Re: [keycloak-dev] A newly added Hardcoded Role mapper ignores users that have already logged in before

Hi everybody, Sorry for taking so long to respond. What would be the semantics of “owner”? From a mapper perspective, they just offer to act on first logins (“importNewUser()”) or subsequent logins (“updateBrokeredUser()”). Every mapper would act on first login, so this logically means there are only two options left: to update on subsequent logins or not. But maybe I am missing something here? Best regards, Sebastian Mit freundlichen Grüßen / Best regards Dr.-Ing. Sebastian Schuster Open Source Services (INST-CSS/BSV-OS2) Bosch Software Innovations GmbH | Ullsteinstr. 128 | 12109 Berlin | GERMANY | www.bosch-si.com Tel. +49 30 726112-485 | Mobil +49 152 02177668 | Telefax +49 30 726112-100 | Sebastian.Schuster@bosch-si.com<mailto:Sebastian.Schuster@bosch-si.com> Sitz: Berlin, Registergericht: Amtsgericht Charlottenburg; HRB 148411 B Aufsichtsratsvorsitzender: Dr.-Ing. Thorsten Lücke; Geschäftsführung: Dr. Stefan Ferber, Michael Hahn, Dr. Aleksandar Mitrovic ​ Von: Stian Thorgersen <sthorger(a)redhat.com&gt; Gesendet: Freitag, 27. September 2019 12:57 An: Marek Posolda <mposolda(a)redhat.com&gt; Cc: EXTERNAL Thiele Frank (TNG, INST-CSS/BSV-OS2) <external.Frank.Thiele(a)bosch-si.com>; Schuster Sebastian (INST-CSS/BSV-OS2) <Sebastian.Schuster(a)bosch-si.com>; keycloak-dev(a)lists.jboss.org Betreff: Re: [keycloak-dev] A newly added Hardcoded Role mapper ignores users that have already logged in before On Thu, 26 Sep 2019 at 11:52, Marek Posolda <mposolda@redhat.com<mailto:mposolda@redhat.com>> wrote: On 25. 09. 19 14:23, Stian Thorgersen wrote: Thinking about this some more. I think all mappers for identity brokering should have an option to select sync-mode: * import - only update on first import * owner - if the idp is the owner of the user (the user has not been modified within KC, and only has a single IdP link) * force - always update The mappers that today has a behaviour that doesn't match one of the above could have an option "legacy". On the IdP config itself there should be a default sync-mode user for mappers that haven't explicitly set the sync-mode. The default value should be import. +1 for this part. With the small note that not all mapper implementations may need this option. For example UsernameTemplateMapper. What does UsernameTemplateMapper actually do? Perhaps for those where it doesn't make sense to have an option it should display the sync-mode, but in non-editable mode? The next piece of the puzzle would be to prevent editing of values that shouldn't be possible to edit locally. For user attributes that should be driven by User Profile SPI, where it would be somehow possible to for example say don't allow editing this attribute if one of the following IdPs. I'm working on a design proposal for the User Profile SPI currently, so we can add that as a requirement there. The same feature could be used for User Federation providers. For roles it is a bit harder, but would be nice to somehow be able to flag what roles are managed by IdP/user-storage and which are not. Perhaps we could add some metadata to the role mapping for that. Yes, will be great to have this capability on User Profile SPI. I am not sure about metadata for role/group mappings. There will be some added complexity in that and not sure it deserves the benefits - also considering that role/group memberships is something, which is editable just by administrator and not by user himself in account console. Remember that very often there will be one person configuring Keycloak and setting up the IdPs and a completely different person managing users and roles/groups. I think it would be nice to be able to show that a role/group mapping is managed externally, but I agree it's not the top priority. Regarding User Profile SPI, I can imagine that it will be nice to have the ability to automatically update the metadata/requirements based on some events? For example when we add support for "temporary" users registered through IDP, those users will be typically read-only. So I can imagine that when administrator creates IDP with temporary users, there will be an event, which will automatically add rules to User Profile SPI that all attributes of such user should be read-only. Similarly when you have read-only LDAP provider or SSSD provider, you may also want auto-created User Profile metadata specifying that users should be non-editable. I am not sure whether to count with this in the User Profile SPI design or not? Not sure. Right now I've not really covered this part in the user profile design as I haven't really thought it through. I'm kinda thinking about 2 options. First 1 where attributes can have different config based on the source, and another where you define different user profiles. The latter is probably simpler and cleaner. Let's move further discussion on this part elsewhere though. Marek Would be great to start on a design proposal around this, so we can have it documented the way it should work. Once we have that and have agreed on the approach I don't mind having PRs for individual mappers merged as I appreciate the fact that for this case you want a solution for hardcoded role mapper quickly without having to do all of the work for all the other mappers. On Wed, 25 Sep 2019 at 12:36, Stian Thorgersen <sthorger@redhat.com<mailto:sthorger@redhat.com>> wrote: https://issues.jboss.org/browse/KEYCLOAK-8690 Adding to my point that we need a consistent solution/strategy for all mappers. On Wed, 25 Sep 2019 at 12:32, Stian Thorgersen <sthorger@redhat.com<mailto:sthorger@redhat.com>> wrote: On Wed, 25 Sep 2019 at 09:42, EXTERNAL Thiele Frank (TNG, INST-CSS/BSV-OS2) <external.Frank.Thiele@bosch-si.com<mailto:external.Frank.Thiele@bosch-si.com>> wrote: Hi, That is an interesting point. I checked some mappers: - AttributeToRoleMapper handles the update like the import with the exception that in case of an update, the role is deleted if the attribute is no longer present (I call this for now inverted logic). - ClaimToRoleMapper and ExternalKeycloakRoleToRoleMapper handle an update with the inverted logic only – so they don’t set but only delete the role. - HardcodeRoleMapper fully ignores updates whereas it could at least do it the same way as AttributeToRoleMapper. - UserAttributeMapper is even more complex… So the currently used IdentityProviderMapper implementations are very inconsistent and hopefully documented and understood well for and by the end users. It is not documented and I doubt anyone can understand how it will function. This is my concern when we have "random" things happening in each mapper without an overall consistent plan. All I am saying is that it will become a breaking change to globally define this behavior as there are nowadays several, conflicting modes implemented. Due to that I would like to emphasize that the flag introduction (“handleUpdateToo”) still seems as the solution with the lowest friction. Adding a flag to an individual mapper is just a solution to your problem and it doesn't address the wider issue. It is basically a work-around for your use-case, and is introducing yet another behaviour on top of the already inconsistent behaviour we have today. Of course we need anything new that is added to be able to match the current behaviour, which will be more and more difficult the more random switches and config options we have in mappers. So we really do need to have a proper solution in thought out, then figure out how to address it. Mit freundlichen Grüßen / Best regards Frank Thiele Open Source Services 2 - Product Group Customer Success Services (INST-CSS/BSV-OS2) Bosch Software Innovations GmbH | Ullsteinstr. 128 | 12109 Berlin | GERMANY | www.bosch-si.com<http://www.bosch-si.com> Tel. +49 30 726112-0 | Fax +49 30 726112-100 | external.Frank.Thiele@bosch-si.com<mailto:external.Frank.Thiele@bosch-si.com> Sitz: Berlin, Registergericht: Amtsgericht Charlottenburg; HRB 148411 B Aufsichtsratsvorsitzender: Dr.-Ing. Thorsten Lücke; Geschäftsführung: Dr. Stefan Ferber, Michael Hahn, Dr. Aleksandar Mitrovic Von: Stian Thorgersen <sthorger@redhat.com<mailto:sthorger@redhat.com>> Gesendet: Mittwoch, 25. September 2019 08:38 An: Marek Posolda <mposolda@redhat.com<mailto:mposolda@redhat.com>> Cc: Schuster Sebastian (INST-CSS/BSV-OS2) <Sebastian.Schuster@bosch-si.com<mailto:Sebastian.Schuster@bosch-si.com>>; EXTERNAL Thiele Frank (TNG, INST-CSS/BSV-OS2) <external.Frank.Thiele@bosch-si.com<mailto:external.Frank.Thiele@bosch-si.com>>; keycloak-dev@lists.jboss.org<mailto:keycloak-dev@lists.jboss.org> Betreff: Re: [keycloak-dev] A newly added Hardcoded Role mapper ignores users that have already logged in before Adding config options on a single mappers is not really a great solution. We need to make sure there is a consistent approach throughout. We probably don't have consistent and predictable behaviour today, but I would rather not make it worse by introducing random config options on mappers. Main question is if this should be controlled on individual mappers or if it should mainly be a config on the identity provider. Having the config on the identity provider would make more sense as it would be simpler to configure and it would avoid corner cases. There's probably at least 3 different modes for identity brokering that should be supported: 1) Import only - User is only imported if it doesn't exist. If user already exists nothing is updated. 2) Sync - Allow changes to the user within Keycloak, but also sync changes from external IdP 3) External - Do not allow any changes to the user within Keycloak as the user should be fully managed from the external IdP Option 1 is trivial. Option 2 can be very complicated. Take the example of the hardcoded role for instance. User first logs in, the role is added. An admin then removes the role from the user. User logs in again and the role is re-added. Same example can be applied to last name for instance. User logs in. 6 months later the user changes the last name in Keycloak account console, but then next day when they re-login the last name is changed back. Option 3 is relatively trivial, but would need some tweaks within Keycloak. A user that is fully externally managed should not be able to use Keycloak account console and should be view-only in the admin console. On Mon, 23 Sep 2019 at 22:21, Marek Posolda <mposolda@redhat.com<mailto:mposolda@redhat.com>> wrote: Makes sense to me. From me +1 for this. Marek On 20. 09. 19 15:57, Schuster Sebastian (INST-CSS/BSV-OS2) wrote: