One of the other PHP libraries I’ve been working for Linkey is a PHP library that makes working with other OAuth 2.0 identity providers “stupidly easy”. I think I’ve done that and it’s time to announce the initial release – https://github.com/lncd/OAuth2-client.

So lets say you want to allow users to sign-in to their Facebook account:

$provider = new \OAuth2\Client\Provider\Facebook(array(
    'clientId'  =>  'XXXXXXXX',
    'clientSecret'  =>  'XXXXXXXX',
    'redirectUri'   =>  'http://your-registered-redirect-uri/'
));

if ( ! isset($_GET['code'])) {

    // If we don't have an authorization code then get one
    $provider->authorize();

} else {

    try {

        // Try to get an access token (using the authorization code grant)
        $t = $provider->getAccessToken('authorization_code', array('code' => $_GET['code']));

        try {

            // We got an access token, let's now get the user's details
            $userDetails = $provider->getUserDetails($t);

            foreach ($userDetails as $attribute => $value) {
                var_dump($attribute, $value) . PHP_EOL . PHP_EOL;
            }

        } catch (Exception $e) {

            // Failed to get user details

        }

    } catch (Exception $e) {

        // Failed to get access token

    }
}

Simple right? If you take out the try/catch statements then it essentially boils down to this:

$provider = new \OAuth2\Client\Provider\<provider name>(array(
    'clientId'  =>  'XXXXXXXX',
    'clientSecret'  =>  'XXXXXXXX',
    'redirectUri'   =>  'http://your-registered-redirect-uri/'
));

if ( ! isset($_GET['code'])) {

    $provider->authorize();

} else {

    $token = $provider->getAccessToken('authorization_code', array('code' => $_GET['code']));
    $userDetails = $provider->getUserDetails($token);
}

The library automatically manages the state parameter to help mitigate cross-site request forgery attacks (where supported by the end-IdP).

At the time of writing there is built in support for Facebook, Google and Github but adding support for other identity providers is trivial – you just need to extend the IdentityProvider class.

I will add support for more providers soon. There also aren’t any unit tests currently but they are coming.

The library is hooked up to Packagist so just add "lncd/oauth2-client": “*” to your composer.json file.

On Tuesday I presented at the first ever PHP North East conference in Newcastle. Hosted in the fantastic Tyneside Cinema (which is the last surviving Newsreel theatre still operating as a cinema full-time in the UK), it was a very well organised event and I’m incredibly grateful to the PHP North East group for accepting my talk proposal.

My topic was “API driven development: Eating your own dog food”, I’ve embedded my slides below:

Unfortunately they weren’t able to video my talk so my slides may seem a little out of context but hopefully you can get the overall gist of what I talked about.

Once again, I’d like to say thank you to Anthony Sterling and the PHPNE crew for putting on a wonderful event.

Someone has posted a Github Gist with all of the client identifiers and secret keys for the official Twitter clients on various platforms.

This just highlights why it is imperitive that you don’t screw up like Facebook did by allowing wildcard client redirect URIs.

The client redirect URIs must be absolute and stored on the authorisation server.

So what can Twitter do about this?

They could release new binarys for all of their clients with new keys and invalidate the old keys, but clients that haven’t been updated will suddenly stop working and leave users confused and frustrated. Also it will be trivial for the new keys to be leaked too.

Alternatively they could come up with some sort of mechanism to allow the official clients to download new keys on the fly but then these could be captured with a man-in-the-middle attack.

But is it actually a problem?

First, it is important to recognise that this is really only a problem for desktop/mobile app clients (assuming that is that web server based clients themselves keep their keys secure – the differnece being that someone would have to break into a web server to get to them).

Therefore as long as Twitter aren’t giving their own applications additional functionality that 3rd party clients can’t replicate then there isn’t too much of a problem because with this equal playing field scenario it could be Tweetbot or MetroTwit we’re talking about.

Having said that, the official iOS mobile client at least does allow you to register a new account, so these leaked keys could be used to create spam accounts.

With 3rd party Twitter clients being limited to 100,000 access tokens this could tempt some to make use of these keys however if they were caught then they risk being named and shamed by the media and that in turn could also result in developer accounts being banned by Apple and Google from their app stores.

Is OAuth to blame?

No, the OAuth protocol is behaving as it designed to here. That is, if a 3rd party client presented a correctly formed request with these valid keys and the correct redirect URI then, assuming Twitter doesn’t have some extra security in place to validate requests using their keys (such as validating user-agent strings for example), the authorisation server will assume the request is genuine.

OAuth by it’s nature is a very flexible standard and can adapted to work in many different scenarios. The core specification describes four authorisation grants:

  • Authorisation code grant
  • Implicit grant
  • Resource owner credentials grant
  • Client credentials grant

The specification also details another grant called the refresh token grant.

Furthermore there are a number of other grants that have gone through the IETF ratification process (none of which at the time of writing have been formally standardised):

  • Message authentication code (MAC) tokens
  • SAML 2.0 Bearer Assertion Profiles
  • JSON web token grant

The end goal of each of these grants (except the refresh token grant) is for the client application to have an access token (which represents a user’s permission for the client to access their data) which it can use to authenticate a request to an API endpoint.

In this post I’m going to describe each of the above grants and their appropriate use cases.

As a refresher here is a quick glossary of OAuth terms (taken from the core spec):

  • Resource owner (a.k.a. the User) – An entity capable of granting access to a protected resource. When the resource owner is a person, it is referred to as an end-user.
  • Resource server (a.k.a. the API server) – The server hosting the protected resources, capable of accepting and responding to protected resource requests using access tokens.
  • Client – An application making protected resource requests on behalf of the resource owner and with its authorisation. The term client does not imply any particular implementation characteristics (e.g. whether the application executes on a server, a desktop, or other devices).
  • Authorisation server – The server issuing access tokens to the client after successfully authenticating the resource owner and obtaining authorisation.

Authorisation code grant (section 4.1)

The authorisation code grant is the grant that most people think of when OAuth is described.

If you’ve ever signed into a website or application with your Twitter/Facebook/Google/(insert major Internet company here) account then you’ll have experienced using this grant.

Essentially a user will click on a “sign in with Facebook” (or other IdP) and then be redirected from the application/website (the “client”) to the IdP authorisation server. The user will then sign in to the IdP with their credentials, and then – if they haven’t already – authorise the client to allow it to use the user’s data (such as their name, email address, etc). If they authorise the request the user will be redirected back to the client with a token (called the authorisation code) in the query string (e.g. http://client.com/redirect?code=XYZ123) which the client will capture and exchange for an access token in the background.

This grant is suitable where the resource owner is a user and they are using a client which is allows a user to interact with a website in a browser. An obvious example is the client being another website, but desktop applications such as Spotify or Reeder use embedded browsers.

Some mobile applications use this flow and again use an embedded browser (or redirect the user to the native browser and then are redirected back to the app using a custom protocol).

In this grant the access token is kept private from the resource owner.

If you have a mobile application that is for your own service (such as the official Spotify or Facebook apps on iOS) it isn’t appropriate to use this grant as the app itself should already be trusted by your authorisation server and so the _resource owner credentials grant would be more appropriate.

Implicit grant (section 4.2)

The implicit grant is similar to the authentication code grant described above. The user will be redirected in a browser to the IdP authorisation server, sign in, authorise the request but instead of being returned to the client with an authentication code they are redirected with an access token straight away.

The purpose of the implicit grant is for use by clients which are not capable of keeping the client’s own credentials secret; for example a JavaScript only application.

If you decide to implement this grant then you must be aware that the access token should be treated as “public knowledge” (like a public RSA key) and therefore it must have a very limited permissions when interacting with the API server. For example an access token that was granted using the authentication code grant could have permission to be used to delete resources owned by the user, however an access token granted through the implicit flow should only be able to “read” resources and never perform any destructive operations.

Resource owner credentials grant (section 4.3)

When this grant is implemented the client itself will ask the user for their username and password (as opposed to being redirected to an IdP authorisation server to authenticate) and then send these to the authorisation server along with the client’s own credentials. If the authentication is successful then the client will be issued with an access token.

This grant is suitable for trusted clients such as a service’s own mobile client (for example Spotify’s iOS app). You could also use this in software where it’s not easy to implement the authorisation code – for example we bolted this authorisation grant into OwnCloud so we could retrieve details about a user that we couldn’t access over LDAP from the university’s Active Directory server.

Client credentials grant (section 4.4)

This grant is similar to the resource owner credentials grant except only the client’s credentials are used to authenticate a request for an access token. Again this grant should only be allowed to be used by trusted clients.

This grant is suitable for machine-to-machine authentication, for example for use in a cron job which is performing maintenance tasks over an API. Another example would be a client making requests to an API that don’t require user’s permission.

When someone visits a member of staff’s page on the University of Lincoln staff directory the website uses it’s own access token (that was generated using this grant) to authenticate a request to the API server to get the data about the member of staff that is used to build the page. When a member of staff signs in to update their profile however their own access token is used to retrieve and update their data. Therefore there is a good separation of concerns and we can easily restrict permissions that each type of access token has.

Refresh token grant (section 1.5)

The OAuth 2.0 specification details a fifth grant which can be used to “refresh” (i.e. renew) an access token which has expired.

Authorisation servers which support this grant will also issue a “refresh token” when it returns an access token to a client. When the access token expires instead of sending the user back through the authorisation code grant the client can use to the refresh token to retrieve a new access token with the same permissions as the old one.

My problem with the grant is that it means the client has to maintain state of each token and then either on a cron job keep access tokens up to date or when it tries to make a request and it fails then go and update the access token and repeat the request.

I personally prefer to issue access tokens that last longer than the user’s session cookie so that when they next sign in they’ll be issued a new token anyway.


The extension grants

The following grants are currently going through the standardisation process.

N.B. My explanation from here on is how ** I ** (as a developer) am interpreting the specifications – all three of the following grants specs need further work in my opinion to better explain their purpose and how they compare to the core grants.

MAC token grant (draft 3 at time of writing)

The MAC token grant can be used alongside another grant (the specification describes using it alongside the authentication code grant) to further improve the security of requests to the API server by including the addition of a MAC (message authentication code) signature along with the access token in order to both authenticate the request and prove the identity of the client making the request (because it prevents tampering and forgery). In some ways it is similar to the OAuth 1.0 request process.

When the authorisation server returns an access token to a client it also includes a key which the client uses to generate the MAC signature. The MAC signature is generated by combining the parameters of the request and then hashing them against the key.

This grant could be used when additional security measures are needed to ensure that the client querying an API is definitely who it is identifying itself as. It can also prevent access tokens being used by unauthorised clients (if an access token has been compromised by a man in the middle attack) because a valid signature (which can only be generated by the client which knows the MAC key associated with the access token) is required for each request.

SAML 2.0 Bearer Assertion Profiles (draft 15 at time of writing)

A SAML assertion is an XML payload which contains a security token. They are issued by identity providers and consumed by a service provider who relies on its content to identify the assertion’s subject for security-related purposes. It is quite literally an assertion as to who someone (or what something) is.

This grant allows a client to exchange an existing SAML assertion for an access token, meaning the user doesn’t have to authenticate again with an authorisation server. In some ways it resembles the concept of the refresh token grant.

An example use case of this grant would be a publishers website receiving an assertion from the UK Federation (after a user from a UK education institution has authenticated with their institution’s own resource server), converting the assertion into an access token and then querying the users institution’s API server to request additional details about the user.

JSON web token grant (draft 6 at time of writing)

The JSON web token grant is similar to the SAML assertion grant but uses JSON as the serialisation format instead of XML. This means that a JWT (JSON web token) can comfortable fit inside a HTTP authorization header.

A potential use case could be a web based client application that also has a mobile version, the web version could somehow share the existing access token with the mobile application which in turn requests its own access token. Possibly? Maybe? I’ll admit I can’t think of a better example.

Last week self-proclaimed web security evangelist Egor Homakov published a blog post entitled How we hacked Facebook with OAuth2 and Chrome bugs. Naturally this led to a new round of “OAuth 2.0 is unsecure” comments on news articles that linked to the post.

If you actually read through the post however, the part of the attack that involves OAuth is this bit (emphasis from original text):

2) OAuth2 is… quite unsafe auth framework. Gigantic attack surface, all parameters are passed in URL. I will write a separate post about OAuth1 vs OAuth2 in a few weeks. Threat Model is bigger than in official docs.
In August 2012 I wrote a lot about common vulnerabilities-by-design and even proposed to fix it: OAuth2.a.

We used 2 bugs: dynamic redirect_uri and dynamic response_type parameter.
response_type=code is the most secure authentication flow, because end user never sees his access_token. But response_type is basically a parameter in authorize URL. By replacing response_type=code to response_type=token,signed_request we receive both token and code on our redirect_uri.

redirect_uri can be not only app’s domain, but facebook.com domain is also allowed.
In our exploit we used response_type=token,signed_request&redirect_uri=FB_PATH where FB_PATH was a specially crafted URL to disclose these values..

If an OAuth 2.0 authorisation server is correctly configured then it should, in addition to numerous other checks, ensure that there is only one value in the response_type parameter (e.g. “code” – if using the authorization code grant) and the server should store whitelisted redirect URIs for each client on the server so that attacks can’t be launched with dynamic URIs.

I suspect partly why this has happened to Facebook is because when they launched their Graph APIs back in 2010 they secured them with draft 5 of the OAuth 2.0 specification. As the specification changed over it’s 27 drafts they already had clients hardcoded with specific parameters that couldn’t be updated easily and so therefore they had to make a choice to allow the above parameters that were used in the attack to accept dynamic and non-canon values.

The OAuth 2.0 server library I’ve developed is not vulnerable to the aforementioned attack as it strictly validates and verifies every parameter that is sent to it and redirect URIs are whitelisted on the server.