If we can agree that it is useless from a security perspective (and I think we agree on that by now), then an auth server must not ever issue an access token to a user agent client without interacting with the user (no "immediate" mode issuing of access tokens) so the user is aware that some anonymous client on the web page is trying to access their data, since there's no assurance whatsoever about which client that is, and whether the user trusts it.
If anyone disagrees, please explain how you can justify it, in light of the fact that leaving it as-is means once one client is authorized, the whole world is authorized by just pretending to be that one client. It might as well be a "make my data public" switch at the resource server to authorize a client such that it gets an access token directly without a client secret. (Not the muddy the waters, but again, I know client secrets don't add value when the client is on an uncontrolled machine. Please don't let that technical hurdle stop you from justifying the status quo in draft 9 of the spec). -- Andrew Arnott "I [may] not agree with what you have to say, but I'll defend to the death your right to say it." - S. G. Tallentyre On Mon, Jul 5, 2010 at 10:27 PM, William Mills <[email protected]> wrote: > It's not verifiable, but it is as useful in this case as a "user agent" > string. Not usefule formt he security perspective, but has some utility in > application tracking. > > ------------------------------ > *From:* [email protected] [mailto:[email protected]] *On Behalf > Of *Andrew Arnott > *Sent:* Saturday, July 03, 2010 12:16 PM > *To:* Eran Hammer-Lahav > *Cc:* OAuth WG ([email protected]) > *Subject:* Re: [OAUTH-WG] Security of user agent clients (WAS: End user > authresponse code-and-token's scope parameter) > > (this sounds sarcastic, but I'm no being sarcastic... it's a serious > question/challenge)... > > Why not just remove the client_id parameter from the user-agent flow? It's > absolutely meaningless to security. It's only perceivable benefit is that > the auth server can possible display to the user the client being authorized > or the list of previously authorized clients. But again, that's totally > meaningless if not verified. > > -- > Andrew Arnott > "I [may] not agree with what you have to say, but I'll defend to the death > your right to say it." - S. G. Tallentyre > > > On Sat, Jul 3, 2010 at 12:12 PM, Andrew Arnott <[email protected]>wrote: > >> On Fri, Jul 2, 2010 at 9:12 PM, Eran Hammer-Lahav >> <[email protected]>wrote: >> >>> You are putting too much weight on the value of redirection URI >>> registration. Since the same problem exists between the user-agent script >>> and the server-side component used in the user-agent profile, anyone can >>> imitate that flow. In most cases, the redirection URI will simply include a >>> script that will pass the parameter to the **parent** window which can >>> be anything. If the server-based page is an active page, it still needs to >>> communicate with the user-agent, which again, can pretend to be that. >>> >> Good point. I wasn't imagining that far through it. Not to deny that >> most redirection URLs may just return a straight script that passes it to >> the parent window, but that seems pretty irresponsible to the host that's >> doing it -- because of the enormous security hole it opens up. Some client >> is trusted enough to authorize (by the user and auth server), receives an >> access token it wasn't expecting, and then blindly forwards it onto whoever >> wants it. Yikes. I sure hope these access tokens have *very* limited >> scope (like, I'd prefer none myself). >> >> I was about to describe how an active server could alleviate that by >> signing the original request using the state parameter, but I realized I'm >> solving a problem that the web server flow (or nowadays that the >> authorization code mode as I guess it's called) already solves. So I guess >> I just need to bite the bullet and accept that the user agent flow is >> *totally >> *insecure for web clients, and thus very special care must be taken to >> enable native clients without opening up the web client hole. >> >> Sorry to sound so negative about it. Enabling this scenario seems really >> important to me as well -- I just am against doing it when it can't be >> secured in some way. Seriously... as is, once you authorize any client, >> you've authorized them all, including clients not even running on your >> computer. I guess this answers how I was visiting a site a few days ago and >> was amazed that it knew who I was (via Facebook) and was displaying my name >> and photo when I never logged into the site. I was shocked it knew who I >> was. Enter living in Incognito mode from now on. >> > >
_______________________________________________ OAuth mailing list [email protected] https://www.ietf.org/mailman/listinfo/oauth
