Hi all,

I discussed OAuth with some of the security experts here at Deutsche 
Telekom. We came up w/ an idea for enhancing refresh token handling I 
would like to discuss in the WG.

Assumption: refresh tokens have a very long duration (month to 
unlimited) and are stored at the client in a persistent storage (e.g. 
database or filesystem)
Question: How can token theft automatically be detected/prevented?

Let me detail the problem using three scenarios:

1) Client is a web site
Let's assume an attacker breaks into the server and gets access all 
tokens stored there along with the client's secret. Abuse could be 
prevented by changing the client secret, but this requires to _detect_ 
the theft first!

2) Mobile application/malware on same device
Let's assume an application on a mobile device, which stores refresh 
token(s) in the device's file system. A malware on the same device could 
read the token(s) from long-term storage and send it to a server under 
the attackers control. One could prevent such an attack by using a 
client secret (probably burned into the application), but that's not a 
very reliable measure. Filesystem security could be used to prevent 
eavesdropping, but its quality varies very much among mobile operating 
systems. The token could be invalidated by the user, after he/she 
realized the theft using some user self care function. But again, the 
theft has to be detected.

3) Mobile application/cloning device software and configuration
Someone could clone a device (operating system, configuration, 
applications and refresh tokens/client secrets) e.g. using a debug API 
when the user has left the device unattended. According to our experts, 
this does not take that much time and is typically not detectable 
afterwards. The clone could be run on another device identical  in 
construction and could be used to access all applications on behalf of 
the legitimate end user.
As a possible countermeasure, one could bind a refresh token to a device 
specific properties, e.g. the IMEI - International Mobile Station 
Equipment Identity. But not every device has a 3G module and access to 
this data might not be possible on some platforms.

The impact of all attacks illustrated above strongly depends on the 
power associated with a token. As soon as the attacker is able to spend 
the end user's money or gets access to medical records it's getting 
serious!

All countermeasures mentioned on the scenarios are either out of reach 
of the Oauth spec, platform specific, or not very effective. What we 
would like to achive is an automatic detection and at best prevention of 
such attacks.

We came up with the following idea:

The authorization server automatically creates a new refresh token with 
every "refresh" request (section 4) and invalidated the previous refresh 
token. The client must use this new token for the next "refresh" 
request. All attempts to obtain access tokens with an invalidated 
refresh token are refused. Additionally, the authorization server should 
hold a history of when the refresh token has been used. The client 
application behavior depends on who uses the refresh token first after 
it has been stolen.

a) If the end user uses the refresh token first, the stolen token is 
invalidated automatically, any attempts of the attacker to obtain an 
access token are refused by the authorization server.
b) Attacker uses the token first: The end users application attempts to 
obtain an access token using the (already invalidated) refresh token are 
refused. The client should notify the user of the problem and recomend 
him/her to check its application authorizations (refresh tokens) in a 
user self care portal. There, the user should have acces to information 
on when the token has been used the last time and therewith detect any 
odd behavior. The user could then revoke the token and/or alarm its 
providers helpdesk.

We think the proposed change to the API would significantly contribute 
to the overall security level of OAuth.

Thoughts?

regards,
Torsten.