Open redirect in rfc6749 aka 'The OAuth 2.0 Authorization Framework'

tl;dr The Internet Bug Bounty rewarded me with a bounty for an Open Redirect in  rfc6749 aka 'The OAuth 2.0 Authorization Framework' .









Here the long version.

The Introduction 

 

Several months ago I did realize that if you want to implement an OAuth Authorization Server and  follow verbatim the OAuth core spec you might end up having an Open Redirect.
Now there is still some debate about this class of vulnerability since often they are relatively benign but not always (as we can see later).
Despite all at that point I notified the OAuth working group. There was some longish discussion but eventually (almost) all in the list agreed that this was somehow an issue (no where near the end of the world :)).

The Issue

 

Section 4.1.2.1 of the OAuth specification says:

   If the request fails due to a missing, invalid, or mismatching
   redirection URI, or if the client identifier is missing or invalid,
   the authorization server SHOULD inform the resource owner of the
   error and MUST NOT automatically redirect the user-agent to the
   invalid redirection URI. 

   If the resource owner denies the access request or if the request
   fails for reasons other than a missing or invalid redirection URI,
   the authorization server informs the client by adding the following
   parameters to the query component of the redirection URI using the...

Now let's assume an attacker:

• Registers a new client to the victim.com provider.
• Registers a redirect uri like attacker.com.

Then the attacker can craft a special URI of the form

http://victim.com/authorize?response_type=code&client_id=bc88FitX1298KPj2WS259BBMa9_KCfL3&scope=WRONG_SCOPE&redirect_uri=http://attacker.com

according to Section 4.1.2.1 this should redirect back to attacker.com (without any user interaction, ever...)!!! Here we use the a wrong scope parameter but any reasons other than a missing or invalid redirection URI would had make the trick....
Now according to my dictionary this is an open redirect :p

Some live example of real providers that exhibit this behavior:

• Facebook: https://graph.facebook.com/oauth/authorize?response_type=code&client_id=1621835668046481&redirect_uri=http://www.attacker.com/&scope=WRONG_SCOPE
• Github: https://github.com/login/oauth/authorize?response_type=code&redirect_uri=http://attacker.com2&client_id=e2ddb90328315c367b11
• Microsoft: https://login.live.com/oauth20_authorize.srf?response_type=code&redirect_uri=http://attacker.com&client_id=000000004C12822C

A real cool special case is Moves (just click and enjoy :p)

Update: it seems that Facebook doesn't allow anymore redirect to data:text/html (hence the link below is broken now), more to come on this topic... ;)

https://api.moves-app.com/oauth/v1/authorize?response_type=code&client_id=bc88FitX1298KPj2WS259BBMa9_KCfL3&redirect_uri=data%3Atext%2Fhtml%2Ca&state=<script>alert('hi')</script>

As  (almost) usual Google did it right. Namely Google return 400 with the cause of the error..

400. That’s an error.

Error: invalid_scope

Some requested scopes were invalid. {invalid=[l]}

As we will see this is only one of the possible mitigation.

The Vulnerability

 

Now you might argue that this is ONLY an open redirect and there is not much you can do with it right? :)
Well well well ...

All I need is....an open redirect

Sometimes in order to accomplish some sort of attack you need to have an open redirect. This is only one small part of the chain but an essential one. And what can it be better (from the attacker perspective) if an OAuth provider gives you one :D ? If you do not believe me look like Andris Atteka used this very own issue as part of his attack to steal an access token.

Lassie come home (again)

In some of my previous post I have highlighted the importance for an Authorization Server to
use exact matching against registered redirect uri to validate the redirect_uri parameter. 
Indeed John Bradley realized how some attacker can chain the relaxed redirect uri validation with the open redirect described here to steal an access token.
He pointed out a possible attack scenario in the OAuth mailing list thread. 
You can find more details about this specific attack in the security addendum draft.

The Mitigation

 

John Bradley,  Hannes Tschofenig and me came up with a draft for an OAuth security addendum that should provide better advice to implementers.
The mitigations are described in section 2.3 and they are rather simple, either :

• Respond with an HTTP 400 (Bad Request) status code. 
• Perform a redirect to an intermediate URI under the control of the Authorization Server to clear referer information 

As usual comments are welcome....