Firstly - AIAs are required and necessary so we need to do those
regardless. That means the first version of the account console can and
should leverage those for all actions regardless with updating credentials.
In the future we can discuss extending on it that's fine, but it should be
out of scope for this round. Defining duplicated UIs, REST APIs, testing,
etc. will be additional work that we don't have time to do now regardless.
Just creating a flexible REST API around credential updates will be hard
enough and time consuming.
With regards to UXD impact of AIAs vs direct actions. As I said before not
all credentials can be updated directly through a REST API, so you would
end up with a mix of some credentials updated directly in account console
and some through AIAs. This is a very minor UXD issue and we are talking
about actions not done regularly by users. We should focus our energy on
improving UXD around AIAs rather than alternative approaches. So I'd rather
see discussions around improving AIA experience.
With regards to security, there's two issues. First if someone gets a hold
of a bearer token they should not be able to hijack someones account. If we
allow a access token to change credentials it is very easy to completely
hijack an account. Secondly as we're talking about an SSO solution it's
important that an app has only access to what it needs to have access to.
That means no applications should have direct access to users credentials,
which they would need to have to be able to update through a REST API.
With regards to custom authenticators. AIAs allow implementing a custom
authenticator with FreeMarker templates and Java code. Without AIAs you
also need to implement React stuff. Worse is that you need to duplicate the
effort on login and testing, which for some authenticators can be quite
complex.
With regards to flexbility AIAs allows a lot more flexiblity into flows
around updating credentials. This enables us in the future to flexibly
build flows around how credentials are updated. For instance adding a OTP
may mean proving that your are the user by authenticating. Updating
password may require a different way. We will also want to have this
flexible so it can be configured/changed on a per-case basis. For instance
some may want to allow updating password if you can enter the existing
password, some may want to send an SMS with a code, etc.. This is stuff
that the authentication flows need to support, but the account console
doesn't and it will be very time consuming (and duplicating effort) to
allow that in both places.
Finally, bear in mind that redirects to the auth server really is expected.
We have already today people that want to build their own login pages in
their apps, just because they don't get used to the new way of doing
things, or the UXD/design team has created some wireframes that does it
with an embedded login form.
On Tue, 27 Aug 2019 at 12:50, Václav Muzikář <vmuzikar(a)redhat.com> wrote:
Thanks Stian for your reply!
On Tue, Aug 27, 2019 at 11:04 AM Stian Thorgersen <sthorger(a)redhat.com>
wrote:
> We've had this discussion many times already, so not sure why this is
> coming up again.
>
I must have somehow missed all of the discussions, I'm sorry for that.
Let me add a few comments below.
>
> On Thu, 22 Aug 2019, 17:40 Václav Muzikář, <vmuzikar(a)redhat.com> wrote:
>
>> Hello,
>> I wanted to discuss the new Account Console which is in development and
>> how it handles password updates and authenticators configuration (e.g.
>> TOTP).
>>
>> Currently, when a user wants to update their password [see attached
>> A.png] or configure authenticator [B.png] they cannot do it directly from
>> the Account Console. They need to be redirected back to Keycloak and
>> perform it there in form of Required Action [C.png, D.png] – which is
>> technically the login screen (it uses the same template). This also means
>> there will be no REST endpoints for changing passwords or managing
>> authenticators.
>>
>> IMHO this is not the best approach either for users or developers
>> (who'll be using Keycloak with their app). Let me explain why.
>>
>> - It's not really user friendly. Redirecting from one application to
>> seemingly completely different application is never good and can be
>> confusing. I can imagine some less tech savvy users could even feel "in
>> danger" – one second you're securely managing your account and then
you are
>> asked to change your password in some other application (that suspiciously
>> uses the same design as login screen).
>>
>> I disagree - it depends on how it's done. With a SSO solution regardless
> you are using redirects to login so doing redirects for other account
> related activities is expected.
>
Maybe this is just my opinion – I'm no UXD expert (maybe we should ask
some for their professional opinion) – but I bet I'm not the only one who
find this a bit clumsy and confusing.
>
>
>> - This approach doesn't feel seamless. You can change e.g. your
>> name, email, etc. directly from the Account Console. But not password and
>> authenticators.
>>
>>
>> - It's not an industry standard. I haven't ever encountered anywhere
>> that I'd have to be redirected from some profile managing app to
completely
>> different app just to change password. Again – it's confusing for the
users
>> to do something differently than they're used to and what they expect.
>>
>> Google used this approach a lot. I've also seen it elsewhere.
>
I'm not sure what you mean by Google used it. Doesn't seem to me like
that. They allow to change passwords directly inside their "Account
Console" – see
https://youtu.be/H2XteMSN_kU
Yes, they redirect user to the SSO to reauthenticate them. But this is
basically what OIDC's promp=login does, it's no way near AIA. User doesn't
change their password through the login screen.
>
>
>> - Network traffic overhead. People accessing the Account Console
>> will often use some limited and slow internet connection. Needless
>> redirecting back and forth and loading the application again takes data and
>> time.
>>
>> There's very little overhead here. Images and such are already cached.
>
>>
>> - Missing REST endpoints limit developers. With a proper REST API
>> devs could e.g. fully integrate the Account Console functionality into
>> their own app (and effectively replacing the Console by their app). This
>> also means no CLI apps support. And I don't think REST API is less secure
>> than Required Actions – either way you're sending an HTTP request and how
>> it's secure depends on the same factors for both.
>>
>> Required actions is for complex actions that would be costly and time
> consuming to implement in your own apps. Some actions is not possible such
> as webauthn and some actions such as password updates is not wanted as it
> provides access to passwords in clear text to other applications. It also
> opens up for more flexible requirements for some actions such as enter
> password to update OTP. Verify email to update password. Etc.
>
I believe that could be up to the developer if they decide to fully rely
on Keycloak Account Console regarding sensitive data or if they want to
proxy it through their own account management app (and therefore fully and
seamlessly integrate it) and use REST API to update it in Keycloak.
>
>
>> - We're stripping out core Account Console functionality. It was
>> always a central place to manage your account. Now it can do what? Change
>> your email and manage apps access? (We could as well replace it with bunch
>> of links to Required Actions. :P There're are already Required Actions
for
>> changing email and name.)
>>
>> I can see however one advantage of this approach. There's only one place
>> where users can change their passwords and authenticators – no need to
>> implement it second time when it's already implemented as Required Actions.
>> But this is actually not entirely true as e.g. password changing process
>> (incl. password policies etc.) is implemented in Admin Console too so there
>> needs to be some shared logic.
>> In general this approach practically benefits "only" the
implementation
>> complexity – it's easier to do it this way and therefore less error prone.
>>
>
> It benefits implementation as well as custom authenticators. It's not
> even possible to create a rest endpoint for everything such as web authn.
> It would be very complex with custom flows and authenticators to even do a
> custom account app. It increases security. Etc. Etc.
>
Yes, I understand that e.g. configuring web authn is not possible directly
in the Account Console. But at least managing the authenticators (seeing
the list of authenticators, removing them, ...) should be possible through
Account Console (and REST API) no matter how complex is the process of
adding that authenticator.
Also, isn't the new Account Console designed to be deeply
extended/modified by developers? Complex flows shouldn't be a problem. Or
am I wrong?
Maybe a silly question – but how exactly is AIA safer than REST API? Are
we implying that a REST API secured by Keycloak is insecure?
Yes, I know that with REST API we potentially give opportunity to 3rd
party apps to use and exploit it more easier that AIA – but after all
everything can be exploited. But a REST API especially in combination with
Keycloak AuthZ services is secured quite enough, isn't it?
>
>
>> WDYT? Should we keep the current approach?
>>
>> Thank you!
>>
>> V.
>>
>> --
>> Václav Muzikář
>> Senior Quality Engineer
>> Keycloak / Red Hat Single Sign-On
>> Red Hat Czech s.r.o.
>>
>
--
Václav Muzikář
Senior Quality Engineer
Keycloak / Red Hat Single Sign-On
Red Hat Czech s.r.o.