all of Alice's emails.
7. Book service sends book to Alice.

The diagram below represents the same transaction, but in a different
way. (One of the diagrams will be clearer to the reader, but it
depends on the reader which.)

~~~plantuml
@startuml
actor Alice
entity "Book service" as Booksite
database "Email service" as Email

Alice -> Booksite: Want book!
Booksite -> Email: May I get emails?
Email -> Alice: OK to give emails to Booksite?
Alice -> Email: Sure!
Email -> Booksite: Have an access token
Booksite -> Email: Emails please
Booksite -> Alice: Your book, bitte
@enduml
~~~

## Token types

OAuth has two types of tokens: access and refresh. Access tokens are
used by the application to access resource providers. Refresh tokens
are used by the application to request a new access token from the
identity provider.

OAuth does not specify the format of the tokens, and leaves it to the
identity provider. The resource provider needs a way to verify that a
token is valid. The validation mechanism is also not specified by
OAuth. However, the [JSON Web Token][] (JWT) specification is commonly
used. This will be covered in detail in the OIDC part of this
document.

### Access tokens

An access token is sufficient to access data on a resource provider,
on its own. A valid access token itself represents all the
authorization to access a protected resource via a resource provider.
Every access token granted on behalf of a resource owner does not
necessarily grant full and complete access to every resource owned by
end user, but may be limited in some way: a particular token may give
access only to a particular resource, or only certain kinds of access
(read vs write vs delete). Access tokens may also have a limited life
time, after which the resource provider refuses to accept them.


Access tokens are "bearer tokens", which means that any person or
software with a copy of the token can use it, within the constraints
encoded in the token itself. As such, access tokens are to be
considered sensitive data, just like all other credentials.

### Refresh tokens

A refresh token is used by the application to get a new access token,
without requiring the end user to grant one. When the end user grants
a token it is an interactive process, and may involve
re-authenticating them. This means it can be quite tedious and
irritating, if it happens often.

An application may need a new access token for various reasons. For
example:

* If access tokens expire, the application will use its refresh token
  to get a new access token.

* An access token may be revoked, but a refresh token may allow the
  application to get a new one. The old access token revocation may be
  due to the resource provider noticing suspicious activity. For
  example, if the access token might be tied to a specific IP address,
  but when a mobile application moves to a new network, its address
  changes, invalidating the old access token.
  
  Revoking access tokens can be tricky, as the revocation needs to be
  communicated to all the resource providers. For this reason, access
  tokens often have a short life time. Refresh tokens are easier to
  revoke, as only the identity provider needs to know about the
  revocation. Thus a combination of short-lived access tokens and
  longer-lived refresh tokens allows fairly rapid, but not instant,
  revocations in a simple manner, without making the end user
  re-authenticate themselves many times an hour.

* The application may need to give an access token to another
  application, but give them less access. In our book printing
  example, the usual email application Alice uses to process all her
  email might have a button to order a book, and the application could
  automatically get a read-only access token to give to the book
  printing service. The application could use a refresh token to get
  the extra read-only access token without Alice having to interact
  again with the identity provider directly.

When the application uses a refresh token, the identity provider
validates it and, if everything is OK, responds with a new access
token, and possibly a new refresh token.

The sequence diagram shows the protocol flow for when Alice's web mail
application orders a book for her.

~~~plantuml
@startuml
hide footbox

actor Alice
entity "Web mail \n application" as Webmail
entity "Identity provider" as IDP
entity "Book service" as Booksite
database "Email service" as Email

Alice -> Webmail: Want book!
Webmail -> IDP: Give read-only access token, \n here is my refresh token
IDP -> Webmail: Your access token \n and a new refresh token
Webmail -> Booksite: Send Alice book, here is a read-only access token
Booksite -> Email: Give Alice's emails, here's my access token
Email -> Booksite: The emails you asked for
Booksite -> Alice: Your book!
@enduml
~~~


# The OIDC 1.0 protocol: authorization code

FIXME: write this

## JSON Web Token format

[JSON Web Token]: https://tools.ietf.org/html/rfc7519

# Acknowledgement

Thank you to Ivan Dolgov and Pyry Heiskanen for reviewing merge
requests and general support when writing this document.

# References

---
title: "OAuth2 and OpenID Connect: protocols and acceptance criteria"
author: Lars Wirzenius
documentclass: report
bibliography: bibliography.yaml
bindings: 
  - yuck.yaml
classes:
  - json
...