[keycloak-dev] Improve back-button and refreshes in authenticators?

Tue Mar 21 02:28:26 EDT 2017

On 20 March 2017 at 18:04, Bill Burke <bburke at redhat.com> wrote:

>
>
> On 3/20/17 3:54 AM, Stian Thorgersen wrote:
>
> Let's go with the "page expired" page then,
>
> The current behavior is "sticky" in that any back button or page refresh
> just renders the current Authenticator in the flow.  What sucks with that
> is that you have to have a cancel button on each page in order to restart
> the flow.
>
>
> but we wouldn't need it for the refresh page would we? That can just
> re-display the same thing again.
>
>
> Depends on the implementation of flow processing.  This was what I was
> trying to explain before.
>
> Remember that refresh re-executes the request to original URL.  IF you set
> cache control headers, back button will also re-execute the request.   With
> current impl, code is in the URL, so with a refresh or back button
> re-execution the runtime knows the request is a bad one, so it just finds
> your current place in the flow and re-renders it.  Without a any code in
> the URL, the runtime would not know whether your POST was an old POST
> triggered by a browser button, or if you were posting to the correct and
> current Authenticator.
>

So we have to have something in the URL that's clear to me now. We should
just add a step number or something rather than the code we have now (the
codes are expensive to create).

>
>
>
> +1 I kinda thought about writing authenticators, but you're right that
> should be made as simple as possible and we should ideally handle all of
> this outside the authenticators themselves. Something that would be
> impossible if users could step backwards/forwards in the flow as you'd have
> to have "roll back" events or something so authenticators could undo stuff
> if needed.
>
> You could have a stack and push and pop on it as you went forwards and
> backwards.
>

That would only work if there are no changes outside authentication
session, but we already have things like required actions that change the
user directly. We would also need some option on authenticators and
required actions whether or not they can be replayed or something like
that. Sounds complicated...

>
>
>
>
> On 17 March 2017 at 16:21, Bill Burke <bburke at redhat.com> wrote:
>
>> The best user experience would be that the user can click the
>> back/forward/refresh buttons as they wanted and things would work as they
>> would expect, i.e. if you're on the OTP page, you could click the back
>> button and go to the username/password page and re-enter username
>> password.  I didn't implement support for this approach as its really
>> freakin hard to get right.
>> The real question is, do we want to support going backwards or forwards
>> in a flow?  If we don't, then there are considerations and limitations on
>> what we can do for the user experience which I was trying to get at
>> before.  Specifically:
>>
>> * Is it possible to determine the difference between the back, forward,
>> or refresh button event?
>>
>> That is the question I struggled the most with when implementing auth
>> flow processing.
>>
>> I'd say we set the bar low and minimally try to provide this experience:
>>
>> * If back, forward, or refresh button is pushed, show the "Page expired"
>> page you suggested before Stian.
>>
>> Anything different than that will be dependent on the limitations of the
>> browser and our auth flow implementation.
>>
>> There's also another type of user experience you aren't considering that
>> should be involved with the discussion.  Specifically the user experience
>> of writing an Authenticator.  You want the Authenticator development to be
>> simple...you also want to make sure that its implemented in a way that
>> developers can't make mistakes and introduce security holes by accident.
>> All this stuff is tied together which makes the problem really complex.
>>
>>
>> On 3/17/17 3:45 AM, Stian Thorgersen wrote:
>>
>> I repeat:
>>
>> Before we discuss implementation though, let's figure out what the ideal
>> user experience would be then figure out how to implement it. What about:
>>
>> * Refresh just works
>> * Back button will display a nicer page, something like "Page has
>> expired. To restart the login process click here. To continue the login
>> process click here.". Or Back button could just go to the start of the flow
>> always.
>> * Resubmitting forms will just display the page above
>> * No need to do redirects. Redirects is bad for performance, but also has
>> twice the response time which is not good from a usability perspective
>>
>> Is this the optimal user experience? Or should we do something else?
>>
>> On 17 March 2017 at 08:44, Stian Thorgersen <sthorger at redhat.com> wrote:
>>
>>> Can we please get back to discussing what the best user experience is
>>> first. Then we can discuss implementations?
>>>
>>> On 16 March 2017 at 18:37, Bill Burke <bburke at redhat.com> wrote:
>>>
>>>>
>>>>
>>>> On 3/16/17 10:50 AM, Marek Posolda wrote:
>>>>
>>>> On 16/03/17 15:27, Bill Burke wrote:
>>>>
>>>> * Hidden field in a form is not a good approach.  Its very brittle and
>>>> will not work in every situation.  So huge -1 there.
>>>>
>>>> * browser back button is not required to resubmit the HTTP request as
>>>> the page can be rendered from cache.  Therefore you couldn't have a "Page
>>>> Expired" page displayed when the back button is pressed without setting the
>>>> header "Cache-Control: no-store, must-revalidate, max-age=0"
>>>>
>>>> Maybe we can do some javascript stuff like this:
>>>> http://stackoverflow.com/questions/9046184/reload-the-site-w
>>>> hen-reached-via-browsers-back-button
>>>>
>>>> But that would mean that we will need to inject some common javascript
>>>> stuff into every HTML form displayed by authentication SPI. Could we rely
>>>> on that?
>>>>
>>>> I don't think this is a good approach as Authenticator develoeprs would
>>>> have to do the same thing.
>>>>
>>>>
>>>>
>>>> * Furthermore, without some type of code/information within the URL,
>>>> you also wouldn't know if somebody clicked the back button or not or
>>>> whether this was a page refresh or some other GET request.
>>>>
>>>> Once we have the cookie with loginSessionID, we can lookup the
>>>> loginSession. And loginSession will contain last code (same like
>>>> clientSession now) and last authenticator. Then we just need to compare the
>>>> code from the loginSession with the code from request. If it matches, we
>>>> are good. If it doesn't match, it's likely the refresh of some previous
>>>> page and in that case, we can just redirect to last authenticator.
>>>>
>>>> This is the current behavior, but instead of using a cookie, the "code"
>>>> is stored in the URL.
>>>>
>>>> With only a cookie though and no URL information, you won't know the
>>>> different between a Back Button and a Page Refresh for GET requests.  For
>>>> POST requests, you won't be able to tell the differencee between a Back
>>>> Button, Page Refresh, or whether the POST is targeted to an actual
>>>> Authenticator.
>>>>
>>>> The more I think about it, things should probably stay the way it
>>>> currently is, with improvements on user experience.  I think we can support
>>>> what Stian suggested with the current implementation.
>>>>
>>>>
>>>> Not sure if we also need to track all codes, so we are able to distinct
>>>> between the "expired" code, and between the "false" code, which was never
>>>> valid and was possibly used by some attacker for CSRF. Maybe we can sign
>>>> codes with HMAC, so we can verify if it is "expired" or "false" code
>>>> without need to track the list of last codes.
>>>>
>>>>
>>>> This has been done in the past.  Then it was switched to using the same
>>>> code throughout the whole flow, then Stian switched it to changing the code
>>>> throughout the flow.  I don't know if he uses a hash or not.
>>>>
>>>> Bill
>>>>
>>>
>>>
>>
>>
>
>