Re: [OAUTH-WG] First Draft of OAuth 2.1

[Torsten Lodderstedt <[e-mail]>]

> On 12. Mar 2020, at 23:14, Vittorio Bertocci <[e-mail]> wrote:
> 
> Rotation can be used to detect leakage, right? Client credentials offer more guarantees, but unless you are using private JWTs with a non exportable certificate as client cred, a classic client secret _could_ technically leak. Having rotation would alert you if someone got a hold on those. Admittedly it’s a stretch, but not entirely inconceivable.

I would rather recommend use of public key based client authentication methods instead of RT rotation. 

> 
> On Thu, Mar 12, 2020 at 13:57 Torsten Lodderstedt <torsten=[e-mail]> wrote:
> Then why are you rotating refresh tokens? 
> 
>> Am 12.03.2020 um 20:48 schrieb Pedro Igor Silva <[e-mail]>:
>> 
>> 
>> A confidential client, as per the `web application` definition in Section `2.1.  Client Types`.
>> 
>> On Thu, Mar 12, 2020 at 4:39 PM Torsten Lodderstedt <[e-mail]> wrote:
>> Is that a public client?
>> 
>>> Am 12.03.2020 um 20:32 schrieb Pedro Igor Silva <[e-mail]>:
>>> 
>>> 
>>> I agree with you and recently, we had to deal with an issue where a `web application` using rotation (as defined by the draft) was having issues to refresh tokens due to multiple concurrent requests at the moment a token is about to expire or already expired.. We had to add some controls to deal with concurrency and additional complexity + performance penalties. And for such clients, I was not sure whether or not rotation makes sense.
>>> 
>>> On Thu, Mar 12, 2020 at 4:05 PM Vittorio Bertocci <Vittorio=[e-mail]> wrote:
>>> Thanks for the clarification, Torsten.
>>> I believe it's the first time I see use of client credentials positioned as sender constraint; if the intent is saying that confidential clients should use their credentials when redeeming refresh tokens, I am of course in agreement but I think the language should be clearer and state the above explicitly.
>>> 
>>> Re: failure frequency, I know of scenarios were the designers added rotation by default, and after a while it was turned to opt in because of the frequency of errors and impact on user experience/call center.
>>> I really believe that putting this as a MUST is justified only for exceedingly vulnerable situations, like SPAs.
>>> Native/desktop clients should be free to decide whether they want to opt in without loosing compliance. Just my 2 C
>>> 
>>> On Thu, Mar 12, 2020 at 11:58 AM Torsten Lodderstedt <torsten=[e-mail]> wrote:
>>> Hi,
>>> 
>>> sender constraining refresh tokens for confidential client means client authentication + check the binding of the refresh token with the respective client id. I don’t think this is new as RFC6759 already required ASs to check this binding. Assuming backends are generally confidential clients also means no rotation and no cache synchronization needed.
>>> 
>>> Rotation should be used for frontends, e.g. native apps and only if there is there no other option. If a refresh fails, the app must go through the authorization process again. That’s inconvenient so the question is how often this happens. What I can say, I have never seen customer complaining in several years of operation of ASs with refresh token rotation (including replay detection) for native apps with millions of users.
>>> 
>>> best regards,
>>> Torsten.
>>> 
>>>> Am 12.03..2020 um 19:24 schrieb Vittorio Bertocci <Vittorio=[e-mail]..org>:
>>>> 
>>>> 
>>>> Hey guys,
>>>> thanks for putting this together.
>>>> I am concerned with the real world impact of imposing sender constraint | rotation as a MUST on refresh tokens in every scenario.
>>>> Sender constraint isn't immediately actionable - we just had the discussion for dPOP, hence I won't go in the details here.
>>>> Rotation isn't something that can be added without significant impact on development and runtime experiences:
>>>> 	• on distributed scenarios, it introduces the need to serialize access to shared caches
>>>> 	• network failures can lead to impact on experience- stranding clients which fail to receive RTn+1 during RTn redemption in a limbo where user interaction might become necessary, disrupting experience or functionality for scenarios where the user isn't available to respond.
>>>> 
>>>> 
>>>> 
>>>> On Wed, Mar 11, 2020 at 5:28 PM Aaron Parecki <aaron@parecki..com> wrote:
>>>> I'm happy to share that Dick and Torsten and I have published a first
>>>> draft of OAuth 2.1. We've taken the feedback from the discussions on
>>>> the list and incorporated that into the draft.
>>>> 
>>>> https://tools.ietf.org/html/draft-parecki-oauth-v2-1-01
>>>> 
>>>> A summary of the differences between this draft and OAuth 2.0 can be
>>>> found in section 12, and I've copied them here below.
>>>> 
>>>> > This draft consolidates the functionality in OAuth 2.0 (RFC6749),
>>>> > OAuth 2.0 for Native Apps (RFC8252), Proof Key for Code Exchange
>>>> > (RFC7636), OAuth 2.0 for Browser-Based Apps
>>>> > (I-D.ietf-oauth-browser-based-apps), OAuth Security Best Current
>>>> > Practice (I-D.ietf-oauth-security-topics), and Bearer Token Usage
>>>> > (RFC6750).
>>>> >
>>>> >   Where a later draft updates or obsoletes functionality found in the
>>>> >   original [RFC6749], that functionality in this draft is updated with
>>>> >   the normative changes described in a later draft, or removed
>>>> >   entirely.
>>>> >
>>>> >   A non-normative list of changes from OAuth 2.0 is listed below:
>>>> >
>>>> >   *  The authorization code grant is extended with the functionality
>>>> >      from PKCE ([RFC7636]) such that the only method of using the
>>>> >      authorization code grant according to this specification requires
>>>> >      the addition of the PKCE mechanism
>>>> >
>>>> >   *  Redirect URIs must be compared using exact string matching as per
>>>> >      Section 4.1.3 of [I-D.ietf-oauth-security-topics]
>>>> >
>>>> >   *  The Implicit grant ("response_type=token") is omitted from this
>>>> >      specification as per Section 2.1.2 of
>>>> >      [I-D.ietf-oauth-security-topics]
>>>> >
>>>> >   *  The Resource Owner Password Credentials grant is omitted from this
>>>> >      specification as per Section 2.4 of
>>>> >      [I-D.ietf-oauth-security-topics]
>>>> >
>>>> >   *  Bearer token usage omits the use of bearer tokens in the query
>>>> >      string of URIs as per Section 4.3.2 of
>>>> >      [I-D.ietf-oauth-security-topics]
>>>> >
>>>> >   *  Refresh tokens must either be sender-constrained or one-time use
>>>> >      as per Section 4.12.2 of [I-D.ietf-oauth-security-topics]
>>>> 
>>>> https://tools.ietf.org/html/draft-parecki-oauth-v2-1-01#section-12
>>>> 
>>>> I'm excited for the direction this is taking, and it has been a
>>>> pleasure working with Dick and Torsten on this so far. My hope is that
>>>> this first draft can serve as a good starting point for our future
>>>> discussions!
>>>> 
>>>> ----
>>>> Aaron Parecki
>>>> aaronparecki.com
>>>> @aaronpk
>>>> 
>>>> P.S. This notice was also posted at
>>>> https://aaronparecki.com/2020/03/11/14/oauth-2-1
>>>> 
>>>> _______________________________________________
>>>> OAuth mailing list
>>>> [e-mail]
>>>> https://www.ietf.org/mailman/listinfo/oauth
>>>> _______________________________________________
>>>> OAuth mailing list
>>>> [e-mail]
>>>> https://www.ietf.org/mailman/listinfo/oauth
>>> _______________________________________________
>>> OAuth mailing list
>>> [e-mail]
>>> https://www.ietf.org/mailman/listinfo/oauth
>>> _______________________________________________
>>> OAuth mailing list
>>> [e-mail]
>>> https://www.ietf.org/mailman/listinfo/oauth