86
votes

We have a website where the only way to login and authenticate yourself with the site is with Facebook (this was not my choice). The first time you login with Facebook, an account gets automatically created for you.

We now want to create an iPhone application for our site and also a public API for others to use our service.

This question is about how to authenticate with our website from the app/API and is broken into 2 parts:

  1. What is the correct way to handle REST authentication from an API to a website which only uses Facebook OAuth as an authentication method?

I have read and researched a lot about standard methods of authentication for REST API. We can't use such methods as Basic Auth over HTTPS, as there are no credentials for a user as such. Something like this seems to be only for authenticating applications using the API.

Currently, the best way I can think is you hit an /authorize end-point on our API, it redirects to Facebook OAuth, then redirects back to the site and provides a 'token' which the user of the API can use to authenticate subsequent requests.

  1. For an official application that we create, we wouldn't necessarily need to use the public API in the same way. What would be the best way then to talk to our website and authenticate users?

I understand (I think) how to authenticate 3rd-party applications that are using our API, using API (public) keys and secret (private) keys. However, when it comes to authenticating the user who is using the app, I am getting rather confused about how to go about it when the only way we have to authenticate a user is Facebook.

I feel like I'm missing something very obvious, or don't fully understand how public REST APIs should work, so any advice and help would be greatly appreciated.

3
I have similar question at stackoverflow.com/questions/30230482/… that talks about some of theseJVK

3 Answers

100
votes

UPDATE: see below

I've been thinking hard about this question too. It's not entirely clear to me yet but here's the route I am thinking of going. I am creating a REST API an my users only auth with Facebook connect.

On the CLIENT:

  1. Use the Facebook API to login and get an OAUTH2 code.
  2. Exchange this code for an access token.
  3. In every call to my custom API I'll include the Facebook user id and the access token.

On the API (for every method that requires user authentication):

  1. Make a request to the /me Facebook graph using the access token from above.
  2. Verify that the Facebook user id returned matches the user id passed to my API from above.
  3. If the access token has expired additional communication is required.

I have yet to test this. How does it sound?

--- Update: July 27th, 2014 to answer question ---

I only use the above exchange once upon login. Once I determine which user is logging in, I create my own access token, and that token is used from that point going forward. So the new flow looks like this...

On the CLIENT:

  1. Use the Facebook API to login and get an OAUTH2 code.
  2. Exchange this code for an access token.
  3. Request an access token from my API, including the Facebook token as a parameter

On the API

  1. Receive access token request.
  2. Make a request to the /me Facebook graph using the facebook access token
  3. Verify that the Facebook user exists and match to a user in my database
  4. Create my own access token, save it and return it to the client to be used from this point forward
15
votes

This is my implementation using JWTs (JSON Web Tokens), basically similar to Chris' updated answer. I have used Facebook JS SDK and JWT.

Here's my implementation.

  1. Client: Use Facebook JS SDK to log in and get the access token.

  2. Client: Request JWT from my API by calling /verify-access-token endpoint.

  3. MyAPI: Receives access token, verify it by calling /me endpoint of Facebook API.

  4. MyAPI: If access token is valid, finds the user from database, logs in the user if exist. Create a JWT with required fields as payload, set an expiry, sign with the secret key and send back to the client.

  5. Client: Stores the JWT in local storage.

  6. Client: Sends the token (the JWT from step 5) along with the request for the next API call.

  7. MyAPI: validate the token with the secret key, if token is valid, exchange the token for a new one, send it back to the client along with the API response. (No external API calls for verification of the token here after) [if the token is invalid/expired request client to authenticate again and repeat from 1]

  8. Client Replaces the stored token with the new one and use it for the next API call. Once the token expiry is met, the token expires revoking access to API.

Every token is used once.

Read more answers about security and JWT

How secure is JWT

If you can decode JWT how are they secure?

JSON Web Tokens (JWT) as user identification and authentication tokens

5
votes

I am trying to answer the same question and have been going through a lot of reading recently...

I won't have "the" answer but things are getting a little clearer for me. Have you read the comments in the article you mentioned? I found them really interesting and helpful.

As a result, and in the light of how things have evolved since the first article has been written, here's what I think I'll do:

  • HTTPS everywhere — this allows you to forget about HMAC, signing, nonce, ...

  • Use OAuth2:

    • When authentication requests come from my own apps/website, use this 'trick' (or a variation of it) described in a reply to the article mentioned before.

    • In my case, I have two types of users: those with classic login/password credentials and those who have signed up with Facebook Connect.
      So I'd provide a regular login form with a "Login with Facebook" button. If the user logs in with his "classic" credentials, I'd just send these to my OAuth2 endpoint with a grant_type=password.
      If he chooses to log in via Facebook, I think that would be a two-steps process:

      • First, use Facebook iOS SDK to open an FBSession
      • When that's done and the app is given back control, there should be a way to get a Facebook ID for that user. I'd send this ID alone to my OAuth2 endpoint with an extension grant understood by my server as "using an FB User ID".

Please note that I am still heavily researching on all this stuff, so that might not be a perfect answer... maybe not even a correct one! But I think that would make for a good starting point. The idea of using an "extension grant" for the Facebook authentication might involve having to register it to do things properly? I'm not quite sure.

Anyway, I hope I was able to help you even a bit, and that at least it can start a discussion to find the best solution to this problem :)

Update
The Facebook login is not a solution as pointed in the comments: anybody could send an arbitrary user ID and log in as this user on the API.

What about doing it like this:

  • Show a login form with a "Facebook login" button
  • If this login method is chosen, act kinda like the Facebook SDK: open a web page from your authentication server, which will initiate the Facebook login.
  • Once the user has logged in, Facebook will use your redirect URL to confirm; make that URL point to another endpoint of your authentication server (possibly with an extra parameter indicating the call came from an app?)
  • When the authentication endpoint is hit, the authentication can securely identify the user, retain its FB User ID/FB Session and return an access token to your app using a custom URL scheme, just like the Facebook SDK would do

Looks better?