Problem with nonces and HTTP GET

[Brian Kissel]

+1 Allen, here's what I get on HuffPo, not very compelling and probably a trigger to "Cancel" to most users.  We need to fix this ASAP!

[cid:image001.png at 01CA9B52.06B74CA0]

Cheers,

Brian
___________

Brian Kissel<http://www.linkedin.com/pub/0/10/254>
CEO - JanRain, Inc.
bkissel at janrain.com
Mobile: 503.342.2668 | Fax: 503.296.5502
519 SW 3rd Ave. Suite 600  Portland, OR 97204

Increase registrations, engage users, and grow your brand with RPX.  Learn more at www.rpxnow.com<http://www.rpxnow.com/>

From: openid-specs-bounces at lists.openid.net [mailto:openid-specs-bounces at lists.openid.net] On Behalf Of Allen Tom
Sent: Friday, January 22, 2010 10:43 AM
To: Andrew Arnott; John Bradley; Breno de Medeiros; specs
Subject: Re: Problem with nonces and HTTP GET

The SSL security warning is a really terrible UX, and I agree that it doesn't make sense to warn on POST but not on GET.

Yahoo is running into the 2KB limit (and the associated SSL warning) with alarming frequency and it's really hurting OpenID relative to the proprietary SSO solutions.

For a real live example of how the giant AX names are hurting OpenID, see http://www.huffingtonpost.com - click on the Login link, then the "Connect with Yahoo" button. This kicks off the Hybrid OpenID+Oauth+AX flow which requires a POST response - forcing the user to click through a security warning to complete the sign in flow. The non-OpenID SSO choices (Facebook/Twitter/GFC) do not have this issue.

With regards to changing browsers to not display SSL warnings for POST, or relying on smart OpenID clients - we really need a solution right now, since the proprietary alternatives are rapidly being adopted.

WRT the nonce - I think it would make more sense for RPs to just check the timestamp, and allow replay for a "narrow" window, like 10 minutes. There are many legitimate reasons why a request could be replayed - intermediate proxy servers might do weird things, the user might hit reload/back/forward etc.

Allen


On 1/22/10 10:06 AM, "Andrew Arnott" <andrewarnott at gmail.com> wrote:
Ideally we could use POST, but avoid the browser warning that information is crossing the SSL world into the non-SSL world.  This might be arguable anyway since sending information can be done with GET or POST, so why warn for POST and not for GET?  If we can get browsers to not warn for POST we're gold.

Alternatively, and perhaps more likely, if we're moving in the direction of smart client browser (plugins), and these have been shown to benefit from extensions to the OpenID spec, perhaps we can leverage these to always use POST without displaying the warning to the user somehow.
--
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 Fri, Jan 22, 2010 at 9:14 AM, John Bradley <john.bradley at wingaa.com> wrote:
The big problem with POST is RP's that use non-ssl endpoints.

One possibility is that the IdP could look at the return_to and discover if it is safe to use POST.

In SAML SSO POST is the most common way to return the token.

The other option is artifact binding.  That way the nonce is not in the GET,  though you probably wind up with the same effect if the RP tries to resolve the artifact more than once.

John B.
On 2010-01-22, at 12:39 PM, Andrew Arnott wrote:
HTTP GET is supposed to be completely effect-free on the server.  But nonces in OpenID messages violate that aspect of the HTTP spec, since any subsequent GET with the same positive assertion will (or should) fail.  I speculate that some random login failures on StackOverflow <http://meta.stackoverflow.com/questions/32247/cant-login-to-so-with-openid-the-signature-verification-failed/36583#36583>  may be caused because a browser, an accelerator plugin, or a proxy attempted to repeat the assertion-carrying GET request (since that's supposed to be safe), and a subsequent request is the one whose response is displayed in the browser, failing user login.
 <http://meta.stackoverflow.com/questions/32247/cant-login-to-so-with-openid-the-signature-verification-failed/36583#36583>
POST is a better fit with the HTTP spec for how the message is actually processed on the server.  I know lately we've been looking for ways to cram more data into < 2KB payloads so we can get off POST and onto GET since the user experience is better.  But I wonder if we can put our heads together and figure out how to have our cake and eat it too with this nonce problem.  This error doesn't come up often, but it can come up, apparently does come up, and is a natural side-effect of the way OpenID communicates.

Any ideas?

--
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
_______________________________________________
specs mailing list
specs at lists.openid.net
http://lists.openid.net/mailman/listinfo/openid-specs


________________________________
_______________________________________________
specs mailing list
specs at lists.openid.net
http://lists.openid.net/mailman/listinfo/openid-specs
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openid.net/pipermail/openid-specs/attachments/20100122/1cf9487b/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image001.png
Type: image/png
Size: 14306 bytes
Desc: image001.png
URL: <http://lists.openid.net/pipermail/openid-specs/attachments/20100122/1cf9487b/attachment.png>