Re: [OAUTH-WG] A Scope Attack against OAuth 2.0

[Phil Hunt <[e-mail]>]

I'm still not seeing how "user choice" can become an "attack"

While a user might hack the request or a phishing attempt might generate a link that requests an unusual scope combination, it shouldn't matter since the authorization server should allow the user to decline specific scopes if they want to / or the authorization server may choose to override specific combinations of scope.

The client has to be prepared to accept in all cases that it may not get what it asks for. It may for example be entirely refused. If the client doesn't get what it asked for then it can choose to generate an error to the user.

A common case is many apps will connect to social nets and request "ReadProfile" and "UpdateStatus". Yet the user might not want to grant "UpdateStatus". The client app has to handle this gracefully.

Finally scope is a right granted to the client for access to a resource -- not the other way around as your attack suggests.

Phil

@independentid
www.independentid.com
[e-mail]





On 2012-02-23, at 3:55 PM, Wenjie Lin wrote:

> We assume that the authorization server is trustworthy and won’t do anything that violates the spec. We want to prevent attacks by the client on the user, as well as the attacks by the user on the client.
> 
> The scope attack is by the user on the client. From the point view of the user it is not an attack. However, the client takes it as an attack; the user violates the service agreement by the client who may have no knowledge about it, unless the authorization server always sends him the authorized scope information when granting an access token, as we've proposed, and the client can figure out the difference/violation and decide how to deal with it.
>  
> When the client identifies the difference between his specified scope and that in the access token, he can choose to abort or continue the service, or take other actions. This is an implementation issue and is beyond the scope of OAuth. There are other possible anomalies from implementations, such as revocation of the scope by the user. However, they are not well specified in the current spec and we take them as an implementation rather than a protocol issue.
>  
> There might be applications where the client would not perceive scope attack as an attack if he could allow any changes of the service agreement scope by the user. If OAuth were aimed for such clients only, it would significantly limit the applicability of the protocol. 
> 
> We appreciate the insightful discussions on how the client could check and handle the scope differences and beyond. They are extremely helpful for the implementers.
> 
> -W. Lin and D. Lee
> 
> On Tue, Feb 21, 2012 at 4:59 PM, John Bradley <[e-mail]> wrote:
> Yes,  OpenID Connect deals with the issue by using a signed request extension in the cases where the client needs to be certain the request is only from the legitimate client and not tampered with.
> 
> As you observe anyone can send a request the client_id and redirect_uri are not secret in any way.
> 
> Though a well behaved client should be using state to check for XSRF attacks. (that is a real attack)
> 
> Signed requests have uses,  but are not core to OAuth.
> 
> John B.
> On 2012-02-21, at 9:38 PM, André DeMarre wrote:
> 
> > The consensus seems to be, and I agree, that this shouldn't be
> > considered an "attack," but that's really just nomenclature. I do
> > concede that there is a spec issue here that I failed to appreciate at
> > first. Where draft-ietf-oauth-v2-23 section 3.3 says "If the issued
> > access token scope is different from the one requested by the client,"
> > it suggests that the authorization server knows what was requested by
> > the client. That isn't strictly true; it only knows what was in the
> > authorization request, not who put it there. For the client to truly
> > know what the auth server wants, (1) it would need to communicate
> > directly with the client, (2) the client would need to preregister its
> > desired scope, or (3) the client would need to sign a message that the
> > user passes to the authorization endpoint. 1 and 3 are clearly not
> > compatible with the spec, 2 might be; but that's not the point.
> >
> > A more correct statement in section 3.3 would be "If the issued access
> > token scope is different from the one in the authorization request..."
> >
> > Regards,
> > Andre DeMarre
> >
> > On Tue, Feb 21, 2012 at 3:01 PM, John Bradley <[e-mail]> wrote:
> >>
> >> On 2012-02-21, at 7:32 PM, Nicholas Devenish wrote:
> >>
> >>>
> >>> On 21 Feb 2012, at 21:59, John Bradley wrote:
> >>>> This 'attack'  is one that only works with badly designed clients that are making unwarranted assumptions about the behaviour of the Authorization server.
> >>>
> >>> Unwarranted assumptions? The spec in 3.3 says:
> >>>
> >>> "If the issued access token scope is different from the one requested by the client, the authorization server MUST include the "scope" response parameter to inform the client of the actual scope granted."
> >>>
> >>> - It says MUST; therefore any server that doesn't do this is non-compliant?
> >>> - It says scope different from the one requested by the *client*. The possibility that the resource owner intercepts this request and changes it doesn't seem to be considered here (it is not strictly the clients request if that happens)
> >>> - The purpose seems to be that the client should be told if the scope changes; this would be important if the client requires a certain scope level to work (though this could be solved more generally in many other ways that wouldn't be in the scope of the oauth spec)
> >>>
> >>> Thus, assuming that the server is stating compliance, isn't the assumption completely warranted?
> >>
> >> No the authorization server may at any time for any reason remove a scope from a granted access_token or refresh_token.
> >>
> >> Reporting back changes in scopes granted along with the access_token is a convenience not a security feature.
> >>
> >> Assuming it is a security feature and those scopes will continue to be valid for the token after granting is a bad design given the OAuth 2 spec.
> >>>
> >>>> The only way for a client to know if a token has a scope it to try it, or use a introspection endpoint.  End of story.
> >>>
> >>> An introspection endpoint obviously isn't part of the specification, so isn't relevant to the discussion (though it solves the discussed facebook issue).
> >>>
> >>> You are right though, that the only way for a client to know for sure is to try to use it; Even if the spec mandated always returning the scope to the client, the user could always intercept the return redirection (assuming a non-confidential client) and change the scope there.
> >>>
> >>> Perhaps MUST should change to SHOULD, given that this essentially seems unenforceable?
> >>
> >> A SHOULD may lead people to the conclusion that it is secure.   I am happy with saying it is not secure the only want to know is to have the client be prepared to deal with tokens that do not contain the desired scope when used.   That is the only 100% solution.
> >>
> >> John B.
> >>
> >>
> >>
> >>
> >> _______________________________________________
> >> OAuth mailing list
> >> [e-mail]
> >> https://www.ietf.org/mailman/listinfo/oauth
> >>
> 
> 
> 
> 
> -- 
> Wenjie Lin
> _______________________________________________
> OAuth mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/oauth