[Openid-specs-ab] Defining a Hardened (Mix-up and Cut-and-Paste Proof) OpenID Connect Profile

William Denniss wdenniss at google.com
Thu Apr 14 16:22:36 UTC 2016


On Thu, Apr 14, 2016 at 3:45 AM, Torsten Lodderstedt <
torsten at lodderstedt.net> wrote:

> Hi all,
>
> I would like to take the perspective of an OP (with a reasonable number of
> active RPs), which currently only supports code. The same endpoint serves
> OpenId Connect, plain OAUth and combined requests.
>
> Let me first summarize what I understand about the proposed profile, which
> is supposed to stop mix up and code injection/CnP:
> - response type: "code id_token"
> - response_mode: "form_post"
> - Clients must use nonces
>
> extensions to OP:
> - issue ID token in the front channel
>
- add c_hash to id token
>

I would rephrase these two extensions as "supporting the hybrid flow",
c_hash is a REQUIRED claim for the hybrid flow, the conformance test ID for
this is: OP-IDToken-c_hash

- add new response mode form post
>
> extensions every RP:
> - support receiving form post
> - add several checks in the front channel (before exchanging the code):
> -- verify id token (iss, aud, signature)
> -- verify c_hash
> -- verify nonce is the same as send in request
>
> Am I missing anything?
>

While every RP should probably verify the nonce to avoid cut-and-paste, my
understand is that only a subset need to protect against mix-up, as only a
certain type of RP is actually vulnerable to that attack. Please somebody
correct me if I'm wrong, I'm keen to know what the profile of a vulnerable
RP is.


> Apperently, this solution would not work for plain OAuth. Moreover, I
> personally think this significantly adds complexity (and cost) to both
> ends, OP as well as RPs, in comparison to other potential mitigations.
> Experience shows that complexity is the enemy of correct/reliable
> implementations, which is esp. dangerous when it comes to security for mass
> market/internet services.
>

We have a solution in Connect. We should document that solution.

The Connect solution requires no changes to Connect. The proposal here is
simply documenting what people should do if they want to mitigate these
potential attacks.

I don't believe the Connect WG should be constrained to solve problems
without using the features of Connect.

So from my perspective, there are two questions:
> 1) Can we find a solution for OIDC and OAuth (for code)?
>

That solution may well be: use Connect.  At least, if you wanted to solve
it today without any spec changes or extensions, that would be my advice.


> 2) Can we make it a simpler one?
>

I'm listening.


>
> best regards,
> Torsten.
>
>
> Am 13.04.2016 um 06:22 schrieb Mike Jones:
>
> It is always mandatory for the server to include the nonce in the ID Token
> if provided in the request and it is likewise always mandatory for the RP
> to validate the nonce, if present in the ID Token.  For all response_type
> values other than “code”, including a nonce in the request is also
> mandatory.
>
>
>
>                                                           -- Mike
>
>
>
> *From:* Openid-specs-ab [mailto:openid-specs-ab-bounces at lists.openid.net
> <openid-specs-ab-bounces at lists.openid.net>] *On Behalf Of *John Bradley
> *Sent:* Tuesday, April 12, 2016 5:31 PM
> *To:* William Denniss <wdenniss at google.com> <wdenniss at google.com>
> *Cc:* openid-specs-ab at lists.openid.net
> *Subject:* Re: [Openid-specs-ab] Defining a Hardened (Mix-up and
> Cut-and-Paste Proof) OpenID Connect Profile
>
>
>
> If they send it and don’t detect if it has changed from the request then
> that is a definite error.
>
>
>
> If they don’t send it, it should be a warning as there are some cases with
> a simple preconfigured client that not checking may be OK.
>
>
>
> I personally argued at the time to always check it, so I am not going to
> push back very hard on making it mandatory to check other than it is a
> change to the spec.
>
>
>
> John B.
>
>
>
> On Apr 12, 2016, at 5:34 PM, William Denniss <wdenniss at google.com> wrote:
>
>
>
>
>
> On Tue, Apr 12, 2016 at 2:28 PM, John Bradley < <ve7jtb at ve7jtb.com>
> ve7jtb at ve7jtb.com> wrote:
>
> The AS is all-ready required to include nonce in the id_token if it is
> present in the request.  That is a MUST in the spec.
>
> That should be a test all-ready.
>
>
>
> There is: *ID Token has nonce when requested for code flow [Basic]
> (OP-nonce-code)*
>
>
>
> So for this, it's more that we need to add an RP test to ensure that RPs
> are actually sending and verifying nonce.
>
>
>
> Basically fragment encoding is not a good idea any more other than for JS
> in the browser or for native apps using view controllers or system browsers.
>
>
>
> Servers really should support the form post response mode.
>
>
>
> A client that wants to be secure can send nonce and use “code id_token” if
> they want offline or “token id_token” if they don’t want offline.
>
>
>
> That is secure.
>
>
>
> There are other mitigations we talked about in OAuth with returning the
> client_id and iss as parameters rather than returning a id_token.
>
>
>
> The client really only needs to check those two values in the front
> channel, but I would not want to change the spec to say that.
>
>
>
> +1 to all the above
>
>
>
>
>
>
>
>
>
> John B.
>
>
>
> On Apr 12, 2016, at 4:58 PM, William Denniss < <wdenniss at google.com>
> wdenniss at google.com> wrote:
>
>
>
>
> On Tue, Apr 12, 2016 at 12:50 PM, Hans Zandbelt <
> <hzandbelt at pingidentity.com>hzandbelt at pingidentity.com> wrote:
>
> from an implementers/deployment standpoint I would argue that if the
> id_token is delivered in the front channel, I may just as well get the
> access token from there too rather than having to deal with the overhead of
> calling the token endpoint, esp. since at_hash is there to protect integrity
>
> so I'd favor response_mode=form_post, response_type="id_token token" over
> hybrid
>
> I realize that this doesn't leverage the possibility of using a client
> secret to get the access token but it doesn't really seem to matter
> security-wise at this point as the id_token was delivered without
> leveraging it as well
>
> anything wrong with that line of reasoning?
>
>
>
> 1) Server won't have offline access with "id_token token". 2) The primary
> case we are dealing with today is the code flow, and we are really just
> suggesting clients add security checks onto their existing flows by
> leveraging id_token.
>
>
>
> It's true that you could do many of the same checks with "id_token token"
> to verify that the access token was issued by the correct party and avoid
> cut-and-paste and mix-up, if you only needed a once-off access token.
>
>
>
> You raise an interesting point regarding client confidentiality.
>
>
>
> On 4/12/16 9:23 PM, William Denniss wrote:
>
> Good point.
>
> Regarding the OP tests, the following are relevant to mitigate the
> cut-and-paste and mix-up attacks:
>
>  1. ID Token has nonce when requested for code flow [Basic] (OP-nonce-code)
>  2. Request with response_mode=form_post [Extra] (OP-Response-form_post)
>
> 1) is important for preventing cut-and-paste (the id token needs to
> contain the 'nonce')
> 2) is important for preventing mix-up as it means the redirect endpoint
> gets the id_token on the response at the server, as opposed to in the
> URI fragment.
>
> Unfortunately, form_post is optional for OPs, and sending the nonce on
> the code flow is optional for RPs (though fortunately to it is
> compulsory for OPs to support thanks to OP-nonce-code).
>
> We could add an hardened OP test for:
> – Forcing nonce to be present on the code flow
>
> We should definitely have RP tests for:
> – sending and validating nonce on the code flow
> – validating the c_hash, iss, aud on the hybrid flow
>
> How we would profile these tests I'm not sure; would they go in the
> Basic testing profile, or in a new Hardened one?  We could move
> OP-Response-form_post to the Basic profile if we wanted to be
> opinionated, or define a new profile.
>
> The good news is that supporting the features required to mitigate
> mix-up & cut-and-paste is not all that hard to do in Connect.
>
>
> On Tue, Apr 12, 2016 at 10:46 AM, Nick Roy < <nroy at internet2.edu>
> nroy at internet2.edu
> <mailto: <nroy at internet2.edu>nroy at internet2.edu>> wrote:
>
>     Would it be possible to check for the secure behavior in Roland's
>     test suite and either not certify non-mitigating implementations, or
>     offer a risk mitigation add-on cert for those that do the right thing?
>
>     Nick
>
>     From: Openid-specs-ab < <openid-specs-ab-bounces at lists.openid.net>
> openid-specs-ab-bounces at lists.openid.net
>     <mailto: <openid-specs-ab-bounces at lists.openid.net>
> openid-specs-ab-bounces at lists.openid.net>> on behalf of
>     William Denniss < <wdenniss at google.com>wdenniss at google.com <mailto:
> <wdenniss at google.com>wdenniss at google.com>>
>     Date: Tuesday, April 12, 2016 at 11:01 AM
>     To: " <openid-specs-ab at lists.openid.net>
> openid-specs-ab at lists.openid.net
>     <mailto: <openid-specs-ab at lists.openid.net>
> openid-specs-ab at lists.openid.net>"
>     < <openid-specs-ab at lists.openid.net>openid-specs-ab at lists.openid.net
>     <mailto: <openid-specs-ab at lists.openid.net>
> openid-specs-ab at lists.openid.net>>
>     Subject: [Openid-specs-ab] Defining a Hardened (Mix-up and
>     Cut-and-Paste Proof) OpenID Connect Profile
>
>     One item that came out of the discussions on the sidelines of IETF95
>     with folk from this WG (specifically Nat, Mike, John, Brian and
>     myself) was the need for the Connect community to respond to the
>     recently < <http://arxiv.org/abs/1508.04324v2/>
> http://arxiv.org/abs/1508.04324v2/> documented
>     < <http://arxiv.org/abs/1601.01229v2/>
> http://arxiv.org/abs/1601.01229v2/> security threats.
>
>     Connect is actually in a much stronger place for mitigating these
>     attacks (as noted in the papers themselves) than pure OAuth, with
>     the id_token offering a cryptographic binding of the code to the
>     issuer, audience and session identifier (nonce).
>
>     However, certain steps need to be followed, for example using
>     'nonce' with the code flow (which is optional to implement for
>     clients) to protect against cut-and-paste, and using the form-post
>     response type with the hybrid flow to verify that the code was
>     issued by the expected IdP, to ensure the code is exchanged at the
>     correct token endpoint (mitigating mix-up).
>
>     We discussed last week creating a profile of Connect that recommends
>     those practices to mitigate these classes of attack as a response to
>     the security researchers' findings. I wanted to share that
>     suggestion with the list, and continue the conversation.
>
>
>
>
> _______________________________________________
> Openid-specs-ab mailing list
> <Openid-specs-ab at lists.openid.net>Openid-specs-ab at lists.openid.net
> <http://lists.openid.net/mailman/listinfo/openid-specs-ab>
> http://lists.openid.net/mailman/listinfo/openid-specs-ab
>
>
> --
> Hans Zandbelt              | Sr. Technical Architect
> hzandbelt at pingidentity.com | Ping Identity
>
>
>
> _______________________________________________
> Openid-specs-ab mailing list
> Openid-specs-ab at lists.openid.net
> http://lists.openid.net/mailman/listinfo/openid-specs-ab
>
>
>
>
> _______________________________________________
> Openid-specs-ab mailing listOpenid-specs-ab at lists.openid.nethttp://lists.openid.net/mailman/listinfo/openid-specs-ab
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openid.net/pipermail/openid-specs-ab/attachments/20160414/d27f841b/attachment-0001.html>


More information about the Openid-specs-ab mailing list