[security-dev] Meaningful relationships

Tue Oct 16 16:17:21 EDT 2012

On Oct 16, 2012, at 9:18 PM, Shane Bryzak <sbryzak at redhat.com> wrote:

> On 17/10/12 03:43, Boleslaw Dawidowicz wrote:
>> I like this. One of biggest mistakes I made in the original model was to define separate notion of user-group relationship.  Design where direct relationship is just another type of role (either 'member' or null) is much more elegant. 
>> 
>> The only concern is to keep Group tree and LDAP use case in mind.
>> 
>> https://github.com/picketlink/picketlink/blob/master/idm/api/src/main/java/org/picketlink/idm/model/Group.java
>> 
>> Path id is in place to ensure name uniqueness in the tree. You can have a/b/a/b/a - which result in 3 groups with name "a" and 2 groups with name "b". Then question is if this path should be resolved on the fly from IdentityMembership (huge performance cost and implementation nightmare) or stored as an id in IdentityObject table during object creation. Second option is better however name also persists relationships between groups from the start. I have chosen first approach in PLIDM 1.x and this was my second biggest regret - mostly because of performance cost.
> 
> The LDAP use case is a good point - I'll need to think about it some more, but I think we can make the path validation an implementation detail.  I agree we should go with the second option for the LDAP implementation (because of it's strict tree structure), however I'm thinking for JPA (in which we don't have the same restrictions) we should possibly allow a Group to be a member of more than one parent group.  It's conceivable that there's use cases that require this, however I'm not totally sold on the idea and would like to hear some feedback.

There are few other issues to keep in mind. 

Like I tried to explain before in separate email I suggest to not decouple LDAP and JPA use cases too much. From my experience LDAP store is too limited to be used in separation. Then identity store federation is a bad idea - it seems flexible but has significant cost. Good approach is to just use LDAP identity store implementation to pull data and sync into. Then just query RDBMS implementation - because SQL is way more flexible and efficient to do complex queries. Actually often use case is that even if you have data in LDAP you still want to store in DB some info that doesn't fit LDAP schema.

All in all don't think about separate LDAP use case - assume that you will want to sync LDAP data into JPA store and map same structure there. Safest and most typical structure is probably 2-3 levels deep tree.

Another point for LDAP to JPA synchronisation. There is no way to implement LDAP store with efficient support of some queries. Some of operations that are easy to optimise in SQL will force you to download whole directory content and filter locally in case of LDAP. 

For graph group structure (many parents) I really beg you to not do this… This is officially my third biggest regret in 1.x design ;) I'm still eager to see any good real life use case. And even if you identify one this has significant cost. Having just a tree structure makes the whole model LDAP compatible which is huge gain.

Additionally when considering group path resolution remember about complex queries. Even simple role resolutions can easily lead into significant number of subqueries. If you add store federation or any additional constraints it gets worse. 

> 
>> 
>> Bolek. 
>> 
>> On Oct 15, 2012, at 10:27 AM, Shane Bryzak <sbryzak at redhat.com> wrote:
>> 
>>> I should clarify something.  It is entirely possible for a user to have a role in a group without being a member of that group.  One of the good use cases that someone from the team informed me about previously is an administrator for a group of doctors.  The membership scenario would look like this:
>>> 
>>> IdentityMembership
>>> -------------------------
>>> MEMBER = Bill (User)
>>> GROUP = Doctors
>>> ROLE = Admin
>>> 
>>> In this case, Bill (the user) would not be a member of the Doctors group himself, he would simply be an administrator for the group.  If he were to be a member of the group (as well as an Administrator) then that would require the following additional record:
>>> 
>>> IdentityMembership
>>> -------------------------
>>> MEMBER = Bill (User)
>>> GROUP = Doctors
>>> ROLE = null
>>> 
>>> So, in a nutshell - if a Role is specified, it means the member has that role for the specified group, however the member is not an actual member of the group themselves.  Hope that makes sense!
>>> 
>>> 
>>> On 15/10/12 18:19, Shane Bryzak wrote:
>>>> No, not that kind.  I'm currently reviewing the database schema for the identity management module - in the previous version of PicketLink we had quite a good design [1] that was a little abstract, but met all the requirements well.  Here's a summary of the key tables:
>>>> 
>>>> IdentityObject - this table would contain both User and Group records
>>>> IdentityObjectRelationship - models the relationship between User and Group, i.e. Group memberships
>>>> IdentityObjectRelationshipName - this table is a special one that contained the names for "named relationships".  A named relationship can effectively be thought of as a Role, (and was also modelled in the IdentityObjectRelationship table) for example "John" (User) is a "Manager" (Role, the "named" bit of the relationship) in "Head Office" (Group) - see [2] for more details.
>>>> 
>>>> With the introduction of application roles we need to jig this design a little bit.  I was thinking of keeping IdentityObject essentially the same, with the exception that it would also be used to contain Roles, as well as Users and Groups.  Instead of the IdentityObjectRelationship table though, I propose we go with the following slightly less abstract design:
>>>> 
>>>> IdentityMembership
>>>> -------------------------
>>>> MEMBER
>>>> GROUP
>>>> ROLE
>>>> 
>>>> This basically allows us to make any IdentityType (User, Group or Role) a member of a Group or Role, or both.  Here's a few scenarios:
>>>> 
>>>> 1. John is a part of the accounting group.
>>>> 
>>>> IdentityMembership
>>>> -------------------------
>>>> MEMBER = John (User)
>>>> GROUP = accounting
>>>> ROLE = null
>>>> 
>>>> 2. The Manager group is a subgroup of the Employee group.
>>>> 
>>>> IdentityMembership
>>>> -------------------------
>>>> MEMBER = Manager (Group)
>>>> GROUP = Employee
>>>> ROLE = null
>>>> 
>>>> 3. Kevin is an administrator for the Manager group
>>>> 
>>>> IdentityMembership
>>>> -------------------------
>>>> MEMBER = Kevin (User)
>>>> GROUP = Manager
>>>> ROLE = Admin
>>>> 
>>>> 4. Kelly is a superuser (which is an application role)
>>>> 
>>>> IdentityMembership
>>>> -------------------------
>>>> MEMBER = Kelly (User)
>>>> GROUP = null
>>>> ROLE = Superuser
>>>> 
>>>> With the above examples in mind, this now leads into the "meaningful relationships" theme - can anyone think of any other meaningful security relationships that cannot be modelled with this design?  I'm not really looking to make the design "future proof" as such, but I would like to ensure we cover all currently known scenarios / use cases.  Comments and feedback welcome of course.
>>>> 
>>>> 
>>>> [1] http://anonsvn.jboss.org/repos/picketlink/idm/downloads/docs/1.0.0.GA/ReferenceGuide/en-US/html_single/index.html#spi_model
>>>> [2] http://anonsvn.jboss.org/repos/picketlink/idm/downloads/docs/1.0.0.GA/ReferenceGuide/en-US/html_single/index.html#d0e342 
>>>> 
>>>> _______________________________________________
>>>> security-dev mailing list
>>>> security-dev at lists.jboss.org
>>>> https://lists.jboss.org/mailman/listinfo/security-dev
>>> 
>>> 
>>> _______________________________________________
>>> security-dev mailing list
>>> security-dev at lists.jboss.org
>>> https://lists.jboss.org/mailman/listinfo/security-dev
>> 
> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.jboss.org/pipermail/security-dev/attachments/20121016/fb64f608/attachment-0001.html 