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.
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.
It has been a busy few weeks lately culminating in a work trip to the USA last week. This trip was quite beneficial as, when one is stuck on a plane for 9 hours each way, or completely disconnected from the world 8300ft up the Colorado Rockies in a 1970s cabin with no phone signal, TV or Internet it allows for lots of thinking.
I’ve come up with a list of things that I want to achieve for Linkey by the end of Semester 1 (end of January ’13):
- Complete the first draft of a case study looking at institutional uses of OAuth.
- Invite our program manager, Chris Brown, for a site visit and spend the day looking over the case study and planning the last few months of the project.
- Formally release the code I have been working on, including documentation, unit tests and tutorials.
- Look at developing a .well-known standard for describing OAuth endpoints and enabling discovery (more on this in a future post).
- Work with my colleagues in ICT to start implementing a joint OAuth and Microsoft UAG web single sign-on implementation.
I’ve been asked to work with the Orbital team to try and develop an authentication endpoint that will allow University of Lincoln staff and Siemens staff to securely sign-in to applications that have been developed by that project, but with each member of staff using their own institutional/company credentials.
Eran Hammer – the former OAuth 2.0 specification editor – recently spoke at RealtimeConf about why he left the working group and what he perceives to be the main problems with OAuth 2.0.
RealtimeConf – “OAuth 2.0 – Looking Back and Moving On” by Eran Hammer from &yet on Vimeo.
Last week, Paul and I sat down to list the IT resources that the library offer to staff and students are and how they authenticate users. We also identified services that allow us to customise the sign in experience in order to achieve at least stage one of our strategy for consistent sign in:
1) To ensure a single, consistent identity for each person, all library (and ICT) applications that we operate internally must have Active Directory sign-in instead of local databases. Almost all of our applications achieve this already.
(source Linkey bid document – section 3.1)
Before proceeding there is some jargon to explain:
- SAM – A SAM (security accounts manager) ID is a user’s AD username. For staff this will generally be the first letter of their forename + their surname – e.g. for me this is abilbie. For students this is their student number – e.g. 01234567.
- Password – This a user’s AD password.
- Employee ID – for staff this is the ID number from our employee database. For students this is the same as their SAM ID.
- SafeCom PIN – SafeCom is our printing authentication service we use here at the university. Each staff and student has their own PIN number which they use with their employee ID.
This is the list we came up with:
- library.lincoln.ac.uk – Hosted on our WordPress platform. We can completely customise every aspect of the site. Users sign in with their SAM ID and password.
- Blackboard – Our VLE. If a user is signed in it will provide infomation about their library account plus they can pay any fines. Digitised material stored is also stored in and access is granted based on a students’ course. Users sign in with their SAM ID and password.
- Clio – inter-library loans management website. Users sign in with their SAM ID and password. We can completely customise the experience.
- EPrints – our research repository. EPrints is open source so we can change anything. Users sign in with their SAM ID and password.
- EZProxy – allows for e-resources access based on IP address proxying. EZProxy is LDAP capable however it can inherit Blackboard/SharePoint authentication sessions. We can completely customise the experience.
- Open Athens LA (Local Authentication) – Enables users to securely access to online resources. Provides Athens->Athens, and Athens/SAML->UK Federation->SAML authentication. Users sign in with their SAM ID and password. This should be installed in the coming weeks. We think we can customise the experience.
- Resources that have their own username + password. A very subset of online resources the library subscribe don’t work with Athens or EZProxy and so give us a username and password to use. Users are directed to these services through our SharePoint site and are authenticate with these services in two ways:
- The user is a presented with a 401 dialog and they have to manually enter the username and password.
- Thin Clients – The GCW library on the Brayford campus over the summer had a number of thin client computers installed. When physically in front of the computer users’ authenticate with their SAM ID and password but because these machines use virtual instances we could provide a “desktop in the browser” experience via UAG.
- Find it @ Lincoln – A search engine of journals and databases that we subscribe to that is hosted remotely by Ebsco. Authentication is delegated to EZProxy.
- RefWorks – Allows users to collect references. Authentication is via Athens. We can’t customise the look and feel of RefWorks itself but we hope we can customise how our Open Athens LA looks.
- Aspire – Reading List software provided by Talis. Authentication is via Athens. See above note about customising Athens LA.
- Journals A to Z / OpenURL resolver – As above
- Databases – A combination of Athens and EZProxy authentication.
- Printing – The physical printers in the library require authentication using a user’s employee ID and their PIN number.
- Self service kiosks – These kiosks authenticate users by requiring them to scan their staff/student card (which has a code 39 barcode of their employee ID and then typing in their PIN onscreen. We can’t customise this experience.
- Print top up kiosks – These kiosks authenticate users by requiring them to type in their SAM ID and password. We can’t customise this experience.
Visually of this looks like:
When the UAG is installed we should be able to easily hook up services that require SAM ID and password over LDAP. Other services that use alternative authentication such as HIP will require us to write some middleware that will translate between SAM ID + password to employee ID + PIN.
Assuming it is as simple as that, then when these services are hooked up to the UAG then the map will look like this:
The model assumes that eventually the UAG will be like a Gateway for users to access most resources.
LDAP inject means that once a user has authenticated with the UAG their SAM ID and password will be stored in a session, and then when the user visits a service that uses LDAP authentication to the AD, UAG will inject their username and password into the sign-in form and click the submit button for them. At the end of the day single sign-in is essentially a user experience which takes some of the pain with accessing resources away from them.