Would it be possible to consider adding this to the list of security considerations? Of course, the spec cannot cover all possible security threats, but this appears to be a realistic one which could easily be exploited if overlooked by developers (evident in the lack of scraping defense mechanisms in many applications). Is it too late to make such an amendment to the draft?
Thank you, Niv On Tue, Jul 26, 2011 at 02:40, Niv Steingarten <nivst...@gmail.com> wrote: > Yes, I believe the vast majority of user-agents would block this kind > of request if it originated from a JavaScript XMLHttpRequest or such. > However, there are still scenarios in which a user-agent based client > could manipulate this vulnerability. > > For example, the client could launch the GET request to the > authorization server via an <img> HTML tag, taking the form of a CSRF. > Since it's a blind attack, the client would likely not receive the > contents of the web-page. However, this request is still necessary > from the client since it has the side-effect of creating an access > token (or other temporary token) on the authorization server. Since it > is highly likely that a malicious client has a priori knowledge of the > structure of the authorization page and form, it does not need the > response in order to advance to the next step, and can simply send the > fake request to 'Allow' itself to access the resource owner's > resources. > > I believe this attack could be made very hard by either including a > CAPTCHA, as you suggested, or including some kind of token or nonce in > the submission of the form (which would still not prevent the attack > if the authorization server doesn't enforce same origin policy). > > Niv > > > > On Tue, Jul 26, 2011 at 02:10, Torsten Lodderstedt > <tors...@lodderstedt.net> wrote: >> >> Hi Niv, >> >> thank you for posting this to the list. I think you are right with your >> threat description. One question: shouldn't the browser already prevent the >> request to the authorization endpoint because of the same origin policy (or >> CORS restrictions)? >> >> Apart from that, a similar attack can be performed by a native applicication >> (using an embedded browser). This kind of attack could not be prevented >> using HTTP features but by enforcing a real user interaction (password >> input, CAPTCHA). >> >> regards, >> Torsten. >> >> Am 25.07.2011 18:27, schrieb Niv Steingarten: >> >> Forwarded as per Eran's request. >> A couple of corrections to my original email: >> 1. By AJAX, I mean, AJAX like techniques (if the user agent does not enforce >> same origin policy). >> 2. When saying POST to '/authorize_callback' -- it may well be GET, if the >> authorization server mishandles the request. >> Thank you, >> Niv >> >> >> ---------- Forwarded message ---------- >> From: Eran Hammer-Lahav <e...@hueniverse.com> >> Date: Tue, Jul 26, 2011 at 01:21 >> Subject: RE: Several typos in -20 and a possible security consideration >> To: Niv Steingarten <nivst...@gmail.com> >> >> >> Please forward this message to the oauth list at oa...@ieft.org. >> >> >> >> Thanks, >> >> >> >> EHL >> >> >> >> From: Niv Steingarten [mailto:nivst...@gmail.com] >> Sent: Monday, July 25, 2011 2:52 PM >> To: draft-ietf-oauth...@tools.ietf.org >> Subject: Several typos in -20 and a possible security consideration >> >> >> >> Hello, >> >> >> >> I've noticed a couple of typos in -20: >> >> >> >> Section 6 (page 41): Under 'The authorization server MUST', the second >> bullet should end with the word "and", and the third bullet should end with >> a full-stop. >> >> Section 10.2 (first paragraph): "...keep is client credentials confidential" >> should be "...keep *its* client credentials confidential". >> >> >> >> Regarding the security consideration -- >> >> >> >> I might be missing something, but I saw there are references to clickjacking >> and to client impersonation, but I haven't seen any reference to possible >> resource owner impersonation. >> >> For example, in the implicit grant flow, a malicious client could send a >> request to the authorization endpoint via, say, AJAX, and receive the markup >> of the page asking the resource owner to authorize the client (assuming the >> resource owner is signed in and no resource owner authentication is >> required). Then, in a poorly designed authorization endpoint, the 'Allow' >> button might be the submission button of a form whose target is >> '/authorize_callback' on the authz server. Then, it may possible for the >> malicious client to simply POST to '/authorize_callback' and authorize >> itself without any resource owner intervention or knowledge that the process >> has even taken place. This, of course, can be mitigated in most modern >> browsers if the authorization server verifies the source of the request >> using the HTTP referrer header. >> >> >> >> Thanks for your time and for the fantastic work on OAuth, >> >> >> >> -- >> >> Niv Steingarten >> >> >> >> T: +972-54-5828088 >> >> E: nivst...@gmail.com >> >> W: http://nivstein.com >> >> >> >> >> -- >> Niv Steingarten >> T: +972-54-5828088 >> E: nivst...@gmail.com >> W: http://nivstein.com >> >> _______________________________________________ >> OAuth mailing list >> OAuth@ietf.org >> https://www.ietf.org/mailman/listinfo/oauth > _______________________________________________ OAuth mailing list OAuth@ietf.org https://www.ietf.org/mailman/listinfo/oauth
